3 * This file is part of LyX, the document processor.
4 * Licence details can be found in the file COPYING.
6 * \author Lars Gullik Bjønnes
8 * Full author contact details are available in file CREDITS.
10 * A collection of unicode conversion functions, using iconv.
15 #include "support/unicode.h"
16 #include "support/debug.h"
18 #include <QThreadStorage>
34 #ifdef WORDS_BIGENDIAN
35 char const * utf16_codeset = "UTF16-BE";
37 char const * utf16_codeset = "UTF16-LE";
45 #ifdef WORDS_BIGENDIAN
46 char const * ucs4_codeset = "UCS-4BE";
48 char const * ucs4_codeset = "UCS-4LE";
52 struct IconvProcessor::Handler {
53 // assumes cd is valid
54 explicit Handler(iconv_t const cd) : cd(cd) {}
56 if (iconv_close(cd) == -1)
57 LYXERR0("Error returned from iconv_close(" << errno << ')');
63 IconvProcessor::IconvProcessor(string tocode, string fromcode)
64 : tocode_(move(tocode)), fromcode_(move(fromcode))
68 bool IconvProcessor::init()
72 iconv_t cd = iconv_open(tocode_.c_str(), fromcode_.c_str());
73 if (cd != (iconv_t)(-1)) {
74 h_ = make_unique<Handler>(cd);
77 lyxerr << "Error returned from iconv_open" << endl;
80 lyxerr << "EINVAL The conversion from " << fromcode_ << " to "
81 << tocode_ << " is not supported by the implementation."
85 lyxerr << "\tSome other error: " << errno << endl;
92 int IconvProcessor::convert(char const * buf, size_t buflen,
93 char * outbuf, size_t maxoutsize)
101 char ICONV_CONST * inbuf = const_cast<char ICONV_CONST *>(buf);
102 size_t inbytesleft = buflen;
103 size_t outbytesleft = maxoutsize;
105 int res = iconv(h_->cd, &inbuf, &inbytesleft, &outbuf, &outbytesleft);
107 // flush out remaining data. This is needed because iconv sometimes
108 // holds back chars in the stream, waiting for a combination character
109 // (see e.g. http://sources.redhat.com/bugzilla/show_bug.cgi?id=1124)
110 iconv(h_->cd, nullptr, nullptr, &outbuf, &outbytesleft);
113 //lyxerr << "Inbytesleft: " << inbytesleft << endl;
114 //lyxerr << "Outbytesleft: " << outbytesleft << endl;
117 // Everything went well.
118 return maxoutsize - outbytesleft;
120 // There are some errors in the conversion
121 lyxerr << "Error returned from iconv" << endl;
124 lyxerr << "E2BIG There is not sufficient room at *outbuf." << endl;
128 lyxerr << (errno == EINVAL
129 ? "EINVAL An incomplete "
130 : "EILSEQ An invalid ")
131 << "multibyte sequence has been encountered in the input.\n"
132 << "When converting from " << fromcode_
133 << " to " << tocode_ << ".\n";
134 lyxerr << "Input:" << hex;
135 for (size_t i = 0; i < buflen; ++i) {
136 // char may be signed, avoid output of
137 // something like 0xffffffc2
139 *reinterpret_cast<unsigned char const *>(buf + i);
140 lyxerr << " 0x" << (unsigned int)b;
142 lyxerr << dec << endl;
145 lyxerr << "\tSome other error: " << errno << endl;
148 // We got an error so we close down the conversion engine
157 template<typename RetType, typename InType>
159 iconv_convert(IconvProcessor & processor, InType const * buf, size_t buflen)
162 return vector<RetType>();
164 char const * inbuf = reinterpret_cast<char const *>(buf);
165 size_t inbytesleft = buflen * sizeof(InType);
167 static QThreadStorage<std::vector<char> *> static_outbuf;
168 if (!static_outbuf.hasLocalData())
169 static_outbuf.setLocalData(new std::vector<char>(32768));
170 std::vector<char> & outbuf = *static_outbuf.localData();
171 // The number of UCS4 code points in buf is at most inbytesleft.
172 // The output encoding will use at most
173 // max_encoded_bytes(pimpl_->tocode_) per UCS4 code point.
174 size_t maxoutbufsize = max_encoded_bytes(processor.to()) * inbytesleft;
175 if (outbuf.size() < maxoutbufsize)
176 outbuf.resize(maxoutbufsize);
178 int bytes = processor.convert(inbuf, inbytesleft, &outbuf[0], outbuf.size());
181 // FIXME Maybe throw an exception and handle that in the caller?
182 return vector<RetType>();
184 RetType const * tmp = reinterpret_cast<RetType const *>(&outbuf[0]);
185 return vector<RetType>(tmp, tmp + bytes / sizeof(RetType));
191 IconvProcessor & utf8ToUcs4()
193 static QThreadStorage<IconvProcessor *> processor;
194 if (!processor.hasLocalData())
195 processor.setLocalData(new IconvProcessor(ucs4_codeset, "UTF-8"));
196 return *processor.localData();
200 vector<char_type> utf8_to_ucs4(vector<char> const & utf8str)
203 return vector<char_type>();
205 return utf8_to_ucs4(&utf8str[0], utf8str.size());
210 utf8_to_ucs4(char const * utf8str, size_t ls)
212 return iconv_convert<char_type>(utf8ToUcs4(), utf8str, ls);
217 utf16_to_ucs4(unsigned short const * s, size_t ls)
219 static QThreadStorage<IconvProcessor *> processor;
220 if (!processor.hasLocalData())
221 processor.setLocalData(new IconvProcessor(ucs4_codeset, utf16_codeset));
222 return iconv_convert<char_type>(*processor.localData(), s, ls);
226 vector<unsigned short>
227 ucs4_to_utf16(char_type const * s, size_t ls)
229 static QThreadStorage<IconvProcessor *> processor;
230 if (!processor.hasLocalData())
231 processor.setLocalData(new IconvProcessor(utf16_codeset, ucs4_codeset));
232 return iconv_convert<unsigned short>(*processor.localData(), s, ls);
236 IconvProcessor & ucs4ToUtf8()
238 static QThreadStorage<IconvProcessor *> processor;
239 if (!processor.hasLocalData())
240 processor.setLocalData(new IconvProcessor("UTF-8", ucs4_codeset));
241 return *processor.localData();
246 IconvProcessor & getProc(map<string, IconvProcessor> & processors,
247 string const & encoding, bool to)
249 string const & fromcode = to ? ucs4_codeset : encoding;
250 string const & tocode = to ? encoding : ucs4_codeset;
251 map<string, IconvProcessor>::iterator const it = processors.find(encoding);
252 if (it == processors.end()) {
253 IconvProcessor p(fromcode, tocode);
254 return processors.insert(make_pair(encoding, move(p))).first->second;
263 ucs4_to_utf8(char_type c)
265 return iconv_convert<char>(ucs4ToUtf8(), &c, 1);
270 ucs4_to_utf8(vector<char_type> const & ucs4str)
273 return vector<char>();
275 return ucs4_to_utf8(&ucs4str[0], ucs4str.size());
280 ucs4_to_utf8(char_type const * ucs4str, size_t ls)
282 return iconv_convert<char>(ucs4ToUtf8(), ucs4str, ls);
287 eightbit_to_ucs4(char const * s, size_t ls, string const & encoding)
289 static QThreadStorage<map<string, IconvProcessor> *> static_processors;
290 if (!static_processors.hasLocalData())
291 static_processors.setLocalData(new map<string, IconvProcessor>);
292 map<string, IconvProcessor> & processors = *static_processors.localData();
293 IconvProcessor & processor = getProc(processors, encoding, true);
294 return iconv_convert<char_type>(processor, s, ls);
300 map<string, IconvProcessor> & ucs4To8bitProcessors()
302 static QThreadStorage<map<string, IconvProcessor> *> processors;
303 if (!processors.hasLocalData())
304 processors.setLocalData(new map<string, IconvProcessor>);
305 return *processors.localData();
311 ucs4_to_eightbit(char_type const * ucs4str, size_t ls, string const & encoding)
313 map<string, IconvProcessor> & processors(ucs4To8bitProcessors());
314 IconvProcessor & processor = getProc(processors, encoding, false);
315 return iconv_convert<char>(processor, ucs4str, ls);
319 char ucs4_to_eightbit(char_type ucs4, string const & encoding)
321 map<string, IconvProcessor> & processors(ucs4To8bitProcessors());
322 IconvProcessor & processor = getProc(processors, encoding, false);
324 int const bytes = processor.convert((char *)(&ucs4), 4, &out, 1);
331 void ucs4_to_multibytes(char_type ucs4, vector<char> & out,
332 string const & encoding)
334 static QThreadStorage<map<string, IconvProcessor> *> static_processors;
335 if (!static_processors.hasLocalData())
336 static_processors.setLocalData(new map<string, IconvProcessor>);
337 map<string, IconvProcessor> & processors = *static_processors.localData();
338 IconvProcessor & processor = getProc(processors, encoding, false);
340 int bytes = processor.convert((char *)(&ucs4), 4, &out[0], 4);
347 int max_encoded_bytes(std::string const & encoding)
349 // FIXME: this information should be transferred to lib/encodings
350 // UTF8 uses at most 4 bytes to represent one UCS4 code point
351 // (see RFC 3629). RFC 2279 specifies 6 bytes, but that
352 // information is outdated, and RFC 2279 has been superseded by
354 // The CJK encodings use (different) multibyte representation as well.
355 // All other encodings encode one UCS4 code point in one byte
356 // (and can therefore only encode a subset of UCS4)
357 // Furthermore, all encodings that use shifting (like SJIS) do not work with
358 // iconv_codecvt_facet.
359 if (encoding == "UTF-8" ||
361 encoding == "EUC-TW")
363 else if (encoding == "EUC-JP")
365 else if (encoding == "ISO-2022-JP")
367 else if (encoding == "BIG5" ||
368 encoding == "EUC-KR" ||
369 encoding == "EUC-CN" ||
370 encoding == "SJIS" ||