X-Git-Url: https://git.saurik.com/wxWidgets.git/blobdiff_plain/cd5b511a70590655c50676cbb7ca8192ff0acc35..d485bda109d5ef0fef36a3f737549e9b9f54baab:/src/common/fmapbase.cpp?ds=sidebyside diff --git a/src/common/fmapbase.cpp b/src/common/fmapbase.cpp index 4ad527ddfe..5693f8c0ad 100644 --- a/src/common/fmapbase.cpp +++ b/src/common/fmapbase.cpp @@ -1,12 +1,12 @@ /////////////////////////////////////////////////////////////////////////////// -// Name: common/fmapbase.cpp +// Name: src/common/fmapbase.cpp // Purpose: wxFontMapperBase class implementation // Author: Vadim Zeitlin // Modified by: // Created: 21.06.2003 (extracted from common/fontmap.cpp) // RCS-ID: $Id$ // Copyright: (c) 1999-2003 Vadim Zeitlin -// License: wxWindows license +// Licence: wxWindows licence /////////////////////////////////////////////////////////////////////////////// // ============================================================================ @@ -30,18 +30,19 @@ #include "wx/app.h" #include "wx/log.h" #include "wx/intl.h" + #include "wx/module.h" + #include "wx/wxcrtvararg.h" #endif //WX_PRECOMP -#if defined(__WXMSW__) - #include "wx/msw/private.h" // includes windows.h for LOGFONT - #include "wx/msw/winundef.h" +#if defined(__WINDOWS__) + #include "wx/msw/private.h" // includes windows.h for LOGFONT + #include "wx/msw/winundef.h" #endif #include "wx/fontmap.h" #include "wx/fmappriv.h" #include "wx/apptrait.h" -#include "wx/module.h" // wxMemoryConfig uses wxFileConfig #if wxUSE_CONFIG && wxUSE_FILECONFIG @@ -54,7 +55,7 @@ // ---------------------------------------------------------------------------- // encodings supported by GetEncodingDescription -static wxFontEncoding gs_encodings[] = +static const wxFontEncoding gs_encodings[] = { wxFONTENCODING_ISO8859_1, wxFONTENCODING_ISO8859_2, @@ -73,6 +74,7 @@ static wxFontEncoding gs_encodings[] = wxFONTENCODING_ISO8859_15, wxFONTENCODING_KOI8, wxFONTENCODING_KOI8_U, + wxFONTENCODING_CP866, wxFONTENCODING_CP874, wxFONTENCODING_CP932, wxFONTENCODING_CP936, @@ -86,24 +88,63 @@ static wxFontEncoding gs_encodings[] = wxFONTENCODING_CP1255, wxFONTENCODING_CP1256, wxFONTENCODING_CP1257, + wxFONTENCODING_CP1258, + wxFONTENCODING_CP1361, wxFONTENCODING_CP437, wxFONTENCODING_UTF7, wxFONTENCODING_UTF8, - wxFONTENCODING_UTF16, wxFONTENCODING_UTF16BE, wxFONTENCODING_UTF16LE, - wxFONTENCODING_UTF32, wxFONTENCODING_UTF32BE, wxFONTENCODING_UTF32LE, wxFONTENCODING_EUC_JP, wxFONTENCODING_DEFAULT, - wxFONTENCODING_BIG5, - wxFONTENCODING_SHIFT_JIS, - wxFONTENCODING_GB2312, + wxFONTENCODING_ISO2022_JP, + + wxFONTENCODING_MACROMAN, + wxFONTENCODING_MACJAPANESE, + wxFONTENCODING_MACCHINESETRAD, + wxFONTENCODING_MACKOREAN, + wxFONTENCODING_MACARABIC, + wxFONTENCODING_MACHEBREW, + wxFONTENCODING_MACGREEK, + wxFONTENCODING_MACCYRILLIC, + wxFONTENCODING_MACDEVANAGARI, + wxFONTENCODING_MACGURMUKHI, + wxFONTENCODING_MACGUJARATI, + wxFONTENCODING_MACORIYA, + wxFONTENCODING_MACBENGALI, + wxFONTENCODING_MACTAMIL, + wxFONTENCODING_MACTELUGU, + wxFONTENCODING_MACKANNADA, + wxFONTENCODING_MACMALAJALAM, + wxFONTENCODING_MACSINHALESE, + wxFONTENCODING_MACBURMESE, + wxFONTENCODING_MACKHMER, + wxFONTENCODING_MACTHAI, + wxFONTENCODING_MACLAOTIAN, + wxFONTENCODING_MACGEORGIAN, + wxFONTENCODING_MACARMENIAN, + wxFONTENCODING_MACCHINESESIMP, + wxFONTENCODING_MACTIBETAN, + wxFONTENCODING_MACMONGOLIAN, + wxFONTENCODING_MACETHIOPIC, + wxFONTENCODING_MACCENTRALEUR, + wxFONTENCODING_MACVIATNAMESE, + wxFONTENCODING_MACARABICEXT, + wxFONTENCODING_MACSYMBOL, + wxFONTENCODING_MACDINGBATS, + wxFONTENCODING_MACTURKISH, + wxFONTENCODING_MACCROATIAN, + wxFONTENCODING_MACICELANDIC, + wxFONTENCODING_MACROMANIAN, + wxFONTENCODING_MACCELTIC, + wxFONTENCODING_MACGAELIC, + wxFONTENCODING_MACKEYBOARD }; // the descriptions for them -static const wxChar* gs_encodingDescs[] = +static const char* const gs_encodingDescs[] = { wxTRANSLATE( "Western European (ISO-8859-1)" ), wxTRANSLATE( "Central European (ISO-8859-2)" ), @@ -122,11 +163,12 @@ static const wxChar* gs_encodingDescs[] = wxTRANSLATE( "Western European with Euro (ISO-8859-15)" ), wxTRANSLATE( "KOI8-R" ), wxTRANSLATE( "KOI8-U" ), + wxTRANSLATE( "Windows/DOS OEM Cyrillic (CP 866)" ), wxTRANSLATE( "Windows Thai (CP 874)" ), - wxTRANSLATE( "Windows Japanese (CP 932)" ), - wxTRANSLATE( "Windows Chinese Simplified (CP 936)" ), + wxTRANSLATE( "Windows Japanese (CP 932) or Shift-JIS" ), + wxTRANSLATE( "Windows Chinese Simplified (CP 936) or GB-2312" ), wxTRANSLATE( "Windows Korean (CP 949)" ), - wxTRANSLATE( "Windows Chinese Traditional (CP 950)" ), + wxTRANSLATE( "Windows Chinese Traditional (CP 950) or Big-5" ), wxTRANSLATE( "Windows Central European (CP 1250)" ), wxTRANSLATE( "Windows Cyrillic (CP 1251)" ), wxTRANSLATE( "Windows Western European (CP 1252)" ), @@ -135,42 +177,88 @@ static const wxChar* gs_encodingDescs[] = wxTRANSLATE( "Windows Hebrew (CP 1255)" ), wxTRANSLATE( "Windows Arabic (CP 1256)" ), wxTRANSLATE( "Windows Baltic (CP 1257)" ), + wxTRANSLATE( "Windows Vietnamese (CP 1258)" ), + wxTRANSLATE( "Windows Johab (CP 1361)" ), wxTRANSLATE( "Windows/DOS OEM (CP 437)" ), wxTRANSLATE( "Unicode 7 bit (UTF-7)" ), wxTRANSLATE( "Unicode 8 bit (UTF-8)" ), +#ifdef WORDS_BIGENDIAN wxTRANSLATE( "Unicode 16 bit (UTF-16)" ), - wxTRANSLATE( "Unicode 16 bit Big Endian (UTF-16BE)" ), wxTRANSLATE( "Unicode 16 bit Little Endian (UTF-16LE)" ), wxTRANSLATE( "Unicode 32 bit (UTF-32)" ), - wxTRANSLATE( "Unicode 32 bit Big Endian (UTF-32BE)" ), wxTRANSLATE( "Unicode 32 bit Little Endian (UTF-32LE)" ), +#else // WORDS_BIGENDIAN + wxTRANSLATE( "Unicode 16 bit Big Endian (UTF-16BE)" ), + wxTRANSLATE( "Unicode 16 bit (UTF-16)" ), + wxTRANSLATE( "Unicode 32 bit Big Endian (UTF-32BE)" ), + wxTRANSLATE( "Unicode 32 bit (UTF-32)" ), +#endif // WORDS_BIGENDIAN wxTRANSLATE( "Extended Unix Codepage for Japanese (EUC-JP)" ), wxTRANSLATE( "US-ASCII" ), - wxTRANSLATE( "BIG5" ), - wxTRANSLATE( "SHIFT-JIS" ), - wxTRANSLATE( "GB-2312" ), + wxTRANSLATE( "ISO-2022-JP" ), + + wxTRANSLATE( "MacRoman" ), + wxTRANSLATE( "MacJapanese" ), + wxTRANSLATE( "MacChineseTrad" ), + wxTRANSLATE( "MacKorean" ), + wxTRANSLATE( "MacArabic" ), + wxTRANSLATE( "MacHebrew" ), + wxTRANSLATE( "MacGreek" ), + wxTRANSLATE( "MacCyrillic" ), + wxTRANSLATE( "MacDevanagari" ), + wxTRANSLATE( "MacGurmukhi" ), + wxTRANSLATE( "MacGujarati" ), + wxTRANSLATE( "MacOriya" ), + wxTRANSLATE( "MacBengali" ), + wxTRANSLATE( "MacTamil" ), + wxTRANSLATE( "MacTelugu" ), + wxTRANSLATE( "MacKannada" ), + wxTRANSLATE( "MacMalayalam" ), + wxTRANSLATE( "MacSinhalese" ), + wxTRANSLATE( "MacBurmese" ), + wxTRANSLATE( "MacKhmer" ), + wxTRANSLATE( "MacThai" ), + wxTRANSLATE( "MacLaotian" ), + wxTRANSLATE( "MacGeorgian" ), + wxTRANSLATE( "MacArmenian" ), + wxTRANSLATE( "MacChineseSimp" ), + wxTRANSLATE( "MacTibetan" ), + wxTRANSLATE( "MacMongolian" ), + wxTRANSLATE( "MacEthiopic" ), + wxTRANSLATE( "MacCentralEurRoman" ), + wxTRANSLATE( "MacVietnamese" ), + wxTRANSLATE( "MacExtArabic" ), + wxTRANSLATE( "MacSymbol" ), + wxTRANSLATE( "MacDingbats" ), + wxTRANSLATE( "MacTurkish" ), + wxTRANSLATE( "MacCroatian" ), + wxTRANSLATE( "MacIcelandic" ), + wxTRANSLATE( "MacRomanian" ), + wxTRANSLATE( "MacCeltic" ), + wxTRANSLATE( "MacGaelic" ), + wxTRANSLATE( "MacKeyboardGlyphs" ) }; // and the internal names (these are not translated on purpose!) -static const wxChar* gs_encodingNames[WXSIZEOF(gs_encodingDescs)][10] = +static const wxChar* const gs_encodingNames[][9] = { - // names from the first column are used under Linux, from the second one - // under Solaris and IRIX and from the third one -- under AIX - { wxT( "ISO-8859-1" ), wxT( "ISO8859-1" ), wxT( "8859-1" ), wxT("iso_8859_1"), NULL }, - { wxT( "ISO-8859-2" ), wxT( "ISO8859-2" ), wxT( "8859-2" ), NULL }, - { wxT( "ISO-8859-3" ), wxT( "ISO8859-3" ), wxT( "8859-3" ), NULL }, - { wxT( "ISO-8859-4" ), wxT( "ISO8859-4" ), wxT( "8859-4" ), NULL }, - { wxT( "ISO-8859-5" ), wxT( "ISO8859-5" ), wxT( "8859-5" ), NULL }, - { wxT( "ISO-8859-6" ), wxT( "ISO8859-6" ), wxT( "8859-6" ), NULL }, - { wxT( "ISO-8859-7" ), wxT( "ISO8859-7" ), wxT( "8859-7" ), NULL }, - { wxT( "ISO-8859-8" ), wxT( "ISO8859-8" ), wxT( "8859-8" ), NULL }, - { wxT( "ISO-8859-9" ), wxT( "ISO8859-9" ), wxT( "8859-9" ), NULL }, - { wxT( "ISO-8859-10" ), wxT( "ISO8859-10" ), wxT( "8859-10" ), NULL }, - { wxT( "ISO-8859-11" ), wxT( "ISO8859-11" ), wxT( "8859-11" ), NULL }, - { wxT( "ISO-8859-12" ), wxT( "ISO8859-12" ), wxT( "8859-12" ), NULL }, - { wxT( "ISO-8859-13" ), wxT( "ISO8859-13" ), wxT( "8859-13" ), NULL }, - { wxT( "ISO-8859-14" ), wxT( "ISO8859-14" ), wxT( "8859-14" ), NULL }, - { wxT( "ISO-8859-15" ), wxT( "ISO8859-15" ), wxT( "8859-15" ), NULL }, + // names from the columns correspond to these OS: + // Linux Solaris and IRIX HP-UX AIX + { wxT("ISO-8859-1"), wxT("ISO8859-1"), wxT("iso88591"), wxT("8859-1"), wxT("iso_8859_1"), NULL }, + { wxT("ISO-8859-2"), wxT("ISO8859-2"), wxT("iso88592"), wxT("8859-2"), NULL }, + { wxT("ISO-8859-3"), wxT("ISO8859-3"), wxT("iso88593"), wxT("8859-3"), NULL }, + { wxT("ISO-8859-4"), wxT("ISO8859-4"), wxT("iso88594"), wxT("8859-4"), NULL }, + { wxT("ISO-8859-5"), wxT("ISO8859-5"), wxT("iso88595"), wxT("8859-5"), NULL }, + { wxT("ISO-8859-6"), wxT("ISO8859-6"), wxT("iso88596"), wxT("8859-6"), NULL }, + { wxT("ISO-8859-7"), wxT("ISO8859-7"), wxT("iso88597"), wxT("8859-7"), NULL }, + { wxT("ISO-8859-8"), wxT("ISO8859-8"), wxT("iso88598"), wxT("8859-8"), NULL }, + { wxT("ISO-8859-9"), wxT("ISO8859-9"), wxT("iso88599"), wxT("8859-9"), NULL }, + { wxT("ISO-8859-10"), wxT("ISO8859-10"), wxT("iso885910"), wxT("8859-10"), NULL }, + { wxT("ISO-8859-11"), wxT("ISO8859-11"), wxT("iso885911"), wxT("8859-11"), NULL }, + { wxT("ISO-8859-12"), wxT("ISO8859-12"), wxT("iso885912"), wxT("8859-12"), NULL }, + { wxT("ISO-8859-13"), wxT("ISO8859-13"), wxT("iso885913"), wxT("8859-13"), NULL }, + { wxT("ISO-8859-14"), wxT("ISO8859-14"), wxT("iso885914"), wxT("8859-14"), NULL }, + { wxT("ISO-8859-15"), wxT("ISO8859-15"), wxT("iso885915"), wxT("8859-15"), NULL }, // although koi8-ru is not strictly speaking the same as koi8-r, // they are similar enough to make mapping it to koi8 better than @@ -178,29 +266,38 @@ static const wxChar* gs_encodingNames[WXSIZEOF(gs_encodingDescs)][10] = { wxT( "KOI8-R" ), wxT( "KOI8-RU" ), NULL }, { wxT( "KOI8-U" ), NULL }, - { wxT( "WINDOWS-874" ), wxT( "CP-874" ), NULL }, - { wxT( "WINDOWS-932" ), wxT( "CP-932" ), NULL }, - { wxT( "WINDOWS-936" ), wxT( "CP-936" ), NULL }, - { wxT( "WINDOWS-949" ), wxT( "CP-949" ), wxT( "EUC-KR" ), wxT( "eucKR" ), wxT( "euc_kr" ), NULL }, - { wxT( "WINDOWS-950" ), wxT( "CP-950" ), NULL }, - { wxT( "WINDOWS-1250" ),wxT( "CP-1250" ), NULL }, - { wxT( "WINDOWS-1251" ),wxT( "CP-1251" ), NULL }, - { wxT( "WINDOWS-1252" ),wxT( "CP-1252" ), wxT("IBM-1252"), NULL }, - { wxT( "WINDOWS-1253" ),wxT( "CP-1253" ), NULL }, - { wxT( "WINDOWS-1254" ),wxT( "CP-1254" ), NULL }, - { wxT( "WINDOWS-1255" ),wxT( "CP-1255" ), NULL }, - { wxT( "WINDOWS-1256" ),wxT( "CP-1256" ), NULL }, - { wxT( "WINDOWS-1257" ),wxT( "CP-1257" ), NULL }, - { wxT( "WINDOWS-437" ), wxT( "CP-437" ), NULL }, - - { wxT( "UTF-7" ), NULL }, - { wxT( "UTF-8" ), NULL }, - { wxT( "UTF-16" ), wxT("UCS-2"), wxT("UCS2"), NULL }, - { wxT( "UTF-16BE" ), wxT("UCS-2BE"), NULL }, - { wxT( "UTF-16LE" ), wxT("UCS-2LE"), NULL }, - { wxT( "UTF-32" ), wxT( "UCS-4" ), wxT("UCS4"), NULL }, - { wxT( "UTF-32BE" ), wxT( "UCS-4BE" ), NULL }, - { wxT( "UTF-32LE" ), wxT( "UCS-4LE" ), NULL }, + { wxT( "WINDOWS-866" ), wxT( "CP866" ), NULL }, + + { wxT( "WINDOWS-874" ), wxT( "CP874" ), wxT( "MS874" ), wxT( "IBM-874" ), NULL }, + { wxT( "WINDOWS-932" ), wxT( "CP932" ), wxT( "MS932" ), wxT( "IBM-932" ), wxT( "SJIS" ), wxT( "SHIFT-JIS" ), wxT( "SHIFT_JIS" ), NULL }, + { wxT( "WINDOWS-936" ), wxT( "CP936" ), wxT( "MS936" ), wxT( "IBM-936" ), wxT( "GB2312" ), wxT( "gbk" ),wxT( "GBK" ), NULL }, + { wxT( "WINDOWS-949" ), wxT( "CP949" ), wxT( "MS949" ), wxT( "IBM-949" ), wxT( "EUC-KR" ), wxT( "eucKR" ), wxT( "euc_kr" ), NULL }, + { wxT( "WINDOWS-950" ), wxT( "CP950" ), wxT( "MS950" ), wxT( "IBM-950" ), wxT( "BIG5" ), wxT( "BIG-5" ), wxT( "BIG-FIVE" ), NULL }, + { wxT( "WINDOWS-1250" ),wxT( "CP1250" ),wxT( "MS1250" ),wxT( "IBM-1250" ),NULL }, + { wxT( "WINDOWS-1251" ),wxT( "CP1251" ),wxT( "MS1251" ),wxT( "IBM-1251" ),NULL }, + { wxT( "WINDOWS-1252" ),wxT( "CP1252" ),wxT( "MS1252" ),wxT( "IBM-1252" ),NULL }, + { wxT( "WINDOWS-1253" ),wxT( "CP1253" ),wxT( "MS1253" ),wxT( "IBM-1253" ),NULL }, + { wxT( "WINDOWS-1254" ),wxT( "CP1254" ),wxT( "MS1254" ),wxT( "IBM-1254" ),NULL }, + { wxT( "WINDOWS-1255" ),wxT( "CP1255" ),wxT( "MS1255" ),wxT( "IBM-1255" ),NULL }, + { wxT( "WINDOWS-1256" ),wxT( "CP1256" ),wxT( "MS1256" ),wxT( "IBM-1256" ),NULL }, + { wxT( "WINDOWS-1257" ),wxT( "CP1257" ),wxT( "MS1257" ),wxT( "IBM-1257" ),NULL }, + { wxT( "WINDOWS-1258" ),wxT( "CP1258" ),wxT( "MS1258" ),wxT( "IBM-1258" ),NULL }, + { wxT( "WINDOWS-1361" ),wxT( "CP1361" ),wxT( "MS1361" ),wxT( "IBM-1361" ), wxT( "JOHAB" ), NULL }, + { wxT( "WINDOWS-437" ), wxT( "CP437" ), wxT( "MS437" ), wxT( "IBM-437" ), NULL }, + + { wxT( "UTF-7" ), wxT("UTF7"), NULL }, + { wxT( "UTF-8" ), wxT("UTF8"), NULL }, +#ifdef WORDS_BIGENDIAN + { wxT( "UTF-16BE" ), wxT("UTF16BE"), wxT("UCS-2BE"), wxT("UCS2BE"), wxT("UTF-16"), wxT("UTF16"), wxT("UCS-2"), wxT("UCS2"), NULL }, + { wxT( "UTF-16LE" ), wxT("UTF16LE"), wxT("UCS-2LE"), wxT("UCS2LE"), NULL }, + { wxT( "UTF-32BE" ), wxT("UTF32BE"), wxT("UCS-4BE" ), wxT("UTF-32"), wxT("UTF32"), wxT("UCS-4"), wxT("UCS4"), NULL }, + { wxT( "UTF-32LE" ), wxT("UTF32LE"), wxT("UCS-4LE"), wxT("UCS4LE"), NULL }, +#else // WORDS_BIGENDIAN + { wxT("UTF-16BE"), wxT("UTF16BE"), wxT("UCS-2BE"), wxT("UCS2BE"), NULL }, + { wxT("UTF-16LE"), wxT("UTF16LE"), wxT("UCS-2LE"), wxT("UTF-16"), wxT("UTF16"), wxT("UCS-2"), wxT("UCS2"), NULL }, + { wxT("UTF-32BE"), wxT("UTF32BE"), wxT("UCS-4BE"), wxT("UCS4BE"), NULL }, + { wxT("UTF-32LE"), wxT("UTF32LE"), wxT("UCS-4LE"), wxT("UCS4LE"), wxT("UTF-32"), wxT("UTF32"), wxT("UCS-4"), wxT("UCS4"), NULL }, +#endif // WORDS_BIGENDIAN { wxT( "EUC-JP" ), wxT( "eucJP" ), wxT( "euc_jp" ), wxT( "IBM-eucJP" ), NULL }, @@ -208,9 +305,49 @@ static const wxChar* gs_encodingNames[WXSIZEOF(gs_encodingDescs)][10] = { wxT( "US-ASCII" ), wxT( "ASCII" ), wxT("C"), wxT("POSIX"), wxT("ANSI_X3.4-1968"), wxT("646"), wxT("roman8"), wxT( "" ), NULL }, - { wxT( "BIG5" ), NULL }, - { wxT( "SJIS" ), wxT( "SHIFT-JIS" ), wxT( "SHIFT_JIS" ), NULL }, - { wxT( "GB2312" ), NULL }, + { wxT( "ISO-2022-JP" ), NULL }, + + + { wxT( "MacRoman" ), NULL }, + { wxT( "MacJapanese" ), NULL }, + { wxT( "MacChineseTrad" ), NULL }, + { wxT( "MacKorean" ), NULL }, + { wxT( "MacArabic" ), NULL }, + { wxT( "MacHebrew" ), NULL }, + { wxT( "MacGreek" ), NULL }, + { wxT( "MacCyrillic" ), NULL }, + { wxT( "MacDevanagari" ), NULL }, + { wxT( "MacGurmukhi" ), NULL }, + { wxT( "MacGujarati" ), NULL }, + { wxT( "MacOriya" ), NULL }, + { wxT( "MacBengali" ), NULL }, + { wxT( "MacTamil" ), NULL }, + { wxT( "MacTelugu" ), NULL }, + { wxT( "MacKannada" ), NULL }, + { wxT( "MacMalayalam" ), NULL }, + { wxT( "MacSinhalese" ), NULL }, + { wxT( "MacBurmese" ), NULL }, + { wxT( "MacKhmer" ), NULL }, + { wxT( "MacThai" ), NULL }, + { wxT( "MacLaotian" ), NULL }, + { wxT( "MacGeorgian" ), NULL }, + { wxT( "MacArmenian" ), NULL }, + { wxT( "MacChineseSimp" ), NULL }, + { wxT( "MacTibetan" ), NULL }, + { wxT( "MacMongolian" ), NULL }, + { wxT( "MacEthiopic" ), NULL }, + { wxT( "MacCentralEurRoman" ), NULL }, + { wxT( "MacVietnamese" ), NULL }, + { wxT( "MacExtArabic" ), NULL }, + { wxT( "MacSymbol" ), NULL }, + { wxT( "MacDingbats" ), NULL }, + { wxT( "MacTurkish" ), NULL }, + { wxT( "MacCroatian" ), NULL }, + { wxT( "MacIcelandic" ), NULL }, + { wxT( "MacRomanian" ), NULL }, + { wxT( "MacCeltic" ), NULL }, + { wxT( "MacGaelic" ), NULL }, + { wxT( "MacKeyboardGlyphs" ), NULL } }; wxCOMPILE_TIME_ASSERT( WXSIZEOF(gs_encodingDescs) == WXSIZEOF(gs_encodings), EncodingsArraysNotInSync ); @@ -225,8 +362,25 @@ class wxFontMapperModule : public wxModule { public: wxFontMapperModule() : wxModule() { } - virtual bool OnInit() { return true; } - virtual void OnExit() { delete (wxFontMapperBase*)wxFontMapperBase::Set(NULL); } + + virtual bool OnInit() + { + // a dummy wxFontMapperBase object could have been created during the + // program startup before wxApp was created, we have to delete it to + // allow creating the real font mapper next time it is needed now that + // we can create it (when the modules are initialized, wxApp object + // already exists) + wxFontMapperBase *fm = wxFontMapperBase::Get(); + if ( fm && fm->IsDummy() ) + wxFontMapperBase::Reset(); + + return true; + } + + virtual void OnExit() + { + wxFontMapperBase::Reset(); + } DECLARE_DYNAMIC_CLASS(wxFontMapperModule) }; @@ -247,22 +401,18 @@ wxFontMapper *wxFontMapperBase::sm_instance = NULL; wxFontMapperBase::wxFontMapperBase() { #if wxUSE_CONFIG && wxUSE_FILECONFIG - m_config = NULL; - m_configIsDummy = false; + m_configDummy = NULL; #endif // wxUSE_CONFIG } wxFontMapperBase::~wxFontMapperBase() { #if wxUSE_CONFIG && wxUSE_FILECONFIG - if ( m_configIsDummy ) - delete m_config; + if ( m_configDummy ) + delete m_configDummy; #endif // wxUSE_CONFIG } -bool wxFontMapperBase::IsWxFontMapper() -{ return false; } - /* static */ wxFontMapperBase *wxFontMapperBase::Get() { @@ -274,7 +424,7 @@ wxFontMapperBase *wxFontMapperBase::Get() sm_instance = traits->CreateFontMapper(); wxASSERT_MSG( sm_instance, - _T("wxAppTraits::CreateFontMapper() failed") ); + wxT("wxAppTraits::CreateFontMapper() failed") ); } if ( !sm_instance ) @@ -296,21 +446,39 @@ wxFontMapper *wxFontMapperBase::Set(wxFontMapper *mapper) return old; } +/* static */ +void wxFontMapperBase::Reset() +{ + if ( sm_instance ) + { + // we need a cast as wxFontMapper is not fully declared here and so the + // compiler can't know that it derives from wxFontMapperBase (but + // run-time behaviour will be correct because the dtor is virtual) + delete (wxFontMapperBase *)sm_instance; + sm_instance = NULL; + } +} + #if wxUSE_CONFIG && wxUSE_FILECONFIG // ---------------------------------------------------------------------------- // config usage customisation // ---------------------------------------------------------------------------- + +static wxString gs_defaultConfigPath(FONTMAPPER_ROOT_PATH); + /* static */ -const wxChar *wxFontMapperBase::GetDefaultConfigPath() +const wxString& wxFontMapperBase::GetDefaultConfigPath() { - return FONTMAPPER_ROOT_PATH; + // NB: we return const wxString& and not wxString for compatibility + // with 2.8 that returned const wxChar* + return gs_defaultConfigPath; } void wxFontMapperBase::SetConfigPath(const wxString& prefix) { - wxCHECK_RET( !prefix.IsEmpty() && prefix[0] == wxCONFIG_PATH_SEPARATOR, + wxCHECK_RET( !prefix.empty() && prefix[0] == wxCONFIG_PATH_SEPARATOR, wxT("an absolute path should be given to wxFontMapper::SetConfigPath()") ); m_configRootPath = prefix; @@ -322,42 +490,23 @@ void wxFontMapperBase::SetConfigPath(const wxString& prefix) wxConfigBase *wxFontMapperBase::GetConfig() { - if ( !m_config ) - { - // try the default - m_config = wxConfig::Get(false /*don't create on demand*/ ); - - if ( !m_config ) - { - // we still want to have a config object because otherwise we would - // keep asking the user the same questions in the interactive mode, - // so create a dummy config which won't write to any files/registry - // but will allow us to remember the results of the questions at - // least during this run - m_config = new wxMemoryConfig; - m_configIsDummy = true; - // VS: we can't call wxConfig::Set(m_config) here because that would - // disable automatic wxConfig instance creation if this code was - // called before wxApp::OnInit (this happens in wxGTK -- it sets - // default wxFont encoding in wxApp::Initialize()) - } - } + wxConfigBase *config = wxConfig::Get(false); - if ( m_configIsDummy && wxConfig::Get(false) != NULL ) + // If there is no global configuration, use an internal memory configuration + if ( !config ) { - // VS: in case we created dummy m_config (see above), we want to switch back - // to the real one as soon as one becomes available. - delete m_config; - m_config = wxConfig::Get(false); - m_configIsDummy = false; - // FIXME: ideally, we should add keys from dummy config to the real one now, + if ( !m_configDummy ) + m_configDummy = new wxMemoryConfig; + config = m_configDummy; + + // FIXME: ideally, we should add keys from dummy config to a real one later, // but it is a low-priority task because typical wxWin application // either doesn't use wxConfig at all or creates wxConfig object in // wxApp::OnInit(), before any real interaction with the user takes // place... } - return m_config; + return config; } const wxString& wxFontMapperBase::GetConfigPath() @@ -384,7 +533,7 @@ bool wxFontMapperBase::ChangePath(const wxString& pathNew, wxString *pathOld) *pathOld = config->GetPath(); wxString path = GetConfigPath(); - if ( path.IsEmpty() || path.Last() != wxCONFIG_PATH_SEPARATOR ) + if ( path.empty() || path.Last() != wxCONFIG_PATH_SEPARATOR ) { path += wxCONFIG_PATH_SEPARATOR; } @@ -467,7 +616,7 @@ wxFontMapperBase::NonInteractiveCharsetToEncoding(const wxString& charset) config->SetPath(FONTMAPPER_CHARSET_ALIAS_PATH); wxString alias = config->Read(charset); - if ( !alias.IsEmpty() ) + if ( !alias.empty() ) { // yes, we do - use it instead cs = alias; @@ -486,7 +635,7 @@ wxFontMapperBase::NonInteractiveCharsetToEncoding(const wxString& charset) // discard the optional quotes if ( !cs.empty() ) { - if ( cs[0u] == _T('"') && cs.Last() == _T('"') ) + if ( cs[0u] == wxT('"') && cs.Last() == wxT('"') ) { cs = wxString(cs.c_str(), cs.length() - 1); } @@ -494,7 +643,7 @@ wxFontMapperBase::NonInteractiveCharsetToEncoding(const wxString& charset) for ( size_t i = 0; i < WXSIZEOF(gs_encodingNames); ++i ) { - for ( const wxChar** encName = gs_encodingNames[i]; *encName; ++encName ) + for ( const wxChar* const* encName = gs_encodingNames[i]; *encName; ++encName ) { if ( cs.CmpNoCase(*encName) == 0 ) return gs_encodings[i]; @@ -506,18 +655,14 @@ wxFontMapperBase::NonInteractiveCharsetToEncoding(const wxString& charset) if ( cs.Left(3) == wxT("ISO") ) { // the dash is optional (or, to be exact, it is not, but - // several brokenmails "forget" it) + // several broken programs "forget" it) const wxChar *p = cs.c_str() + 3; if ( *p == wxT('-') ) p++; - // printf( "iso %s\n", (const char*) cs.ToAscii() ); - unsigned int value; if ( wxSscanf(p, wxT("8859-%u"), &value) == 1 ) { - // printf( "value %d\n", (int)value ); - // make it 0 based and check that it is strictly positive in // the process (no such thing as iso8859-0 encoding) if ( (value-- > 0) && @@ -537,8 +682,6 @@ wxFontMapperBase::NonInteractiveCharsetToEncoding(const wxString& charset) unsigned int value; if ( wxSscanf(p, wxT("8859-%u"), &value) == 1 ) { - // printf( "value %d\n", (int)value ); - // make it 0 based and check that it is strictly positive in // the process (no such thing as iso8859-0 encoding) if ( (value-- > 0) && @@ -590,6 +733,10 @@ wxFontMapperBase::NonInteractiveCharsetToEncoding(const wxString& charset) switch ( value ) { + case 866: + encoding = wxFONTENCODING_CP866; + break; + case 874: encoding = wxFONTENCODING_CP874; break; @@ -609,6 +756,14 @@ wxFontMapperBase::NonInteractiveCharsetToEncoding(const wxString& charset) case 950: encoding = wxFONTENCODING_CP950; break; + + case 1258: + encoding = wxFONTENCODING_CP1258; + break; + + case 1361: + encoding = wxFONTENCODING_CP1361; + break; } } } @@ -629,7 +784,7 @@ size_t wxFontMapperBase::GetSupportedEncodingsCount() wxFontEncoding wxFontMapperBase::GetEncoding(size_t n) { wxCHECK_MSG( n < WXSIZEOF(gs_encodings), wxFONTENCODING_SYSTEM, - _T("wxFontMapper::GetEncoding(): invalid index") ); + wxT("wxFontMapper::GetEncoding(): invalid index") ); return gs_encodings[n]; } @@ -685,17 +840,17 @@ wxString wxFontMapperBase::GetEncodingName(wxFontEncoding encoding) /* static */ const wxChar** wxFontMapperBase::GetAllEncodingNames(wxFontEncoding encoding) { - static const wxChar* dummy[] = { NULL }; + static const wxChar* const dummy[] = { NULL }; for ( size_t i = 0; i < WXSIZEOF(gs_encodingNames); i++ ) { if ( gs_encodings[i] == encoding ) { - return gs_encodingNames[i]; + return const_cast(gs_encodingNames[i]); } } - return dummy; + return const_cast(dummy); } /* static */ @@ -705,7 +860,7 @@ wxFontEncoding wxFontMapperBase::GetEncodingFromName(const wxString& name) for ( size_t i = 0; i < count; i++ ) { - for ( const wxChar** encName = gs_encodingNames[i]; *encName; ++encName ) + for ( const wxChar* const* encName = gs_encodingNames[i]; *encName; ++encName ) { if ( name.CmpNoCase(*encName) == 0 ) return gs_encodings[i]; @@ -716,4 +871,3 @@ wxFontEncoding wxFontMapperBase::GetEncodingFromName(const wxString& name) } #endif // wxUSE_FONTMAP -