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 in
90 // the buffer is a wide char, it gets truncated (see bugs 5216,
91 // 5280, and also 5489). As a workaround, we append a null
92 // char and then remove it from output after the conversion.
93 intern_type * from_new = 0;
94 if (*(from_end - 1) >= 0x80) {
95 size_t len = from_end - from;
96 from_new = new intern_type[len + 1];
97 memcpy(from_new, from, len * sizeof(intern_type));
99 from_end = from_new + len + 1;
103 size_t inbytesleft = (from_end - from) * sizeof(intern_type);
104 size_t outbytesleft = (to_end - to) * sizeof(extern_type);
107 result const retval = do_iconv(out_cd_,
108 reinterpret_cast<char const **>(&from_next),
109 &inbytesleft, &to_next, &outbytesleft);
110 #if WORKAROUND_ICONV_BUG
111 // Remove from output the null char that we inserted at the end
112 // of the input buffer in order to circumvent an iconv bug.
116 if (retval == base::error) {
118 "Error %d returned from iconv when converting from %s to %s: %s\n",
119 errno, ucs4_codeset, encoding_.c_str(),
121 fputs("Converted input:", stderr);
122 for (intern_type const * i = from; i < from_next; ++i) {
123 unsigned int const c = *i;
124 fprintf(stderr, " 0x%04x", c);
126 unsigned int const c = *from_next;
127 fprintf(stderr, "\nStopped at: 0x%04x\n", c);
128 fputs("Unconverted input:", stderr);
129 for (intern_type const * i = from_next + 1; i < from_end; ++i) {
130 unsigned int const c = *i;
131 fprintf(stderr, " 0x%04x", c);
133 fputs("\nConverted output:", stderr);
134 for (extern_type const * i = to; i < to_next; ++i) {
135 // extern_type may be signed, avoid output of
136 // something like 0xffffffc2
137 unsigned int const c =
138 *reinterpret_cast<unsigned char const *>(i);
139 fprintf(stderr, " 0x%02x", c);
144 #if WORKAROUND_ICONV_BUG
149 virtual result do_unshift(state_type &, extern_type * to,
150 extern_type *, extern_type *& to_next) const
152 // utf8 does not use shifting
156 virtual result do_in(state_type &,
157 extern_type const * from, extern_type const * from_end,
158 extern_type const *& from_next,
159 intern_type * to, intern_type * to_end,
160 intern_type *& to_next) const
162 size_t inbytesleft = (from_end - from) * sizeof(extern_type);
163 size_t outbytesleft = (to_end - to) * sizeof(intern_type);
166 result const retval = do_iconv(in_cd_, &from_next, &inbytesleft,
167 reinterpret_cast<char **>(&to_next),
169 if (retval == base::error) {
171 "Error %d returned from iconv when converting from %s to %s: %s\n",
172 errno, encoding_.c_str(), ucs4_codeset,
174 fputs("Converted input:", stderr);
175 for (extern_type const * i = from; i < from_next; ++i) {
176 // extern_type may be signed, avoid output of
177 // something like 0xffffffc2
178 unsigned int const c =
179 *reinterpret_cast<unsigned char const *>(i);
180 fprintf(stderr, " 0x%02x", c);
182 unsigned int const c =
183 *reinterpret_cast<unsigned char const *>(from_next);
184 fprintf(stderr, "\nStopped at: 0x%02x\n", c);
185 fputs("Unconverted input:", stderr);
186 for (extern_type const * i = from_next + 1; i < from_end; ++i) {
187 unsigned int const c =
188 *reinterpret_cast<unsigned char const *>(i);
189 fprintf(stderr, " 0x%02x", c);
191 fputs("\nConverted output:", stderr);
192 for (intern_type const * i = to; i < to_next; ++i) {
193 unsigned int const c = *i;
194 fprintf(stderr, " 0x%02x", c);
201 virtual int do_encoding() const throw()
205 virtual bool do_always_noconv() const throw()
209 virtual int do_length(state_type & /*state*/, extern_type const * from,
210 extern_type const * end, size_t max) const
212 // The docs are a bit unclear about this method.
213 // It seems that we should calculate the actual length of the
214 // converted sequence, but that would not make sense, since
215 // once could just do the conversion directly.
216 // Therefore we just return the number of unconverted
217 // characters, since that is the best guess we can do.
219 intern_type * to = new intern_type[max];
220 intern_type * to_end = to + max;
221 intern_type * to_next = to;
222 extern_type const * from_next = from;
223 do_in(state, from, end, from_next, to, to_end, to_next);
227 size_t const length = end - from;
228 return min(length, max);
231 virtual int do_max_length() const throw()
233 // FIXME: this information should be transferred to lib/encodings
234 // UTF8 uses at most 4 bytes to represent one UCS4 code point
235 // (see RFC 3629). RFC 2279 specifies 6 bytes, but that
236 // information is outdated, and RFC 2279 has been superseded by
238 // The CJK encodings use (different) multibyte representation as well.
239 // All other encodings encode one UCS4 code point in one byte
240 // (and can therefore only encode a subset of UCS4)
241 // Note that BIG5 and SJIS do not work with LaTeX (see lib/encodings).
242 // Furthermore, all encodings that use shifting (like SJIS) do not work with
243 // iconv_codecvt_facet.
244 if (encoding_ == "UTF-8" ||
246 encoding_ == "EUC-TW")
248 else if (encoding_ == "EUC-JP" ||
249 encoding_ == "ISO-2022-JP")
251 else if (encoding_ == "BIG5" ||
252 encoding_ == "EUC-KR" ||
253 encoding_ == "EUC-CN" ||
254 encoding_ == "SJIS" ||
261 /// Do the actual conversion. The interface is equivalent to that of
262 /// iconv() (but const correct).
263 inline base::result do_iconv(iconv_t cd, char const ** from,
264 size_t * inbytesleft, char ** to, size_t * outbytesleft) const
266 char const * const to_start = *to;
267 size_t converted = iconv(cd, const_cast<char ICONV_CONST **>(from),
268 inbytesleft, to, outbytesleft);
269 if (converted == (size_t)(-1)) {
273 return base::partial;
285 /// The narrow encoding
295 void setEncoding(Ios & ios, string const & encoding, ios_base::openmode mode)
297 // We must imbue the stream before openening the file
299 locale locale(global, new iconv_codecvt_facet(encoding, mode));
304 const char * iconv_codecvt_facet_exception::what() const throw()
306 return "iconv problem in iconv_codecvt_facet initialization";
310 ifdocstream::ifdocstream(string const & encoding) : base()
312 setEncoding(*this, encoding, in);
316 ifdocstream::ifdocstream(const char* s, ios_base::openmode mode,
317 string const & encoding)
320 setEncoding(*this, encoding, in);
325 ofdocstream::ofdocstream(): base()
327 setEncoding(*this, "UTF-8", out);
331 ofdocstream::ofdocstream(const char* s, ios_base::openmode mode,
332 string const & encoding)
335 setEncoding(*this, encoding, out);
340 void ofdocstream::reset(string const & encoding)
342 setEncoding(*this, encoding, out);
347 SetEnc setEncoding(string const & encoding)
349 return SetEnc(encoding);
353 odocstream & operator<<(odocstream & os, SetEnc e)
355 if (has_facet<iconv_codecvt_facet>(os.rdbuf()->getloc())) {
356 // This stream must be a file stream, since we never imbue
357 // any other stream with a locale having a iconv_codecvt_facet.
358 // Flush the stream so that all pending output is written
359 // with the old encoding.
361 locale locale(os.rdbuf()->getloc(),
362 new iconv_codecvt_facet(e.encoding, ios_base::out));
363 // FIXME Does changing the codecvt facet of an open file
364 // stream always work? It does with gcc 4.1, but I have read
365 // somewhere that it does not with MSVC.
366 // What does the standard say?
373 //CHECKME: I just copied the code above, and have no idea whether it
374 //is correct... (JMarc)
375 idocstream & operator<<(idocstream & is, SetEnc e)
377 if (has_facet<iconv_codecvt_facet>(is.rdbuf()->getloc())) {
378 // This stream must be a file stream, since we never imbue
379 // any other stream with a locale having a iconv_codecvt_facet.
380 // Flush the stream so that all pending output is written
381 // with the old encoding.
383 locale locale(is.rdbuf()->getloc(),
384 new iconv_codecvt_facet(e.encoding, ios_base::in));
385 // FIXME Does changing the codecvt facet of an open file
386 // stream always work? It does with gcc 4.1, but I have read
387 // somewhere that it does not with MSVC.
388 // What does the standard say?
395 #if ! defined(USE_WCHAR_T)
396 odocstream & operator<<(odocstream & os, char c)
405 #if ! defined(USE_WCHAR_T) && defined(__GNUC__)
406 // We get undefined references to these virtual methods. This looks like
407 // a bug in gcc. The implementation here does not do anything useful, since
408 // it is overriden in iconv_codecvt_facet.
411 template<> codecvt<lyx::char_type, char, mbstate_t>::result
412 codecvt<lyx::char_type, char, mbstate_t>::do_out(
413 mbstate_t &, const lyx::char_type *, const lyx::char_type *,
414 const lyx::char_type *&, char *, char *, char *&) const
420 template<> codecvt<lyx::char_type, char, mbstate_t>::result
421 codecvt<lyx::char_type, char, mbstate_t>::do_unshift(
422 mbstate_t &, char *, char *, char *&) const
428 template<> codecvt<lyx::char_type, char, mbstate_t>::result
429 codecvt<lyx::char_type, char, mbstate_t>::do_in(
430 mbstate_t &, const char *, const char *, const char *&,
431 lyx::char_type *, lyx::char_type *, lyx::char_type *&) const
438 int codecvt<lyx::char_type, char, mbstate_t>::do_encoding() const throw()
445 bool codecvt<lyx::char_type, char, mbstate_t>::do_always_noconv() const throw()
450 #if __GNUC__ == 3 && __GNUC_MINOR__ < 4
453 int codecvt<lyx::char_type, char, mbstate_t>::do_length(
454 mbstate_t const &, const char *, const char *, size_t) const
462 int codecvt<lyx::char_type, char, mbstate_t>::do_length(
463 mbstate_t &, const char *, const char *, size_t) const
471 int codecvt<lyx::char_type, char, mbstate_t>::do_max_length() const throw()