///////////////////////////////////////////////////////////////////////////// // Name: wx/private/wxprintf.h // Purpose: wxWidgets wxPrintf() implementation // Author: Ove Kaven // Modified by: Ron Lee, Francesco Montorsi // Created: 09/04/99 // Copyright: (c) wxWidgets copyright // Licence: wxWindows licence ///////////////////////////////////////////////////////////////////////////// #ifndef _WX_PRIVATE_WXPRINTF_H_ #define _WX_PRIVATE_WXPRINTF_H_ // --------------------------------------------------------------------------- // headers and macros // --------------------------------------------------------------------------- #include "wx/crt.h" #include "wx/log.h" #include "wx/utils.h" #include // prefer snprintf over sprintf #if defined(__VISUALC__) || \ (defined(__BORLANDC__) && __BORLANDC__ >= 0x540) #define system_sprintf(buff, max, flags, data) \ ::_snprintf(buff, max, flags, data) #elif defined(HAVE_SNPRINTF) #define system_sprintf(buff, max, flags, data) \ ::snprintf(buff, max, flags, data) #else // NB: at least sprintf() should always be available // since 'max' is not used in this case, wxVsnprintf() should always // ensure that 'buff' is big enough for all common needs // (see wxMAX_SVNPRINTF_FLAGBUFFER_LEN and wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN) #define system_sprintf(buff, max, flags, data) \ ::sprintf(buff, flags, data) #define SYSTEM_SPRINTF_IS_UNSAFE #endif // --------------------------------------------------------------------------- // printf format string parsing // --------------------------------------------------------------------------- // some limits of our implementation #define wxMAX_SVNPRINTF_ARGUMENTS 64 #define wxMAX_SVNPRINTF_FLAGBUFFER_LEN 32 #define wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN 512 // the conversion specifiers accepted by wxCRT_VsnprintfW enum wxPrintfArgType { wxPAT_INVALID = -1, wxPAT_INT, // %d, %i, %o, %u, %x, %X wxPAT_LONGINT, // %ld, etc #ifdef wxLongLong_t wxPAT_LONGLONGINT, // %Ld, etc #endif wxPAT_SIZET, // %zd, etc wxPAT_DOUBLE, // %e, %E, %f, %g, %G wxPAT_LONGDOUBLE, // %le, etc wxPAT_POINTER, // %p wxPAT_CHAR, // %hc (in ANSI mode: %c, too) wxPAT_WCHAR, // %lc (in Unicode mode: %c, too) wxPAT_PCHAR, // %s (related to a char *) wxPAT_PWCHAR, // %s (related to a wchar_t *) wxPAT_NINT, // %n wxPAT_NSHORTINT, // %hn wxPAT_NLONGINT, // %ln wxPAT_STAR // '*' used for width or precision }; // an argument passed to wxCRT_VsnprintfW union wxPrintfArg { int pad_int; // %d, %i, %o, %u, %x, %X long int pad_longint; // %ld, etc #ifdef wxLongLong_t wxLongLong_t pad_longlongint; // %Ld, etc #endif size_t pad_sizet; // %zd, etc double pad_double; // %e, %E, %f, %g, %G long double pad_longdouble; // %le, etc void *pad_pointer; // %p char pad_char; // %hc (in ANSI mode: %c, too) wchar_t pad_wchar; // %lc (in Unicode mode: %c, too) void *pad_str; // %s int *pad_nint; // %n short int *pad_nshortint; // %hn long int *pad_nlongint; // %ln }; // helper for converting string into either char* or wchar_t* depending // on the type of wxPrintfConvSpec instantiation: template struct wxPrintfStringHelper {}; template<> struct wxPrintfStringHelper { typedef const wxWX2MBbuf ConvertedType; static ConvertedType Convert(const wxString& s) { return s.mb_str(); } }; template<> struct wxPrintfStringHelper { typedef const wxWX2WCbuf ConvertedType; static ConvertedType Convert(const wxString& s) { return s.wc_str(); } }; // Contains parsed data relative to a conversion specifier given to // wxCRT_VsnprintfW and parsed from the format string // NOTE: in C++ there is almost no difference between struct & classes thus // there is no performance gain by using a struct here... template class wxPrintfConvSpec { public: // the position of the argument relative to this conversion specifier size_t m_pos; // the type of this conversion specifier wxPrintfArgType m_type; // the minimum and maximum width // when one of this var is set to -1 it means: use the following argument // in the stack as minimum/maximum width for this conversion specifier int m_nMinWidth, m_nMaxWidth; // does the argument need to the be aligned to left ? bool m_bAlignLeft; // pointer to the '%' of this conversion specifier in the format string // NOTE: this points somewhere in the string given to the Parse() function - // it's task of the caller ensure that memory is still valid ! const CharType *m_pArgPos; // pointer to the last character of this conversion specifier in the // format string // NOTE: this points somewhere in the string given to the Parse() function - // it's task of the caller ensure that memory is still valid ! const CharType *m_pArgEnd; // a little buffer where formatting flags like #+\.hlqLz are stored by Parse() // for use in Process() char m_szFlags[wxMAX_SVNPRINTF_FLAGBUFFER_LEN]; public: // we don't declare this as a constructor otherwise it would be called // automatically and we don't want this: to be optimized, wxCRT_VsnprintfW // calls this function only on really-used instances of this class. void Init(); // Parses the first conversion specifier in the given string, which must // begin with a '%'. Returns false if the first '%' does not introduce a // (valid) conversion specifier and thus should be ignored. bool Parse(const CharType *format); // Process this conversion specifier and puts the result in the given // buffer. Returns the number of characters written in 'buf' or -1 if // there's not enough space. int Process(CharType *buf, size_t lenMax, wxPrintfArg *p, size_t written); // Loads the argument of this conversion specifier from given va_list. bool LoadArg(wxPrintfArg *p, va_list &argptr); private: // An helper function of LoadArg() which is used to handle the '*' flag void ReplaceAsteriskWith(int w); }; template void wxPrintfConvSpec::Init() { m_nMinWidth = 0; m_nMaxWidth = 0xFFFF; m_pos = 0; m_bAlignLeft = false; m_pArgPos = m_pArgEnd = NULL; m_type = wxPAT_INVALID; memset(m_szFlags, 0, sizeof(m_szFlags)); // this character will never be removed from m_szFlags array and // is important when calling sprintf() in wxPrintfConvSpec::Process() ! m_szFlags[0] = '%'; } template bool wxPrintfConvSpec::Parse(const CharType *format) { bool done = false; // temporary parse data size_t flagofs = 1; bool in_prec, // true if we found the dot in some previous iteration prec_dot; // true if the dot has been already added to m_szFlags int ilen = 0; m_bAlignLeft = in_prec = prec_dot = false; m_pArgPos = m_pArgEnd = format; do { #define CHECK_PREC \ if (in_prec && !prec_dot) \ { \ m_szFlags[flagofs++] = '.'; \ prec_dot = true; \ } // what follows '%'? const CharType ch = *(++m_pArgEnd); switch ( ch ) { case wxT('\0'): return false; // not really an argument case wxT('%'): return false; // not really an argument case wxT('#'): case wxT('0'): case wxT(' '): case wxT('+'): case wxT('\''): CHECK_PREC m_szFlags[flagofs++] = char(ch); break; case wxT('-'): CHECK_PREC m_bAlignLeft = true; m_szFlags[flagofs++] = char(ch); break; case wxT('.'): // don't use CHECK_PREC here to avoid warning about the value // assigned to prec_dot inside it being never used (because // overwritten just below) from Borland in release build if (in_prec && !prec_dot) m_szFlags[flagofs++] = '.'; in_prec = true; prec_dot = false; m_nMaxWidth = 0; // dot will be auto-added to m_szFlags if non-negative // number follows break; case wxT('h'): ilen = -1; CHECK_PREC m_szFlags[flagofs++] = char(ch); break; case wxT('l'): // NB: it's safe to use flagofs-1 as flagofs always start from 1 if (m_szFlags[flagofs-1] == 'l') // 'll' modifier is the same as 'L' or 'q' ilen = 2; else ilen = 1; CHECK_PREC m_szFlags[flagofs++] = char(ch); break; case wxT('q'): case wxT('L'): ilen = 2; CHECK_PREC m_szFlags[flagofs++] = char(ch); break; #ifdef __WINDOWS__ // under Windows we support the special '%I64' notation as longlong // integer conversion specifier for MSVC compatibility // (it behaves exactly as '%lli' or '%Li' or '%qi') case wxT('I'): if (*(m_pArgEnd+1) == wxT('6') && *(m_pArgEnd+2) == wxT('4')) { m_pArgEnd++; m_pArgEnd++; ilen = 2; CHECK_PREC m_szFlags[flagofs++] = char(ch); m_szFlags[flagofs++] = '6'; m_szFlags[flagofs++] = '4'; break; } // else: fall-through, 'I' is MSVC equivalent of C99 'z' #endif // __WINDOWS__ case wxT('z'): case wxT('Z'): // 'z' is C99 standard for size_t and ptrdiff_t, 'Z' was used // for this purpose in libc5 and by wx <= 2.8 ilen = 3; CHECK_PREC m_szFlags[flagofs++] = char(ch); break; case wxT('*'): if (in_prec) { CHECK_PREC // tell Process() to use the next argument // in the stack as maxwidth... m_nMaxWidth = -1; } else { // tell Process() to use the next argument // in the stack as minwidth... m_nMinWidth = -1; } // save the * in our formatting buffer... // will be replaced later by Process() m_szFlags[flagofs++] = char(ch); break; case wxT('1'): case wxT('2'): case wxT('3'): case wxT('4'): case wxT('5'): case wxT('6'): case wxT('7'): case wxT('8'): case wxT('9'): { int len = 0; CHECK_PREC while ( (*m_pArgEnd >= CharType('0')) && (*m_pArgEnd <= CharType('9')) ) { m_szFlags[flagofs++] = char(*m_pArgEnd); len = len*10 + (*m_pArgEnd - wxT('0')); m_pArgEnd++; } if (in_prec) m_nMaxWidth = len; else m_nMinWidth = len; m_pArgEnd--; // the main loop pre-increments n again } break; case wxT('$'): // a positional parameter (e.g. %2$s) ? { if (m_nMinWidth <= 0) break; // ignore this formatting flag as no // numbers are preceding it // remove from m_szFlags all digits previously added do { flagofs--; } while (m_szFlags[flagofs] >= '1' && m_szFlags[flagofs] <= '9'); // re-adjust the offset making it point to the // next free char of m_szFlags flagofs++; m_pos = m_nMinWidth; m_nMinWidth = 0; } break; case wxT('d'): case wxT('i'): case wxT('o'): case wxT('u'): case wxT('x'): case wxT('X'): CHECK_PREC m_szFlags[flagofs++] = char(ch); if (ilen == 0) m_type = wxPAT_INT; else if (ilen == -1) // NB: 'short int' value passed through '...' // is promoted to 'int', so we have to get // an int from stack even if we need a short m_type = wxPAT_INT; else if (ilen == 1) m_type = wxPAT_LONGINT; else if (ilen == 2) #ifdef wxLongLong_t m_type = wxPAT_LONGLONGINT; #else // !wxLongLong_t m_type = wxPAT_LONGINT; #endif // wxLongLong_t/!wxLongLong_t else if (ilen == 3) m_type = wxPAT_SIZET; done = true; break; case wxT('e'): case wxT('E'): case wxT('f'): case wxT('g'): case wxT('G'): CHECK_PREC m_szFlags[flagofs++] = char(ch); if (ilen == 2) m_type = wxPAT_LONGDOUBLE; else m_type = wxPAT_DOUBLE; done = true; break; case wxT('p'): m_type = wxPAT_POINTER; m_szFlags[flagofs++] = char(ch); done = true; break; case wxT('c'): if (ilen == -1) { // in Unicode mode %hc == ANSI character // and in ANSI mode, %hc == %c == ANSI... m_type = wxPAT_CHAR; } else if (ilen == 1) { // in ANSI mode %lc == Unicode character // and in Unicode mode, %lc == %c == Unicode... m_type = wxPAT_WCHAR; } else { #if wxUSE_UNICODE // in Unicode mode, %c == Unicode character m_type = wxPAT_WCHAR; #else // in ANSI mode, %c == ANSI character m_type = wxPAT_CHAR; #endif } done = true; break; case wxT('s'): if (ilen == -1) { // Unicode mode wx extension: we'll let %hs mean non-Unicode // strings (when in ANSI mode, %s == %hs == ANSI string) m_type = wxPAT_PCHAR; } else if (ilen == 1) { // in Unicode mode, %ls == %s == Unicode string // in ANSI mode, %ls == Unicode string m_type = wxPAT_PWCHAR; } else { #if wxUSE_UNICODE m_type = wxPAT_PWCHAR; #else m_type = wxPAT_PCHAR; #endif } done = true; break; case wxT('n'): if (ilen == 0) m_type = wxPAT_NINT; else if (ilen == -1) m_type = wxPAT_NSHORTINT; else if (ilen >= 1) m_type = wxPAT_NLONGINT; done = true; break; default: // bad format, don't consider this an argument; // leave it unchanged return false; } if (flagofs == wxMAX_SVNPRINTF_FLAGBUFFER_LEN) { wxLogDebug(wxT("Too many flags specified for a single conversion specifier!")); return false; } } while (!done); return true; // parsing was successful } template void wxPrintfConvSpec::ReplaceAsteriskWith(int width) { char temp[wxMAX_SVNPRINTF_FLAGBUFFER_LEN]; // find the first * in our flag buffer char *pwidth = strchr(m_szFlags, '*'); wxCHECK_RET(pwidth, wxT("field width must be specified")); // save what follows the * (the +1 is to skip the asterisk itself!) strcpy(temp, pwidth+1); if (width < 0) { pwidth[0] = wxT('-'); pwidth++; } // replace * with the actual integer given as width #ifndef SYSTEM_SPRINTF_IS_UNSAFE int maxlen = (m_szFlags + wxMAX_SVNPRINTF_FLAGBUFFER_LEN - pwidth) / sizeof(*m_szFlags); #endif int offset = system_sprintf(pwidth, maxlen, "%d", abs(width)); // restore after the expanded * what was following it strcpy(pwidth+offset, temp); } template bool wxPrintfConvSpec::LoadArg(wxPrintfArg *p, va_list &argptr) { // did the '*' width/precision specifier was used ? if (m_nMaxWidth == -1) { // take the maxwidth specifier from the stack m_nMaxWidth = va_arg(argptr, int); if (m_nMaxWidth < 0) m_nMaxWidth = 0; else ReplaceAsteriskWith(m_nMaxWidth); } if (m_nMinWidth == -1) { // take the minwidth specifier from the stack m_nMinWidth = va_arg(argptr, int); ReplaceAsteriskWith(m_nMinWidth); if (m_nMinWidth < 0) { m_bAlignLeft = !m_bAlignLeft; m_nMinWidth = -m_nMinWidth; } } switch (m_type) { case wxPAT_INT: p->pad_int = va_arg(argptr, int); break; case wxPAT_LONGINT: p->pad_longint = va_arg(argptr, long int); break; #ifdef wxLongLong_t case wxPAT_LONGLONGINT: p->pad_longlongint = va_arg(argptr, wxLongLong_t); break; #endif // wxLongLong_t case wxPAT_SIZET: p->pad_sizet = va_arg(argptr, size_t); break; case wxPAT_DOUBLE: p->pad_double = va_arg(argptr, double); break; case wxPAT_LONGDOUBLE: p->pad_longdouble = va_arg(argptr, long double); break; case wxPAT_POINTER: p->pad_pointer = va_arg(argptr, void *); break; case wxPAT_CHAR: p->pad_char = (char)va_arg(argptr, int); // char is promoted to int when passed through '...' break; case wxPAT_WCHAR: p->pad_wchar = (wchar_t)va_arg(argptr, int); // char is promoted to int when passed through '...' break; case wxPAT_PCHAR: case wxPAT_PWCHAR: p->pad_str = va_arg(argptr, void *); break; case wxPAT_NINT: p->pad_nint = va_arg(argptr, int *); break; case wxPAT_NSHORTINT: p->pad_nshortint = va_arg(argptr, short int *); break; case wxPAT_NLONGINT: p->pad_nlongint = va_arg(argptr, long int *); break; case wxPAT_STAR: // this will be handled as part of the next argument return true; case wxPAT_INVALID: default: return false; } return true; // loading was successful } template int wxPrintfConvSpec::Process(CharType *buf, size_t lenMax, wxPrintfArg *p, size_t written) { // buffer to avoid dynamic memory allocation each time for small strings; // note that this buffer is used only to hold results of number formatting, // %s directly writes user's string in buf, without using szScratch char szScratch[wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN]; size_t lenScratch = 0, lenCur = 0; #define APPEND_CH(ch) \ { \ if ( lenCur == lenMax ) \ return -1; \ \ buf[lenCur++] = ch; \ } switch ( m_type ) { case wxPAT_INT: lenScratch = system_sprintf(szScratch, wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN, m_szFlags, p->pad_int); break; case wxPAT_LONGINT: lenScratch = system_sprintf(szScratch, wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN, m_szFlags, p->pad_longint); break; #ifdef wxLongLong_t case wxPAT_LONGLONGINT: lenScratch = system_sprintf(szScratch, wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN, m_szFlags, p->pad_longlongint); break; #endif // SIZEOF_LONG_LONG case wxPAT_SIZET: lenScratch = system_sprintf(szScratch, wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN, m_szFlags, p->pad_sizet); break; case wxPAT_LONGDOUBLE: lenScratch = system_sprintf(szScratch, wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN, m_szFlags, p->pad_longdouble); break; case wxPAT_DOUBLE: lenScratch = system_sprintf(szScratch, wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN, m_szFlags, p->pad_double); break; case wxPAT_POINTER: lenScratch = system_sprintf(szScratch, wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN, m_szFlags, p->pad_pointer); break; case wxPAT_CHAR: case wxPAT_WCHAR: { wxUniChar ch; if (m_type == wxPAT_CHAR) ch = p->pad_char; else // m_type == wxPAT_WCHAR ch = p->pad_wchar; CharType val = ch; size_t i; if (!m_bAlignLeft) for (i = 1; i < (size_t)m_nMinWidth; i++) APPEND_CH(wxT(' ')); APPEND_CH(val); if (m_bAlignLeft) for (i = 1; i < (size_t)m_nMinWidth; i++) APPEND_CH(wxT(' ')); } break; case wxPAT_PCHAR: case wxPAT_PWCHAR: { wxString s; if ( !p->pad_str ) { if ( m_nMaxWidth >= 6 ) s = wxT("(null)"); } else if (m_type == wxPAT_PCHAR) s.assign(static_cast(p->pad_str)); else // m_type == wxPAT_PWCHAR s.assign(static_cast(p->pad_str)); typename wxPrintfStringHelper::ConvertedType strbuf( wxPrintfStringHelper::Convert(s)); // at this point we are sure that m_nMaxWidth is positive or // null (see top of wxPrintfConvSpec::LoadArg) int len = wxMin((unsigned int)m_nMaxWidth, wxStrlen(strbuf)); int i; if (!m_bAlignLeft) { for (i = len; i < m_nMinWidth; i++) APPEND_CH(wxT(' ')); } len = wxMin((unsigned int)len, lenMax-lenCur); wxStrncpy(buf+lenCur, strbuf, len); lenCur += len; if (m_bAlignLeft) { for (i = len; i < m_nMinWidth; i++) APPEND_CH(wxT(' ')); } } break; case wxPAT_NINT: *p->pad_nint = written; break; case wxPAT_NSHORTINT: *p->pad_nshortint = (short int)written; break; case wxPAT_NLONGINT: *p->pad_nlongint = written; break; case wxPAT_INVALID: default: return -1; } // if we used system's sprintf() then we now need to append the s_szScratch // buffer to the given one... switch (m_type) { case wxPAT_INT: case wxPAT_LONGINT: #ifdef wxLongLong_t case wxPAT_LONGLONGINT: #endif case wxPAT_SIZET: case wxPAT_LONGDOUBLE: case wxPAT_DOUBLE: case wxPAT_POINTER: wxASSERT(lenScratch < wxMAX_SVNPRINTF_SCRATCHBUFFER_LEN); // NB: 1) we can compare lenMax (for CharType*, i.e. possibly // wchar_t*) with lenScratch (char*) because this code is // formatting integers and that will have the same length // even in UTF-8 (the only case when char* length may be // more than wchar_t* length of the same string) // 2) wxStrncpy converts the 2nd argument to 1st argument's // type transparently if their types differ, so this code // works for both instantiations if (lenMax < lenScratch) { // fill output buffer and then return -1 wxStrncpy(buf, szScratch, lenMax); return -1; } wxStrncpy(buf, szScratch, lenScratch); lenCur += lenScratch; break; default: break; // all other cases were completed previously } return lenCur; } // helper that parses format string template struct wxPrintfConvSpecParser { typedef wxPrintfConvSpec ConvSpec; wxPrintfConvSpecParser(const CharType *fmt) { nargs = 0; posarg_present = nonposarg_present = false; memset(pspec, 0, sizeof(pspec)); // parse the format string for ( const CharType *toparse = fmt; *toparse != wxT('\0'); toparse++ ) { // skip everything except format specifications if ( *toparse != '%' ) continue; // also skip escaped percent signs if ( toparse[1] == '%' ) { toparse++; continue; } ConvSpec *spec = &specs[nargs]; spec->Init(); // attempt to parse this format specification if ( !spec->Parse(toparse) ) continue; // advance to the end of this specifier toparse = spec->m_pArgEnd; // special handling for specifications including asterisks: we need // to reserve an extra slot (or two if asterisks were used for both // width and precision) in specs array in this case if ( const char *f = strchr(spec->m_szFlags, '*') ) { unsigned numAsterisks = 1; if ( strchr(++f, '*') ) numAsterisks++; for ( unsigned n = 0; n < numAsterisks; n++ ) { if ( ++nargs == wxMAX_SVNPRINTF_ARGUMENTS ) break; // TODO: we need to support specifiers of the form "%2$*1$s" // (this is the same as "%*s") as if any positional arguments // are used all asterisks must be positional as well but this // requires a lot of changes in this code (basically we'd need // to rewrite Parse() to return "*" and conversion itself as // separate entries) if ( posarg_present ) { wxFAIL_MSG ( wxString::Format ( "Format string \"%s\" uses both positional " "parameters and '*' but this is not currently " "supported by this implementation, sorry.", fmt ) ); } specs[nargs] = *spec; // make an entry for '*' and point to it from pspec spec->Init(); spec->m_type = wxPAT_STAR; pspec[nargs - 1] = spec; spec = &specs[nargs]; } // If we hit the maximal number of arguments inside the inner // loop, break out of the outer one as well. if ( nargs == wxMAX_SVNPRINTF_ARGUMENTS ) break; } // check if this is a positional or normal argument if ( spec->m_pos > 0 ) { // the positional arguments start from number 1 so we need // to adjust the index spec->m_pos--; posarg_present = true; } else // not a positional argument... { spec->m_pos = nargs; nonposarg_present = true; } // this conversion specifier is tied to the pos-th argument... pspec[spec->m_pos] = spec; if ( ++nargs == wxMAX_SVNPRINTF_ARGUMENTS ) break; } // warn if we lost any arguments (the program probably will crash // anyhow because of stack corruption...) if ( nargs == wxMAX_SVNPRINTF_ARGUMENTS ) { wxFAIL_MSG ( wxString::Format ( "wxVsnprintf() currently supports only %d arguments, " "but format string \"%s\" defines more of them.\n" "You need to change wxMAX_SVNPRINTF_ARGUMENTS and " "recompile if more are really needed.", fmt, wxMAX_SVNPRINTF_ARGUMENTS ) ); } } // total number of valid elements in specs unsigned nargs; // all format specifications in this format string in order of their // appearance (which may be different from arguments order) ConvSpec specs[wxMAX_SVNPRINTF_ARGUMENTS]; // pointer to specs array element for the N-th argument ConvSpec *pspec[wxMAX_SVNPRINTF_ARGUMENTS]; // true if any positional/non-positional parameters are used bool posarg_present, nonposarg_present; }; #undef APPEND_CH #undef CHECK_PREC #endif // _WX_PRIVATE_WXPRINTF_H_