Commit | Line | Data |
---|---|---|
24d56127 | 1 | /* Character set conversion with error handling. |
f0007cad | 2 | Copyright (C) 2001-2012 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. */ | |
0f00f2c3 LC |
55 | # if ((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) && !defined __UCLIBC__) \ |
56 | && !defined _LIBICONV_VERSION | |
8912421c LC |
57 | if (c_strcasecmp (from_codeset, "EUC-KR") == 0 |
58 | || c_strcasecmp (to_codeset, "EUC-KR") == 0) | |
59 | { | |
60 | errno = EINVAL; | |
61 | return -1; | |
62 | } | |
63 | # endif | |
64 | ||
65 | cd = iconv_open (to_codeset, from_codeset); | |
66 | ||
67 | if (STRCASEEQ (from_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0)) | |
68 | cd1 = (iconv_t)(-1); | |
69 | else | |
70 | { | |
71 | cd1 = iconv_open ("UTF-8", from_codeset); | |
72 | if (cd1 == (iconv_t)(-1)) | |
1cd4fffc LC |
73 | { |
74 | int saved_errno = errno; | |
75 | if (cd != (iconv_t)(-1)) | |
76 | iconv_close (cdp->cd); | |
77 | errno = saved_errno; | |
78 | return -1; | |
79 | } | |
8912421c LC |
80 | } |
81 | ||
82 | if (STRCASEEQ (to_codeset, "UTF-8", 'U','T','F','-','8',0,0,0,0) | |
0f00f2c3 LC |
83 | # if (((__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2) \ |
84 | && !defined __UCLIBC__) \ | |
85 | || _LIBICONV_VERSION >= 0x0105 | |
8912421c LC |
86 | || c_strcasecmp (to_codeset, "UTF-8//TRANSLIT") == 0 |
87 | # endif | |
88 | ) | |
89 | cd2 = (iconv_t)(-1); | |
90 | else | |
91 | { | |
92 | cd2 = iconv_open (to_codeset, "UTF-8"); | |
93 | if (cd2 == (iconv_t)(-1)) | |
1cd4fffc LC |
94 | { |
95 | int saved_errno = errno; | |
96 | if (cd1 != (iconv_t)(-1)) | |
97 | iconv_close (cd1); | |
98 | if (cd != (iconv_t)(-1)) | |
99 | iconv_close (cd); | |
100 | errno = saved_errno; | |
101 | return -1; | |
102 | } | |
8912421c LC |
103 | } |
104 | ||
105 | cdp->cd = cd; | |
106 | cdp->cd1 = cd1; | |
107 | cdp->cd2 = cd2; | |
108 | return 0; | |
109 | } | |
110 | ||
111 | int | |
112 | iconveh_close (const iconveh_t *cd) | |
113 | { | |
114 | if (cd->cd2 != (iconv_t)(-1) && iconv_close (cd->cd2) < 0) | |
115 | { | |
116 | /* Return -1, but preserve the errno from iconv_close. */ | |
117 | int saved_errno = errno; | |
118 | if (cd->cd1 != (iconv_t)(-1)) | |
1cd4fffc | 119 | iconv_close (cd->cd1); |
8912421c | 120 | if (cd->cd != (iconv_t)(-1)) |
1cd4fffc | 121 | iconv_close (cd->cd); |
8912421c LC |
122 | errno = saved_errno; |
123 | return -1; | |
124 | } | |
125 | if (cd->cd1 != (iconv_t)(-1) && iconv_close (cd->cd1) < 0) | |
126 | { | |
127 | /* Return -1, but preserve the errno from iconv_close. */ | |
128 | int saved_errno = errno; | |
129 | if (cd->cd != (iconv_t)(-1)) | |
1cd4fffc | 130 | iconv_close (cd->cd); |
8912421c LC |
131 | errno = saved_errno; |
132 | return -1; | |
133 | } | |
134 | if (cd->cd != (iconv_t)(-1) && iconv_close (cd->cd) < 0) | |
135 | return -1; | |
136 | return 0; | |
137 | } | |
24d56127 LC |
138 | |
139 | /* iconv_carefully is like iconv, except that it stops as soon as it encounters | |
140 | a conversion error, and it returns in *INCREMENTED a boolean telling whether | |
141 | it has incremented the input pointers past the error location. */ | |
0f00f2c3 | 142 | # if !defined _LIBICONV_VERSION && !(defined __GLIBC__ && !defined __UCLIBC__) |
24d56127 LC |
143 | /* Irix iconv() inserts a NUL byte if it cannot convert. |
144 | NetBSD iconv() inserts a question mark if it cannot convert. | |
145 | Only GNU libiconv and GNU libc are known to prefer to fail rather | |
146 | than doing a lossy conversion. */ | |
147 | static size_t | |
148 | iconv_carefully (iconv_t cd, | |
1cd4fffc LC |
149 | const char **inbuf, size_t *inbytesleft, |
150 | char **outbuf, size_t *outbytesleft, | |
151 | bool *incremented) | |
24d56127 LC |
152 | { |
153 | const char *inptr = *inbuf; | |
154 | const char *inptr_end = inptr + *inbytesleft; | |
155 | char *outptr = *outbuf; | |
156 | size_t outsize = *outbytesleft; | |
157 | const char *inptr_before; | |
158 | size_t res; | |
159 | ||
160 | do | |
161 | { | |
162 | size_t insize; | |
163 | ||
164 | inptr_before = inptr; | |
165 | res = (size_t)(-1); | |
166 | ||
167 | for (insize = 1; inptr + insize <= inptr_end; insize++) | |
1cd4fffc LC |
168 | { |
169 | res = iconv (cd, | |
170 | (ICONV_CONST char **) &inptr, &insize, | |
171 | &outptr, &outsize); | |
172 | if (!(res == (size_t)(-1) && errno == EINVAL)) | |
173 | break; | |
174 | /* iconv can eat up a shift sequence but give EINVAL while attempting | |
175 | to convert the first character. E.g. libiconv does this. */ | |
176 | if (inptr > inptr_before) | |
177 | { | |
178 | res = 0; | |
179 | break; | |
180 | } | |
181 | } | |
24d56127 LC |
182 | |
183 | if (res == 0) | |
1cd4fffc LC |
184 | { |
185 | *outbuf = outptr; | |
186 | *outbytesleft = outsize; | |
187 | } | |
24d56127 LC |
188 | } |
189 | while (res == 0 && inptr < inptr_end); | |
190 | ||
191 | *inbuf = inptr; | |
192 | *inbytesleft = inptr_end - inptr; | |
193 | if (res != (size_t)(-1) && res > 0) | |
194 | { | |
195 | /* iconv() has already incremented INPTR. We cannot go back to a | |
1cd4fffc LC |
196 | previous INPTR, otherwise the state inside CD would become invalid, |
197 | if FROM_CODESET is a stateful encoding. So, tell the caller that | |
198 | *INBUF has already been incremented. */ | |
24d56127 LC |
199 | *incremented = (inptr > inptr_before); |
200 | errno = EILSEQ; | |
201 | return (size_t)(-1); | |
202 | } | |
203 | else | |
204 | { | |
205 | *incremented = false; | |
206 | return res; | |
207 | } | |
208 | } | |
209 | # else | |
210 | # define iconv_carefully(cd, inbuf, inbytesleft, outbuf, outbytesleft, incremented) \ | |
211 | (*(incremented) = false, \ | |
212 | iconv (cd, (ICONV_CONST char **) (inbuf), inbytesleft, outbuf, outbytesleft)) | |
213 | # endif | |
214 | ||
215 | /* iconv_carefully_1 is like iconv_carefully, except that it stops after | |
216 | converting one character or one shift sequence. */ | |
217 | static size_t | |
218 | iconv_carefully_1 (iconv_t cd, | |
1cd4fffc LC |
219 | const char **inbuf, size_t *inbytesleft, |
220 | char **outbuf, size_t *outbytesleft, | |
221 | bool *incremented) | |
24d56127 LC |
222 | { |
223 | const char *inptr_before = *inbuf; | |
224 | const char *inptr = inptr_before; | |
225 | const char *inptr_end = inptr_before + *inbytesleft; | |
226 | char *outptr = *outbuf; | |
227 | size_t outsize = *outbytesleft; | |
228 | size_t res = (size_t)(-1); | |
229 | size_t insize; | |
230 | ||
231 | for (insize = 1; inptr_before + insize <= inptr_end; insize++) | |
232 | { | |
233 | inptr = inptr_before; | |
234 | res = iconv (cd, | |
1cd4fffc LC |
235 | (ICONV_CONST char **) &inptr, &insize, |
236 | &outptr, &outsize); | |
24d56127 | 237 | if (!(res == (size_t)(-1) && errno == EINVAL)) |
1cd4fffc | 238 | break; |
24d56127 | 239 | /* iconv can eat up a shift sequence but give EINVAL while attempting |
1cd4fffc | 240 | to convert the first character. E.g. libiconv does this. */ |
24d56127 | 241 | if (inptr > inptr_before) |
1cd4fffc LC |
242 | { |
243 | res = 0; | |
244 | break; | |
245 | } | |
24d56127 LC |
246 | } |
247 | ||
248 | *inbuf = inptr; | |
249 | *inbytesleft = inptr_end - inptr; | |
0f00f2c3 | 250 | # if !defined _LIBICONV_VERSION && !(defined __GLIBC__ && !defined __UCLIBC__) |
24d56127 LC |
251 | /* Irix iconv() inserts a NUL byte if it cannot convert. |
252 | NetBSD iconv() inserts a question mark if it cannot convert. | |
253 | Only GNU libiconv and GNU libc are known to prefer to fail rather | |
254 | than doing a lossy conversion. */ | |
255 | if (res != (size_t)(-1) && res > 0) | |
256 | { | |
257 | /* iconv() has already incremented INPTR. We cannot go back to a | |
1cd4fffc LC |
258 | previous INPTR, otherwise the state inside CD would become invalid, |
259 | if FROM_CODESET is a stateful encoding. So, tell the caller that | |
260 | *INBUF has already been incremented. */ | |
24d56127 LC |
261 | *incremented = (inptr > inptr_before); |
262 | errno = EILSEQ; | |
263 | return (size_t)(-1); | |
264 | } | |
265 | # endif | |
266 | ||
267 | if (res != (size_t)(-1)) | |
268 | { | |
269 | *outbuf = outptr; | |
270 | *outbytesleft = outsize; | |
271 | } | |
272 | *incremented = false; | |
273 | return res; | |
274 | } | |
275 | ||
276 | /* utf8conv_carefully is like iconv, except that | |
277 | - it converts from UTF-8 to UTF-8, | |
278 | - it stops as soon as it encounters a conversion error, and it returns | |
279 | in *INCREMENTED a boolean telling whether it has incremented the input | |
280 | pointers past the error location, | |
281 | - if one_character_only is true, it stops after converting one | |
282 | character. */ | |
283 | static size_t | |
284 | utf8conv_carefully (bool one_character_only, | |
1cd4fffc LC |
285 | const char **inbuf, size_t *inbytesleft, |
286 | char **outbuf, size_t *outbytesleft, | |
287 | bool *incremented) | |
24d56127 LC |
288 | { |
289 | const char *inptr = *inbuf; | |
290 | size_t insize = *inbytesleft; | |
291 | char *outptr = *outbuf; | |
292 | size_t outsize = *outbytesleft; | |
293 | size_t res; | |
294 | ||
295 | res = 0; | |
296 | do | |
297 | { | |
298 | ucs4_t uc; | |
299 | int n; | |
300 | int m; | |
301 | ||
302 | n = u8_mbtoucr (&uc, (const uint8_t *) inptr, insize); | |
303 | if (n < 0) | |
1cd4fffc LC |
304 | { |
305 | errno = (n == -2 ? EINVAL : EILSEQ); | |
306 | n = u8_mbtouc (&uc, (const uint8_t *) inptr, insize); | |
307 | inptr += n; | |
308 | insize -= n; | |
309 | res = (size_t)(-1); | |
310 | *incremented = true; | |
311 | break; | |
312 | } | |
24d56127 | 313 | if (outsize == 0) |
1cd4fffc LC |
314 | { |
315 | errno = E2BIG; | |
316 | res = (size_t)(-1); | |
317 | *incremented = false; | |
318 | break; | |
319 | } | |
24d56127 LC |
320 | m = u8_uctomb ((uint8_t *) outptr, uc, outsize); |
321 | if (m == -2) | |
1cd4fffc LC |
322 | { |
323 | errno = E2BIG; | |
324 | res = (size_t)(-1); | |
325 | *incremented = false; | |
326 | break; | |
327 | } | |
24d56127 LC |
328 | inptr += n; |
329 | insize -= n; | |
330 | if (m == -1) | |
1cd4fffc LC |
331 | { |
332 | errno = EILSEQ; | |
333 | res = (size_t)(-1); | |
334 | *incremented = true; | |
335 | break; | |
336 | } | |
24d56127 LC |
337 | outptr += m; |
338 | outsize -= m; | |
339 | } | |
340 | while (!one_character_only && insize > 0); | |
341 | ||
342 | *inbuf = inptr; | |
343 | *inbytesleft = insize; | |
344 | *outbuf = outptr; | |
345 | *outbytesleft = outsize; | |
346 | return res; | |
347 | } | |
348 | ||
349 | static int | |
350 | mem_cd_iconveh_internal (const char *src, size_t srclen, | |
1cd4fffc LC |
351 | iconv_t cd, iconv_t cd1, iconv_t cd2, |
352 | enum iconv_ilseq_handler handler, | |
353 | size_t extra_alloc, | |
354 | size_t *offsets, | |
355 | char **resultp, size_t *lengthp) | |
24d56127 LC |
356 | { |
357 | /* When a conversion error occurs, we cannot start using CD1 and CD2 at | |
358 | this point: FROM_CODESET may be a stateful encoding like ISO-2022-KR. | |
359 | Instead, we have to start afresh from the beginning of SRC. */ | |
360 | /* Use a temporary buffer, so that for small strings, a single malloc() | |
361 | call will be sufficient. */ | |
362 | # define tmpbufsize 4096 | |
363 | /* The alignment is needed when converting e.g. to glibc's WCHAR_T or | |
364 | libiconv's UCS-4-INTERNAL encoding. */ | |
365 | union { unsigned int align; char buf[tmpbufsize]; } tmp; | |
366 | # define tmpbuf tmp.buf | |
367 | ||
368 | char *initial_result; | |
369 | char *result; | |
370 | size_t allocated; | |
371 | size_t length; | |
372 | size_t last_length = (size_t)(-1); /* only needed if offsets != NULL */ | |
373 | ||
374 | if (*resultp != NULL && *lengthp >= sizeof (tmpbuf)) | |
375 | { | |
376 | initial_result = *resultp; | |
377 | allocated = *lengthp; | |
378 | } | |
379 | else | |
380 | { | |
381 | initial_result = tmpbuf; | |
382 | allocated = sizeof (tmpbuf); | |
383 | } | |
384 | result = initial_result; | |
385 | ||
386 | /* Test whether a direct conversion is possible at all. */ | |
387 | if (cd == (iconv_t)(-1)) | |
388 | goto indirectly; | |
389 | ||
390 | if (offsets != NULL) | |
391 | { | |
392 | size_t i; | |
393 | ||
394 | for (i = 0; i < srclen; i++) | |
1cd4fffc | 395 | offsets[i] = (size_t)(-1); |
24d56127 LC |
396 | |
397 | last_length = (size_t)(-1); | |
398 | } | |
399 | length = 0; | |
400 | ||
401 | /* First, try a direct conversion, and see whether a conversion error | |
402 | occurs at all. */ | |
403 | { | |
404 | const char *inptr = src; | |
405 | size_t insize = srclen; | |
406 | ||
407 | /* Avoid glibc-2.1 bug and Solaris 2.7-2.9 bug. */ | |
408 | # if defined _LIBICONV_VERSION \ | |
0f00f2c3 LC |
409 | || !(((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) && !defined __UCLIBC__) \ |
410 | || defined __sun) | |
24d56127 LC |
411 | /* Set to the initial state. */ |
412 | iconv (cd, NULL, NULL, NULL, NULL); | |
413 | # endif | |
414 | ||
415 | while (insize > 0) | |
416 | { | |
1cd4fffc LC |
417 | char *outptr = result + length; |
418 | size_t outsize = allocated - extra_alloc - length; | |
419 | bool incremented; | |
420 | size_t res; | |
421 | bool grow; | |
422 | ||
423 | if (offsets != NULL) | |
424 | { | |
425 | if (length != last_length) /* ensure that offset[] be increasing */ | |
426 | { | |
427 | offsets[inptr - src] = length; | |
428 | last_length = length; | |
429 | } | |
430 | res = iconv_carefully_1 (cd, | |
431 | &inptr, &insize, | |
432 | &outptr, &outsize, | |
433 | &incremented); | |
434 | } | |
435 | else | |
436 | /* Use iconv_carefully instead of iconv here, because: | |
437 | - If TO_CODESET is UTF-8, we can do the error handling in this | |
438 | loop, no need for a second loop, | |
439 | - With iconv() implementations other than GNU libiconv and GNU | |
440 | libc, if we use iconv() in a big swoop, checking for an E2BIG | |
441 | return, we lose the number of irreversible conversions. */ | |
442 | res = iconv_carefully (cd, | |
443 | &inptr, &insize, | |
444 | &outptr, &outsize, | |
445 | &incremented); | |
446 | ||
447 | length = outptr - result; | |
448 | grow = (length + extra_alloc > allocated / 2); | |
449 | if (res == (size_t)(-1)) | |
450 | { | |
451 | if (errno == E2BIG) | |
452 | grow = true; | |
453 | else if (errno == EINVAL) | |
454 | break; | |
455 | else if (errno == EILSEQ && handler != iconveh_error) | |
456 | { | |
457 | if (cd2 == (iconv_t)(-1)) | |
458 | { | |
459 | /* TO_CODESET is UTF-8. */ | |
460 | /* Error handling can produce up to 1 byte of output. */ | |
461 | if (length + 1 + extra_alloc > allocated) | |
462 | { | |
463 | char *memory; | |
464 | ||
465 | allocated = 2 * allocated; | |
466 | if (length + 1 + extra_alloc > allocated) | |
467 | abort (); | |
468 | if (result == initial_result) | |
469 | memory = (char *) malloc (allocated); | |
470 | else | |
471 | memory = (char *) realloc (result, allocated); | |
472 | if (memory == NULL) | |
473 | { | |
474 | if (result != initial_result) | |
475 | free (result); | |
476 | errno = ENOMEM; | |
477 | return -1; | |
478 | } | |
479 | if (result == initial_result) | |
480 | memcpy (memory, initial_result, length); | |
481 | result = memory; | |
482 | grow = false; | |
483 | } | |
484 | /* The input is invalid in FROM_CODESET. Eat up one byte | |
485 | and emit a question mark. */ | |
486 | if (!incremented) | |
487 | { | |
488 | if (insize == 0) | |
489 | abort (); | |
490 | inptr++; | |
491 | insize--; | |
492 | } | |
493 | result[length] = '?'; | |
494 | length++; | |
495 | } | |
496 | else | |
497 | goto indirectly; | |
498 | } | |
499 | else | |
500 | { | |
501 | if (result != initial_result) | |
502 | { | |
503 | int saved_errno = errno; | |
504 | free (result); | |
505 | errno = saved_errno; | |
506 | } | |
507 | return -1; | |
508 | } | |
509 | } | |
510 | if (insize == 0) | |
511 | break; | |
512 | if (grow) | |
513 | { | |
514 | char *memory; | |
515 | ||
516 | allocated = 2 * allocated; | |
517 | if (result == initial_result) | |
518 | memory = (char *) malloc (allocated); | |
519 | else | |
520 | memory = (char *) realloc (result, allocated); | |
521 | if (memory == NULL) | |
522 | { | |
523 | if (result != initial_result) | |
524 | free (result); | |
525 | errno = ENOMEM; | |
526 | return -1; | |
527 | } | |
528 | if (result == initial_result) | |
529 | memcpy (memory, initial_result, length); | |
530 | result = memory; | |
531 | } | |
24d56127 LC |
532 | } |
533 | } | |
534 | ||
535 | /* Now get the conversion state back to the initial state. | |
536 | But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
537 | #if defined _LIBICONV_VERSION \ | |
0f00f2c3 LC |
538 | || !(((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) && !defined __UCLIBC__) \ |
539 | || defined __sun) | |
24d56127 LC |
540 | for (;;) |
541 | { | |
542 | char *outptr = result + length; | |
543 | size_t outsize = allocated - extra_alloc - length; | |
544 | size_t res; | |
545 | ||
546 | res = iconv (cd, NULL, NULL, &outptr, &outsize); | |
547 | length = outptr - result; | |
548 | if (res == (size_t)(-1)) | |
1cd4fffc LC |
549 | { |
550 | if (errno == E2BIG) | |
551 | { | |
552 | char *memory; | |
553 | ||
554 | allocated = 2 * allocated; | |
555 | if (result == initial_result) | |
556 | memory = (char *) malloc (allocated); | |
557 | else | |
558 | memory = (char *) realloc (result, allocated); | |
559 | if (memory == NULL) | |
560 | { | |
561 | if (result != initial_result) | |
562 | free (result); | |
563 | errno = ENOMEM; | |
564 | return -1; | |
565 | } | |
566 | if (result == initial_result) | |
567 | memcpy (memory, initial_result, length); | |
568 | result = memory; | |
569 | } | |
570 | else | |
571 | { | |
572 | if (result != initial_result) | |
573 | { | |
574 | int saved_errno = errno; | |
575 | free (result); | |
576 | errno = saved_errno; | |
577 | } | |
578 | return -1; | |
579 | } | |
580 | } | |
24d56127 | 581 | else |
1cd4fffc | 582 | break; |
24d56127 LC |
583 | } |
584 | #endif | |
585 | ||
586 | /* The direct conversion succeeded. */ | |
587 | goto done; | |
588 | ||
589 | indirectly: | |
590 | /* The direct conversion failed. | |
591 | Use a conversion through UTF-8. */ | |
592 | if (offsets != NULL) | |
593 | { | |
594 | size_t i; | |
595 | ||
596 | for (i = 0; i < srclen; i++) | |
1cd4fffc | 597 | offsets[i] = (size_t)(-1); |
24d56127 LC |
598 | |
599 | last_length = (size_t)(-1); | |
600 | } | |
601 | length = 0; | |
602 | { | |
603 | const bool slowly = (offsets != NULL || handler == iconveh_error); | |
604 | # define utf8bufsize 4096 /* may also be smaller or larger than tmpbufsize */ | |
605 | char utf8buf[utf8bufsize + 1]; | |
606 | size_t utf8len = 0; | |
607 | const char *in1ptr = src; | |
608 | size_t in1size = srclen; | |
609 | bool do_final_flush1 = true; | |
610 | bool do_final_flush2 = true; | |
611 | ||
612 | /* Avoid glibc-2.1 bug and Solaris 2.7-2.9 bug. */ | |
613 | # if defined _LIBICONV_VERSION \ | |
0f00f2c3 LC |
614 | || !(((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) && !defined __UCLIBC__) \ |
615 | || defined __sun) | |
24d56127 LC |
616 | /* Set to the initial state. */ |
617 | if (cd1 != (iconv_t)(-1)) | |
618 | iconv (cd1, NULL, NULL, NULL, NULL); | |
619 | if (cd2 != (iconv_t)(-1)) | |
620 | iconv (cd2, NULL, NULL, NULL, NULL); | |
621 | # endif | |
622 | ||
623 | while (in1size > 0 || do_final_flush1 || utf8len > 0 || do_final_flush2) | |
624 | { | |
1cd4fffc LC |
625 | char *out1ptr = utf8buf + utf8len; |
626 | size_t out1size = utf8bufsize - utf8len; | |
627 | bool incremented1; | |
628 | size_t res1; | |
629 | int errno1; | |
630 | ||
631 | /* Conversion step 1: from FROM_CODESET to UTF-8. */ | |
632 | if (in1size > 0) | |
633 | { | |
634 | if (offsets != NULL | |
635 | && length != last_length) /* ensure that offset[] be increasing */ | |
636 | { | |
637 | offsets[in1ptr - src] = length; | |
638 | last_length = length; | |
639 | } | |
640 | if (cd1 != (iconv_t)(-1)) | |
641 | { | |
642 | if (slowly) | |
643 | res1 = iconv_carefully_1 (cd1, | |
644 | &in1ptr, &in1size, | |
645 | &out1ptr, &out1size, | |
646 | &incremented1); | |
647 | else | |
648 | res1 = iconv_carefully (cd1, | |
649 | &in1ptr, &in1size, | |
650 | &out1ptr, &out1size, | |
651 | &incremented1); | |
652 | } | |
653 | else | |
654 | { | |
655 | /* FROM_CODESET is UTF-8. */ | |
656 | res1 = utf8conv_carefully (slowly, | |
657 | &in1ptr, &in1size, | |
658 | &out1ptr, &out1size, | |
659 | &incremented1); | |
660 | } | |
661 | } | |
662 | else if (do_final_flush1) | |
663 | { | |
664 | /* Now get the conversion state of CD1 back to the initial state. | |
665 | But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
24d56127 | 666 | # if defined _LIBICONV_VERSION \ |
0f00f2c3 LC |
667 | || !(((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) && !defined __UCLIBC__) \ |
668 | || defined __sun) | |
1cd4fffc LC |
669 | if (cd1 != (iconv_t)(-1)) |
670 | res1 = iconv (cd1, NULL, NULL, &out1ptr, &out1size); | |
671 | else | |
24d56127 | 672 | # endif |
1cd4fffc LC |
673 | res1 = 0; |
674 | do_final_flush1 = false; | |
675 | incremented1 = true; | |
676 | } | |
677 | else | |
678 | { | |
679 | res1 = 0; | |
680 | incremented1 = true; | |
681 | } | |
682 | if (res1 == (size_t)(-1) | |
683 | && !(errno == E2BIG || errno == EINVAL || errno == EILSEQ)) | |
684 | { | |
685 | if (result != initial_result) | |
686 | { | |
687 | int saved_errno = errno; | |
688 | free (result); | |
689 | errno = saved_errno; | |
690 | } | |
691 | return -1; | |
692 | } | |
693 | if (res1 == (size_t)(-1) | |
694 | && errno == EILSEQ && handler != iconveh_error) | |
695 | { | |
696 | /* The input is invalid in FROM_CODESET. Eat up one byte and | |
697 | emit a question mark. Room for the question mark was allocated | |
698 | at the end of utf8buf. */ | |
699 | if (!incremented1) | |
700 | { | |
701 | if (in1size == 0) | |
702 | abort (); | |
703 | in1ptr++; | |
704 | in1size--; | |
705 | } | |
706 | *out1ptr++ = '?'; | |
707 | res1 = 0; | |
708 | } | |
709 | errno1 = errno; | |
710 | utf8len = out1ptr - utf8buf; | |
711 | ||
712 | if (offsets != NULL | |
713 | || in1size == 0 | |
714 | || utf8len > utf8bufsize / 2 | |
715 | || (res1 == (size_t)(-1) && errno1 == E2BIG)) | |
716 | { | |
717 | /* Conversion step 2: from UTF-8 to TO_CODESET. */ | |
718 | const char *in2ptr = utf8buf; | |
719 | size_t in2size = utf8len; | |
720 | ||
721 | while (in2size > 0 | |
722 | || (in1size == 0 && !do_final_flush1 && do_final_flush2)) | |
723 | { | |
724 | char *out2ptr = result + length; | |
725 | size_t out2size = allocated - extra_alloc - length; | |
726 | bool incremented2; | |
727 | size_t res2; | |
728 | bool grow; | |
729 | ||
730 | if (in2size > 0) | |
731 | { | |
732 | if (cd2 != (iconv_t)(-1)) | |
733 | res2 = iconv_carefully (cd2, | |
734 | &in2ptr, &in2size, | |
735 | &out2ptr, &out2size, | |
736 | &incremented2); | |
737 | else | |
738 | /* TO_CODESET is UTF-8. */ | |
739 | res2 = utf8conv_carefully (false, | |
740 | &in2ptr, &in2size, | |
741 | &out2ptr, &out2size, | |
742 | &incremented2); | |
743 | } | |
744 | else /* in1size == 0 && !do_final_flush1 | |
745 | && in2size == 0 && do_final_flush2 */ | |
746 | { | |
747 | /* Now get the conversion state of CD1 back to the initial | |
748 | state. But avoid glibc-2.1 bug and Solaris 2.7 bug. */ | |
24d56127 | 749 | # if defined _LIBICONV_VERSION \ |
0f00f2c3 LC |
750 | || !(((__GLIBC__ == 2 && __GLIBC_MINOR__ <= 1) && !defined __UCLIBC__) \ |
751 | || defined __sun) | |
1cd4fffc LC |
752 | if (cd2 != (iconv_t)(-1)) |
753 | res2 = iconv (cd2, NULL, NULL, &out2ptr, &out2size); | |
754 | else | |
24d56127 | 755 | # endif |
1cd4fffc LC |
756 | res2 = 0; |
757 | do_final_flush2 = false; | |
758 | incremented2 = true; | |
759 | } | |
760 | ||
761 | length = out2ptr - result; | |
762 | grow = (length + extra_alloc > allocated / 2); | |
763 | if (res2 == (size_t)(-1)) | |
764 | { | |
765 | if (errno == E2BIG) | |
766 | grow = true; | |
767 | else if (errno == EINVAL) | |
768 | break; | |
769 | else if (errno == EILSEQ && handler != iconveh_error) | |
770 | { | |
771 | /* Error handling can produce up to 10 bytes of ASCII | |
772 | output. But TO_CODESET may be UCS-2, UTF-16 or | |
773 | UCS-4, so use CD2 here as well. */ | |
774 | char scratchbuf[10]; | |
775 | size_t scratchlen; | |
776 | ucs4_t uc; | |
777 | const char *inptr; | |
778 | size_t insize; | |
779 | size_t res; | |
780 | ||
781 | if (incremented2) | |
782 | { | |
783 | if (u8_prev (&uc, (const uint8_t *) in2ptr, | |
784 | (const uint8_t *) utf8buf) | |
785 | == NULL) | |
786 | abort (); | |
787 | } | |
788 | else | |
789 | { | |
790 | int n; | |
791 | if (in2size == 0) | |
792 | abort (); | |
793 | n = u8_mbtouc_unsafe (&uc, (const uint8_t *) in2ptr, | |
794 | in2size); | |
795 | in2ptr += n; | |
796 | in2size -= n; | |
797 | } | |
798 | ||
799 | if (handler == iconveh_escape_sequence) | |
800 | { | |
801 | static char hex[16] = "0123456789ABCDEF"; | |
802 | scratchlen = 0; | |
803 | scratchbuf[scratchlen++] = '\\'; | |
804 | if (uc < 0x10000) | |
805 | scratchbuf[scratchlen++] = 'u'; | |
806 | else | |
807 | { | |
808 | scratchbuf[scratchlen++] = 'U'; | |
809 | scratchbuf[scratchlen++] = hex[(uc>>28) & 15]; | |
810 | scratchbuf[scratchlen++] = hex[(uc>>24) & 15]; | |
811 | scratchbuf[scratchlen++] = hex[(uc>>20) & 15]; | |
812 | scratchbuf[scratchlen++] = hex[(uc>>16) & 15]; | |
813 | } | |
814 | scratchbuf[scratchlen++] = hex[(uc>>12) & 15]; | |
815 | scratchbuf[scratchlen++] = hex[(uc>>8) & 15]; | |
816 | scratchbuf[scratchlen++] = hex[(uc>>4) & 15]; | |
817 | scratchbuf[scratchlen++] = hex[uc & 15]; | |
818 | } | |
819 | else | |
820 | { | |
821 | scratchbuf[0] = '?'; | |
822 | scratchlen = 1; | |
823 | } | |
824 | ||
825 | inptr = scratchbuf; | |
826 | insize = scratchlen; | |
827 | if (cd2 != (iconv_t)(-1)) | |
828 | res = iconv (cd2, | |
829 | (ICONV_CONST char **) &inptr, &insize, | |
830 | &out2ptr, &out2size); | |
831 | else | |
832 | { | |
833 | /* TO_CODESET is UTF-8. */ | |
834 | if (out2size >= insize) | |
835 | { | |
836 | memcpy (out2ptr, inptr, insize); | |
837 | out2ptr += insize; | |
838 | out2size -= insize; | |
839 | inptr += insize; | |
840 | insize = 0; | |
841 | res = 0; | |
842 | } | |
843 | else | |
844 | { | |
845 | errno = E2BIG; | |
846 | res = (size_t)(-1); | |
847 | } | |
848 | } | |
849 | length = out2ptr - result; | |
850 | if (res == (size_t)(-1) && errno == E2BIG) | |
851 | { | |
852 | char *memory; | |
853 | ||
854 | allocated = 2 * allocated; | |
855 | if (length + 1 + extra_alloc > allocated) | |
856 | abort (); | |
857 | if (result == initial_result) | |
858 | memory = (char *) malloc (allocated); | |
859 | else | |
860 | memory = (char *) realloc (result, allocated); | |
861 | if (memory == NULL) | |
862 | { | |
863 | if (result != initial_result) | |
864 | free (result); | |
865 | errno = ENOMEM; | |
866 | return -1; | |
867 | } | |
868 | if (result == initial_result) | |
869 | memcpy (memory, initial_result, length); | |
870 | result = memory; | |
871 | grow = false; | |
872 | ||
873 | out2ptr = result + length; | |
874 | out2size = allocated - extra_alloc - length; | |
875 | if (cd2 != (iconv_t)(-1)) | |
876 | res = iconv (cd2, | |
877 | (ICONV_CONST char **) &inptr, | |
878 | &insize, | |
879 | &out2ptr, &out2size); | |
880 | else | |
881 | { | |
882 | /* TO_CODESET is UTF-8. */ | |
883 | if (!(out2size >= insize)) | |
884 | abort (); | |
885 | memcpy (out2ptr, inptr, insize); | |
886 | out2ptr += insize; | |
887 | out2size -= insize; | |
888 | inptr += insize; | |
889 | insize = 0; | |
890 | res = 0; | |
891 | } | |
892 | length = out2ptr - result; | |
893 | } | |
0f00f2c3 | 894 | # if !defined _LIBICONV_VERSION && !(defined __GLIBC__ && !defined __UCLIBC__) |
1cd4fffc LC |
895 | /* Irix iconv() inserts a NUL byte if it cannot convert. |
896 | NetBSD iconv() inserts a question mark if it cannot | |
897 | convert. | |
898 | Only GNU libiconv and GNU libc are known to prefer | |
899 | to fail rather than doing a lossy conversion. */ | |
900 | if (res != (size_t)(-1) && res > 0) | |
901 | { | |
902 | errno = EILSEQ; | |
903 | res = (size_t)(-1); | |
904 | } | |
24d56127 | 905 | # endif |
1cd4fffc LC |
906 | if (res == (size_t)(-1)) |
907 | { | |
908 | /* Failure converting the ASCII replacement. */ | |
909 | if (result != initial_result) | |
910 | { | |
911 | int saved_errno = errno; | |
912 | free (result); | |
913 | errno = saved_errno; | |
914 | } | |
915 | return -1; | |
916 | } | |
917 | } | |
918 | else | |
919 | { | |
920 | if (result != initial_result) | |
921 | { | |
922 | int saved_errno = errno; | |
923 | free (result); | |
924 | errno = saved_errno; | |
925 | } | |
926 | return -1; | |
927 | } | |
928 | } | |
929 | if (!(in2size > 0 | |
930 | || (in1size == 0 && !do_final_flush1 && do_final_flush2))) | |
931 | break; | |
932 | if (grow) | |
933 | { | |
934 | char *memory; | |
935 | ||
936 | allocated = 2 * allocated; | |
937 | if (result == initial_result) | |
938 | memory = (char *) malloc (allocated); | |
939 | else | |
940 | memory = (char *) realloc (result, allocated); | |
941 | if (memory == NULL) | |
942 | { | |
943 | if (result != initial_result) | |
944 | free (result); | |
945 | errno = ENOMEM; | |
946 | return -1; | |
947 | } | |
948 | if (result == initial_result) | |
949 | memcpy (memory, initial_result, length); | |
950 | result = memory; | |
951 | } | |
952 | } | |
953 | ||
954 | /* Move the remaining bytes to the beginning of utf8buf. */ | |
955 | if (in2size > 0) | |
956 | memmove (utf8buf, in2ptr, in2size); | |
957 | utf8len = in2size; | |
958 | } | |
959 | ||
960 | if (res1 == (size_t)(-1)) | |
961 | { | |
962 | if (errno1 == EINVAL) | |
963 | in1size = 0; | |
964 | else if (errno1 == EILSEQ) | |
965 | { | |
966 | if (result != initial_result) | |
967 | free (result); | |
968 | errno = errno1; | |
969 | return -1; | |
970 | } | |
971 | } | |
24d56127 LC |
972 | } |
973 | # undef utf8bufsize | |
974 | } | |
975 | ||
976 | done: | |
977 | /* Now the final memory allocation. */ | |
978 | if (result == tmpbuf) | |
979 | { | |
980 | size_t memsize = length + extra_alloc; | |
24d56127 | 981 | |
9157d901 LC |
982 | if (*resultp != NULL && *lengthp >= memsize) |
983 | result = *resultp; | |
24d56127 | 984 | else |
1cd4fffc | 985 | { |
9157d901 LC |
986 | char *memory; |
987 | ||
988 | memory = (char *) malloc (memsize > 0 ? memsize : 1); | |
989 | if (memory != NULL) | |
990 | result = memory; | |
991 | else | |
992 | { | |
993 | errno = ENOMEM; | |
994 | return -1; | |
995 | } | |
24d56127 | 996 | } |
9157d901 | 997 | memcpy (result, tmpbuf, length); |
24d56127 LC |
998 | } |
999 | else if (result != *resultp && length + extra_alloc < allocated) | |
1000 | { | |
1001 | /* Shrink the allocated memory if possible. */ | |
1002 | size_t memsize = length + extra_alloc; | |
1003 | char *memory; | |
1004 | ||
1005 | memory = (char *) realloc (result, memsize > 0 ? memsize : 1); | |
1006 | if (memory != NULL) | |
1cd4fffc | 1007 | result = memory; |
24d56127 LC |
1008 | } |
1009 | *resultp = result; | |
1010 | *lengthp = length; | |
1011 | return 0; | |
1012 | # undef tmpbuf | |
1013 | # undef tmpbufsize | |
1014 | } | |
1015 | ||
1016 | int | |
1017 | mem_cd_iconveh (const char *src, size_t srclen, | |
1cd4fffc LC |
1018 | const iconveh_t *cd, |
1019 | enum iconv_ilseq_handler handler, | |
1020 | size_t *offsets, | |
1021 | char **resultp, size_t *lengthp) | |
24d56127 | 1022 | { |
8912421c | 1023 | return mem_cd_iconveh_internal (src, srclen, cd->cd, cd->cd1, cd->cd2, |
1cd4fffc | 1024 | handler, 0, offsets, resultp, lengthp); |
24d56127 LC |
1025 | } |
1026 | ||
1027 | char * | |
1028 | str_cd_iconveh (const char *src, | |
1cd4fffc LC |
1029 | const iconveh_t *cd, |
1030 | enum iconv_ilseq_handler handler) | |
24d56127 LC |
1031 | { |
1032 | /* For most encodings, a trailing NUL byte in the input will be converted | |
1033 | to a trailing NUL byte in the output. But not for UTF-7. So that this | |
1034 | function is usable for UTF-7, we have to exclude the NUL byte from the | |
1035 | conversion and add it by hand afterwards. */ | |
1036 | char *result = NULL; | |
1037 | size_t length = 0; | |
1038 | int retval = mem_cd_iconveh_internal (src, strlen (src), | |
1cd4fffc LC |
1039 | cd->cd, cd->cd1, cd->cd2, handler, 1, |
1040 | NULL, &result, &length); | |
24d56127 LC |
1041 | |
1042 | if (retval < 0) | |
1043 | { | |
1044 | if (result != NULL) | |
1cd4fffc LC |
1045 | { |
1046 | int saved_errno = errno; | |
1047 | free (result); | |
1048 | errno = saved_errno; | |
1049 | } | |
24d56127 LC |
1050 | return NULL; |
1051 | } | |
1052 | ||
1053 | /* Add the terminating NUL byte. */ | |
1054 | result[length] = '\0'; | |
1055 | ||
1056 | return result; | |
1057 | } | |
1058 | ||
1059 | #endif | |
1060 | ||
1061 | int | |
1062 | mem_iconveh (const char *src, size_t srclen, | |
1cd4fffc LC |
1063 | const char *from_codeset, const char *to_codeset, |
1064 | enum iconv_ilseq_handler handler, | |
1065 | size_t *offsets, | |
1066 | char **resultp, size_t *lengthp) | |
24d56127 LC |
1067 | { |
1068 | if (srclen == 0) | |
1069 | { | |
1070 | /* Nothing to convert. */ | |
1071 | *lengthp = 0; | |
1072 | return 0; | |
1073 | } | |
1074 | else if (offsets == NULL && c_strcasecmp (from_codeset, to_codeset) == 0) | |
1075 | { | |
1076 | char *result; | |
1077 | ||
1078 | if (*resultp != NULL && *lengthp >= srclen) | |
1cd4fffc | 1079 | result = *resultp; |
24d56127 | 1080 | else |
1cd4fffc LC |
1081 | { |
1082 | result = (char *) malloc (srclen); | |
1083 | if (result == NULL) | |
1084 | { | |
1085 | errno = ENOMEM; | |
1086 | return -1; | |
1087 | } | |
1088 | } | |
24d56127 LC |
1089 | memcpy (result, src, srclen); |
1090 | *resultp = result; | |
1091 | *lengthp = srclen; | |
1092 | return 0; | |
1093 | } | |
1094 | else | |
1095 | { | |
1096 | #if HAVE_ICONV | |
8912421c | 1097 | iconveh_t cd; |
24d56127 LC |
1098 | char *result; |
1099 | size_t length; | |
1100 | int retval; | |
1101 | ||
8912421c | 1102 | if (iconveh_open (to_codeset, from_codeset, &cd) < 0) |
1cd4fffc | 1103 | return -1; |
24d56127 LC |
1104 | |
1105 | result = *resultp; | |
1106 | length = *lengthp; | |
8912421c | 1107 | retval = mem_cd_iconveh (src, srclen, &cd, handler, offsets, |
1cd4fffc | 1108 | &result, &length); |
24d56127 LC |
1109 | |
1110 | if (retval < 0) | |
1cd4fffc LC |
1111 | { |
1112 | /* Close cd, but preserve the errno from str_cd_iconv. */ | |
1113 | int saved_errno = errno; | |
1114 | iconveh_close (&cd); | |
1115 | errno = saved_errno; | |
1116 | } | |
24d56127 | 1117 | else |
1cd4fffc LC |
1118 | { |
1119 | if (iconveh_close (&cd) < 0) | |
1120 | { | |
1121 | /* Return -1, but free the allocated memory, and while doing | |
1122 | that, preserve the errno from iconveh_close. */ | |
1123 | int saved_errno = errno; | |
1124 | if (result != *resultp && result != NULL) | |
1125 | free (result); | |
1126 | errno = saved_errno; | |
1127 | return -1; | |
1128 | } | |
1129 | *resultp = result; | |
1130 | *lengthp = length; | |
1131 | } | |
24d56127 LC |
1132 | return retval; |
1133 | #else | |
1134 | /* This is a different error code than if iconv_open existed but didn't | |
1cd4fffc LC |
1135 | support from_codeset and to_codeset, so that the caller can emit |
1136 | an error message such as | |
1137 | "iconv() is not supported. Installing GNU libiconv and | |
1138 | then reinstalling this package would fix this." */ | |
24d56127 LC |
1139 | errno = ENOSYS; |
1140 | return -1; | |
1141 | #endif | |
1142 | } | |
1143 | } | |
1144 | ||
1145 | char * | |
1146 | str_iconveh (const char *src, | |
1cd4fffc LC |
1147 | const char *from_codeset, const char *to_codeset, |
1148 | enum iconv_ilseq_handler handler) | |
24d56127 LC |
1149 | { |
1150 | if (*src == '\0' || c_strcasecmp (from_codeset, to_codeset) == 0) | |
1151 | { | |
1152 | char *result = strdup (src); | |
1153 | ||
1154 | if (result == NULL) | |
1cd4fffc | 1155 | errno = ENOMEM; |
24d56127 LC |
1156 | return result; |
1157 | } | |
1158 | else | |
1159 | { | |
1160 | #if HAVE_ICONV | |
8912421c | 1161 | iconveh_t cd; |
24d56127 LC |
1162 | char *result; |
1163 | ||
8912421c | 1164 | if (iconveh_open (to_codeset, from_codeset, &cd) < 0) |
1cd4fffc | 1165 | return NULL; |
24d56127 | 1166 | |
8912421c | 1167 | result = str_cd_iconveh (src, &cd, handler); |
24d56127 LC |
1168 | |
1169 | if (result == NULL) | |
1cd4fffc LC |
1170 | { |
1171 | /* Close cd, but preserve the errno from str_cd_iconv. */ | |
1172 | int saved_errno = errno; | |
1173 | iconveh_close (&cd); | |
1174 | errno = saved_errno; | |
1175 | } | |
24d56127 | 1176 | else |
1cd4fffc LC |
1177 | { |
1178 | if (iconveh_close (&cd) < 0) | |
1179 | { | |
1180 | /* Return NULL, but free the allocated memory, and while doing | |
1181 | that, preserve the errno from iconveh_close. */ | |
1182 | int saved_errno = errno; | |
1183 | free (result); | |
1184 | errno = saved_errno; | |
1185 | return NULL; | |
1186 | } | |
1187 | } | |
24d56127 LC |
1188 | return result; |
1189 | #else | |
1190 | /* This is a different error code than if iconv_open existed but didn't | |
1cd4fffc LC |
1191 | support from_codeset and to_codeset, so that the caller can emit |
1192 | an error message such as | |
1193 | "iconv() is not supported. Installing GNU libiconv and | |
1194 | then reinstalling this package would fix this." */ | |
24d56127 LC |
1195 | errno = ENOSYS; |
1196 | return NULL; | |
1197 | #endif | |
1198 | } | |
1199 | } |