You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
tdelibs/kdecore/kcharsets.cpp

665 lines
18 KiB

/* This file is part of the KDE libraries
Copyright (C) 1999 Lars Knoll (knoll@kde.org)
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Library General Public
License as published by the Free Software Foundation; either
version 2 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Library General Public License for more details.
You should have received a copy of the GNU Library General Public License
along with this library; see the file COPYING.LIB. If not, write to
the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor,
Boston, MA 02110-1301, USA.
*/
#include "kcharsets.h"
#include "kqiodevicegzip_p.h"
#include "kentities.c"
#include <kapplication.h>
#include <kglobal.h>
#include <klocale.h>
#include <kconfig.h>
#include <qfontinfo.h>
#include <qstrlist.h>
#include <qfontdatabase.h>
#include <kdebug.h>
#include <qtextcodec.h>
#include <qmap.h>
#include <qcstring.h>
#include <qdir.h>
#include <qregexp.h>
#include <assert.h>
static const char * const language_names[] = {
I18N_NOOP( "Other" ),
I18N_NOOP( "Arabic" ),
I18N_NOOP( "Baltic" ),
I18N_NOOP( "Central European" ),
I18N_NOOP( "Chinese Simplified" ),
I18N_NOOP( "Chinese Traditional" ),
I18N_NOOP( "Cyrillic" ),
I18N_NOOP( "Greek" ),
I18N_NOOP( "Hebrew" ),
I18N_NOOP( "Japanese" ),
I18N_NOOP( "Korean" ),
I18N_NOOP( "Thai" ),
I18N_NOOP( "Turkish" ),
I18N_NOOP( "Western European" ),
I18N_NOOP( "Tamil" ),
I18N_NOOP( "Unicode" ),
I18N_NOOP( "Northern Saami" ),
I18N_NOOP( "Vietnamese" ),
I18N_NOOP( "South-Eastern Europe" )
};
// This list gives the charsets that can be used to display a file given in a certain encoding.
// The list should be in order of preference
static const char* const charsets_for_encoding[] = {
"koi8-r",
"koi8-u",
"iso 8859-1",
"iso 8859-2",
"iso 8859-3",
"iso 8859-4",
"iso 8859-5",
"iso 8859-6",
"iso 8859-7",
"iso 8859-8",
"iso 8859-8-i",
"iso 8859-9",
"iso 8859-11",
"iso 8859-13",
"iso 8859-14",
"iso 8859-15",
"iso 8859-16",
"utf8",
"utf16",
"iso-10646-ucs-2",
"cp 1250",
"cp 1251",
"cp 1252",
"cp 1253",
"cp 1254",
"cp 1255",
"cp 1256",
"cp 1257",
"cp 1258",
"ibm850",
"ibm852",
"ibm866",
"tis620",
"eucjp",
"sjis",
"jis7",
"big5",
"big5-hkscs",
"gbk",
"gb18030",
"gb2312",
"euckr",
"tscii",
// "pt 154",
"winsami2",
"cp 874",
0 }; // extra 0 for end
// 0 other
// 1 Arabic
// 2 Baltic
// 3 Central European
// 4 Chinese Simplified
// 5 Chinese Traditional
// 6 Cyrillic
// 7 Greek
// 8 Hebrew
// 9 Japanese
// 10 Korean
// 11 Thai
// 12 Turkish
// 13 Western European
// 14 Tamil
// 15 Unicode
// 16 Northern Sami
// 17 Vietnamese
// 18 South-Eastern Europe
// ### FIXME KDE4: the name of the encodings should mostly be uppercase
static struct LanguageForEncoding
{
const char* index;
int data;
} const language_for_encoding[] = {
{ "iso 8859-1", 13 },
{ "iso 8859-15", 13 },
{ "iso 8859-14", 13 },
{ "cp 1252", 13 },
{ "ibm850", 13 },
{ "iso 8859-2", 3 },
{ "iso 8859-3", 3 },
{ "iso 8859-4", 2 },
{ "iso 8859-13", 2 },
{ "iso 8859-16", 18 },
{ "cp 1250", 3 },
{ "cp 1254", 12 },
{ "cp 1257", 2 },
{ "ibm852", 3 },
{ "koi8-r", 6 },
{ "iso 8859-5", 6 },
{ "cp 1251", 6 },
{ "koi8-u", 6 },
// { "pt 154", 6 },
{ "ibm866", 6 },
{ "big5", 5 },
{ "big5-hkscs", 5 },
{ "gb18030", 4 },
{ "gbk", 4 },
{ "gb2312", 4 },
{ "euckr", 10 },
{ "sjis", 9 },
{ "jis7", 9 },
{ "eucjp", 9 },
{ "iso 8859-7", 7 },
{ "cp 1253", 7 },
{ "iso 8859-6", 1 },
{ "cp 1256", 1 },
{ "iso 8859-8", 8 },
{ "iso 8859-8-i", 8 },
{ "cp 1255", 8 },
{ "iso 8859-9", 12 },
{ "tis620", 11 },
{ "iso 8859-11", 11 },
{ "cp 874", 11 },
{ "cp 1258", 17 },
{ "tscii", 14 },
{ "utf8", 15 },
{ "utf16", 15 },
{ "utf7", 15 }, // ### FIXME: UTF-7 is not in Qt
{ "ucs2", 15 },
{ "iso-10646-ucs-2", 15 },
{ "winsami2", 16},
{ 0, 0 } };
// defines some different names for codecs that are built into Qt.
static struct Builtin
{
const char* index;
const char* data;
} const builtin[] = {
{ "iso-ir-111", "koi8-r" },
{ "koi8-ru", "koi8-u" }, // ### Qt 3.3 maps it to koi8-r
{ "koi unified", "koi8-r" }, // ### FIXME: Qt 3.3 seems to map this to EUC-KR, so this mapping is too late :-(
// Using ISO-8859-1 for ASCII is an approximation at write
{ "us-ascii", "iso 8859-1" },
{ "usascii", "iso 8859-1" },
{ "ascii", "iso 8859-1" },
{ "x-utf-8", "utf-8" },
{ "x-utf-7", "utf-7" }, // ### FIXME: UTF-7 is not in Qt
{ "unicode-1-1-utf-7", "utf-7" }, // ### FIXME: UTF-7 is not in Qt
{ "utf-16", "iso-10646-ucs-2" },
{ "utf16", "iso-10646-ucs-2" },
{ "ucs2", "iso-10646-ucs-2" },
{ "iso10646-1", "iso-10646-ucs-2" },
{ "gb18030.2000-1", "gb18030" },
{ "gb18030.2000-0", "gb18030" },
{ "gbk-0", "gbk" },
{ "gb2312.1980-0", "gbk" },
{ "gb_2312-80", "gbk" },/* this one is not official, but MS is using it :/ */
{ "x-euc-kr", "euckr" },
{ "jisx0201.1976-0", "eucjp" },
{ "jisx0208.1983-0", "eucjp" },
{ "jisx0208.1990-0", "eucjp" },
{ "jisx0208.1997-0", "eucjp" },
{ "jisx0212.1990-0", "eucjp" },
{ "jisx0213.2000-1", "eucjp" },
{ "jisx0213.2000-2", "eucjp" },
{ "windows850", "ibm850" },
{ "windows866", "ibm866" },
{ "windows1251", "cp 1251" },
{ "windows1252", "cp 1252" },
{ "windows1253", "cp 1253" },
{ "windows1254", "cp 1254" },
{ "windows1255", "cp 1255" },
{ "windows1256", "cp 1256" },
{ "windows1257", "cp 1257" },
{ "windows1258", "cp 1258" },
{ "windows-850", "ibm850" },
{ "windows-866", "ibm866" },
{ "x-windows-850", "ibm850" },
{ "x-windows-866", "ibm866" },
{ "x-windows-1250", "cp 1250" },
{ "x-windows-1251", "cp 1251" },
{ "x-windows-1252", "cp 1252" },
{ "x-windows-1253", "cp 1253" },
{ "x-windows-1254", "cp 1254" },
{ "x-windows-1255", "cp 1255" },
{ "x-windows-1256", "cp 1256" },
{ "x-windows-1257", "cp 1257" },
{ "x-windows-1258", "cp 1258" },
{ "cp819", "iso 8859-1" },
{ "cp850", "ibm850" },
{ "cp866", "ibm866" },
{ "cp-819", "iso 8859-1" },
{ "cp-850", "ibm850" },
{ "cp-866", "ibm866" },
{ "cp-1250", "cp 1250" },
{ "cp-1251", "cp 1251" },
{ "cp-1252", "cp 1252" },
{ "cp-1253", "cp 1253" },
{ "cp-1254", "cp 1254" },
{ "cp-1255", "cp 1255" },
{ "cp-1256", "cp 1256" },
{ "cp-1257", "cp 1257" },
{ "cp-1258", "cp 1258" },
{ "cp-10000", "apple roman" },
{ "x-cp-850", "ibm850" },
{ "x-cp-866", "ibm866" },
{ "x-cp-1250", "cp 1250" },
{ "x-cp-1251", "cp 1251" },
{ "x-cp-1252", "cp 1252" },
{ "x-cp-1253", "cp 1253" },
{ "x-cp-1254", "cp 1254" },
{ "x-cp-1255", "cp 1255" },
{ "x-cp-1256", "cp 1256" },
{ "x-cp-1257", "cp 1257" },
{ "x-cp-1258", "cp 1258" },
{ "x-cp-10000", "apple roman" },
{ "ibm819", "iso 8859-1" },
{ "thai-tis620", "iso 8859-11" },
{ "windows-874", "cp 874" },
{ "windows874", "cp 874" },
{ "x-windows-874", "cp 874" },
{ "x-cp-874", "cp 874" },
{ "ibm 874", "cp 874" },
{ "ibm874", "cp 874" }, // Qt4 name
{ "x-ibm874", "cp 874" },
{ "ksc5601.1987-0", "euckr" },
{ "x-winsami2", "winsami2" },
{ "x-mac-roman", "apple roman" },
{ "macintosh", "apple roman" },
{ "mac", "apple roman" },
{ "csiso2022jp", "jis7" }, // See bug #77243
{ "big5-eten", "big5-hkscs" },
{ "cp950", "big5-hkscs" },
{ 0, 0 }};
// some different names for the encodings defined in the charmaps files.
// even though the charmap file names are all uppercase, the names are all lowercase here.
static struct Aliases
{
const char* index;
const char* data;
} const aliases[] = {
{ "cp852", "ibm852" },
{ "cp-852", "ibm852" },
{ "x-cp-852", "ibm852" },
{ "windows852", "ibm852" },
{ "windows-852", "ibm852" },
{ "x-windows-852", "ibm852" },
{ 0, 0 }};
// some last resort hints in case the charmap file couldn't be found. This gives at least a partial conversion
// and helps making things readable.
// the name used as input here is already converted to the more canonical name as defined in the aliases array.
static struct ConversionHints
{
const char* index;
const char* data;
} const conversion_hints[] = {
{ "cp1250", "iso-8859-2" },
{ "koi8-r", "iso-8859-5" },
{ "koi8-u", "koi8-r" },
// KDE had always "CP 1251" as best fallback to PT 154. Now that Qt does not offer this encoding anymore, it is our fallback.
{ "pt 154", "cp 1251" },
{ "paratype-154", "cp 1251" },
{ "pt-154", "cp 1251" },
{ 0, 0 }};
// search an array of items index/data, index is const char*, data is T, find first matching index
// and return data, or return 0
template< typename T, typename Data >
static Data kcharsets_array_search( const T* start, const char* entry )
{
for( const T* pos = start;
pos->index != 0;
++pos )
if( qstrcmp( pos->index, entry ) == 0 )
return pos->data;
return 0;
}
class KCharsetsPrivate
{
public:
KCharsetsPrivate(KCharsets* _kc)
: codecForNameDict(43, false) // case insensitive
{
db = 0;
kc = _kc;
}
~KCharsetsPrivate()
{
delete db;
}
QFontDatabase *db;
QAsciiDict<QTextCodec> codecForNameDict;
KCharsets* kc;
};
// --------------------------------------------------------------------------
KCharsets::KCharsets()
{
d = new KCharsetsPrivate(this);
}
KCharsets::~KCharsets()
{
delete d;
}
QChar KCharsets::fromEntity(const QString &str)
{
QChar res = QChar::null;
int pos = 0;
if(str[pos] == '&') pos++;
// Check for '&#000' or '&#x0000' sequence
if (str[pos] == '#' && str.length()-pos > 1) {
bool ok;
pos++;
if (str[pos] == 'x' || str[pos] == 'X') {
pos++;
// '&#x0000', hexadeciaml character reference
QString tmp(str.unicode()+pos, str.length()-pos);
res = tmp.toInt(&ok, 16);
} else {
// '&#0000', decimal character reference
QString tmp(str.unicode()+pos, str.length()-pos);
res = tmp.toInt(&ok, 10);
}
return res;
}
const entity *e = kde_findEntity(str.ascii(), str.length());
if(!e)
{
//kdDebug( 0 ) << "unknown entity " << str <<", len = " << str.length() << endl;
return QChar::null;
}
//kdDebug() << "got entity " << str << " = " << e->code << endl;
return QChar(e->code);
}
QChar KCharsets::fromEntity(const QString &str, int &len)
{
// entities are never longer than 8 chars... we start from
// that length and work backwards...
len = 8;
while(len > 0)
{
QString tmp = str.left(len);
QChar res = fromEntity(tmp);
if( res != QChar::null ) return res;
len--;
}
return QChar::null;
}
QString KCharsets::toEntity(const QChar &ch)
{
QString ent;
ent.sprintf("&#0x%x;", ch.unicode());
return ent;
}
QString KCharsets::resolveEntities( const QString &input )
{
QString text = input;
const QChar *p = text.unicode();
const QChar *end = p + text.length();
const QChar *ampersand = 0;
bool scanForSemicolon = false;
for ( ; p < end; ++p ) {
const QChar ch = *p;
if ( ch == '&' ) {
ampersand = p;
scanForSemicolon = true;
continue;
}
if ( ch != ';' || scanForSemicolon == false )
continue;
assert( ampersand );
scanForSemicolon = false;
const QChar *entityBegin = ampersand + 1;
const uint entityLength = p - entityBegin;
if ( entityLength == 0 )
continue;
const QChar entityValue = KCharsets::fromEntity( QConstString( entityBegin, entityLength ).string() );
if ( entityValue.isNull() )
continue;
const uint ampersandPos = ampersand - text.unicode();
text[ (int)ampersandPos ] = entityValue;
text.remove( ampersandPos + 1, entityLength + 1 );
p = text.unicode() + ampersandPos;
end = text.unicode() + text.length();
ampersand = 0;
}
return text;
}
QStringList KCharsets::availableEncodingNames()
{
QStringList available;
for ( const char* const* pos = charsets_for_encoding; *pos; ++pos ) {
//kdDebug(0) << *charsets << " available" << endl;
available.append( QString::fromLatin1( *pos ));
}
return available;
}
QString KCharsets::languageForEncoding( const QString &encoding )
{
int lang = kcharsets_array_search< LanguageForEncoding, int >
( language_for_encoding, encoding.latin1());
return i18n( language_names[lang] );
}
QString KCharsets::encodingForName( const QString &descriptiveName )
{
const int left = descriptiveName.findRev( '(' );
if (left<0) // No parenthesis, so assume it is a normal encoding name
return descriptiveName.stripWhiteSpace();
QString name(descriptiveName.mid(left+1));
const int right = name.findRev( ')' );
if (right<0)
return name;
return name.left(right).stripWhiteSpace();
}
QStringList KCharsets::descriptiveEncodingNames()
{
// As we are sorting, we can directly read the array language_for_encoding
QStringList encodings;
for ( const LanguageForEncoding* pos = language_for_encoding; pos->index; ++pos ) {
const QString name = QString::fromLatin1( pos->index );
const QString description = i18n( language_names[ pos->data ] );
encodings.append( i18n("Descriptive Encoding Name", "%1 ( %2 )"). arg ( description ). arg( name ) );
}
encodings.sort();
return encodings;
}
QTextCodec *KCharsets::codecForName(const QString &n) const
{
bool b;
return codecForName( n, b );
}
QTextCodec *KCharsets::codecForName(const QString &n, bool &ok) const
{
ok = true;
QTextCodec* codec = 0;
// dict lookup is case insensitive anyway
if((codec = d->codecForNameDict[n.isEmpty() ? "->locale<-" : n.latin1()]))
return codec; // cache hit, return
if (n.isEmpty()) {
codec = KGlobal::locale()->codecForEncoding();
d->codecForNameDict.replace("->locale<-", codec);
return codec;
}
QCString name = n.lower().latin1();
QCString key = name;
if (name.right(8) == "_charset")
name.truncate(name.length()-8);
if (name.isEmpty()) {
ok = false;
return QTextCodec::codecForName("iso8859-1");
}
codec = QTextCodec::codecForName(name);
if(codec) {
d->codecForNameDict.replace(key, codec);
return codec;
}
// these codecs are built into Qt, but the name given for the codec is different,
// so QTextCodec did not recognize it.
QCString cname = kcharsets_array_search< Builtin, const char* >( builtin, name.data());
if(!cname.isEmpty())
codec = QTextCodec::codecForName(cname);
if(codec)
{
d->codecForNameDict.replace(key, codec);
return codec;
}
QString dir;
{
KConfigGroupSaver cfgsav( KGlobal::config(), "i18n" );
dir = KGlobal::config()->readPathEntry("i18ndir", QString::fromLatin1("/usr/share/i18n/charmaps"));
}
// these are codecs not included in Qt. They can be build up if the corresponding charmap
// is available in the charmap directory.
cname = kcharsets_array_search< Aliases, const char* >( aliases, name.data());
if(cname.isEmpty())
cname = name;
cname = cname.upper();
const QString basicName = QString::fromLatin1(cname);
kdDebug() << k_funcinfo << endl << " Trying to find " << cname << " in " << dir << endl;
QString charMapFileName;
bool gzipped = false;
QDir qdir(dir);
if (!qdir.exists()) {
// The directory for the charmaps does not even exist... (That is common!)
}
else if (qdir.exists(basicName, false)) {
charMapFileName = basicName;
}
else if (qdir.exists(basicName+".gz", false)) {
charMapFileName = basicName + ".gz";
gzipped = true;
}
else {
// Check if we are asking a code page
// If yes, then check "CP99999" and "IBM99999"
// First we need to find the number of the codepage
QRegExp regexp("^(X-)?(CP|IBM)(-| )?(0-9)+");
if ( regexp.search(basicName) != -1) {
const QString num = regexp.cap(4);
if (num.isEmpty()) {
// No number, not a code page (or something went wrong)
}
else if (qdir.exists("IBM"+num)) {
charMapFileName = "IBM"+num;
}
else if (qdir.exists("IBM"+num+".gz")) {
charMapFileName = "IBM"+num+".gz";
gzipped = true;
}
else if (qdir.exists("CP"+num)) {
charMapFileName = "CP"+num;
}
else if (qdir.exists("CP"+num+".gz")) {
charMapFileName = "CP"+num+".gz";
gzipped = true;
}
}
}
if (gzipped && !charMapFileName.isEmpty()) {
KQIODeviceGZip gzip(dir + "/" + charMapFileName);
if (gzip.open(IO_ReadOnly)) {
kdDebug() << "Loading gzipped charset..." << endl;
codec = QTextCodec::loadCharmap(&gzip);
gzip.close();
}
else
kdWarning() << "Could not open gzipped charset!" << endl;
}
else if (!charMapFileName.isEmpty()) {
codec = QTextCodec::loadCharmapFile(dir + "/" + charMapFileName);
}
if(codec) {
d->codecForNameDict.replace(key, codec);
return codec;
}
// this also failed, the last resort is now to take some compatibility charmap
cname = kcharsets_array_search< ConversionHints, const char* >( conversion_hints, (const char*)name.data() );
if(!cname.isEmpty())
codec = QTextCodec::codecForName(cname);
if(codec) {
d->codecForNameDict.replace(key, codec);
return codec;
}
// could not assign a codec, let's return Latin1
ok = false;
return QTextCodec::codecForName("iso8859-1");
}