3 * This file is part of LyX, the document processor.
4 * Licence details can be found in the file COPYING.
8 * Full author contact details are available in file CREDITS.
13 #include "support/docstream.h"
14 #include "support/unicode.h"
24 using lyx::ucs4_codeset;
28 // We use C IO throughout this file, because the facets might be used with
29 // lyxerr in the future.
32 /// codecvt facet for conversion of UCS4 (internal representation) to UTF8
33 /// (external representation) or vice versa
34 class iconv_codecvt_facet : public codecvt<lyx::char_type, char, mbstate_t>
36 typedef codecvt<lyx::char_type, char, mbstate_t> base;
38 /// Constructor. You have to specify with \p inout whether you want
39 /// to use this facet only for input, only for output or for both.
40 explicit iconv_codecvt_facet(string const & encoding = "UTF-8",
41 ios_base::openmode inout = ios_base::in | ios_base::out,
43 : base(refs), encoding_(encoding)
45 if (inout & ios_base::in) {
46 in_cd_ = iconv_open(ucs4_codeset, encoding.c_str());
47 if (in_cd_ == (iconv_t)(-1)) {
48 fprintf(stderr, "Error %d returned from iconv_open(in_cd_): %s\n",
49 errno, strerror(errno));
51 throw lyx::iconv_codecvt_facet_exception();
54 in_cd_ = (iconv_t)(-1);
55 if (inout & ios_base::out) {
56 out_cd_ = iconv_open(encoding.c_str(), ucs4_codeset);
57 if (out_cd_ == (iconv_t)(-1)) {
58 fprintf(stderr, "Error %d returned from iconv_open(out_cd_): %s\n",
59 errno, strerror(errno));
61 throw lyx::iconv_codecvt_facet_exception();
64 out_cd_ = (iconv_t)(-1);
67 virtual ~iconv_codecvt_facet()
69 if (in_cd_ != (iconv_t)(-1))
70 if (iconv_close(in_cd_) == -1) {
71 fprintf(stderr, "Error %d returned from iconv_close(in_cd_): %s\n",
72 errno, strerror(errno));
75 if (out_cd_ != (iconv_t)(-1))
76 if (iconv_close(out_cd_) == -1) {
77 fprintf(stderr, "Error %d returned from iconv_close(out_cd_): %s\n",
78 errno, strerror(errno));
82 virtual result do_out(state_type &, intern_type const * from,
83 intern_type const * from_end, intern_type const *& from_next,
84 extern_type * to, extern_type * to_end,
85 extern_type *& to_next) const
87 #define WORKAROUND_ICONV_BUG 1
88 #if WORKAROUND_ICONV_BUG
89 // Due to a bug in some iconv versions, when the last char
90 // in the buffer is a wide char and the output encoding is
91 // ISO-2022-JP and we are going to switch to another encoding,
92 // the appropriate escape sequence for changing the character
93 // set is not output (see bugs 5216, 5280, and also 5489).
94 // As a workaround, we append a nul char in order to force
95 // a switch to ASCII, and then remove it from output after
97 intern_type * from_new = 0;
98 intern_type const * from_old = from;
100 if (*(from_end - 1) >= 0x80 && encoding_ == "ISO-2022-JP") {
101 size_t len = from_end - from;
102 from_new = new intern_type[len + 1];
103 memcpy(from_new, from, len * sizeof(intern_type));
105 from_end = from_new + len + 1;
110 size_t inbytesleft = (from_end - from) * sizeof(intern_type);
111 size_t outbytesleft = (to_end - to) * sizeof(extern_type);
112 #if WORKAROUND_ICONV_BUG
113 outbytesleft += extra * sizeof(extern_type);
117 result const retval = do_iconv(out_cd_,
118 reinterpret_cast<char const **>(&from_next),
119 &inbytesleft, &to_next, &outbytesleft);
120 #if WORKAROUND_ICONV_BUG
121 // Remove from output the nul char that we inserted at the end
122 // of the input buffer in order to circumvent an iconv bug.
126 from_next = from_old + (from_next - from);
131 if (retval == base::error) {
133 "Error %d returned from iconv when converting from %s to %s: %s\n",
134 errno, ucs4_codeset, encoding_.c_str(),
136 fputs("Converted input:", stderr);
137 for (intern_type const * i = from; i < from_next; ++i) {
138 unsigned int const c = *i;
139 fprintf(stderr, " 0x%04x", c);
141 unsigned int const c = *from_next;
142 fprintf(stderr, "\nStopped at: 0x%04x\n", c);
143 fputs("Unconverted input:", stderr);
144 for (intern_type const * i = from_next + 1; i < from_end; ++i) {
145 unsigned int const c = *i;
146 fprintf(stderr, " 0x%04x", c);
148 fputs("\nConverted output:", stderr);
149 for (extern_type const * i = to; i < to_next; ++i) {
150 // extern_type may be signed, avoid output of
151 // something like 0xffffffc2
152 unsigned int const c =
153 *reinterpret_cast<unsigned char const *>(i);
154 fprintf(stderr, " 0x%02x", c);
161 virtual result do_unshift(state_type &, extern_type * to,
162 extern_type *, extern_type *& to_next) const
164 // utf8 does not use shifting
168 virtual result do_in(state_type &,
169 extern_type const * from, extern_type const * from_end,
170 extern_type const *& from_next,
171 intern_type * to, intern_type * to_end,
172 intern_type *& to_next) const
174 size_t inbytesleft = (from_end - from) * sizeof(extern_type);
175 size_t outbytesleft = (to_end - to) * sizeof(intern_type);
178 result const retval = do_iconv(in_cd_, &from_next, &inbytesleft,
179 reinterpret_cast<char **>(&to_next),
181 if (retval == base::error) {
183 "Error %d returned from iconv when converting from %s to %s: %s\n",
184 errno, encoding_.c_str(), ucs4_codeset,
186 fputs("Converted input:", stderr);
187 for (extern_type const * i = from; i < from_next; ++i) {
188 // extern_type may be signed, avoid output of
189 // something like 0xffffffc2
190 unsigned int const c =
191 *reinterpret_cast<unsigned char const *>(i);
192 fprintf(stderr, " 0x%02x", c);
194 unsigned int const c =
195 *reinterpret_cast<unsigned char const *>(from_next);
196 fprintf(stderr, "\nStopped at: 0x%02x\n", c);
197 fputs("Unconverted input:", stderr);
198 for (extern_type const * i = from_next + 1; i < from_end; ++i) {
199 unsigned int const c =
200 *reinterpret_cast<unsigned char const *>(i);
201 fprintf(stderr, " 0x%02x", c);
203 fputs("\nConverted output:", stderr);
204 for (intern_type const * i = to; i < to_next; ++i) {
205 unsigned int const c = *i;
206 fprintf(stderr, " 0x%02x", c);
213 virtual int do_encoding() const throw()
217 virtual bool do_always_noconv() const throw()
221 virtual int do_length(state_type & /*state*/, extern_type const * from,
222 extern_type const * end, size_t max) const
224 // The docs are a bit unclear about this method.
225 // It seems that we should calculate the actual length of the
226 // converted sequence, but that would not make sense, since
227 // once could just do the conversion directly.
228 // Therefore we just return the number of unconverted
229 // characters, since that is the best guess we can do.
231 intern_type * to = new intern_type[max];
232 intern_type * to_end = to + max;
233 intern_type * to_next = to;
234 extern_type const * from_next = from;
235 do_in(state, from, end, from_next, to, to_end, to_next);
239 size_t const length = end - from;
240 return min(length, max);
243 virtual int do_max_length() const throw()
245 // FIXME: this information should be transferred to lib/encodings
246 // UTF8 uses at most 4 bytes to represent one UCS4 code point
247 // (see RFC 3629). RFC 2279 specifies 6 bytes, but that
248 // information is outdated, and RFC 2279 has been superseded by
250 // The CJK encodings use (different) multibyte representation as well.
251 // All other encodings encode one UCS4 code point in one byte
252 // (and can therefore only encode a subset of UCS4)
253 // Note that BIG5 and SJIS do not work with LaTeX (see lib/encodings).
254 // Furthermore, all encodings that use shifting (like SJIS) do not work with
255 // iconv_codecvt_facet.
256 if (encoding_ == "UTF-8" ||
258 encoding_ == "EUC-TW")
260 else if (encoding_ == "EUC-JP" ||
261 encoding_ == "ISO-2022-JP")
263 else if (encoding_ == "BIG5" ||
264 encoding_ == "EUC-KR" ||
265 encoding_ == "EUC-CN" ||
266 encoding_ == "SJIS" ||
273 /// Do the actual conversion. The interface is equivalent to that of
274 /// iconv() (but const correct).
275 inline base::result do_iconv(iconv_t cd, char const ** from,
276 size_t * inbytesleft, char ** to, size_t * outbytesleft) const
278 char const * const to_start = *to;
279 size_t converted = iconv(cd, const_cast<char ICONV_CONST **>(from),
280 inbytesleft, to, outbytesleft);
281 if (converted == (size_t)(-1)) {
285 return base::partial;
297 /// The narrow encoding
307 void setEncoding(Ios & ios, string const & encoding, ios_base::openmode mode)
309 // We must imbue the stream before openening the file
311 locale locale(global, new iconv_codecvt_facet(encoding, mode));
316 const char * iconv_codecvt_facet_exception::what() const throw()
318 return "iconv problem in iconv_codecvt_facet initialization";
322 ifdocstream::ifdocstream(string const & encoding) : base()
324 setEncoding(*this, encoding, in);
328 ifdocstream::ifdocstream(const char* s, ios_base::openmode mode,
329 string const & encoding)
332 setEncoding(*this, encoding, in);
337 ofdocstream::ofdocstream(): base()
339 setEncoding(*this, "UTF-8", out);
343 ofdocstream::ofdocstream(const char* s, ios_base::openmode mode,
344 string const & encoding)
347 setEncoding(*this, encoding, out);
352 void ofdocstream::reset(string const & encoding)
354 setEncoding(*this, encoding, out);
359 SetEnc setEncoding(string const & encoding)
361 return SetEnc(encoding);
365 odocstream & operator<<(odocstream & os, SetEnc e)
367 if (has_facet<iconv_codecvt_facet>(os.rdbuf()->getloc())) {
368 // This stream must be a file stream, since we never imbue
369 // any other stream with a locale having a iconv_codecvt_facet.
370 // Flush the stream so that all pending output is written
371 // with the old encoding.
373 locale locale(os.rdbuf()->getloc(),
374 new iconv_codecvt_facet(e.encoding, ios_base::out));
375 // FIXME Does changing the codecvt facet of an open file
376 // stream always work? It does with gcc 4.1, but I have read
377 // somewhere that it does not with MSVC.
378 // What does the standard say?
385 //CHECKME: I just copied the code above, and have no idea whether it
386 //is correct... (JMarc)
387 idocstream & operator<<(idocstream & is, SetEnc e)
389 if (has_facet<iconv_codecvt_facet>(is.rdbuf()->getloc())) {
390 // This stream must be a file stream, since we never imbue
391 // any other stream with a locale having a iconv_codecvt_facet.
392 // Flush the stream so that all pending output is written
393 // with the old encoding.
395 locale locale(is.rdbuf()->getloc(),
396 new iconv_codecvt_facet(e.encoding, ios_base::in));
397 // FIXME Does changing the codecvt facet of an open file
398 // stream always work? It does with gcc 4.1, but I have read
399 // somewhere that it does not with MSVC.
400 // What does the standard say?
407 #if ! defined(USE_WCHAR_T)
408 odocstream & operator<<(odocstream & os, char c)
417 #if ! defined(USE_WCHAR_T) && defined(__GNUC__)
418 // We get undefined references to these virtual methods. This looks like
419 // a bug in gcc. The implementation here does not do anything useful, since
420 // it is overriden in iconv_codecvt_facet.
423 template<> codecvt<lyx::char_type, char, mbstate_t>::result
424 codecvt<lyx::char_type, char, mbstate_t>::do_out(
425 mbstate_t &, const lyx::char_type *, const lyx::char_type *,
426 const lyx::char_type *&, char *, char *, char *&) const
432 template<> codecvt<lyx::char_type, char, mbstate_t>::result
433 codecvt<lyx::char_type, char, mbstate_t>::do_unshift(
434 mbstate_t &, char *, char *, char *&) const
440 template<> codecvt<lyx::char_type, char, mbstate_t>::result
441 codecvt<lyx::char_type, char, mbstate_t>::do_in(
442 mbstate_t &, const char *, const char *, const char *&,
443 lyx::char_type *, lyx::char_type *, lyx::char_type *&) const
450 int codecvt<lyx::char_type, char, mbstate_t>::do_encoding() const throw()
457 bool codecvt<lyx::char_type, char, mbstate_t>::do_always_noconv() const throw()
462 #if __GNUC__ == 3 && __GNUC_MINOR__ < 4
465 int codecvt<lyx::char_type, char, mbstate_t>::do_length(
466 mbstate_t const &, const char *, const char *, size_t) const
474 int codecvt<lyx::char_type, char, mbstate_t>::do_length(
475 mbstate_t &, const char *, const char *, size_t) const
483 int codecvt<lyx::char_type, char, mbstate_t>::do_max_length() const throw()