Commit | Line | Data |
---|---|---|
24d56127 | 1 | /* Character set conversion with error handling. |
8912421c | 2 | Copyright (C) 2001-2009 Free Software Foundation, Inc. |
24d56127 LC |
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 | ||
8912421c LC |
43 | /* The caller must provide an iconveh_t, not just an iconv_t, because when a |
44 | conversion error occurs, we may have to determine the Unicode representation | |
45 | of the inconvertible character. */ | |
46 | ||
47 | int | |
48 | iconveh_open (const char *to_codeset, const char *from_codeset, iconveh_t *cdp) | |
49 | { | |
50 | iconv_t cd; | |
51 | iconv_t cd1; | |
52 | iconv_t cd2; | |
53 | ||
54 | /* Avoid glibc-2.1 bug with EUC-KR. */ | |
55 | # if (__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) && !defined _LIBICONV_VERSION | |
56 | if (c_strcasecmp (from_codeset, "EUC-KR") == 0 | |
57 | || c_strcasecmp (to_codeset, "EUC-KR") == 0) | |
58 | { | |
59 | errno = EINVAL; | |
60 | return -1; | |
61 | } | |
62 | # endif | |
63 | ||
64 | cd = iconv_open (to_codeset, from_codeset); | |
65 | ||
66 | if (STRCASEEQ (from_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0)) | |
67 | cd1 = (iconv_t)(-1); | |
68 | else | |
69 | { | |
70 | cd1 = iconv_open ("UTF-8", from_codeset); | |
71 | if (cd1 == (iconv_t)(-1)) | |
72 | { | |
73 | int saved_errno = errno; | |
74 | if (cd != (iconv_t)(-1)) | |
75 | iconv_close (cdp->cd); | |
76 | errno = saved_errno; | |
77 | return -1; | |
78 | } | |
79 | } | |
80 | ||
81 | if (STRCASEEQ (to_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0) | |
82 | # if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 || _LIBICONV_VERSION >= 0x0105 | |
83 | || c_strcasecmp (to_codeset, "UTF-8//TRANSLIT") == 0 | |
84 | # endif | |
85 | ) | |
86 | cd2 = (iconv_t)(-1); | |
87 | else | |
88 | { | |
89 | cd2 = iconv_open (to_codeset, "UTF-8"); | |
90 | if (cd2 == (iconv_t)(-1)) | |
91 | { | |
92 | int saved_errno = errno; | |
93 | if (cd1 != (iconv_t)(-1)) | |
94 | iconv_close (cd1); | |
95 | if (cd != (iconv_t)(-1)) | |
96 | iconv_close (cd); | |
97 | errno = saved_errno; | |
98 | return -1; | |
99 | } | |
100 | } | |
101 | ||
102 | cdp->cd = cd; | |
103 | cdp->cd1 = cd1; | |
104 | cdp->cd2 = cd2; | |
105 | return 0; | |
106 | } | |
107 | ||
108 | int | |
109 | iconveh_close (const iconveh_t *cd) | |
110 | { | |
111 | if (cd->cd2 != (iconv_t)(-1) && iconv_close (cd->cd2) < 0) | |
112 | { | |
113 | /* Return -1, but preserve the errno from iconv_close. */ | |
114 | int saved_errno = errno; | |
115 | if (cd->cd1 != (iconv_t)(-1)) | |
116 | iconv_close (cd->cd1); | |
117 | if (cd->cd != (iconv_t)(-1)) | |
118 | iconv_close (cd->cd); | |
119 | errno = saved_errno; | |
120 | return -1; | |
121 | } | |
122 | if (cd->cd1 != (iconv_t)(-1) && iconv_close (cd->cd1) < 0) | |
123 | { | |
124 | /* Return -1, but preserve the errno from iconv_close. */ | |
125 | int saved_errno = errno; | |
126 | if (cd->cd != (iconv_t)(-1)) | |
127 | iconv_close (cd->cd); | |
128 | errno = saved_errno; | |
129 | return -1; | |
130 | } | |
131 | if (cd->cd != (iconv_t)(-1) && iconv_close (cd->cd) < 0) | |
132 | return -1; | |
133 | return 0; | |
134 | } | |
24d56127 LC |
135 | |
136 | /* iconv_carefully is like iconv, except that it stops as soon as it encounters | |
137 | a conversion error, and it returns in *INCREMENTED a boolean telling whether | |
138 | it has incremented the input pointers past the error location. */ | |
139 | # if !defined _LIBICONV_VERSION && !defined __GLIBC__ | |
140 | /* Irix iconv() inserts a NUL byte if it cannot convert. | |
141 | NetBSD iconv() inserts a question mark if it cannot convert. | |
142 | Only GNU libiconv and GNU libc are known to prefer to fail rather | |
143 | than doing a lossy conversion. */ | |
144 | static size_t | |
145 | iconv_carefully (iconv_t cd, | |
146 | const char **inbuf, size_t *inbytesleft, | |
147 | char **outbuf, size_t *outbytesleft, | |
148 | bool *incremented) | |
149 | { | |
150 | const char *inptr = *inbuf; | |
151 | const char *inptr_end = inptr + *inbytesleft; | |
152 | char *outptr = *outbuf; | |
153 | size_t outsize = *outbytesleft; | |
154 | const char *inptr_before; | |
155 | size_t res; | |
156 | ||
157 | do | |
158 | { | |
159 | size_t insize; | |
160 | ||
161 | inptr_before = inptr; | |
162 | res = (size_t)(-1); | |
163 | ||
164 | for (insize = 1; inptr + insize <= inptr_end; insize++) | |
165 | { | |
166 | res = iconv (cd, | |
167 | (ICONV_CONST char **) &inptr, &insize, | |
168 | &outptr, &outsize); | |
169 | if (!(res == (size_t)(-1) && errno == EINVAL)) | |
170 | break; | |
171 | /* iconv can eat up a shift sequence but give EINVAL while attempting | |
172 | to convert the first character. E.g. libiconv does this. */ | |
173 | if (inptr > inptr_before) | |
174 | { | |
175 | res = 0; | |
176 | break; | |
177 | } | |
178 | } | |
179 | ||
180 | if (res == 0) | |
181 | { | |
182 | *outbuf = outptr; | |
183 | *outbytesleft = outsize; | |
184 | } | |
185 | } | |
186 | while (res == 0 && inptr < inptr_end); | |
187 | ||
188 | *inbuf = inptr; | |
189 | *inbytesleft = inptr_end - inptr; | |
190 | if (res != (size_t)(-1) && res > 0) | |
191 | { | |
192 | /* iconv() has already incremented INPTR. We cannot go back to a | |
193 | previous INPTR, otherwise the state inside CD would become invalid, | |
194 | if FROM_CODESET is a stateful encoding. So, tell the caller that | |
195 | *INBUF has already been incremented. */ | |
196 | *incremented = (inptr > inptr_before); | |
197 | errno = EILSEQ; | |
198 | return (size_t)(-1); | |
199 | } | |
200 | else | |
201 | { | |
202 | *incremented = false; | |
203 | return res; | |
204 | } | |
205 | } | |
206 | # else | |
207 | # define iconv_carefully(cd, inbuf, inbytesleft, outbuf, outbytesleft, incremented) \ | |
208 | (*(incremented) = false, \ | |
209 | iconv (cd, (ICONV_CONST char **) (inbuf), inbytesleft, outbuf, outbytesleft)) | |
210 | # endif | |
211 | ||
212 | /* iconv_carefully_1 is like iconv_carefully, except that it stops after | |
213 | converting one character or one shift sequence. */ | |
214 | static size_t | |
215 | iconv_carefully_1 (iconv_t cd, | |
216 | const char **inbuf, size_t *inbytesleft, | |
217 | char **outbuf, size_t *outbytesleft, | |
218 | bool *incremented) | |
219 | { | |
220 | const char *inptr_before = *inbuf; | |
221 | const char *inptr = inptr_before; | |
222 | const char *inptr_end = inptr_before + *inbytesleft; | |
223 | char *outptr = *outbuf; | |
224 | size_t outsize = *outbytesleft; | |
225 | size_t res = (size_t)(-1); | |
226 | size_t insize; | |
227 | ||
228 | for (insize = 1; inptr_before + insize <= inptr_end; insize++) | |
229 | { | |
230 | inptr = inptr_before; | |
231 | res = iconv (cd, | |
232 | (ICONV_CONST char **) &inptr, &insize, | |
233 | &outptr, &outsize); | |
234 | if (!(res == (size_t)(-1) && errno == EINVAL)) | |
235 | break; | |
236 | /* iconv can eat up a shift sequence but give EINVAL while attempting | |
237 | to convert the first character. E.g. libiconv does this. */ | |
238 | if (inptr > inptr_before) | |
239 | { | |
240 | res = 0; | |
241 | break; | |
242 | } | |
243 | } | |
244 | ||
245 | *inbuf = inptr; | |
246 | *inbytesleft = inptr_end - inptr; | |
247 | # if !defined _LIBICONV_VERSION && !defined __GLIBC__ | |
248 | /* Irix iconv() inserts a NUL byte if it cannot convert. | |
249 | NetBSD iconv() inserts a question mark if it cannot convert. | |
250 | Only GNU libiconv and GNU libc are known to prefer to fail rather | |
251 | than doing a lossy conversion. */ | |
252 | if (res != (size_t)(-1) && res > 0) | |
253 | { | |
254 | /* iconv() has already incremented INPTR. We cannot go back to a | |
255 | previous INPTR, otherwise the state inside CD would become invalid, | |
256 | if FROM_CODESET is a stateful encoding. So, tell the caller that | |
257 | *INBUF has already been incremented. */ | |
258 | *incremented = (inptr > inptr_before); | |
259 | errno = EILSEQ; | |
260 | return (size_t)(-1); | |
261 | } | |
262 | # endif | |
263 | ||
264 | if (res != (size_t)(-1)) | |
265 | { | |
266 | *outbuf = outptr; | |
267 | *outbytesleft = outsize; | |
268 | } | |
269 | *incremented = false; | |
270 | return res; | |
271 | } | |
272 | ||
273 | /* utf8conv_carefully is like iconv, except that | |
274 | - it converts from UTF-8 to UTF-8, | |
275 | - it stops as soon as it encounters a conversion error, and it returns | |
276 | in *INCREMENTED a boolean telling whether it has incremented the input | |
277 | pointers past the error location, | |
278 | - if one_character_only is true, it stops after converting one | |
279 | character. */ | |
280 | static size_t | |
281 | utf8conv_carefully (bool one_character_only, | |
282 | const char **inbuf, size_t *inbytesleft, | |
283 | char **outbuf, size_t *outbytesleft, | |
284 | bool *incremented) | |
285 | { | |
286 | const char *inptr = *inbuf; | |
287 | size_t insize = *inbytesleft; | |
288 | char *outptr = *outbuf; | |
289 | size_t outsize = *outbytesleft; | |
290 | size_t res; | |
291 | ||
292 | res = 0; | |
293 | do | |
294 | { | |
295 | ucs4_t uc; | |
296 | int n; | |
297 | int m; | |
298 | ||
299 | n = u8_mbtoucr (&uc, (const uint8_t *) inptr, insize); | |
300 | if (n < 0) | |
301 | { | |
302 | errno = (n == -2 ? EINVAL : EILSEQ); | |
303 | n = u8_mbtouc (&uc, (const uint8_t *) inptr, insize); | |
304 | inptr += n; | |
305 | insize -= n; | |
306 | res = (size_t)(-1); | |
307 | *incremented = true; | |
308 | break; | |
309 | } | |
310 | if (outsize == 0) | |
311 | { | |
312 | errno = E2BIG; | |
313 | res = (size_t)(-1); | |
314 | *incremented = false; | |
315 | break; | |
316 | } | |
317 | m = u8_uctomb ((uint8_t *) outptr, uc, outsize); | |
318 | if (m == -2) | |
319 | { | |
320 | errno = E2BIG; | |
321 | res = (size_t)(-1); | |
322 | *incremented = false; | |
323 | break; | |
324 | } | |
325 | inptr += n; | |
326 | insize -= n; | |
327 | if (m == -1) | |
328 | { | |
329 | errno = EILSEQ; | |
330 | res = (size_t)(-1); | |
331 | *incremented = true; | |
332 | break; | |
333 | } | |
334 | outptr += m; | |
335 | outsize -= m; | |
336 | } | |
337 | while (!one_character_only && insize > 0); | |
338 | ||
339 | *inbuf = inptr; | |
340 | *inbytesleft = insize; | |
341 | *outbuf = outptr; | |
342 | *outbytesleft = outsize; | |
343 | return res; | |
344 | } | |
345 | ||
346 | static int | |
347 | mem_cd_iconveh_internal (const char *src, size_t srclen, | |
348 | iconv_t cd, iconv_t cd1, iconv_t cd2, | |
349 | enum iconv_ilseq_handler handler, | |
350 | size_t extra_alloc, | |
351 | size_t *offsets, | |
352 | char **resultp, size_t *lengthp) | |
353 | { | |
354 | /* When a conversion error occurs, we cannot start using CD1 and CD2 at | |
355 | this point: FROM_CODESET may be a stateful encoding like ISO-2022-KR. | |
356 | Instead, we have to start afresh from the beginning of SRC. */ | |
357 | /* Use a temporary buffer, so that for small strings, a single malloc() | |
358 | call will be sufficient. */ | |
359 | # define tmpbufsize 4096 | |
360 | /* The alignment is needed when converting e.g. to glibc's WCHAR_T or | |
361 | libiconv's UCS-4-INTERNAL encoding. */ | |
362 | union { unsigned int align; char buf[tmpbufsize]; } tmp; | |
363 | # define tmpbuf tmp.buf | |
364 | ||
365 | char *initial_result; | |
366 | char *result; | |
367 | size_t allocated; | |
368 | size_t length; | |
369 | size_t last_length = (size_t)(-1); /* only needed if offsets != NULL */ | |
370 | ||
371 | if (*resultp != NULL && *lengthp >= sizeof (tmpbuf)) | |
372 | { | |
373 | initial_result = *resultp; | |
374 | allocated = *lengthp; | |
375 | } | |
376 | else | |
377 | { | |
378 | initial_result = tmpbuf; | |
379 | allocated = sizeof (tmpbuf); | |
380 | } | |
381 | result = initial_result; | |
382 | ||
383 | /* Test whether a direct conversion is possible at all. */ | |
384 | if (cd == (iconv_t)(-1)) | |
385 | goto indirectly; | |
386 | ||
387 | if (offsets != NULL) | |
388 | { | |
389 | size_t i; | |
390 | ||
391 | for (i = 0; i < srclen; i++) | |
392 | offsets[i] = (size_t)(-1); | |
393 | ||
394 | last_length = (size_t)(-1); | |
395 | } | |
396 | length = 0; | |
397 | ||
398 | /* First, try a direct conversion, and see whether a conversion error | |
399 | occurs at all. */ | |
400 | { | |
401 | const char *inptr = src; | |
402 | size_t insize = srclen; | |
403 | ||
404 | /* Avoid glibc-2.1 bug and Solaris 2.7-2.9 bug. */ | |
405 | # if defined _LIBICONV_VERSION \ | |
406 | || !((__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) || defined __sun) | |
407 | /* Set to the initial state. */ | |
408 | iconv (cd, NULL, NULL, NULL, NULL); | |
409 | # endif | |
410 | ||
411 | while (insize > 0) | |
412 | { | |
413 | char *outptr = result + length; | |
414 | size_t outsize = allocated - extra_alloc - length; | |
415 | bool incremented; | |
416 | size_t res; | |
417 | bool grow; | |
418 | ||
419 | if (offsets != NULL) | |
420 | { | |
421 | if (length != last_length) /* ensure that offset[] be increasing */ | |
422 | { | |
423 | offsets[inptr - src] = length; | |
424 | last_length = length; | |
425 | } | |
426 | res = iconv_carefully_1 (cd, | |
427 | &inptr, &insize, | |
428 | &outptr, &outsize, | |
429 | &incremented); | |
430 | } | |
431 | else | |
432 | /* Use iconv_carefully instead of iconv here, because: | |
433 | - If TO_CODESET is UTF-8, we can do the error handling in this | |
434 | loop, no need for a second loop, | |
435 | - With iconv() implementations other than GNU libiconv and GNU | |
436 | libc, if we use iconv() in a big swoop, checking for an E2BIG | |
437 | return, we lose the number of irreversible conversions. */ | |
438 | res = iconv_carefully (cd, | |
439 | &inptr, &insize, | |
440 | &outptr, &outsize, | |
441 | &incremented); | |
442 | ||
443 | length = outptr - result; | |
444 | grow = (length + extra_alloc > allocated / 2); | |
445 | if (res == (size_t)(-1)) | |
446 | { | |
447 | if (errno == E2BIG) | |
448 | grow = true; | |
449 | else if (errno == EINVAL) | |
450 | break; | |
451 | else if (errno == EILSEQ && handler != iconveh_error) | |
452 | { | |
453 | if (cd2 == (iconv_t)(-1)) | |
454 | { | |
455 | /* TO_CODESET is UTF-8. */ | |
456 | /* Error handling can produce up to 1 byte of output. */ | |
457 | if (length + 1 + extra_alloc > allocated) | |
458 | { | |
459 | char *memory; | |
460 | ||
461 | allocated = 2 * allocated; | |
462 | if (length + 1 + extra_alloc > allocated) | |
463 | abort (); | |
464 | if (result == initial_result) | |
465 | memory = (char *) malloc (allocated); | |
466 | else | |
467 | memory = (char *) realloc (result, allocated); | |
468 | if (memory == NULL) | |
469 | { | |
470 | if (result != initial_result) | |
471 | free (result); | |
472 | errno = ENOMEM; | |
473 | return -1; | |
474 | } | |
475 | if (result == initial_result) | |
476 | memcpy (memory, initial_result, length); | |
477 | result = memory; | |
478 | grow = false; | |
479 | } | |
480 | /* The input is invalid in FROM_CODESET. Eat up one byte | |
481 | and emit a question mark. */ | |
482 | if (!incremented) | |
483 | { | |
484 | if (insize == 0) | |
485 | abort (); | |
486 | inptr++; | |
487 | insize--; | |
488 | } | |
489 | result[length] = '?'; | |
490 | length++; | |
491 | } | |
492 | else | |
493 | goto indirectly; | |
494 | } | |
495 | else | |
496 | { | |
497 | if (result != initial_result) | |
498 | { | |
499 | int saved_errno = errno; | |
500 | free (result); | |
501 | errno = saved_errno; | |
502 | } | |
503 | return -1; | |
504 | } | |
505 | } | |
506 | if (insize == 0) | |
507 | break; | |
508 | if (grow) | |
509 | { | |
510 | char *memory; | |
511 | ||
512 | allocated = 2 * allocated; | |
513 | if (result == initial_result) | |
514 | memory = (char *) malloc (allocated); | |
515 | else | |
516 | memory = (char *) realloc (result, allocated); | |
517 | if (memory == NULL) | |
518 | { | |
519 | if (result != initial_result) | |
520 | free (result); | |
521 | errno = ENOMEM; | |
522 | return -1; | |
523 | } | |
524 | if (result == initial_result) | |
525 | memcpy (memory, initial_result, length); | |
526 | result = memory; | |
527 | } | |
528 | } | |
529 | } | |
530 | ||
531 | /* Now get the conversion state back to the initial state. | |
532 | But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
533 | #if defined _LIBICONV_VERSION \ | |
534 | || !((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) || defined __sun) | |
535 | for (;;) | |
536 | { | |
537 | char *outptr = result + length; | |
538 | size_t outsize = allocated - extra_alloc - length; | |
539 | size_t res; | |
540 | ||
541 | res = iconv (cd, NULL, NULL, &outptr, &outsize); | |
542 | length = outptr - result; | |
543 | if (res == (size_t)(-1)) | |
544 | { | |
545 | if (errno == E2BIG) | |
546 | { | |
547 | char *memory; | |
548 | ||
549 | allocated = 2 * allocated; | |
550 | if (result == initial_result) | |
551 | memory = (char *) malloc (allocated); | |
552 | else | |
553 | memory = (char *) realloc (result, allocated); | |
554 | if (memory == NULL) | |
555 | { | |
556 | if (result != initial_result) | |
557 | free (result); | |
558 | errno = ENOMEM; | |
559 | return -1; | |
560 | } | |
561 | if (result == initial_result) | |
562 | memcpy (memory, initial_result, length); | |
563 | result = memory; | |
564 | } | |
565 | else | |
566 | { | |
567 | if (result != initial_result) | |
568 | { | |
569 | int saved_errno = errno; | |
570 | free (result); | |
571 | errno = saved_errno; | |
572 | } | |
573 | return -1; | |
574 | } | |
575 | } | |
576 | else | |
577 | break; | |
578 | } | |
579 | #endif | |
580 | ||
581 | /* The direct conversion succeeded. */ | |
582 | goto done; | |
583 | ||
584 | indirectly: | |
585 | /* The direct conversion failed. | |
586 | Use a conversion through UTF-8. */ | |
587 | if (offsets != NULL) | |
588 | { | |
589 | size_t i; | |
590 | ||
591 | for (i = 0; i < srclen; i++) | |
592 | offsets[i] = (size_t)(-1); | |
593 | ||
594 | last_length = (size_t)(-1); | |
595 | } | |
596 | length = 0; | |
597 | { | |
598 | const bool slowly = (offsets != NULL || handler == iconveh_error); | |
599 | # define utf8bufsize 4096 /* may also be smaller or larger than tmpbufsize */ | |
600 | char utf8buf[utf8bufsize + 1]; | |
601 | size_t utf8len = 0; | |
602 | const char *in1ptr = src; | |
603 | size_t in1size = srclen; | |
604 | bool do_final_flush1 = true; | |
605 | bool do_final_flush2 = true; | |
606 | ||
607 | /* Avoid glibc-2.1 bug and Solaris 2.7-2.9 bug. */ | |
608 | # if defined _LIBICONV_VERSION \ | |
609 | || !((__GLIBC__ - 0 == 2 && __GLIBC_MINOR__ - 0 <= 1) || defined __sun) | |
610 | /* Set to the initial state. */ | |
611 | if (cd1 != (iconv_t)(-1)) | |
612 | iconv (cd1, NULL, NULL, NULL, NULL); | |
613 | if (cd2 != (iconv_t)(-1)) | |
614 | iconv (cd2, NULL, NULL, NULL, NULL); | |
615 | # endif | |
616 | ||
617 | while (in1size > 0 || do_final_flush1 || utf8len > 0 || do_final_flush2) | |
618 | { | |
619 | char *out1ptr = utf8buf + utf8len; | |
620 | size_t out1size = utf8bufsize - utf8len; | |
621 | bool incremented1; | |
622 | size_t res1; | |
623 | int errno1; | |
624 | ||
625 | /* Conversion step 1: from FROM_CODESET to UTF-8. */ | |
626 | if (in1size > 0) | |
627 | { | |
628 | if (offsets != NULL | |
629 | && length != last_length) /* ensure that offset[] be increasing */ | |
630 | { | |
631 | offsets[in1ptr - src] = length; | |
632 | last_length = length; | |
633 | } | |
634 | if (cd1 != (iconv_t)(-1)) | |
635 | { | |
636 | if (slowly) | |
637 | res1 = iconv_carefully_1 (cd1, | |
638 | &in1ptr, &in1size, | |
639 | &out1ptr, &out1size, | |
640 | &incremented1); | |
641 | else | |
642 | res1 = iconv_carefully (cd1, | |
643 | &in1ptr, &in1size, | |
644 | &out1ptr, &out1size, | |
645 | &incremented1); | |
646 | } | |
647 | else | |
648 | { | |
649 | /* FROM_CODESET is UTF-8. */ | |
650 | res1 = utf8conv_carefully (slowly, | |
651 | &in1ptr, &in1size, | |
652 | &out1ptr, &out1size, | |
653 | &incremented1); | |
654 | } | |
655 | } | |
656 | else if (do_final_flush1) | |
657 | { | |
658 | /* Now get the conversion state of CD1 back to the initial state. | |
659 | But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
660 | # if defined _LIBICONV_VERSION \ | |
661 | || !((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) || defined __sun) | |
662 | if (cd1 != (iconv_t)(-1)) | |
663 | res1 = iconv (cd1, NULL, NULL, &out1ptr, &out1size); | |
664 | else | |
665 | # endif | |
666 | res1 = 0; | |
667 | do_final_flush1 = false; | |
668 | incremented1 = true; | |
669 | } | |
670 | else | |
671 | { | |
672 | res1 = 0; | |
673 | incremented1 = true; | |
674 | } | |
675 | if (res1 == (size_t)(-1) | |
676 | && !(errno == E2BIG || errno == EINVAL || errno == EILSEQ)) | |
677 | { | |
678 | if (result != initial_result) | |
679 | { | |
680 | int saved_errno = errno; | |
681 | free (result); | |
682 | errno = saved_errno; | |
683 | } | |
684 | return -1; | |
685 | } | |
686 | if (res1 == (size_t)(-1) | |
687 | && errno == EILSEQ && handler != iconveh_error) | |
688 | { | |
689 | /* The input is invalid in FROM_CODESET. Eat up one byte and | |
690 | emit a question mark. Room for the question mark was allocated | |
691 | at the end of utf8buf. */ | |
692 | if (!incremented1) | |
693 | { | |
694 | if (in1size == 0) | |
695 | abort (); | |
696 | in1ptr++; | |
697 | in1size--; | |
698 | } | |
8912421c LC |
699 | *out1ptr++ = '?'; |
700 | res1 = 0; | |
24d56127 LC |
701 | } |
702 | errno1 = errno; | |
703 | utf8len = out1ptr - utf8buf; | |
704 | ||
705 | if (offsets != NULL | |
706 | || in1size == 0 | |
707 | || utf8len > utf8bufsize / 2 | |
708 | || (res1 == (size_t)(-1) && errno1 == E2BIG)) | |
709 | { | |
710 | /* Conversion step 2: from UTF-8 to TO_CODESET. */ | |
711 | const char *in2ptr = utf8buf; | |
712 | size_t in2size = utf8len; | |
713 | ||
714 | while (in2size > 0 | |
715 | || (in1size == 0 && !do_final_flush1 && do_final_flush2)) | |
716 | { | |
717 | char *out2ptr = result + length; | |
718 | size_t out2size = allocated - extra_alloc - length; | |
719 | bool incremented2; | |
720 | size_t res2; | |
721 | bool grow; | |
722 | ||
723 | if (in2size > 0) | |
724 | { | |
725 | if (cd2 != (iconv_t)(-1)) | |
726 | res2 = iconv_carefully (cd2, | |
727 | &in2ptr, &in2size, | |
728 | &out2ptr, &out2size, | |
729 | &incremented2); | |
730 | else | |
731 | /* TO_CODESET is UTF-8. */ | |
732 | res2 = utf8conv_carefully (false, | |
733 | &in2ptr, &in2size, | |
734 | &out2ptr, &out2size, | |
735 | &incremented2); | |
736 | } | |
737 | else /* in1size == 0 && !do_final_flush1 | |
738 | && in2size == 0 && do_final_flush2 */ | |
739 | { | |
740 | /* Now get the conversion state of CD1 back to the initial | |
741 | state. But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
742 | # if defined _LIBICONV_VERSION \ | |
743 | || !((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) || defined __sun) | |
744 | if (cd2 != (iconv_t)(-1)) | |
745 | res2 = iconv (cd2, NULL, NULL, &out2ptr, &out2size); | |
746 | else | |
747 | # endif | |
748 | res2 = 0; | |
749 | do_final_flush2 = false; | |
750 | incremented2 = true; | |
751 | } | |
752 | ||
753 | length = out2ptr - result; | |
754 | grow = (length + extra_alloc > allocated / 2); | |
755 | if (res2 == (size_t)(-1)) | |
756 | { | |
757 | if (errno == E2BIG) | |
758 | grow = true; | |
759 | else if (errno == EINVAL) | |
760 | break; | |
761 | else if (errno == EILSEQ && handler != iconveh_error) | |
762 | { | |
763 | /* Error handling can produce up to 10 bytes of ASCII | |
764 | output. But TO_CODESET may be UCS-2, UTF-16 or | |
765 | UCS-4, so use CD2 here as well. */ | |
766 | char scratchbuf[10]; | |
767 | size_t scratchlen; | |
768 | ucs4_t uc; | |
769 | const char *inptr; | |
770 | size_t insize; | |
771 | size_t res; | |
772 | ||
773 | if (incremented2) | |
774 | { | |
775 | if (u8_prev (&uc, (const uint8_t *) in2ptr, | |
776 | (const uint8_t *) utf8buf) | |
777 | == NULL) | |
778 | abort (); | |
779 | } | |
780 | else | |
781 | { | |
782 | int n; | |
783 | if (in2size == 0) | |
784 | abort (); | |
785 | n = u8_mbtouc_unsafe (&uc, (const uint8_t *) in2ptr, | |
786 | in2size); | |
787 | in2ptr += n; | |
788 | in2size -= n; | |
789 | } | |
790 | ||
791 | if (handler == iconveh_escape_sequence) | |
792 | { | |
793 | static char hex[16] = "0123456789ABCDEF"; | |
794 | scratchlen = 0; | |
795 | scratchbuf[scratchlen++] = '\\'; | |
796 | if (uc < 0x10000) | |
797 | scratchbuf[scratchlen++] = 'u'; | |
798 | else | |
799 | { | |
800 | scratchbuf[scratchlen++] = 'U'; | |
801 | scratchbuf[scratchlen++] = hex[(uc>>28) & 15]; | |
802 | scratchbuf[scratchlen++] = hex[(uc>>24) & 15]; | |
803 | scratchbuf[scratchlen++] = hex[(uc>>20) & 15]; | |
804 | scratchbuf[scratchlen++] = hex[(uc>>16) & 15]; | |
805 | } | |
806 | scratchbuf[scratchlen++] = hex[(uc>>12) & 15]; | |
807 | scratchbuf[scratchlen++] = hex[(uc>>8) & 15]; | |
808 | scratchbuf[scratchlen++] = hex[(uc>>4) & 15]; | |
809 | scratchbuf[scratchlen++] = hex[uc & 15]; | |
810 | } | |
811 | else | |
812 | { | |
813 | scratchbuf[0] = '?'; | |
814 | scratchlen = 1; | |
815 | } | |
816 | ||
817 | inptr = scratchbuf; | |
818 | insize = scratchlen; | |
819 | if (cd2 != (iconv_t)(-1)) | |
820 | res = iconv (cd2, | |
821 | (ICONV_CONST char **) &inptr, &insize, | |
822 | &out2ptr, &out2size); | |
823 | else | |
824 | { | |
825 | /* TO_CODESET is UTF-8. */ | |
826 | if (out2size >= insize) | |
827 | { | |
828 | memcpy (out2ptr, inptr, insize); | |
829 | out2ptr += insize; | |
830 | out2size -= insize; | |
831 | inptr += insize; | |
832 | insize = 0; | |
833 | res = 0; | |
834 | } | |
835 | else | |
836 | { | |
837 | errno = E2BIG; | |
838 | res = (size_t)(-1); | |
839 | } | |
840 | } | |
841 | length = out2ptr - result; | |
842 | if (res == (size_t)(-1) && errno == E2BIG) | |
843 | { | |
844 | char *memory; | |
845 | ||
846 | allocated = 2 * allocated; | |
847 | if (length + 1 + extra_alloc > allocated) | |
848 | abort (); | |
849 | if (result == initial_result) | |
850 | memory = (char *) malloc (allocated); | |
851 | else | |
852 | memory = (char *) realloc (result, allocated); | |
853 | if (memory == NULL) | |
854 | { | |
855 | if (result != initial_result) | |
856 | free (result); | |
857 | errno = ENOMEM; | |
858 | return -1; | |
859 | } | |
860 | if (result == initial_result) | |
861 | memcpy (memory, initial_result, length); | |
862 | result = memory; | |
863 | grow = false; | |
864 | ||
865 | out2ptr = result + length; | |
866 | out2size = allocated - extra_alloc - length; | |
867 | if (cd2 != (iconv_t)(-1)) | |
868 | res = iconv (cd2, | |
869 | (ICONV_CONST char **) &inptr, | |
870 | &insize, | |
871 | &out2ptr, &out2size); | |
872 | else | |
873 | { | |
874 | /* TO_CODESET is UTF-8. */ | |
875 | if (!(out2size >= insize)) | |
876 | abort (); | |
877 | memcpy (out2ptr, inptr, insize); | |
878 | out2ptr += insize; | |
879 | out2size -= insize; | |
880 | inptr += insize; | |
881 | insize = 0; | |
882 | res = 0; | |
883 | } | |
884 | length = out2ptr - result; | |
885 | } | |
886 | # if !defined _LIBICONV_VERSION && !defined __GLIBC__ | |
887 | /* Irix iconv() inserts a NUL byte if it cannot convert. | |
888 | NetBSD iconv() inserts a question mark if it cannot | |
889 | convert. | |
890 | Only GNU libiconv and GNU libc are known to prefer | |
891 | to fail rather than doing a lossy conversion. */ | |
892 | if (res != (size_t)(-1) && res > 0) | |
893 | { | |
894 | errno = EILSEQ; | |
895 | res = (size_t)(-1); | |
896 | } | |
897 | # endif | |
898 | if (res == (size_t)(-1)) | |
899 | { | |
900 | /* Failure converting the ASCII replacement. */ | |
901 | if (result != initial_result) | |
902 | { | |
903 | int saved_errno = errno; | |
904 | free (result); | |
905 | errno = saved_errno; | |
906 | } | |
907 | return -1; | |
908 | } | |
909 | } | |
910 | else | |
911 | { | |
912 | if (result != initial_result) | |
913 | { | |
914 | int saved_errno = errno; | |
915 | free (result); | |
916 | errno = saved_errno; | |
917 | } | |
918 | return -1; | |
919 | } | |
920 | } | |
921 | if (!(in2size > 0 | |
922 | || (in1size == 0 && !do_final_flush1 && do_final_flush2))) | |
923 | break; | |
924 | if (grow) | |
925 | { | |
926 | char *memory; | |
927 | ||
928 | allocated = 2 * allocated; | |
929 | if (result == initial_result) | |
930 | memory = (char *) malloc (allocated); | |
931 | else | |
932 | memory = (char *) realloc (result, allocated); | |
933 | if (memory == NULL) | |
934 | { | |
935 | if (result != initial_result) | |
936 | free (result); | |
937 | errno = ENOMEM; | |
938 | return -1; | |
939 | } | |
940 | if (result == initial_result) | |
941 | memcpy (memory, initial_result, length); | |
942 | result = memory; | |
943 | } | |
944 | } | |
945 | ||
946 | /* Move the remaining bytes to the beginning of utf8buf. */ | |
947 | if (in2size > 0) | |
948 | memmove (utf8buf, in2ptr, in2size); | |
949 | utf8len = in2size; | |
950 | } | |
951 | ||
952 | if (res1 == (size_t)(-1)) | |
953 | { | |
954 | if (errno1 == EINVAL) | |
955 | in1size = 0; | |
956 | else if (errno1 == EILSEQ) | |
957 | { | |
958 | if (result != initial_result) | |
959 | free (result); | |
960 | errno = errno1; | |
961 | return -1; | |
962 | } | |
963 | } | |
964 | } | |
965 | # undef utf8bufsize | |
966 | } | |
967 | ||
968 | done: | |
969 | /* Now the final memory allocation. */ | |
970 | if (result == tmpbuf) | |
971 | { | |
972 | size_t memsize = length + extra_alloc; | |
973 | char *memory; | |
974 | ||
975 | memory = (char *) malloc (memsize > 0 ? memsize : 1); | |
976 | if (memory != NULL) | |
977 | { | |
978 | memcpy (memory, tmpbuf, length); | |
979 | result = memory; | |
980 | } | |
981 | else | |
982 | { | |
983 | errno = ENOMEM; | |
984 | return -1; | |
985 | } | |
986 | } | |
987 | else if (result != *resultp && length + extra_alloc < allocated) | |
988 | { | |
989 | /* Shrink the allocated memory if possible. */ | |
990 | size_t memsize = length + extra_alloc; | |
991 | char *memory; | |
992 | ||
993 | memory = (char *) realloc (result, memsize > 0 ? memsize : 1); | |
994 | if (memory != NULL) | |
995 | result = memory; | |
996 | } | |
997 | *resultp = result; | |
998 | *lengthp = length; | |
999 | return 0; | |
1000 | # undef tmpbuf | |
1001 | # undef tmpbufsize | |
1002 | } | |
1003 | ||
1004 | int | |
1005 | mem_cd_iconveh (const char *src, size_t srclen, | |
8912421c | 1006 | const iconveh_t *cd, |
24d56127 LC |
1007 | enum iconv_ilseq_handler handler, |
1008 | size_t *offsets, | |
1009 | char **resultp, size_t *lengthp) | |
1010 | { | |
8912421c LC |
1011 | return mem_cd_iconveh_internal (src, srclen, cd->cd, cd->cd1, cd->cd2, |
1012 | handler, 0, offsets, resultp, lengthp); | |
24d56127 LC |
1013 | } |
1014 | ||
1015 | char * | |
1016 | str_cd_iconveh (const char *src, | |
8912421c | 1017 | const iconveh_t *cd, |
24d56127 LC |
1018 | enum iconv_ilseq_handler handler) |
1019 | { | |
1020 | /* For most encodings, a trailing NUL byte in the input will be converted | |
1021 | to a trailing NUL byte in the output. But not for UTF-7. So that this | |
1022 | function is usable for UTF-7, we have to exclude the NUL byte from the | |
1023 | conversion and add it by hand afterwards. */ | |
1024 | char *result = NULL; | |
1025 | size_t length = 0; | |
1026 | int retval = mem_cd_iconveh_internal (src, strlen (src), | |
8912421c LC |
1027 | cd->cd, cd->cd1, cd->cd2, handler, 1, |
1028 | NULL, &result, &length); | |
24d56127 LC |
1029 | |
1030 | if (retval < 0) | |
1031 | { | |
1032 | if (result != NULL) | |
1033 | { | |
1034 | int saved_errno = errno; | |
1035 | free (result); | |
1036 | errno = saved_errno; | |
1037 | } | |
1038 | return NULL; | |
1039 | } | |
1040 | ||
1041 | /* Add the terminating NUL byte. */ | |
1042 | result[length] = '\0'; | |
1043 | ||
1044 | return result; | |
1045 | } | |
1046 | ||
1047 | #endif | |
1048 | ||
1049 | int | |
1050 | mem_iconveh (const char *src, size_t srclen, | |
1051 | const char *from_codeset, const char *to_codeset, | |
1052 | enum iconv_ilseq_handler handler, | |
1053 | size_t *offsets, | |
1054 | char **resultp, size_t *lengthp) | |
1055 | { | |
1056 | if (srclen == 0) | |
1057 | { | |
1058 | /* Nothing to convert. */ | |
1059 | *lengthp = 0; | |
1060 | return 0; | |
1061 | } | |
1062 | else if (offsets == NULL && c_strcasecmp (from_codeset, to_codeset) == 0) | |
1063 | { | |
1064 | char *result; | |
1065 | ||
1066 | if (*resultp != NULL && *lengthp >= srclen) | |
1067 | result = *resultp; | |
1068 | else | |
1069 | { | |
1070 | result = (char *) malloc (srclen); | |
1071 | if (result == NULL) | |
1072 | { | |
1073 | errno = ENOMEM; | |
1074 | return -1; | |
1075 | } | |
1076 | } | |
1077 | memcpy (result, src, srclen); | |
1078 | *resultp = result; | |
1079 | *lengthp = srclen; | |
1080 | return 0; | |
1081 | } | |
1082 | else | |
1083 | { | |
1084 | #if HAVE_ICONV | |
8912421c | 1085 | iconveh_t cd; |
24d56127 LC |
1086 | char *result; |
1087 | size_t length; | |
1088 | int retval; | |
1089 | ||
8912421c LC |
1090 | if (iconveh_open (to_codeset, from_codeset, &cd) < 0) |
1091 | return -1; | |
24d56127 LC |
1092 | |
1093 | result = *resultp; | |
1094 | length = *lengthp; | |
8912421c | 1095 | retval = mem_cd_iconveh (src, srclen, &cd, handler, offsets, |
24d56127 LC |
1096 | &result, &length); |
1097 | ||
1098 | if (retval < 0) | |
1099 | { | |
8912421c | 1100 | /* Close cd, but preserve the errno from str_cd_iconv. */ |
24d56127 | 1101 | int saved_errno = errno; |
8912421c | 1102 | iconveh_close (&cd); |
24d56127 LC |
1103 | errno = saved_errno; |
1104 | } | |
1105 | else | |
1106 | { | |
8912421c | 1107 | if (iconveh_close (&cd) < 0) |
24d56127 LC |
1108 | { |
1109 | /* Return -1, but free the allocated memory, and while doing | |
8912421c | 1110 | that, preserve the errno from iconveh_close. */ |
24d56127 LC |
1111 | int saved_errno = errno; |
1112 | if (result != *resultp && result != NULL) | |
1113 | free (result); | |
1114 | errno = saved_errno; | |
1115 | return -1; | |
1116 | } | |
1117 | *resultp = result; | |
1118 | *lengthp = length; | |
1119 | } | |
1120 | return retval; | |
1121 | #else | |
1122 | /* This is a different error code than if iconv_open existed but didn't | |
1123 | support from_codeset and to_codeset, so that the caller can emit | |
1124 | an error message such as | |
1125 | "iconv() is not supported. Installing GNU libiconv and | |
1126 | then reinstalling this package would fix this." */ | |
1127 | errno = ENOSYS; | |
1128 | return -1; | |
1129 | #endif | |
1130 | } | |
1131 | } | |
1132 | ||
1133 | char * | |
1134 | str_iconveh (const char *src, | |
1135 | const char *from_codeset, const char *to_codeset, | |
1136 | enum iconv_ilseq_handler handler) | |
1137 | { | |
1138 | if (*src == '\0' || c_strcasecmp (from_codeset, to_codeset) == 0) | |
1139 | { | |
1140 | char *result = strdup (src); | |
1141 | ||
1142 | if (result == NULL) | |
1143 | errno = ENOMEM; | |
1144 | return result; | |
1145 | } | |
1146 | else | |
1147 | { | |
1148 | #if HAVE_ICONV | |
8912421c | 1149 | iconveh_t cd; |
24d56127 LC |
1150 | char *result; |
1151 | ||
8912421c LC |
1152 | if (iconveh_open (to_codeset, from_codeset, &cd) < 0) |
1153 | return NULL; | |
24d56127 | 1154 | |
8912421c | 1155 | result = str_cd_iconveh (src, &cd, handler); |
24d56127 LC |
1156 | |
1157 | if (result == NULL) | |
1158 | { | |
8912421c | 1159 | /* Close cd, but preserve the errno from str_cd_iconv. */ |
24d56127 | 1160 | int saved_errno = errno; |
8912421c | 1161 | iconveh_close (&cd); |
24d56127 LC |
1162 | errno = saved_errno; |
1163 | } | |
1164 | else | |
1165 | { | |
8912421c | 1166 | if (iconveh_close (&cd) < 0) |
24d56127 LC |
1167 | { |
1168 | /* Return NULL, but free the allocated memory, and while doing | |
8912421c | 1169 | that, preserve the errno from iconveh_close. */ |
24d56127 LC |
1170 | int saved_errno = errno; |
1171 | free (result); | |
1172 | errno = saved_errno; | |
1173 | return NULL; | |
1174 | } | |
1175 | } | |
1176 | return result; | |
1177 | #else | |
1178 | /* This is a different error code than if iconv_open existed but didn't | |
1179 | support from_codeset and to_codeset, so that the caller can emit | |
1180 | an error message such as | |
1181 | "iconv() is not supported. Installing GNU libiconv and | |
1182 | then reinstalling this package would fix this." */ | |
1183 | errno = ENOSYS; | |
1184 | return NULL; | |
1185 | #endif | |
1186 | } | |
1187 | } |