/////////////////////////////////////////////////////////////////////////////// // Name: tests/strings/strings.cpp // Purpose: wxString unit test // Author: Vadim Zeitlin, Wlodzimierz ABX Skiba // Created: 2004-04-19 // RCS-ID: $Id$ // Copyright: (c) 2004 Vadim Zeitlin, Wlodzimierz Skiba /////////////////////////////////////////////////////////////////////////////// // ---------------------------------------------------------------------------- // headers // ---------------------------------------------------------------------------- #include "testprec.h" #ifdef __BORLANDC__ #pragma hdrstop #endif #ifndef WX_PRECOMP #include "wx/wx.h" #endif // WX_PRECOMP // ---------------------------------------------------------------------------- // test class // ---------------------------------------------------------------------------- class StringTestCase : public CppUnit::TestCase { public: StringTestCase(); private: CPPUNIT_TEST_SUITE( StringTestCase ); CPPUNIT_TEST( String ); CPPUNIT_TEST( PChar ); CPPUNIT_TEST( Format ); CPPUNIT_TEST( FormatUnicode ); CPPUNIT_TEST( Constructors ); CPPUNIT_TEST( StaticConstructors ); CPPUNIT_TEST( Extraction ); CPPUNIT_TEST( Trim ); CPPUNIT_TEST( Find ); CPPUNIT_TEST( Replace ); CPPUNIT_TEST( Match ); CPPUNIT_TEST( CaseChanges ); CPPUNIT_TEST( Compare ); CPPUNIT_TEST( CompareNoCase ); CPPUNIT_TEST( Contains ); CPPUNIT_TEST( ToLong ); CPPUNIT_TEST( ToULong ); #ifdef wxLongLong_t CPPUNIT_TEST( ToLongLong ); CPPUNIT_TEST( ToULongLong ); #endif // wxLongLong_t CPPUNIT_TEST( ToDouble ); CPPUNIT_TEST( FromDouble ); CPPUNIT_TEST( StringBuf ); CPPUNIT_TEST( UTF8Buf ); CPPUNIT_TEST( CStrDataTernaryOperator ); CPPUNIT_TEST( CStrDataOperators ); CPPUNIT_TEST( CStrDataImplicitConversion ); CPPUNIT_TEST( ExplicitConversion ); CPPUNIT_TEST( IndexedAccess ); CPPUNIT_TEST( BeforeAndAfter ); CPPUNIT_TEST( ScopedBuffers ); CPPUNIT_TEST_SUITE_END(); void String(); void PChar(); void Format(); void FormatUnicode(); void Constructors(); void StaticConstructors(); void Extraction(); void Trim(); void Find(); void Replace(); void Match(); void CaseChanges(); void Compare(); void CompareNoCase(); void Contains(); void ToLong(); void ToULong(); #ifdef wxLongLong_t void ToLongLong(); void ToULongLong(); #endif // wxLongLong_t void ToDouble(); void FromDouble(); void StringBuf(); void UTF8Buf(); void CStrDataTernaryOperator(); void DoCStrDataTernaryOperator(bool cond); void CStrDataOperators(); void CStrDataImplicitConversion(); void ExplicitConversion(); void IndexedAccess(); void BeforeAndAfter(); void ScopedBuffers(); DECLARE_NO_COPY_CLASS(StringTestCase) }; // register in the unnamed registry so that these tests are run by default CPPUNIT_TEST_SUITE_REGISTRATION( StringTestCase ); // also include in its own registry so that these tests can be run alone CPPUNIT_TEST_SUITE_NAMED_REGISTRATION( StringTestCase, "StringTestCase" ); StringTestCase::StringTestCase() { } void StringTestCase::String() { wxString a, b, c; a.reserve (128); b.reserve (128); c.reserve (128); for (int i = 0; i < 2; ++i) { a = wxT("Hello"); b = wxT(" world"); c = wxT("! How'ya doin'?"); a += b; a += c; c = wxT("Hello world! What's up?"); CPPUNIT_ASSERT( c != a ); } } void StringTestCase::PChar() { wxChar a [128]; wxChar b [128]; wxChar c [128]; for (int i = 0; i < 2; ++i) { wxStrcpy (a, wxT("Hello")); wxStrcpy (b, wxT(" world")); wxStrcpy (c, wxT("! How'ya doin'?")); wxStrcat (a, b); wxStrcat (a, c); wxStrcpy (c, wxT("Hello world! What's up?")); CPPUNIT_ASSERT( wxStrcmp (c, a) != 0 ); } } void StringTestCase::Format() { wxString s1,s2; s1.Printf(wxT("%03d"), 18); CPPUNIT_ASSERT( s1 == wxString::Format(wxT("%03d"), 18) ); s2.Printf(wxT("Number 18: %s\n"), s1.c_str()); CPPUNIT_ASSERT( s2 == wxString::Format(wxT("Number 18: %s\n"), s1.c_str()) ); static const size_t lengths[] = { 1, 512, 1024, 1025, 2048, 4096, 4097 }; for ( size_t n = 0; n < WXSIZEOF(lengths); n++ ) { const size_t len = lengths[n]; wxString s(wxT('Z'), len); CPPUNIT_ASSERT_EQUAL( len, wxString::Format(wxT("%s"), s.c_str()).length()); } CPPUNIT_ASSERT_EQUAL ( "two one", wxString::Format(wxT("%2$s %1$s"), wxT("one"), wxT("two")) ); } void StringTestCase::FormatUnicode() { #if wxUSE_UNICODE const char *UNICODE_STR = "Iestat\xC4\xAB %i%i"; //const char *UNICODE_STR = "Iestat\xCC\x84 %i%i"; wxString fmt = wxString::FromUTF8(UNICODE_STR); wxString s = wxString::Format(fmt, 1, 1); wxString expected(fmt); expected.Replace("%i", "1"); CPPUNIT_ASSERT_EQUAL( expected, s ); #endif // wxUSE_UNICODE } void StringTestCase::Constructors() { CPPUNIT_ASSERT_EQUAL( "", wxString('Z', 0) ); CPPUNIT_ASSERT_EQUAL( "Z", wxString('Z') ); CPPUNIT_ASSERT_EQUAL( "ZZZZ", wxString('Z', 4) ); CPPUNIT_ASSERT_EQUAL( "Hell", wxString("Hello", 4) ); CPPUNIT_ASSERT_EQUAL( "Hello", wxString("Hello", 5) ); #if wxUSE_UNICODE CPPUNIT_ASSERT_EQUAL( L"", wxString(L'Z', 0) ); CPPUNIT_ASSERT_EQUAL( L"Z", wxString(L'Z') ); CPPUNIT_ASSERT_EQUAL( L"ZZZZ", wxString(L'Z', 4) ); CPPUNIT_ASSERT_EQUAL( L"Hell", wxString(L"Hello", 4) ); CPPUNIT_ASSERT_EQUAL( L"Hello", wxString(L"Hello", 5) ); #endif // wxUSE_UNICODE CPPUNIT_ASSERT_EQUAL( 0, wxString(wxString(), 17).length() ); #if wxUSE_UNICODE_UTF8 // This string has 3 characters (, and ), not 4 when using UTF-8 // locale! if ( wxConvLibc.IsUTF8() ) { wxString s3("h\xc3\xa9llo", 4); CPPUNIT_ASSERT_EQUAL( 3, s3.length() ); CPPUNIT_ASSERT_EQUAL( 'l', (char)s3[2] ); } #endif // wxUSE_UNICODE_UTF8 static const char *s = "?really!"; const char *start = wxStrchr(s, 'r'); const char *end = wxStrchr(s, '!'); CPPUNIT_ASSERT_EQUAL( "really", wxString(start, end) ); // test if creating string from NULL C pointer works: CPPUNIT_ASSERT_EQUAL( "", wxString((const char *)NULL) ); } void StringTestCase::StaticConstructors() { CPPUNIT_ASSERT_EQUAL( "", wxString::FromAscii("") ); CPPUNIT_ASSERT_EQUAL( "", wxString::FromAscii("Hello", 0) ); CPPUNIT_ASSERT_EQUAL( "Hell", wxString::FromAscii("Hello", 4) ); CPPUNIT_ASSERT_EQUAL( "Hello", wxString::FromAscii("Hello", 5) ); CPPUNIT_ASSERT_EQUAL( "Hello", wxString::FromAscii("Hello") ); // FIXME: this doesn't work currently but should! //CPPUNIT_ASSERT_EQUAL( 1, wxString::FromAscii("", 1).length() ); CPPUNIT_ASSERT_EQUAL( "", wxString::FromUTF8("") ); CPPUNIT_ASSERT_EQUAL( "", wxString::FromUTF8("Hello", 0) ); CPPUNIT_ASSERT_EQUAL( "Hell", wxString::FromUTF8("Hello", 4) ); CPPUNIT_ASSERT_EQUAL( "Hello", wxString::FromUTF8("Hello", 5) ); CPPUNIT_ASSERT_EQUAL( "Hello", wxString::FromUTF8("Hello") ); CPPUNIT_ASSERT_EQUAL( 2, wxString::FromUTF8("h\xc3\xa9llo", 3).length() ); //CPPUNIT_ASSERT_EQUAL( 1, wxString::FromUTF8("", 1).length() ); } void StringTestCase::Extraction() { wxString s(wxT("Hello, world!")); CPPUNIT_ASSERT( wxStrcmp( s.c_str() , wxT("Hello, world!") ) == 0 ); CPPUNIT_ASSERT( wxStrcmp( s.Left(5).c_str() , wxT("Hello") ) == 0 ); CPPUNIT_ASSERT( wxStrcmp( s.Right(6).c_str() , wxT("world!") ) == 0 ); CPPUNIT_ASSERT( wxStrcmp( s(3, 5).c_str() , wxT("lo, w") ) == 0 ); CPPUNIT_ASSERT( wxStrcmp( s.Mid(3).c_str() , wxT("lo, world!") ) == 0 ); CPPUNIT_ASSERT( wxStrcmp( s.substr(3, 5).c_str() , wxT("lo, w") ) == 0 ); CPPUNIT_ASSERT( wxStrcmp( s.substr(3).c_str() , wxT("lo, world!") ) == 0 ); #if wxUSE_UNICODE static const char *germanUTF8 = "Oberfl\303\244che"; wxString strUnicode(wxString::FromUTF8(germanUTF8)); CPPUNIT_ASSERT( strUnicode.Mid(0, 10) == strUnicode ); CPPUNIT_ASSERT( strUnicode.Mid(7, 2) == "ch" ); #endif // wxUSE_UNICODE wxString rest; #define TEST_STARTS_WITH(prefix, correct_rest, result) \ CPPUNIT_ASSERT_EQUAL(result, s.StartsWith(prefix, &rest)); \ if ( result ) \ CPPUNIT_ASSERT_EQUAL(correct_rest, rest) TEST_STARTS_WITH( wxT("Hello"), wxT(", world!"), true ); TEST_STARTS_WITH( wxT("Hello, "), wxT("world!"), true ); TEST_STARTS_WITH( wxT("Hello, world!"), wxT(""), true ); TEST_STARTS_WITH( wxT("Hello, world!!!"), wxT(""), false ); TEST_STARTS_WITH( wxT(""), wxT("Hello, world!"), true ); TEST_STARTS_WITH( wxT("Goodbye"), wxT(""), false ); TEST_STARTS_WITH( wxT("Hi"), wxT(""), false ); #undef TEST_STARTS_WITH rest = "Hello world"; CPPUNIT_ASSERT( rest.StartsWith("Hello ", &rest) ); CPPUNIT_ASSERT_EQUAL("world", rest); #define TEST_ENDS_WITH(suffix, correct_rest, result) \ CPPUNIT_ASSERT_EQUAL(result, s.EndsWith(suffix, &rest)); \ if ( result ) \ CPPUNIT_ASSERT_EQUAL(correct_rest, rest) TEST_ENDS_WITH( wxT(""), wxT("Hello, world!"), true ); TEST_ENDS_WITH( wxT("!"), wxT("Hello, world"), true ); TEST_ENDS_WITH( wxT(", world!"), wxT("Hello"), true ); TEST_ENDS_WITH( wxT("ello, world!"), wxT("H"), true ); TEST_ENDS_WITH( wxT("Hello, world!"), wxT(""), true ); TEST_ENDS_WITH( wxT("very long string"), wxT(""), false ); TEST_ENDS_WITH( wxT("?"), wxT(""), false ); TEST_ENDS_WITH( wxT("Hello, world"), wxT(""), false ); TEST_ENDS_WITH( wxT("Gello, world!"), wxT(""), false ); #undef TEST_ENDS_WITH } void StringTestCase::Trim() { #define TEST_TRIM( str , dir , result ) \ CPPUNIT_ASSERT( wxString(str).Trim(dir) == result ) TEST_TRIM( wxT(" Test "), true, wxT(" Test") ); TEST_TRIM( wxT(" "), true, wxT("") ); TEST_TRIM( wxT(" "), true, wxT("") ); TEST_TRIM( wxT(""), true, wxT("") ); TEST_TRIM( wxT(" Test "), false, wxT("Test ") ); TEST_TRIM( wxT(" "), false, wxT("") ); TEST_TRIM( wxT(" "), false, wxT("") ); TEST_TRIM( wxT(""), false, wxT("") ); #undef TEST_TRIM } void StringTestCase::Find() { #define TEST_FIND( str , start , result ) \ CPPUNIT_ASSERT( wxString(str).find(wxT("ell"), start) == result ); TEST_FIND( wxT("Well, hello world"), 0, 1 ); TEST_FIND( wxT("Well, hello world"), 6, 7 ); TEST_FIND( wxT("Well, hello world"), 9, wxString::npos ); #undef TEST_FIND } void StringTestCase::Replace() { #define TEST_REPLACE( original , pos , len , replacement , result ) \ { \ wxString s = original; \ s.replace( pos , len , replacement ); \ CPPUNIT_ASSERT_EQUAL( result, s ); \ } TEST_REPLACE( wxT("012-AWORD-XYZ"), 4, 5, wxT("BWORD"), wxT("012-BWORD-XYZ") ); TEST_REPLACE( wxT("increase"), 0, 2, wxT("de"), wxT("decrease") ); TEST_REPLACE( wxT("wxWindow"), 8, 0, wxT("s"), wxT("wxWindows") ); TEST_REPLACE( wxT("foobar"), 3, 0, wxT("-"), wxT("foo-bar") ); TEST_REPLACE( wxT("barfoo"), 0, 6, wxT("foobar"), wxT("foobar") ); #define TEST_NULLCHARREPLACE( o , olen, pos , len , replacement , r, rlen ) \ { \ wxString s(o,olen); \ s.replace( pos , len , replacement ); \ CPPUNIT_ASSERT_EQUAL( wxString(r,rlen), s ); \ } TEST_NULLCHARREPLACE( wxT("null\0char"), 9, 5, 1, wxT("d"), wxT("null\0dhar"), 9 ); #define TEST_WXREPLACE( o , olen, olds, news, all, r, rlen ) \ { \ wxString s(o,olen); \ s.Replace( olds, news, all ); \ CPPUNIT_ASSERT_EQUAL( wxString(r,rlen), s ); \ } TEST_WXREPLACE( wxT("null\0char"), 9, wxT("c"), wxT("de"), true, wxT("null\0dehar"), 10 ); TEST_WXREPLACE( wxT("null\0dehar"), 10, wxT("de"), wxT("c"), true, wxT("null\0char"), 9 ); TEST_WXREPLACE( "life", 4, "f", "", false, "lie", 3 ); TEST_WXREPLACE( "life", 4, "f", "", true, "lie", 3 ); TEST_WXREPLACE( "life", 4, "fe", "ve", true, "live", 4 ); TEST_WXREPLACE( "xx", 2, "x", "yy", true, "yyyy", 4 ); TEST_WXREPLACE( "xxx", 3, "xx", "z", true, "zx", 2 ); #undef TEST_WXREPLACE #undef TEST_NULLCHARREPLACE #undef TEST_REPLACE } void StringTestCase::Match() { #define TEST_MATCH( s1 , s2 , result ) \ CPPUNIT_ASSERT( wxString(s1).Matches(s2) == result ) TEST_MATCH( wxT("foobar"), wxT("foo*"), true ); TEST_MATCH( wxT("foobar"), wxT("*oo*"), true ); TEST_MATCH( wxT("foobar"), wxT("*bar"), true ); TEST_MATCH( wxT("foobar"), wxT("??????"), true ); TEST_MATCH( wxT("foobar"), wxT("f??b*"), true ); TEST_MATCH( wxT("foobar"), wxT("f?b*"), false ); TEST_MATCH( wxT("foobar"), wxT("*goo*"), false ); TEST_MATCH( wxT("foobar"), wxT("*foo"), false ); TEST_MATCH( wxT("foobarfoo"), wxT("*foo"), true ); TEST_MATCH( wxT(""), wxT("*"), true ); TEST_MATCH( wxT(""), wxT("?"), false ); #undef TEST_MATCH } void StringTestCase::CaseChanges() { wxString s1(wxT("Hello!")); wxString s1u(s1); wxString s1l(s1); s1u.MakeUpper(); s1l.MakeLower(); CPPUNIT_ASSERT_EQUAL( wxT("HELLO!"), s1u ); CPPUNIT_ASSERT_EQUAL( wxT("hello!"), s1l ); wxString s2u, s2l; s2u.MakeUpper(); s2l.MakeLower(); CPPUNIT_ASSERT_EQUAL( "", s2u ); CPPUNIT_ASSERT_EQUAL( "", s2l ); wxString s3("good bye"); CPPUNIT_ASSERT_EQUAL( "Good bye", s3.Capitalize() ); s3.MakeCapitalized(); CPPUNIT_ASSERT_EQUAL( "Good bye", s3 ); CPPUNIT_ASSERT_EQUAL( "Abc", wxString("ABC").Capitalize() ); CPPUNIT_ASSERT_EQUAL( "", wxString().Capitalize() ); } void StringTestCase::Compare() { wxString s1 = wxT("AHH"); wxString eq = wxT("AHH"); wxString neq1 = wxT("HAH"); wxString neq2 = wxT("AH"); wxString neq3 = wxT("AHHH"); wxString neq4 = wxT("AhH"); CPPUNIT_ASSERT( s1 == eq ); CPPUNIT_ASSERT( s1 != neq1 ); CPPUNIT_ASSERT( s1 != neq2 ); CPPUNIT_ASSERT( s1 != neq3 ); CPPUNIT_ASSERT( s1 != neq4 ); CPPUNIT_ASSERT( s1 == wxT("AHH") ); CPPUNIT_ASSERT( s1 != wxT("no") ); CPPUNIT_ASSERT( s1 < wxT("AZ") ); CPPUNIT_ASSERT( s1 <= wxT("AZ") ); CPPUNIT_ASSERT( s1 <= wxT("AHH") ); CPPUNIT_ASSERT( s1 > wxT("AA") ); CPPUNIT_ASSERT( s1 >= wxT("AA") ); CPPUNIT_ASSERT( s1 >= wxT("AHH") ); // test comparison with C strings in Unicode build (must work in ANSI as // well, of course): CPPUNIT_ASSERT( s1 == "AHH" ); CPPUNIT_ASSERT( s1 != "no" ); CPPUNIT_ASSERT( s1 < "AZ" ); CPPUNIT_ASSERT( s1 <= "AZ" ); CPPUNIT_ASSERT( s1 <= "AHH" ); CPPUNIT_ASSERT( s1 > "AA" ); CPPUNIT_ASSERT( s1 >= "AA" ); CPPUNIT_ASSERT( s1 >= "AHH" ); // wxString _s1 = wxT("A\0HH"); // wxString _eq = wxT("A\0HH"); // wxString _neq1 = wxT("H\0AH"); // wxString _neq2 = wxT("A\0H"); // wxString _neq3 = wxT("A\0HHH"); // wxString _neq4 = wxT("A\0hH"); s1.insert(1,1,'\0'); eq.insert(1,1,'\0'); neq1.insert(1,1,'\0'); neq2.insert(1,1,'\0'); neq3.insert(1,1,'\0'); neq4.insert(1,1,'\0'); CPPUNIT_ASSERT( s1 == eq ); CPPUNIT_ASSERT( s1 != neq1 ); CPPUNIT_ASSERT( s1 != neq2 ); CPPUNIT_ASSERT( s1 != neq3 ); CPPUNIT_ASSERT( s1 != neq4 ); CPPUNIT_ASSERT( wxString("\n").Cmp(" ") < 0 ); CPPUNIT_ASSERT( wxString("'").Cmp("!") > 0 ); CPPUNIT_ASSERT( wxString("!").Cmp("z") < 0 ); } void StringTestCase::CompareNoCase() { wxString s1 = wxT("AHH"); wxString eq = wxT("AHH"); wxString eq2 = wxT("AhH"); wxString eq3 = wxT("ahh"); wxString neq = wxT("HAH"); wxString neq2 = wxT("AH"); wxString neq3 = wxT("AHHH"); #define CPPUNIT_CNCEQ_ASSERT(s1, s2) CPPUNIT_ASSERT( s1.CmpNoCase(s2) == 0) #define CPPUNIT_CNCNEQ_ASSERT(s1, s2) CPPUNIT_ASSERT( s1.CmpNoCase(s2) != 0) CPPUNIT_CNCEQ_ASSERT( s1, eq ); CPPUNIT_CNCEQ_ASSERT( s1, eq2 ); CPPUNIT_CNCEQ_ASSERT( s1, eq3 ); CPPUNIT_CNCNEQ_ASSERT( s1, neq ); CPPUNIT_CNCNEQ_ASSERT( s1, neq2 ); CPPUNIT_CNCNEQ_ASSERT( s1, neq3 ); // wxString _s1 = wxT("A\0HH"); // wxString _eq = wxT("A\0HH"); // wxString _eq2 = wxT("A\0hH"); // wxString _eq3 = wxT("a\0hh"); // wxString _neq = wxT("H\0AH"); // wxString _neq2 = wxT("A\0H"); // wxString _neq3 = wxT("A\0HHH"); s1.insert(1,1,'\0'); eq.insert(1,1,'\0'); eq2.insert(1,1,'\0'); eq3.insert(1,1,'\0'); neq.insert(1,1,'\0'); neq2.insert(1,1,'\0'); neq3.insert(1,1,'\0'); CPPUNIT_CNCEQ_ASSERT( s1, eq ); CPPUNIT_CNCEQ_ASSERT( s1, eq2 ); CPPUNIT_CNCEQ_ASSERT( s1, eq3 ); CPPUNIT_CNCNEQ_ASSERT( s1, neq ); CPPUNIT_CNCNEQ_ASSERT( s1, neq2 ); CPPUNIT_CNCNEQ_ASSERT( s1, neq3 ); CPPUNIT_ASSERT( wxString("\n").CmpNoCase(" ") < 0 ); CPPUNIT_ASSERT( wxString("'").CmpNoCase("!") > 0); CPPUNIT_ASSERT( wxString("!").Cmp("Z") < 0 ); } void StringTestCase::Contains() { static const struct ContainsData { const wxChar *hay; const wxChar *needle; bool contains; } containsData[] = { { wxT(""), wxT(""), true }, { wxT(""), wxT("foo"), false }, { wxT("foo"), wxT(""), true }, { wxT("foo"), wxT("f"), true }, { wxT("foo"), wxT("o"), true }, { wxT("foo"), wxT("oo"), true }, { wxT("foo"), wxT("ooo"), false }, { wxT("foo"), wxT("oooo"), false }, { wxT("foo"), wxT("fooo"), false }, }; for ( size_t n = 0; n < WXSIZEOF(containsData); n++ ) { const ContainsData& cd = containsData[n]; CPPUNIT_ASSERT_EQUAL( cd.contains, wxString(cd.hay).Contains(cd.needle) ); } } // flags used in ToLongData.flags enum { Number_Ok = 0, Number_Invalid = 1, Number_Unsigned = 2, // if not specified, works for signed conversion Number_Signed = 4, // if not specified, works for unsigned Number_LongLong = 8, // only for long long tests Number_Long = 16 // only for long tests }; static const struct ToLongData { const wxChar *str; #ifdef wxLongLong_t wxLongLong_t value; #else long value; #endif // wxLongLong_t int flags; int base; long LValue() const { return value; } unsigned long ULValue() const { return value; } #ifdef wxLongLong_t wxLongLong_t LLValue() const { return value; } wxULongLong_t ULLValue() const { return (wxULongLong_t)value; } #endif // wxLongLong_t bool IsOk() const { return !(flags & Number_Invalid); } } longData[] = { { wxT("1"), 1, Number_Ok }, { wxT("0"), 0, Number_Ok }, { wxT("a"), 0, Number_Invalid }, { wxT("12345"), 12345, Number_Ok }, { wxT("--1"), 0, Number_Invalid }, { wxT("-1"), -1, Number_Signed | Number_Long }, // this is surprising but consistent with strtoul() behaviour { wxT("-1"), ULONG_MAX, Number_Unsigned | Number_Long }, // this must overflow, even with 64 bit long { wxT("922337203685477580711"), 0, Number_Invalid }, #ifdef wxLongLong_t { wxT("2147483648"), wxLL(2147483648), Number_LongLong }, { wxT("-2147483648"), wxLL(-2147483648), Number_LongLong | Number_Signed }, { wxT("9223372036854775808"), wxULL(9223372036854775808), Number_LongLong | Number_Unsigned }, #endif // wxLongLong_t // Base tests. { wxT("010"), 10, Number_Ok, 10 }, { wxT("010"), 8, Number_Ok, 0 }, { wxT("010"), 8, Number_Ok, 8 }, { wxT("010"), 16, Number_Ok, 16 }, { wxT("0010"), 10, Number_Ok, 10 }, { wxT("0010"), 8, Number_Ok, 0 }, { wxT("0010"), 8, Number_Ok, 8 }, { wxT("0010"), 16, Number_Ok, 16 }, { wxT("0x11"), 0, Number_Invalid, 10 }, { wxT("0x11"), 17, Number_Ok, 0 }, { wxT("0x11"), 0, Number_Invalid, 8 }, { wxT("0x11"), 17, Number_Ok, 16 }, }; void StringTestCase::ToLong() { long l; for ( size_t n = 0; n < WXSIZEOF(longData); n++ ) { const ToLongData& ld = longData[n]; if ( ld.flags & (Number_LongLong | Number_Unsigned) ) continue; // NOTE: unless you're using some exotic locale, ToCLong and ToLong // should behave the same for our test data set: CPPUNIT_ASSERT_EQUAL( ld.IsOk(), wxString(ld.str).ToCLong(&l, ld.base) ); if ( ld.IsOk() ) CPPUNIT_ASSERT_EQUAL( ld.LValue(), l ); CPPUNIT_ASSERT_EQUAL( ld.IsOk(), wxString(ld.str).ToLong(&l, ld.base) ); if ( ld.IsOk() ) CPPUNIT_ASSERT_EQUAL( ld.LValue(), l ); } // special case: check that the output is not modified if the parsing // failed completely l = 17; CPPUNIT_ASSERT( !wxString("foo").ToLong(&l) ); CPPUNIT_ASSERT_EQUAL( 17, l ); // also check that it is modified if we did parse something successfully in // the beginning of the string CPPUNIT_ASSERT( !wxString("9 cats").ToLong(&l) ); CPPUNIT_ASSERT_EQUAL( 9, l ); } void StringTestCase::ToULong() { unsigned long ul; for ( size_t n = 0; n < WXSIZEOF(longData); n++ ) { const ToLongData& ld = longData[n]; if ( ld.flags & (Number_LongLong | Number_Signed) ) continue; // NOTE: unless you're using some exotic locale, ToCLong and ToLong // should behave the same for our test data set: CPPUNIT_ASSERT_EQUAL( ld.IsOk(), wxString(ld.str).ToCULong(&ul, ld.base) ); if ( ld.IsOk() ) CPPUNIT_ASSERT_EQUAL( ld.ULValue(), ul ); CPPUNIT_ASSERT_EQUAL( ld.IsOk(), wxString(ld.str).ToULong(&ul, ld.base) ); if ( ld.IsOk() ) CPPUNIT_ASSERT_EQUAL( ld.ULValue(), ul ); } } #ifdef wxLongLong_t void StringTestCase::ToLongLong() { wxLongLong_t l; for ( size_t n = 0; n < WXSIZEOF(longData); n++ ) { const ToLongData& ld = longData[n]; if ( ld.flags & (Number_Long | Number_Unsigned) ) continue; CPPUNIT_ASSERT_EQUAL( ld.IsOk(), wxString(ld.str).ToLongLong(&l, ld.base) ); if ( ld.IsOk() ) CPPUNIT_ASSERT_EQUAL( ld.LLValue(), l ); } } void StringTestCase::ToULongLong() { wxULongLong_t ul; for ( size_t n = 0; n < WXSIZEOF(longData); n++ ) { const ToLongData& ld = longData[n]; if ( ld.flags & (Number_Long | Number_Signed) ) continue; CPPUNIT_ASSERT_EQUAL( ld.IsOk(), wxString(ld.str).ToULongLong(&ul, ld.base) ); if ( ld.IsOk() ) CPPUNIT_ASSERT_EQUAL( ld.ULLValue(), ul ); } } #endif // wxLongLong_t void StringTestCase::ToDouble() { double d; static const struct ToDoubleData { const wxChar *str; double value; bool ok; } doubleData[] = { { wxT("1"), 1, true }, { wxT("1.23"), 1.23, true }, { wxT(".1"), .1, true }, { wxT("1."), 1, true }, { wxT("1.."), 0, false }, { wxT("0"), 0, true }, { wxT("a"), 0, false }, { wxT("12345"), 12345, true }, { wxT("-1"), -1, true }, { wxT("--1"), 0, false }, { wxT("-3E-5"), -3E-5, true }, { wxT("-3E-abcde5"), 0, false }, }; // test ToCDouble() first: size_t n; for ( n = 0; n < WXSIZEOF(doubleData); n++ ) { const ToDoubleData& ld = doubleData[n]; CPPUNIT_ASSERT_EQUAL( ld.ok, wxString(ld.str).ToCDouble(&d) ); if ( ld.ok ) CPPUNIT_ASSERT_EQUAL( ld.value, d ); } // test ToDouble() now: // NOTE: for the test to be reliable, we need to set the locale explicitly // so that we know the decimal point character to use if (!wxLocale::IsAvailable(wxLANGUAGE_FRENCH)) return; // you should have french support installed to continue this test! wxLocale locale; // don't load default catalog, it may be unavailable: CPPUNIT_ASSERT( locale.Init(wxLANGUAGE_FRENCH, wxLOCALE_DONT_LOAD_DEFAULT) ); static const struct ToDoubleData doubleData2[] = { { wxT("1"), 1, true }, { wxT("1,23"), 1.23, true }, { wxT(",1"), .1, true }, { wxT("1,"), 1, true }, { wxT("1,,"), 0, false }, { wxT("0"), 0, true }, { wxT("a"), 0, false }, { wxT("12345"), 12345, true }, { wxT("-1"), -1, true }, { wxT("--1"), 0, false }, { wxT("-3E-5"), -3E-5, true }, { wxT("-3E-abcde5"), 0, false }, }; for ( n = 0; n < WXSIZEOF(doubleData2); n++ ) { const ToDoubleData& ld = doubleData2[n]; CPPUNIT_ASSERT_EQUAL( ld.ok, wxString(ld.str).ToDouble(&d) ); if ( ld.ok ) CPPUNIT_ASSERT_EQUAL( ld.value, d ); } } void StringTestCase::FromDouble() { static const struct FromDoubleTestData { double value; int prec; const char *str; } testData[] = { { 1.23, -1, "1.23" }, // NB: there are no standards about the minimum exponent width // and newer MSVC versions use 3 digits as minimum exponent // width while GNU libc uses 2 digits as minimum width... #ifdef wxUSING_VC_CRT_IO { -3e-10, -1, "-3e-010" }, #else { -3e-10, -1, "-3e-10" }, #endif { -0.45678, -1, "-0.45678" }, { 1.2345678, 0, "1" }, { 1.2345678, 1, "1.2" }, { 1.2345678, 2, "1.23" }, { 1.2345678, 3, "1.235" }, }; for ( unsigned n = 0; n < WXSIZEOF(testData); n++ ) { const FromDoubleTestData& td = testData[n]; CPPUNIT_ASSERT_EQUAL( td.str, wxString::FromCDouble(td.value, td.prec) ); } if ( !wxLocale::IsAvailable(wxLANGUAGE_FRENCH) ) return; wxLocale locale; CPPUNIT_ASSERT( locale.Init(wxLANGUAGE_FRENCH, wxLOCALE_DONT_LOAD_DEFAULT) ); for ( unsigned m = 0; m < WXSIZEOF(testData); m++ ) { const FromDoubleTestData& td = testData[m]; wxString str(td.str); str.Replace(".", ","); CPPUNIT_ASSERT_EQUAL( str, wxString::FromDouble(td.value, td.prec) ); } } void StringTestCase::StringBuf() { // check that buffer can be used to write into the string wxString s; wxStrcpy(wxStringBuffer(s, 10), wxT("foo")); CPPUNIT_ASSERT_EQUAL(3, s.length()); CPPUNIT_ASSERT(wxT('f') == s[0u]); CPPUNIT_ASSERT(wxT('o') == s[1]); CPPUNIT_ASSERT(wxT('o') == s[2]); { // also check that the buffer initially contains the original string // contents wxStringBuffer buf(s, 10); CPPUNIT_ASSERT_EQUAL( wxT('f'), buf[0] ); CPPUNIT_ASSERT_EQUAL( wxT('o'), buf[1] ); CPPUNIT_ASSERT_EQUAL( wxT('o'), buf[2] ); CPPUNIT_ASSERT_EQUAL( wxT('\0'), buf[3] ); } { wxStringBufferLength buf(s, 10); CPPUNIT_ASSERT_EQUAL( wxT('f'), buf[0] ); CPPUNIT_ASSERT_EQUAL( wxT('o'), buf[1] ); CPPUNIT_ASSERT_EQUAL( wxT('o'), buf[2] ); CPPUNIT_ASSERT_EQUAL( wxT('\0'), buf[3] ); // and check that it can be used to write only the specified number of // characters to the string wxStrcpy(buf, wxT("barrbaz")); buf.SetLength(4); } CPPUNIT_ASSERT_EQUAL(4, s.length()); CPPUNIT_ASSERT(wxT('b') == s[0u]); CPPUNIT_ASSERT(wxT('a') == s[1]); CPPUNIT_ASSERT(wxT('r') == s[2]); CPPUNIT_ASSERT(wxT('r') == s[3]); // check that creating buffer of length smaller than string works, i.e. at // least doesn't crash (it would if we naively copied the entire original // string contents in the buffer) *wxStringBuffer(s, 1) = '!'; } void StringTestCase::UTF8Buf() { #if wxUSE_UNICODE // "czech" in Czech ("cestina"): static const char *textUTF8 = "\304\215e\305\241tina"; static const wchar_t textUTF16[] = {0x10D, 0x65, 0x161, 0x74, 0x69, 0x6E, 0x61, 0}; wxString s; wxStrcpy(wxUTF8StringBuffer(s, 9), textUTF8); CPPUNIT_ASSERT(s == textUTF16); { wxUTF8StringBufferLength buf(s, 20); wxStrcpy(buf, textUTF8); buf.SetLength(5); } CPPUNIT_ASSERT(s == wxString(textUTF16, 0, 3)); #endif // wxUSE_UNICODE } void StringTestCase::CStrDataTernaryOperator() { DoCStrDataTernaryOperator(true); DoCStrDataTernaryOperator(false); } template bool CheckStr(const wxString& expected, T s) { return expected == wxString(s); } void StringTestCase::DoCStrDataTernaryOperator(bool cond) { // test compilation of wxCStrData when used with operator?: (the asserts // are not very important, we're testing if the code compiles at all): wxString s("foo"); const wchar_t *wcStr = L"foo"; CPPUNIT_ASSERT( CheckStr(s, (cond ? s.c_str() : wcStr)) ); CPPUNIT_ASSERT( CheckStr(s, (cond ? s.c_str() : L"foo")) ); CPPUNIT_ASSERT( CheckStr(s, (cond ? wcStr : s.c_str())) ); CPPUNIT_ASSERT( CheckStr(s, (cond ? L"foo" : s.c_str())) ); const char *mbStr = "foo"; CPPUNIT_ASSERT( CheckStr(s, (cond ? s.c_str() : mbStr)) ); CPPUNIT_ASSERT( CheckStr(s, (cond ? s.c_str() : "foo")) ); CPPUNIT_ASSERT( CheckStr(s, (cond ? mbStr : s.c_str())) ); CPPUNIT_ASSERT( CheckStr(s, (cond ? "foo" : s.c_str())) ); wxString empty(""); CPPUNIT_ASSERT( CheckStr(empty, (cond ? empty.c_str() : wxEmptyString)) ); CPPUNIT_ASSERT( CheckStr(empty, (cond ? wxEmptyString : empty.c_str())) ); } void StringTestCase::CStrDataOperators() { wxString s("hello"); CPPUNIT_ASSERT( s.c_str()[0] == 'h' ); CPPUNIT_ASSERT( s.c_str()[1] == 'e' ); // IMPORTANT: at least with the CRT coming with MSVC++ 2008 trying to access // the final character results in an assert failure (with debug CRT) //CPPUNIT_ASSERT( s.c_str()[5] == '\0' ); CPPUNIT_ASSERT( *s.c_str() == 'h' ); CPPUNIT_ASSERT( *(s.c_str() + 2) == 'l' ); //CPPUNIT_ASSERT( *(s.c_str() + 5) == '\0' ); } bool CheckStrChar(const wxString& expected, char *s) { return CheckStr(expected, s); } bool CheckStrWChar(const wxString& expected, wchar_t *s) { return CheckStr(expected, s); } bool CheckStrConstChar(const wxString& expected, const char *s) { return CheckStr(expected, s); } bool CheckStrConstWChar(const wxString& expected, const wchar_t *s) { return CheckStr(expected, s); } void StringTestCase::CStrDataImplicitConversion() { wxString s("foo"); CPPUNIT_ASSERT( CheckStrConstWChar(s, s.c_str()) ); CPPUNIT_ASSERT( CheckStrConstChar(s, s.c_str()) ); // implicit conversion of wxString is not available in STL build #if !wxUSE_STL CPPUNIT_ASSERT( CheckStrConstWChar(s, s) ); CPPUNIT_ASSERT( CheckStrConstChar(s, s) ); #endif } void StringTestCase::ExplicitConversion() { wxString s("foo"); CPPUNIT_ASSERT( CheckStr(s, s.mb_str()) ); CPPUNIT_ASSERT( CheckStrConstChar(s, s.mb_str()) ); CPPUNIT_ASSERT( CheckStrChar(s, s.char_str()) ); CPPUNIT_ASSERT( CheckStr(s, s.wc_str()) ); CPPUNIT_ASSERT( CheckStrConstWChar(s, s.wc_str()) ); CPPUNIT_ASSERT( CheckStrWChar(s, s.wchar_str()) ); } void StringTestCase::IndexedAccess() { wxString s("bar"); CPPUNIT_ASSERT_EQUAL( 'r', (char)s[2] ); // this tests for a possible bug in UTF-8 based wxString implementation: // the 3rd character of the underlying byte string is going to change, but // the 3rd character of wxString should remain the same s[0] = L'\xe9'; CPPUNIT_ASSERT_EQUAL( 'r', (char)s[2] ); } void StringTestCase::BeforeAndAfter() { // Construct a string with 2 equal signs in it by concatenating its three // parts: before the first "=", in between the two "="s and after the last // one. This allows to avoid duplicating the string contents (which has to // be different for Unicode and ANSI builds) in the tests below. #if wxUSE_UNICODE #define FIRST_PART L"letter" #define MIDDLE_PART L"\xe9;\xe7a" #define LAST_PART L"l\xe0" #else // !wxUSE_UNICODE #define FIRST_PART "letter" #define MIDDLE_PART "e;ca" #define LAST_PART "la" #endif // wxUSE_UNICODE/!wxUSE_UNICODE const wxString s(FIRST_PART wxT("=") MIDDLE_PART wxT("=") LAST_PART); wxString r; CPPUNIT_ASSERT_EQUAL( FIRST_PART, s.BeforeFirst('=', &r) ); CPPUNIT_ASSERT_EQUAL( MIDDLE_PART wxT("=") LAST_PART, r ); CPPUNIT_ASSERT_EQUAL( s, s.BeforeFirst('!', &r) ); CPPUNIT_ASSERT_EQUAL( "", r ); CPPUNIT_ASSERT_EQUAL( FIRST_PART wxT("=") MIDDLE_PART, s.BeforeLast('=', &r) ); CPPUNIT_ASSERT_EQUAL( LAST_PART, r ); CPPUNIT_ASSERT_EQUAL( "", s.BeforeLast('!', &r) ); CPPUNIT_ASSERT_EQUAL( s, r ); CPPUNIT_ASSERT_EQUAL( MIDDLE_PART wxT("=") LAST_PART, s.AfterFirst('=') ); CPPUNIT_ASSERT_EQUAL( "", s.AfterFirst('!') ); CPPUNIT_ASSERT_EQUAL( LAST_PART, s.AfterLast('=') ); CPPUNIT_ASSERT_EQUAL( s, s.AfterLast('!') ); #undef LAST_PART #undef MIDDLE_PART #undef FIRST_PART } void StringTestCase::ScopedBuffers() { // wxString relies on efficient buffers, verify they work as they should const char *literal = "Hello World!"; // non-owned buffer points to the string passed to it wxScopedCharBuffer sbuf = wxScopedCharBuffer::CreateNonOwned(literal); CPPUNIT_ASSERT( sbuf.data() == literal ); // a copy of scoped non-owned buffer still points to the same string wxScopedCharBuffer sbuf2(sbuf); CPPUNIT_ASSERT( sbuf.data() == sbuf2.data() ); // but assigning it to wxCharBuffer makes a full copy wxCharBuffer buf(sbuf); CPPUNIT_ASSERT( buf.data() != literal ); CPPUNIT_ASSERT_EQUAL( literal, buf.data() ); wxCharBuffer buf2 = sbuf; CPPUNIT_ASSERT( buf2.data() != literal ); CPPUNIT_ASSERT_EQUAL( literal, buf.data() ); // Check that extending the buffer keeps it NUL-terminated. size_t len = 10; wxCharBuffer buf3(len); CPPUNIT_ASSERT_EQUAL('\0', buf3.data()[len]); wxCharBuffer buf4; buf4.extend(len); CPPUNIT_ASSERT_EQUAL('\0', buf4.data()[len]); wxCharBuffer buf5(5); buf5.extend(len); CPPUNIT_ASSERT_EQUAL('\0', buf5.data()[len]); }