Commit | Line | Data |
---|---|---|
24d56127 LC |
1 | /* Character set conversion with error handling. |
2 | Copyright (C) 2001-2008 Free Software Foundation, Inc. | |
3 | Written by Bruno Haible and Simon Josefsson. | |
4 | ||
5 | This program is free software: you can redistribute it and/or modify | |
6 | it under the terms of the GNU Lesser General Public License as published by | |
7 | the Free Software Foundation; either version 3 of the License, or | |
8 | (at your option) any later version. | |
9 | ||
10 | This program is distributed in the hope that it will be useful, | |
11 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | GNU Lesser General Public License for more details. | |
14 | ||
15 | You should have received a copy of the GNU Lesser General Public License | |
16 | along with this program. If not, see <http://www.gnu.org/licenses/>. */ | |
17 | ||
18 | #include <config.h> | |
19 | ||
20 | /* Specification. */ | |
21 | #include "striconveh.h" | |
22 | ||
23 | #include <errno.h> | |
24 | #include <stdbool.h> | |
25 | #include <stdlib.h> | |
26 | #include <string.h> | |
27 | ||
28 | #if HAVE_ICONV | |
29 | # include <iconv.h> | |
30 | # include "unistr.h" | |
31 | #endif | |
32 | ||
33 | #include "c-strcase.h" | |
34 | #include "c-strcaseeq.h" | |
35 | ||
36 | #ifndef SIZE_MAX | |
37 | # define SIZE_MAX ((size_t) -1) | |
38 | #endif | |
39 | ||
40 | ||
41 | #if HAVE_ICONV | |
42 | ||
43 | /* The caller must provide CD, CD1, CD2, not just CD, because when a conversion | |
44 | error occurs, we may have to determine the Unicode representation of the | |
45 | inconvertible character. */ | |
46 | ||
47 | /* iconv_carefully is like iconv, except that it stops as soon as it encounters | |
48 | a conversion error, and it returns in *INCREMENTED a boolean telling whether | |
49 | it has incremented the input pointers past the error location. */ | |
50 | # if !defined _LIBICONV_VERSION && !defined __GLIBC__ | |
51 | /* Irix iconv() inserts a NUL byte if it cannot convert. | |
52 | NetBSD iconv() inserts a question mark if it cannot convert. | |
53 | Only GNU libiconv and GNU libc are known to prefer to fail rather | |
54 | than doing a lossy conversion. */ | |
55 | static size_t | |
56 | iconv_carefully (iconv_t cd, | |
57 | const char **inbuf, size_t *inbytesleft, | |
58 | char **outbuf, size_t *outbytesleft, | |
59 | bool *incremented) | |
60 | { | |
61 | const char *inptr = *inbuf; | |
62 | const char *inptr_end = inptr + *inbytesleft; | |
63 | char *outptr = *outbuf; | |
64 | size_t outsize = *outbytesleft; | |
65 | const char *inptr_before; | |
66 | size_t res; | |
67 | ||
68 | do | |
69 | { | |
70 | size_t insize; | |
71 | ||
72 | inptr_before = inptr; | |
73 | res = (size_t)(-1); | |
74 | ||
75 | for (insize = 1; inptr + insize <= inptr_end; insize++) | |
76 | { | |
77 | res = iconv (cd, | |
78 | (ICONV_CONST char **) &inptr, &insize, | |
79 | &outptr, &outsize); | |
80 | if (!(res == (size_t)(-1) && errno == EINVAL)) | |
81 | break; | |
82 | /* iconv can eat up a shift sequence but give EINVAL while attempting | |
83 | to convert the first character. E.g. libiconv does this. */ | |
84 | if (inptr > inptr_before) | |
85 | { | |
86 | res = 0; | |
87 | break; | |
88 | } | |
89 | } | |
90 | ||
91 | if (res == 0) | |
92 | { | |
93 | *outbuf = outptr; | |
94 | *outbytesleft = outsize; | |
95 | } | |
96 | } | |
97 | while (res == 0 && inptr < inptr_end); | |
98 | ||
99 | *inbuf = inptr; | |
100 | *inbytesleft = inptr_end - inptr; | |
101 | if (res != (size_t)(-1) && res > 0) | |
102 | { | |
103 | /* iconv() has already incremented INPTR. We cannot go back to a | |
104 | previous INPTR, otherwise the state inside CD would become invalid, | |
105 | if FROM_CODESET is a stateful encoding. So, tell the caller that | |
106 | *INBUF has already been incremented. */ | |
107 | *incremented = (inptr > inptr_before); | |
108 | errno = EILSEQ; | |
109 | return (size_t)(-1); | |
110 | } | |
111 | else | |
112 | { | |
113 | *incremented = false; | |
114 | return res; | |
115 | } | |
116 | } | |
117 | # else | |
118 | # define iconv_carefully(cd, inbuf, inbytesleft, outbuf, outbytesleft, incremented) \ | |
119 | (*(incremented) = false, \ | |
120 | iconv (cd, (ICONV_CONST char **) (inbuf), inbytesleft, outbuf, outbytesleft)) | |
121 | # endif | |
122 | ||
123 | /* iconv_carefully_1 is like iconv_carefully, except that it stops after | |
124 | converting one character or one shift sequence. */ | |
125 | static size_t | |
126 | iconv_carefully_1 (iconv_t cd, | |
127 | const char **inbuf, size_t *inbytesleft, | |
128 | char **outbuf, size_t *outbytesleft, | |
129 | bool *incremented) | |
130 | { | |
131 | const char *inptr_before = *inbuf; | |
132 | const char *inptr = inptr_before; | |
133 | const char *inptr_end = inptr_before + *inbytesleft; | |
134 | char *outptr = *outbuf; | |
135 | size_t outsize = *outbytesleft; | |
136 | size_t res = (size_t)(-1); | |
137 | size_t insize; | |
138 | ||
139 | for (insize = 1; inptr_before + insize <= inptr_end; insize++) | |
140 | { | |
141 | inptr = inptr_before; | |
142 | res = iconv (cd, | |
143 | (ICONV_CONST char **) &inptr, &insize, | |
144 | &outptr, &outsize); | |
145 | if (!(res == (size_t)(-1) && errno == EINVAL)) | |
146 | break; | |
147 | /* iconv can eat up a shift sequence but give EINVAL while attempting | |
148 | to convert the first character. E.g. libiconv does this. */ | |
149 | if (inptr > inptr_before) | |
150 | { | |
151 | res = 0; | |
152 | break; | |
153 | } | |
154 | } | |
155 | ||
156 | *inbuf = inptr; | |
157 | *inbytesleft = inptr_end - inptr; | |
158 | # if !defined _LIBICONV_VERSION && !defined __GLIBC__ | |
159 | /* Irix iconv() inserts a NUL byte if it cannot convert. | |
160 | NetBSD iconv() inserts a question mark if it cannot convert. | |
161 | Only GNU libiconv and GNU libc are known to prefer to fail rather | |
162 | than doing a lossy conversion. */ | |
163 | if (res != (size_t)(-1) && res > 0) | |
164 | { | |
165 | /* iconv() has already incremented INPTR. We cannot go back to a | |
166 | previous INPTR, otherwise the state inside CD would become invalid, | |
167 | if FROM_CODESET is a stateful encoding. So, tell the caller that | |
168 | *INBUF has already been incremented. */ | |
169 | *incremented = (inptr > inptr_before); | |
170 | errno = EILSEQ; | |
171 | return (size_t)(-1); | |
172 | } | |
173 | # endif | |
174 | ||
175 | if (res != (size_t)(-1)) | |
176 | { | |
177 | *outbuf = outptr; | |
178 | *outbytesleft = outsize; | |
179 | } | |
180 | *incremented = false; | |
181 | return res; | |
182 | } | |
183 | ||
184 | /* utf8conv_carefully is like iconv, except that | |
185 | - it converts from UTF-8 to UTF-8, | |
186 | - it stops as soon as it encounters a conversion error, and it returns | |
187 | in *INCREMENTED a boolean telling whether it has incremented the input | |
188 | pointers past the error location, | |
189 | - if one_character_only is true, it stops after converting one | |
190 | character. */ | |
191 | static size_t | |
192 | utf8conv_carefully (bool one_character_only, | |
193 | const char **inbuf, size_t *inbytesleft, | |
194 | char **outbuf, size_t *outbytesleft, | |
195 | bool *incremented) | |
196 | { | |
197 | const char *inptr = *inbuf; | |
198 | size_t insize = *inbytesleft; | |
199 | char *outptr = *outbuf; | |
200 | size_t outsize = *outbytesleft; | |
201 | size_t res; | |
202 | ||
203 | res = 0; | |
204 | do | |
205 | { | |
206 | ucs4_t uc; | |
207 | int n; | |
208 | int m; | |
209 | ||
210 | n = u8_mbtoucr (&uc, (const uint8_t *) inptr, insize); | |
211 | if (n < 0) | |
212 | { | |
213 | errno = (n == -2 ? EINVAL : EILSEQ); | |
214 | n = u8_mbtouc (&uc, (const uint8_t *) inptr, insize); | |
215 | inptr += n; | |
216 | insize -= n; | |
217 | res = (size_t)(-1); | |
218 | *incremented = true; | |
219 | break; | |
220 | } | |
221 | if (outsize == 0) | |
222 | { | |
223 | errno = E2BIG; | |
224 | res = (size_t)(-1); | |
225 | *incremented = false; | |
226 | break; | |
227 | } | |
228 | m = u8_uctomb ((uint8_t *) outptr, uc, outsize); | |
229 | if (m == -2) | |
230 | { | |
231 | errno = E2BIG; | |
232 | res = (size_t)(-1); | |
233 | *incremented = false; | |
234 | break; | |
235 | } | |
236 | inptr += n; | |
237 | insize -= n; | |
238 | if (m == -1) | |
239 | { | |
240 | errno = EILSEQ; | |
241 | res = (size_t)(-1); | |
242 | *incremented = true; | |
243 | break; | |
244 | } | |
245 | outptr += m; | |
246 | outsize -= m; | |
247 | } | |
248 | while (!one_character_only && insize > 0); | |
249 | ||
250 | *inbuf = inptr; | |
251 | *inbytesleft = insize; | |
252 | *outbuf = outptr; | |
253 | *outbytesleft = outsize; | |
254 | return res; | |
255 | } | |
256 | ||
257 | static int | |
258 | mem_cd_iconveh_internal (const char *src, size_t srclen, | |
259 | iconv_t cd, iconv_t cd1, iconv_t cd2, | |
260 | enum iconv_ilseq_handler handler, | |
261 | size_t extra_alloc, | |
262 | size_t *offsets, | |
263 | char **resultp, size_t *lengthp) | |
264 | { | |
265 | /* When a conversion error occurs, we cannot start using CD1 and CD2 at | |
266 | this point: FROM_CODESET may be a stateful encoding like ISO-2022-KR. | |
267 | Instead, we have to start afresh from the beginning of SRC. */ | |
268 | /* Use a temporary buffer, so that for small strings, a single malloc() | |
269 | call will be sufficient. */ | |
270 | # define tmpbufsize 4096 | |
271 | /* The alignment is needed when converting e.g. to glibc's WCHAR_T or | |
272 | libiconv's UCS-4-INTERNAL encoding. */ | |
273 | union { unsigned int align; char buf[tmpbufsize]; } tmp; | |
274 | # define tmpbuf tmp.buf | |
275 | ||
276 | char *initial_result; | |
277 | char *result; | |
278 | size_t allocated; | |
279 | size_t length; | |
280 | size_t last_length = (size_t)(-1); /* only needed if offsets != NULL */ | |
281 | ||
282 | if (*resultp != NULL && *lengthp >= sizeof (tmpbuf)) | |
283 | { | |
284 | initial_result = *resultp; | |
285 | allocated = *lengthp; | |
286 | } | |
287 | else | |
288 | { | |
289 | initial_result = tmpbuf; | |
290 | allocated = sizeof (tmpbuf); | |
291 | } | |
292 | result = initial_result; | |
293 | ||
294 | /* Test whether a direct conversion is possible at all. */ | |
295 | if (cd == (iconv_t)(-1)) | |
296 | goto indirectly; | |
297 | ||
298 | if (offsets != NULL) | |
299 | { | |
300 | size_t i; | |
301 | ||
302 | for (i = 0; i < srclen; i++) | |
303 | offsets[i] = (size_t)(-1); | |
304 | ||
305 | last_length = (size_t)(-1); | |
306 | } | |
307 | length = 0; | |
308 | ||
309 | /* First, try a direct conversion, and see whether a conversion error | |
310 | occurs at all. */ | |
311 | { | |
312 | const char *inptr = src; | |
313 | size_t insize = srclen; | |
314 | ||
315 | /* Avoid glibc-2.1 bug and Solaris 2.7-2.9 bug. */ | |
316 | # if defined _LIBICONV_VERSION \ | |
317 | || !((__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) || defined __sun) | |
318 | /* Set to the initial state. */ | |
319 | iconv (cd, NULL, NULL, NULL, NULL); | |
320 | # endif | |
321 | ||
322 | while (insize > 0) | |
323 | { | |
324 | char *outptr = result + length; | |
325 | size_t outsize = allocated - extra_alloc - length; | |
326 | bool incremented; | |
327 | size_t res; | |
328 | bool grow; | |
329 | ||
330 | if (offsets != NULL) | |
331 | { | |
332 | if (length != last_length) /* ensure that offset[] be increasing */ | |
333 | { | |
334 | offsets[inptr - src] = length; | |
335 | last_length = length; | |
336 | } | |
337 | res = iconv_carefully_1 (cd, | |
338 | &inptr, &insize, | |
339 | &outptr, &outsize, | |
340 | &incremented); | |
341 | } | |
342 | else | |
343 | /* Use iconv_carefully instead of iconv here, because: | |
344 | - If TO_CODESET is UTF-8, we can do the error handling in this | |
345 | loop, no need for a second loop, | |
346 | - With iconv() implementations other than GNU libiconv and GNU | |
347 | libc, if we use iconv() in a big swoop, checking for an E2BIG | |
348 | return, we lose the number of irreversible conversions. */ | |
349 | res = iconv_carefully (cd, | |
350 | &inptr, &insize, | |
351 | &outptr, &outsize, | |
352 | &incremented); | |
353 | ||
354 | length = outptr - result; | |
355 | grow = (length + extra_alloc > allocated / 2); | |
356 | if (res == (size_t)(-1)) | |
357 | { | |
358 | if (errno == E2BIG) | |
359 | grow = true; | |
360 | else if (errno == EINVAL) | |
361 | break; | |
362 | else if (errno == EILSEQ && handler != iconveh_error) | |
363 | { | |
364 | if (cd2 == (iconv_t)(-1)) | |
365 | { | |
366 | /* TO_CODESET is UTF-8. */ | |
367 | /* Error handling can produce up to 1 byte of output. */ | |
368 | if (length + 1 + extra_alloc > allocated) | |
369 | { | |
370 | char *memory; | |
371 | ||
372 | allocated = 2 * allocated; | |
373 | if (length + 1 + extra_alloc > allocated) | |
374 | abort (); | |
375 | if (result == initial_result) | |
376 | memory = (char *) malloc (allocated); | |
377 | else | |
378 | memory = (char *) realloc (result, allocated); | |
379 | if (memory == NULL) | |
380 | { | |
381 | if (result != initial_result) | |
382 | free (result); | |
383 | errno = ENOMEM; | |
384 | return -1; | |
385 | } | |
386 | if (result == initial_result) | |
387 | memcpy (memory, initial_result, length); | |
388 | result = memory; | |
389 | grow = false; | |
390 | } | |
391 | /* The input is invalid in FROM_CODESET. Eat up one byte | |
392 | and emit a question mark. */ | |
393 | if (!incremented) | |
394 | { | |
395 | if (insize == 0) | |
396 | abort (); | |
397 | inptr++; | |
398 | insize--; | |
399 | } | |
400 | result[length] = '?'; | |
401 | length++; | |
402 | } | |
403 | else | |
404 | goto indirectly; | |
405 | } | |
406 | else | |
407 | { | |
408 | if (result != initial_result) | |
409 | { | |
410 | int saved_errno = errno; | |
411 | free (result); | |
412 | errno = saved_errno; | |
413 | } | |
414 | return -1; | |
415 | } | |
416 | } | |
417 | if (insize == 0) | |
418 | break; | |
419 | if (grow) | |
420 | { | |
421 | char *memory; | |
422 | ||
423 | allocated = 2 * allocated; | |
424 | if (result == initial_result) | |
425 | memory = (char *) malloc (allocated); | |
426 | else | |
427 | memory = (char *) realloc (result, allocated); | |
428 | if (memory == NULL) | |
429 | { | |
430 | if (result != initial_result) | |
431 | free (result); | |
432 | errno = ENOMEM; | |
433 | return -1; | |
434 | } | |
435 | if (result == initial_result) | |
436 | memcpy (memory, initial_result, length); | |
437 | result = memory; | |
438 | } | |
439 | } | |
440 | } | |
441 | ||
442 | /* Now get the conversion state back to the initial state. | |
443 | But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
444 | #if defined _LIBICONV_VERSION \ | |
445 | || !((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) || defined __sun) | |
446 | for (;;) | |
447 | { | |
448 | char *outptr = result + length; | |
449 | size_t outsize = allocated - extra_alloc - length; | |
450 | size_t res; | |
451 | ||
452 | res = iconv (cd, NULL, NULL, &outptr, &outsize); | |
453 | length = outptr - result; | |
454 | if (res == (size_t)(-1)) | |
455 | { | |
456 | if (errno == E2BIG) | |
457 | { | |
458 | char *memory; | |
459 | ||
460 | allocated = 2 * allocated; | |
461 | if (result == initial_result) | |
462 | memory = (char *) malloc (allocated); | |
463 | else | |
464 | memory = (char *) realloc (result, allocated); | |
465 | if (memory == NULL) | |
466 | { | |
467 | if (result != initial_result) | |
468 | free (result); | |
469 | errno = ENOMEM; | |
470 | return -1; | |
471 | } | |
472 | if (result == initial_result) | |
473 | memcpy (memory, initial_result, length); | |
474 | result = memory; | |
475 | } | |
476 | else | |
477 | { | |
478 | if (result != initial_result) | |
479 | { | |
480 | int saved_errno = errno; | |
481 | free (result); | |
482 | errno = saved_errno; | |
483 | } | |
484 | return -1; | |
485 | } | |
486 | } | |
487 | else | |
488 | break; | |
489 | } | |
490 | #endif | |
491 | ||
492 | /* The direct conversion succeeded. */ | |
493 | goto done; | |
494 | ||
495 | indirectly: | |
496 | /* The direct conversion failed. | |
497 | Use a conversion through UTF-8. */ | |
498 | if (offsets != NULL) | |
499 | { | |
500 | size_t i; | |
501 | ||
502 | for (i = 0; i < srclen; i++) | |
503 | offsets[i] = (size_t)(-1); | |
504 | ||
505 | last_length = (size_t)(-1); | |
506 | } | |
507 | length = 0; | |
508 | { | |
509 | const bool slowly = (offsets != NULL || handler == iconveh_error); | |
510 | # define utf8bufsize 4096 /* may also be smaller or larger than tmpbufsize */ | |
511 | char utf8buf[utf8bufsize + 1]; | |
512 | size_t utf8len = 0; | |
513 | const char *in1ptr = src; | |
514 | size_t in1size = srclen; | |
515 | bool do_final_flush1 = true; | |
516 | bool do_final_flush2 = true; | |
517 | ||
518 | /* Avoid glibc-2.1 bug and Solaris 2.7-2.9 bug. */ | |
519 | # if defined _LIBICONV_VERSION \ | |
520 | || !((__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) || defined __sun) | |
521 | /* Set to the initial state. */ | |
522 | if (cd1 != (iconv_t)(-1)) | |
523 | iconv (cd1, NULL, NULL, NULL, NULL); | |
524 | if (cd2 != (iconv_t)(-1)) | |
525 | iconv (cd2, NULL, NULL, NULL, NULL); | |
526 | # endif | |
527 | ||
528 | while (in1size > 0 || do_final_flush1 || utf8len > 0 || do_final_flush2) | |
529 | { | |
530 | char *out1ptr = utf8buf + utf8len; | |
531 | size_t out1size = utf8bufsize - utf8len; | |
532 | bool incremented1; | |
533 | size_t res1; | |
534 | int errno1; | |
535 | ||
536 | /* Conversion step 1: from FROM_CODESET to UTF-8. */ | |
537 | if (in1size > 0) | |
538 | { | |
539 | if (offsets != NULL | |
540 | && length != last_length) /* ensure that offset[] be increasing */ | |
541 | { | |
542 | offsets[in1ptr - src] = length; | |
543 | last_length = length; | |
544 | } | |
545 | if (cd1 != (iconv_t)(-1)) | |
546 | { | |
547 | if (slowly) | |
548 | res1 = iconv_carefully_1 (cd1, | |
549 | &in1ptr, &in1size, | |
550 | &out1ptr, &out1size, | |
551 | &incremented1); | |
552 | else | |
553 | res1 = iconv_carefully (cd1, | |
554 | &in1ptr, &in1size, | |
555 | &out1ptr, &out1size, | |
556 | &incremented1); | |
557 | } | |
558 | else | |
559 | { | |
560 | /* FROM_CODESET is UTF-8. */ | |
561 | res1 = utf8conv_carefully (slowly, | |
562 | &in1ptr, &in1size, | |
563 | &out1ptr, &out1size, | |
564 | &incremented1); | |
565 | } | |
566 | } | |
567 | else if (do_final_flush1) | |
568 | { | |
569 | /* Now get the conversion state of CD1 back to the initial state. | |
570 | But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
571 | # if defined _LIBICONV_VERSION \ | |
572 | || !((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) || defined __sun) | |
573 | if (cd1 != (iconv_t)(-1)) | |
574 | res1 = iconv (cd1, NULL, NULL, &out1ptr, &out1size); | |
575 | else | |
576 | # endif | |
577 | res1 = 0; | |
578 | do_final_flush1 = false; | |
579 | incremented1 = true; | |
580 | } | |
581 | else | |
582 | { | |
583 | res1 = 0; | |
584 | incremented1 = true; | |
585 | } | |
586 | if (res1 == (size_t)(-1) | |
587 | && !(errno == E2BIG || errno == EINVAL || errno == EILSEQ)) | |
588 | { | |
589 | if (result != initial_result) | |
590 | { | |
591 | int saved_errno = errno; | |
592 | free (result); | |
593 | errno = saved_errno; | |
594 | } | |
595 | return -1; | |
596 | } | |
597 | if (res1 == (size_t)(-1) | |
598 | && errno == EILSEQ && handler != iconveh_error) | |
599 | { | |
600 | /* The input is invalid in FROM_CODESET. Eat up one byte and | |
601 | emit a question mark. Room for the question mark was allocated | |
602 | at the end of utf8buf. */ | |
603 | if (!incremented1) | |
604 | { | |
605 | if (in1size == 0) | |
606 | abort (); | |
607 | in1ptr++; | |
608 | in1size--; | |
609 | } | |
610 | utf8buf[utf8len++] = '?'; | |
611 | } | |
612 | errno1 = errno; | |
613 | utf8len = out1ptr - utf8buf; | |
614 | ||
615 | if (offsets != NULL | |
616 | || in1size == 0 | |
617 | || utf8len > utf8bufsize / 2 | |
618 | || (res1 == (size_t)(-1) && errno1 == E2BIG)) | |
619 | { | |
620 | /* Conversion step 2: from UTF-8 to TO_CODESET. */ | |
621 | const char *in2ptr = utf8buf; | |
622 | size_t in2size = utf8len; | |
623 | ||
624 | while (in2size > 0 | |
625 | || (in1size == 0 && !do_final_flush1 && do_final_flush2)) | |
626 | { | |
627 | char *out2ptr = result + length; | |
628 | size_t out2size = allocated - extra_alloc - length; | |
629 | bool incremented2; | |
630 | size_t res2; | |
631 | bool grow; | |
632 | ||
633 | if (in2size > 0) | |
634 | { | |
635 | if (cd2 != (iconv_t)(-1)) | |
636 | res2 = iconv_carefully (cd2, | |
637 | &in2ptr, &in2size, | |
638 | &out2ptr, &out2size, | |
639 | &incremented2); | |
640 | else | |
641 | /* TO_CODESET is UTF-8. */ | |
642 | res2 = utf8conv_carefully (false, | |
643 | &in2ptr, &in2size, | |
644 | &out2ptr, &out2size, | |
645 | &incremented2); | |
646 | } | |
647 | else /* in1size == 0 && !do_final_flush1 | |
648 | && in2size == 0 && do_final_flush2 */ | |
649 | { | |
650 | /* Now get the conversion state of CD1 back to the initial | |
651 | state. But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
652 | # if defined _LIBICONV_VERSION \ | |
653 | || !((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) || defined __sun) | |
654 | if (cd2 != (iconv_t)(-1)) | |
655 | res2 = iconv (cd2, NULL, NULL, &out2ptr, &out2size); | |
656 | else | |
657 | # endif | |
658 | res2 = 0; | |
659 | do_final_flush2 = false; | |
660 | incremented2 = true; | |
661 | } | |
662 | ||
663 | length = out2ptr - result; | |
664 | grow = (length + extra_alloc > allocated / 2); | |
665 | if (res2 == (size_t)(-1)) | |
666 | { | |
667 | if (errno == E2BIG) | |
668 | grow = true; | |
669 | else if (errno == EINVAL) | |
670 | break; | |
671 | else if (errno == EILSEQ && handler != iconveh_error) | |
672 | { | |
673 | /* Error handling can produce up to 10 bytes of ASCII | |
674 | output. But TO_CODESET may be UCS-2, UTF-16 or | |
675 | UCS-4, so use CD2 here as well. */ | |
676 | char scratchbuf[10]; | |
677 | size_t scratchlen; | |
678 | ucs4_t uc; | |
679 | const char *inptr; | |
680 | size_t insize; | |
681 | size_t res; | |
682 | ||
683 | if (incremented2) | |
684 | { | |
685 | if (u8_prev (&uc, (const uint8_t *) in2ptr, | |
686 | (const uint8_t *) utf8buf) | |
687 | == NULL) | |
688 | abort (); | |
689 | } | |
690 | else | |
691 | { | |
692 | int n; | |
693 | if (in2size == 0) | |
694 | abort (); | |
695 | n = u8_mbtouc_unsafe (&uc, (const uint8_t *) in2ptr, | |
696 | in2size); | |
697 | in2ptr += n; | |
698 | in2size -= n; | |
699 | } | |
700 | ||
701 | if (handler == iconveh_escape_sequence) | |
702 | { | |
703 | static char hex[16] = "0123456789ABCDEF"; | |
704 | scratchlen = 0; | |
705 | scratchbuf[scratchlen++] = '\\'; | |
706 | if (uc < 0x10000) | |
707 | scratchbuf[scratchlen++] = 'u'; | |
708 | else | |
709 | { | |
710 | scratchbuf[scratchlen++] = 'U'; | |
711 | scratchbuf[scratchlen++] = hex[(uc>>28) & 15]; | |
712 | scratchbuf[scratchlen++] = hex[(uc>>24) & 15]; | |
713 | scratchbuf[scratchlen++] = hex[(uc>>20) & 15]; | |
714 | scratchbuf[scratchlen++] = hex[(uc>>16) & 15]; | |
715 | } | |
716 | scratchbuf[scratchlen++] = hex[(uc>>12) & 15]; | |
717 | scratchbuf[scratchlen++] = hex[(uc>>8) & 15]; | |
718 | scratchbuf[scratchlen++] = hex[(uc>>4) & 15]; | |
719 | scratchbuf[scratchlen++] = hex[uc & 15]; | |
720 | } | |
721 | else | |
722 | { | |
723 | scratchbuf[0] = '?'; | |
724 | scratchlen = 1; | |
725 | } | |
726 | ||
727 | inptr = scratchbuf; | |
728 | insize = scratchlen; | |
729 | if (cd2 != (iconv_t)(-1)) | |
730 | res = iconv (cd2, | |
731 | (ICONV_CONST char **) &inptr, &insize, | |
732 | &out2ptr, &out2size); | |
733 | else | |
734 | { | |
735 | /* TO_CODESET is UTF-8. */ | |
736 | if (out2size >= insize) | |
737 | { | |
738 | memcpy (out2ptr, inptr, insize); | |
739 | out2ptr += insize; | |
740 | out2size -= insize; | |
741 | inptr += insize; | |
742 | insize = 0; | |
743 | res = 0; | |
744 | } | |
745 | else | |
746 | { | |
747 | errno = E2BIG; | |
748 | res = (size_t)(-1); | |
749 | } | |
750 | } | |
751 | length = out2ptr - result; | |
752 | if (res == (size_t)(-1) && errno == E2BIG) | |
753 | { | |
754 | char *memory; | |
755 | ||
756 | allocated = 2 * allocated; | |
757 | if (length + 1 + extra_alloc > allocated) | |
758 | abort (); | |
759 | if (result == initial_result) | |
760 | memory = (char *) malloc (allocated); | |
761 | else | |
762 | memory = (char *) realloc (result, allocated); | |
763 | if (memory == NULL) | |
764 | { | |
765 | if (result != initial_result) | |
766 | free (result); | |
767 | errno = ENOMEM; | |
768 | return -1; | |
769 | } | |
770 | if (result == initial_result) | |
771 | memcpy (memory, initial_result, length); | |
772 | result = memory; | |
773 | grow = false; | |
774 | ||
775 | out2ptr = result + length; | |
776 | out2size = allocated - extra_alloc - length; | |
777 | if (cd2 != (iconv_t)(-1)) | |
778 | res = iconv (cd2, | |
779 | (ICONV_CONST char **) &inptr, | |
780 | &insize, | |
781 | &out2ptr, &out2size); | |
782 | else | |
783 | { | |
784 | /* TO_CODESET is UTF-8. */ | |
785 | if (!(out2size >= insize)) | |
786 | abort (); | |
787 | memcpy (out2ptr, inptr, insize); | |
788 | out2ptr += insize; | |
789 | out2size -= insize; | |
790 | inptr += insize; | |
791 | insize = 0; | |
792 | res = 0; | |
793 | } | |
794 | length = out2ptr - result; | |
795 | } | |
796 | # if !defined _LIBICONV_VERSION && !defined __GLIBC__ | |
797 | /* Irix iconv() inserts a NUL byte if it cannot convert. | |
798 | NetBSD iconv() inserts a question mark if it cannot | |
799 | convert. | |
800 | Only GNU libiconv and GNU libc are known to prefer | |
801 | to fail rather than doing a lossy conversion. */ | |
802 | if (res != (size_t)(-1) && res > 0) | |
803 | { | |
804 | errno = EILSEQ; | |
805 | res = (size_t)(-1); | |
806 | } | |
807 | # endif | |
808 | if (res == (size_t)(-1)) | |
809 | { | |
810 | /* Failure converting the ASCII replacement. */ | |
811 | if (result != initial_result) | |
812 | { | |
813 | int saved_errno = errno; | |
814 | free (result); | |
815 | errno = saved_errno; | |
816 | } | |
817 | return -1; | |
818 | } | |
819 | } | |
820 | else | |
821 | { | |
822 | if (result != initial_result) | |
823 | { | |
824 | int saved_errno = errno; | |
825 | free (result); | |
826 | errno = saved_errno; | |
827 | } | |
828 | return -1; | |
829 | } | |
830 | } | |
831 | if (!(in2size > 0 | |
832 | || (in1size == 0 && !do_final_flush1 && do_final_flush2))) | |
833 | break; | |
834 | if (grow) | |
835 | { | |
836 | char *memory; | |
837 | ||
838 | allocated = 2 * allocated; | |
839 | if (result == initial_result) | |
840 | memory = (char *) malloc (allocated); | |
841 | else | |
842 | memory = (char *) realloc (result, allocated); | |
843 | if (memory == NULL) | |
844 | { | |
845 | if (result != initial_result) | |
846 | free (result); | |
847 | errno = ENOMEM; | |
848 | return -1; | |
849 | } | |
850 | if (result == initial_result) | |
851 | memcpy (memory, initial_result, length); | |
852 | result = memory; | |
853 | } | |
854 | } | |
855 | ||
856 | /* Move the remaining bytes to the beginning of utf8buf. */ | |
857 | if (in2size > 0) | |
858 | memmove (utf8buf, in2ptr, in2size); | |
859 | utf8len = in2size; | |
860 | } | |
861 | ||
862 | if (res1 == (size_t)(-1)) | |
863 | { | |
864 | if (errno1 == EINVAL) | |
865 | in1size = 0; | |
866 | else if (errno1 == EILSEQ) | |
867 | { | |
868 | if (result != initial_result) | |
869 | free (result); | |
870 | errno = errno1; | |
871 | return -1; | |
872 | } | |
873 | } | |
874 | } | |
875 | # undef utf8bufsize | |
876 | } | |
877 | ||
878 | done: | |
879 | /* Now the final memory allocation. */ | |
880 | if (result == tmpbuf) | |
881 | { | |
882 | size_t memsize = length + extra_alloc; | |
883 | char *memory; | |
884 | ||
885 | memory = (char *) malloc (memsize > 0 ? memsize : 1); | |
886 | if (memory != NULL) | |
887 | { | |
888 | memcpy (memory, tmpbuf, length); | |
889 | result = memory; | |
890 | } | |
891 | else | |
892 | { | |
893 | errno = ENOMEM; | |
894 | return -1; | |
895 | } | |
896 | } | |
897 | else if (result != *resultp && length + extra_alloc < allocated) | |
898 | { | |
899 | /* Shrink the allocated memory if possible. */ | |
900 | size_t memsize = length + extra_alloc; | |
901 | char *memory; | |
902 | ||
903 | memory = (char *) realloc (result, memsize > 0 ? memsize : 1); | |
904 | if (memory != NULL) | |
905 | result = memory; | |
906 | } | |
907 | *resultp = result; | |
908 | *lengthp = length; | |
909 | return 0; | |
910 | # undef tmpbuf | |
911 | # undef tmpbufsize | |
912 | } | |
913 | ||
914 | int | |
915 | mem_cd_iconveh (const char *src, size_t srclen, | |
916 | iconv_t cd, iconv_t cd1, iconv_t cd2, | |
917 | enum iconv_ilseq_handler handler, | |
918 | size_t *offsets, | |
919 | char **resultp, size_t *lengthp) | |
920 | { | |
921 | return mem_cd_iconveh_internal (src, srclen, cd, cd1, cd2, handler, 0, | |
922 | offsets, resultp, lengthp); | |
923 | } | |
924 | ||
925 | char * | |
926 | str_cd_iconveh (const char *src, | |
927 | iconv_t cd, iconv_t cd1, iconv_t cd2, | |
928 | enum iconv_ilseq_handler handler) | |
929 | { | |
930 | /* For most encodings, a trailing NUL byte in the input will be converted | |
931 | to a trailing NUL byte in the output. But not for UTF-7. So that this | |
932 | function is usable for UTF-7, we have to exclude the NUL byte from the | |
933 | conversion and add it by hand afterwards. */ | |
934 | char *result = NULL; | |
935 | size_t length = 0; | |
936 | int retval = mem_cd_iconveh_internal (src, strlen (src), | |
937 | cd, cd1, cd2, handler, 1, NULL, | |
938 | &result, &length); | |
939 | ||
940 | if (retval < 0) | |
941 | { | |
942 | if (result != NULL) | |
943 | { | |
944 | int saved_errno = errno; | |
945 | free (result); | |
946 | errno = saved_errno; | |
947 | } | |
948 | return NULL; | |
949 | } | |
950 | ||
951 | /* Add the terminating NUL byte. */ | |
952 | result[length] = '\0'; | |
953 | ||
954 | return result; | |
955 | } | |
956 | ||
957 | #endif | |
958 | ||
959 | int | |
960 | mem_iconveh (const char *src, size_t srclen, | |
961 | const char *from_codeset, const char *to_codeset, | |
962 | enum iconv_ilseq_handler handler, | |
963 | size_t *offsets, | |
964 | char **resultp, size_t *lengthp) | |
965 | { | |
966 | if (srclen == 0) | |
967 | { | |
968 | /* Nothing to convert. */ | |
969 | *lengthp = 0; | |
970 | return 0; | |
971 | } | |
972 | else if (offsets == NULL && c_strcasecmp (from_codeset, to_codeset) == 0) | |
973 | { | |
974 | char *result; | |
975 | ||
976 | if (*resultp != NULL && *lengthp >= srclen) | |
977 | result = *resultp; | |
978 | else | |
979 | { | |
980 | result = (char *) malloc (srclen); | |
981 | if (result == NULL) | |
982 | { | |
983 | errno = ENOMEM; | |
984 | return -1; | |
985 | } | |
986 | } | |
987 | memcpy (result, src, srclen); | |
988 | *resultp = result; | |
989 | *lengthp = srclen; | |
990 | return 0; | |
991 | } | |
992 | else | |
993 | { | |
994 | #if HAVE_ICONV | |
995 | iconv_t cd; | |
996 | iconv_t cd1; | |
997 | iconv_t cd2; | |
998 | char *result; | |
999 | size_t length; | |
1000 | int retval; | |
1001 | ||
1002 | /* Avoid glibc-2.1 bug with EUC-KR. */ | |
1003 | # if (__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) && !defined _LIBICONV_VERSION | |
1004 | if (c_strcasecmp (from_codeset, "EUC-KR") == 0 | |
1005 | || c_strcasecmp (to_codeset, "EUC-KR") == 0) | |
1006 | { | |
1007 | errno = EINVAL; | |
1008 | return -1; | |
1009 | } | |
1010 | # endif | |
1011 | ||
1012 | cd = iconv_open (to_codeset, from_codeset); | |
1013 | ||
1014 | if (STRCASEEQ (from_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0)) | |
1015 | cd1 = (iconv_t)(-1); | |
1016 | else | |
1017 | { | |
1018 | cd1 = iconv_open ("UTF-8", from_codeset); | |
1019 | if (cd1 == (iconv_t)(-1)) | |
1020 | { | |
1021 | int saved_errno = errno; | |
1022 | if (cd != (iconv_t)(-1)) | |
1023 | iconv_close (cd); | |
1024 | errno = saved_errno; | |
1025 | return -1; | |
1026 | } | |
1027 | } | |
1028 | ||
1029 | if (STRCASEEQ (to_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0) | |
1030 | # if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 || _LIBICONV_VERSION >= 0x0105 | |
1031 | || c_strcasecmp (to_codeset, "UTF-8//TRANSLIT") == 0 | |
1032 | # endif | |
1033 | ) | |
1034 | cd2 = (iconv_t)(-1); | |
1035 | else | |
1036 | { | |
1037 | cd2 = iconv_open (to_codeset, "UTF-8"); | |
1038 | if (cd2 == (iconv_t)(-1)) | |
1039 | { | |
1040 | int saved_errno = errno; | |
1041 | if (cd1 != (iconv_t)(-1)) | |
1042 | iconv_close (cd1); | |
1043 | if (cd != (iconv_t)(-1)) | |
1044 | iconv_close (cd); | |
1045 | errno = saved_errno; | |
1046 | return -1; | |
1047 | } | |
1048 | } | |
1049 | ||
1050 | result = *resultp; | |
1051 | length = *lengthp; | |
1052 | retval = mem_cd_iconveh (src, srclen, cd, cd1, cd2, handler, offsets, | |
1053 | &result, &length); | |
1054 | ||
1055 | if (retval < 0) | |
1056 | { | |
1057 | /* Close cd, cd1, cd2, but preserve the errno from str_cd_iconv. */ | |
1058 | int saved_errno = errno; | |
1059 | if (cd2 != (iconv_t)(-1)) | |
1060 | iconv_close (cd2); | |
1061 | if (cd1 != (iconv_t)(-1)) | |
1062 | iconv_close (cd1); | |
1063 | if (cd != (iconv_t)(-1)) | |
1064 | iconv_close (cd); | |
1065 | errno = saved_errno; | |
1066 | } | |
1067 | else | |
1068 | { | |
1069 | if (cd2 != (iconv_t)(-1) && iconv_close (cd2) < 0) | |
1070 | { | |
1071 | /* Return -1, but free the allocated memory, and while doing | |
1072 | that, preserve the errno from iconv_close. */ | |
1073 | int saved_errno = errno; | |
1074 | if (cd1 != (iconv_t)(-1)) | |
1075 | iconv_close (cd1); | |
1076 | if (cd != (iconv_t)(-1)) | |
1077 | iconv_close (cd); | |
1078 | if (result != *resultp && result != NULL) | |
1079 | free (result); | |
1080 | errno = saved_errno; | |
1081 | return -1; | |
1082 | } | |
1083 | if (cd1 != (iconv_t)(-1) && iconv_close (cd1) < 0) | |
1084 | { | |
1085 | /* Return -1, but free the allocated memory, and while doing | |
1086 | that, preserve the errno from iconv_close. */ | |
1087 | int saved_errno = errno; | |
1088 | if (cd != (iconv_t)(-1)) | |
1089 | iconv_close (cd); | |
1090 | if (result != *resultp && result != NULL) | |
1091 | free (result); | |
1092 | errno = saved_errno; | |
1093 | return -1; | |
1094 | } | |
1095 | if (cd != (iconv_t)(-1) && iconv_close (cd) < 0) | |
1096 | { | |
1097 | /* Return -1, but free the allocated memory, and while doing | |
1098 | that, preserve the errno from iconv_close. */ | |
1099 | int saved_errno = errno; | |
1100 | if (result != *resultp && result != NULL) | |
1101 | free (result); | |
1102 | errno = saved_errno; | |
1103 | return -1; | |
1104 | } | |
1105 | *resultp = result; | |
1106 | *lengthp = length; | |
1107 | } | |
1108 | return retval; | |
1109 | #else | |
1110 | /* This is a different error code than if iconv_open existed but didn't | |
1111 | support from_codeset and to_codeset, so that the caller can emit | |
1112 | an error message such as | |
1113 | "iconv() is not supported. Installing GNU libiconv and | |
1114 | then reinstalling this package would fix this." */ | |
1115 | errno = ENOSYS; | |
1116 | return -1; | |
1117 | #endif | |
1118 | } | |
1119 | } | |
1120 | ||
1121 | char * | |
1122 | str_iconveh (const char *src, | |
1123 | const char *from_codeset, const char *to_codeset, | |
1124 | enum iconv_ilseq_handler handler) | |
1125 | { | |
1126 | if (*src == '\0' || c_strcasecmp (from_codeset, to_codeset) == 0) | |
1127 | { | |
1128 | char *result = strdup (src); | |
1129 | ||
1130 | if (result == NULL) | |
1131 | errno = ENOMEM; | |
1132 | return result; | |
1133 | } | |
1134 | else | |
1135 | { | |
1136 | #if HAVE_ICONV | |
1137 | iconv_t cd; | |
1138 | iconv_t cd1; | |
1139 | iconv_t cd2; | |
1140 | char *result; | |
1141 | ||
1142 | /* Avoid glibc-2.1 bug with EUC-KR. */ | |
1143 | # if (__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) && !defined _LIBICONV_VERSION | |
1144 | if (c_strcasecmp (from_codeset, "EUC-KR") == 0 | |
1145 | || c_strcasecmp (to_codeset, "EUC-KR") == 0) | |
1146 | { | |
1147 | errno = EINVAL; | |
1148 | return NULL; | |
1149 | } | |
1150 | # endif | |
1151 | ||
1152 | cd = iconv_open (to_codeset, from_codeset); | |
1153 | ||
1154 | if (STRCASEEQ (from_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0)) | |
1155 | cd1 = (iconv_t)(-1); | |
1156 | else | |
1157 | { | |
1158 | cd1 = iconv_open ("UTF-8", from_codeset); | |
1159 | if (cd1 == (iconv_t)(-1)) | |
1160 | { | |
1161 | int saved_errno = errno; | |
1162 | if (cd != (iconv_t)(-1)) | |
1163 | iconv_close (cd); | |
1164 | errno = saved_errno; | |
1165 | return NULL; | |
1166 | } | |
1167 | } | |
1168 | ||
1169 | if (STRCASEEQ (to_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0) | |
1170 | # if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 || _LIBICONV_VERSION >= 0x0105 | |
1171 | || c_strcasecmp (to_codeset, "UTF-8//TRANSLIT") == 0 | |
1172 | # endif | |
1173 | ) | |
1174 | cd2 = (iconv_t)(-1); | |
1175 | else | |
1176 | { | |
1177 | cd2 = iconv_open (to_codeset, "UTF-8"); | |
1178 | if (cd2 == (iconv_t)(-1)) | |
1179 | { | |
1180 | int saved_errno = errno; | |
1181 | if (cd1 != (iconv_t)(-1)) | |
1182 | iconv_close (cd1); | |
1183 | if (cd != (iconv_t)(-1)) | |
1184 | iconv_close (cd); | |
1185 | errno = saved_errno; | |
1186 | return NULL; | |
1187 | } | |
1188 | } | |
1189 | ||
1190 | result = str_cd_iconveh (src, cd, cd1, cd2, handler); | |
1191 | ||
1192 | if (result == NULL) | |
1193 | { | |
1194 | /* Close cd, cd1, cd2, but preserve the errno from str_cd_iconv. */ | |
1195 | int saved_errno = errno; | |
1196 | if (cd2 != (iconv_t)(-1)) | |
1197 | iconv_close (cd2); | |
1198 | if (cd1 != (iconv_t)(-1)) | |
1199 | iconv_close (cd1); | |
1200 | if (cd != (iconv_t)(-1)) | |
1201 | iconv_close (cd); | |
1202 | errno = saved_errno; | |
1203 | } | |
1204 | else | |
1205 | { | |
1206 | if (cd2 != (iconv_t)(-1) && iconv_close (cd2) < 0) | |
1207 | { | |
1208 | /* Return NULL, but free the allocated memory, and while doing | |
1209 | that, preserve the errno from iconv_close. */ | |
1210 | int saved_errno = errno; | |
1211 | if (cd1 != (iconv_t)(-1)) | |
1212 | iconv_close (cd1); | |
1213 | if (cd != (iconv_t)(-1)) | |
1214 | iconv_close (cd); | |
1215 | free (result); | |
1216 | errno = saved_errno; | |
1217 | return NULL; | |
1218 | } | |
1219 | if (cd1 != (iconv_t)(-1) && iconv_close (cd1) < 0) | |
1220 | { | |
1221 | /* Return NULL, but free the allocated memory, and while doing | |
1222 | that, preserve the errno from iconv_close. */ | |
1223 | int saved_errno = errno; | |
1224 | if (cd != (iconv_t)(-1)) | |
1225 | iconv_close (cd); | |
1226 | free (result); | |
1227 | errno = saved_errno; | |
1228 | return NULL; | |
1229 | } | |
1230 | if (cd != (iconv_t)(-1) && iconv_close (cd) < 0) | |
1231 | { | |
1232 | /* Return NULL, but free the allocated memory, and while doing | |
1233 | that, preserve the errno from iconv_close. */ | |
1234 | int saved_errno = errno; | |
1235 | free (result); | |
1236 | errno = saved_errno; | |
1237 | return NULL; | |
1238 | } | |
1239 | } | |
1240 | return result; | |
1241 | #else | |
1242 | /* This is a different error code than if iconv_open existed but didn't | |
1243 | support from_codeset and to_codeset, so that the caller can emit | |
1244 | an error message such as | |
1245 | "iconv() is not supported. Installing GNU libiconv and | |
1246 | then reinstalling this package would fix this." */ | |
1247 | errno = ENOSYS; | |
1248 | return NULL; | |
1249 | #endif | |
1250 | } | |
1251 | } |