Add 4-byte header to messages so that we can have different message types
[glibc/nacl-glibc.git] / intl / dcigettext.c
blobe0782446e4ab1e9b60cf70b4f39aa91c67e52d76
1 /* Implementation of the internal dcigettext function.
2 Copyright (C) 1995-2005, 2006, 2007, 2008
3 Free Software Foundation, Inc.
4 This file is part of the GNU C Library.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
19 02111-1307 USA. */
21 /* Tell glibc's <string.h> to provide a prototype for mempcpy().
22 This must come before <config.h> because <config.h> may include
23 <features.h>, and once <features.h> has been included, it's too late. */
24 #ifndef _GNU_SOURCE
25 # define _GNU_SOURCE 1
26 #endif
28 #ifdef HAVE_CONFIG_H
29 # include <config.h>
30 #endif
32 #include <sys/types.h>
34 #ifdef __GNUC__
35 # define alloca __builtin_alloca
36 # define HAVE_ALLOCA 1
37 #else
38 # if defined HAVE_ALLOCA_H || defined _LIBC
39 # include <alloca.h>
40 # else
41 # ifdef _AIX
42 #pragma alloca
43 # else
44 # ifndef alloca
45 char *alloca ();
46 # endif
47 # endif
48 # endif
49 #endif
51 #include <errno.h>
52 #ifndef errno
53 extern int errno;
54 #endif
55 #ifndef __set_errno
56 # define __set_errno(val) errno = (val)
57 #endif
59 #include <stddef.h>
60 #include <stdlib.h>
61 #include <string.h>
63 #if defined HAVE_UNISTD_H || defined _LIBC
64 # include <unistd.h>
65 #endif
67 #include <locale.h>
69 #if defined HAVE_SYS_PARAM_H || defined _LIBC
70 # include <sys/param.h>
71 #endif
73 #include "gettextP.h"
74 #include "plural-exp.h"
75 #ifdef _LIBC
76 # include <libintl.h>
77 #else
78 # include "libgnuintl.h"
79 #endif
80 #include "hash-string.h"
82 /* Thread safetyness. */
83 #ifdef _LIBC
84 # include <bits/libc-lock.h>
85 #else
86 /* Provide dummy implementation if this is outside glibc. */
87 # define __libc_lock_define_initialized(CLASS, NAME)
88 # define __libc_lock_lock(NAME)
89 # define __libc_lock_unlock(NAME)
90 # define __libc_rwlock_define_initialized(CLASS, NAME)
91 # define __libc_rwlock_rdlock(NAME)
92 # define __libc_rwlock_unlock(NAME)
93 #endif
95 /* Alignment of types. */
96 #if defined __GNUC__ && __GNUC__ >= 2
97 # define alignof(TYPE) __alignof__ (TYPE)
98 #else
99 # define alignof(TYPE) \
100 ((int) &((struct { char dummy1; TYPE dummy2; } *) 0)->dummy2)
101 #endif
103 /* The internal variables in the standalone libintl.a must have different
104 names than the internal variables in GNU libc, otherwise programs
105 using libintl.a cannot be linked statically. */
106 #if !defined _LIBC
107 # define _nl_default_default_domain libintl_nl_default_default_domain
108 # define _nl_current_default_domain libintl_nl_current_default_domain
109 # define _nl_default_dirname libintl_nl_default_dirname
110 # define _nl_domain_bindings libintl_nl_domain_bindings
111 #endif
113 /* Some compilers, like SunOS4 cc, don't have offsetof in <stddef.h>. */
114 #ifndef offsetof
115 # define offsetof(type,ident) ((size_t)&(((type*)0)->ident))
116 #endif
118 /* @@ end of prolog @@ */
120 #ifdef _LIBC
121 /* Rename the non ANSI C functions. This is required by the standard
122 because some ANSI C functions will require linking with this object
123 file and the name space must not be polluted. */
124 # define getcwd __getcwd
125 # ifndef stpcpy
126 # define stpcpy __stpcpy
127 # endif
128 # define tfind __tfind
129 #else
130 # if !defined HAVE_GETCWD
131 char *getwd ();
132 # define getcwd(buf, max) getwd (buf)
133 # else
134 char *getcwd ();
135 # endif
136 # ifndef HAVE_STPCPY
137 static char *stpcpy PARAMS ((char *dest, const char *src));
138 # endif
139 # ifndef HAVE_MEMPCPY
140 static void *mempcpy PARAMS ((void *dest, const void *src, size_t n));
141 # endif
142 #endif
144 /* Amount to increase buffer size by in each try. */
145 #define PATH_INCR 32
147 /* The following is from pathmax.h. */
148 /* Non-POSIX BSD systems might have gcc's limits.h, which doesn't define
149 PATH_MAX but might cause redefinition warnings when sys/param.h is
150 later included (as on MORE/BSD 4.3). */
151 #if defined _POSIX_VERSION || (defined HAVE_LIMITS_H && !defined __GNUC__)
152 # include <limits.h>
153 #endif
155 #ifndef _POSIX_PATH_MAX
156 # define _POSIX_PATH_MAX 255
157 #endif
159 #if !defined PATH_MAX && defined _PC_PATH_MAX
160 # define PATH_MAX (pathconf ("/", _PC_PATH_MAX) < 1 ? 1024 : pathconf ("/", _PC_PATH_MAX))
161 #endif
163 /* Don't include sys/param.h if it already has been. */
164 #if defined HAVE_SYS_PARAM_H && !defined PATH_MAX && !defined MAXPATHLEN
165 # include <sys/param.h>
166 #endif
168 #if !defined PATH_MAX && defined MAXPATHLEN
169 # define PATH_MAX MAXPATHLEN
170 #endif
172 #ifndef PATH_MAX
173 # define PATH_MAX _POSIX_PATH_MAX
174 #endif
176 /* Whether to support different locales in different threads. */
177 #if defined _LIBC || HAVE_NL_LOCALE_NAME
178 # define HAVE_PER_THREAD_LOCALE
179 #endif
181 /* This is the type used for the search tree where known translations
182 are stored. */
183 struct known_translation_t
185 /* Domain in which to search. */
186 const char *domainname;
188 /* The category. */
189 int category;
191 #ifdef HAVE_PER_THREAD_LOCALE
192 /* Name of the relevant locale category, or "" for the global locale. */
193 const char *localename;
194 #endif
196 /* State of the catalog counter at the point the string was found. */
197 int counter;
199 /* Catalog where the string was found. */
200 struct loaded_l10nfile *domain;
202 /* And finally the translation. */
203 const char *translation;
204 size_t translation_length;
206 /* Pointer to the string in question. */
207 union
209 char appended[ZERO]; /* used if domain != NULL */
210 const char *ptr; /* used if domain == NULL */
212 msgid;
215 /* Root of the search tree with known translations. We can use this
216 only if the system provides the `tsearch' function family. */
217 #if defined HAVE_TSEARCH || defined _LIBC
218 # include <search.h>
220 static void *root;
222 # ifdef _LIBC
223 # define tsearch __tsearch
224 # endif
226 /* Function to compare two entries in the table of known translations. */
227 static int transcmp PARAMS ((const void *p1, const void *p2));
228 static int
229 transcmp (p1, p2)
230 const void *p1;
231 const void *p2;
233 const struct known_translation_t *s1;
234 const struct known_translation_t *s2;
235 int result;
237 s1 = (const struct known_translation_t *) p1;
238 s2 = (const struct known_translation_t *) p2;
240 result = strcmp (s1->domain != NULL ? s1->msgid.appended : s1->msgid.ptr,
241 s2->domain != NULL ? s2->msgid.appended : s2->msgid.ptr);
242 if (result == 0)
244 result = strcmp (s1->domainname, s2->domainname);
245 if (result == 0)
247 #ifdef HAVE_PER_THREAD_LOCALE
248 result = strcmp (s1->localename, s2->localename);
249 if (result == 0)
250 #endif
251 /* We compare the category last (though this is the cheapest
252 operation) since it is hopefully always the same (namely
253 LC_MESSAGES). */
254 result = s1->category - s2->category;
258 return result;
260 #endif
262 /* Name of the default domain used for gettext(3) prior any call to
263 textdomain(3). The default value for this is "messages". */
264 const char _nl_default_default_domain[] attribute_hidden = "messages";
266 /* Value used as the default domain for gettext(3). */
267 const char *_nl_current_default_domain attribute_hidden
268 = _nl_default_default_domain;
270 /* Contains the default location of the message catalogs. */
272 #ifdef _LIBC
273 extern const char _nl_default_dirname[];
274 libc_hidden_proto (_nl_default_dirname)
275 #endif
276 const char _nl_default_dirname[] = LOCALEDIR;
277 #ifdef _LIBC
278 libc_hidden_data_def (_nl_default_dirname)
279 #endif
281 /* List with bindings of specific domains created by bindtextdomain()
282 calls. */
283 struct binding *_nl_domain_bindings;
285 /* Prototypes for local functions. */
286 static char *plural_lookup PARAMS ((struct loaded_l10nfile *domain,
287 unsigned long int n,
288 const char *translation,
289 size_t translation_len))
290 internal_function;
291 static const char *guess_category_value PARAMS ((int category,
292 const char *categoryname))
293 internal_function;
294 #ifdef _LIBC
295 # include "../locale/localeinfo.h"
296 # define category_to_name(category) \
297 _nl_category_names.str + _nl_category_name_idxs[category]
298 #else
299 static const char *category_to_name PARAMS ((int category)) internal_function;
300 #endif
303 /* For those loosing systems which don't have `alloca' we have to add
304 some additional code emulating it. */
305 #ifdef HAVE_ALLOCA
306 /* Nothing has to be done. */
307 # define freea(p) /* nothing */
308 # define ADD_BLOCK(list, address) /* nothing */
309 # define FREE_BLOCKS(list) /* nothing */
310 #else
311 struct block_list
313 void *address;
314 struct block_list *next;
316 # define ADD_BLOCK(list, addr) \
317 do { \
318 struct block_list *newp = (struct block_list *) malloc (sizeof (*newp)); \
319 /* If we cannot get a free block we cannot add the new element to \
320 the list. */ \
321 if (newp != NULL) { \
322 newp->address = (addr); \
323 newp->next = (list); \
324 (list) = newp; \
326 } while (0)
327 # define FREE_BLOCKS(list) \
328 do { \
329 while (list != NULL) { \
330 struct block_list *old = list; \
331 list = list->next; \
332 free (old->address); \
333 free (old); \
335 } while (0)
336 # undef alloca
337 # define alloca(size) (malloc (size))
338 # define freea(p) free (p)
339 #endif /* have alloca */
342 #ifdef _LIBC
343 /* List of blocks allocated for translations. */
344 typedef struct transmem_list
346 struct transmem_list *next;
347 char data[ZERO];
348 } transmem_block_t;
349 static struct transmem_list *transmem_list;
350 #else
351 typedef unsigned char transmem_block_t;
352 #endif
353 #if defined _LIBC || HAVE_ICONV
354 static const char *get_output_charset PARAMS ((struct binding *domainbinding))
355 internal_function;
356 #endif
359 /* Names for the libintl functions are a problem. They must not clash
360 with existing names and they should follow ANSI C. But this source
361 code is also used in GNU C Library where the names have a __
362 prefix. So we have to make a difference here. */
363 #ifdef _LIBC
364 # define DCIGETTEXT __dcigettext
365 #else
366 # define DCIGETTEXT libintl_dcigettext
367 #endif
369 /* Lock variable to protect the global data in the gettext implementation. */
370 #ifdef _LIBC
371 __libc_rwlock_define_initialized (, _nl_state_lock attribute_hidden)
372 #endif
374 /* Checking whether the binaries runs SUID must be done and glibc provides
375 easier methods therefore we make a difference here. */
376 #ifdef _LIBC
377 # define ENABLE_SECURE __libc_enable_secure
378 # define DETERMINE_SECURE
379 #else
380 # ifndef HAVE_GETUID
381 # define getuid() 0
382 # endif
383 # ifndef HAVE_GETGID
384 # define getgid() 0
385 # endif
386 # ifndef HAVE_GETEUID
387 # define geteuid() getuid()
388 # endif
389 # ifndef HAVE_GETEGID
390 # define getegid() getgid()
391 # endif
392 static int enable_secure;
393 # define ENABLE_SECURE (enable_secure == 1)
394 # define DETERMINE_SECURE \
395 if (enable_secure == 0) \
397 if (getuid () != geteuid () || getgid () != getegid ()) \
398 enable_secure = 1; \
399 else \
400 enable_secure = -1; \
402 #endif
404 /* Get the function to evaluate the plural expression. */
405 #include "plural-eval.c"
407 /* Look up MSGID in the DOMAINNAME message catalog for the current
408 CATEGORY locale and, if PLURAL is nonzero, search over string
409 depending on the plural form determined by N. */
410 char *
411 DCIGETTEXT (domainname, msgid1, msgid2, plural, n, category)
412 const char *domainname;
413 const char *msgid1;
414 const char *msgid2;
415 int plural;
416 unsigned long int n;
417 int category;
419 #ifndef HAVE_ALLOCA
420 struct block_list *block_list = NULL;
421 #endif
422 struct loaded_l10nfile *domain;
423 struct binding *binding;
424 const char *categoryname;
425 const char *categoryvalue;
426 char *dirname, *xdomainname;
427 char *single_locale;
428 char *retval;
429 size_t retlen;
430 int saved_errno;
431 #if defined HAVE_TSEARCH || defined _LIBC
432 struct known_translation_t search;
433 struct known_translation_t **foundp = NULL;
434 # ifdef HAVE_PER_THREAD_LOCALE
435 const char *localename;
436 # endif
437 #endif
438 size_t domainname_len;
440 /* If no real MSGID is given return NULL. */
441 if (msgid1 == NULL)
442 return NULL;
444 #ifdef _LIBC
445 if (category < 0 || category >= __LC_LAST || category == LC_ALL)
446 /* Bogus. */
447 return (plural == 0
448 ? (char *) msgid1
449 /* Use the Germanic plural rule. */
450 : n == 1 ? (char *) msgid1 : (char *) msgid2);
451 #endif
453 #ifdef _LIBC
454 __libc_rwlock_define (extern, __libc_setlocale_lock attribute_hidden)
455 __libc_rwlock_rdlock (__libc_setlocale_lock);
456 #endif
458 __libc_rwlock_rdlock (_nl_state_lock);
460 /* If DOMAINNAME is NULL, we are interested in the default domain. If
461 CATEGORY is not LC_MESSAGES this might not make much sense but the
462 definition left this undefined. */
463 if (domainname == NULL)
464 domainname = _nl_current_default_domain;
466 #if defined HAVE_TSEARCH || defined _LIBC
467 /* Try to find the translation among those which we found at
468 some time. */
469 search.domain = NULL;
470 search.msgid.ptr = msgid1;
471 search.domainname = domainname;
472 search.category = category;
473 # ifdef HAVE_PER_THREAD_LOCALE
474 # ifdef _LIBC
475 localename = strdupa (__current_locale_name (category));
476 # endif
477 search.localename = localename;
478 # endif
480 /* Since tfind/tsearch manage a balanced tree, concurrent tfind and
481 tsearch calls can be fatal. */
482 __libc_rwlock_define_initialized (static, tree_lock);
483 __libc_rwlock_rdlock (tree_lock);
485 foundp = (struct known_translation_t **) tfind (&search, &root, transcmp);
487 __libc_rwlock_unlock (tree_lock);
489 if (foundp != NULL && (*foundp)->counter == _nl_msg_cat_cntr)
491 /* Now deal with plural. */
492 if (plural)
493 retval = plural_lookup ((*foundp)->domain, n, (*foundp)->translation,
494 (*foundp)->translation_length);
495 else
496 retval = (char *) (*foundp)->translation;
498 # ifdef _LIBC
499 __libc_rwlock_unlock (__libc_setlocale_lock);
500 # endif
501 __libc_rwlock_unlock (_nl_state_lock);
502 return retval;
504 #endif
506 /* Preserve the `errno' value. */
507 saved_errno = errno;
509 /* See whether this is a SUID binary or not. */
510 DETERMINE_SECURE;
512 /* First find matching binding. */
513 for (binding = _nl_domain_bindings; binding != NULL; binding = binding->next)
515 int compare = strcmp (domainname, binding->domainname);
516 if (compare == 0)
517 /* We found it! */
518 break;
519 if (compare < 0)
521 /* It is not in the list. */
522 binding = NULL;
523 break;
527 if (binding == NULL)
528 dirname = (char *) _nl_default_dirname;
529 else if (binding->dirname[0] == '/')
530 dirname = binding->dirname;
531 else
533 /* We have a relative path. Make it absolute now. */
534 size_t dirname_len = strlen (binding->dirname) + 1;
535 size_t path_max;
536 char *ret;
538 path_max = (unsigned int) PATH_MAX;
539 path_max += 2; /* The getcwd docs say to do this. */
541 for (;;)
543 dirname = (char *) alloca (path_max + dirname_len);
544 ADD_BLOCK (block_list, dirname);
546 __set_errno (0);
547 ret = getcwd (dirname, path_max);
548 if (ret != NULL || errno != ERANGE)
549 break;
551 path_max += path_max / 2;
552 path_max += PATH_INCR;
555 if (ret == NULL)
557 /* We cannot get the current working directory. Don't signal an
558 error but simply return the default string. */
559 FREE_BLOCKS (block_list);
560 __libc_rwlock_unlock (__libc_setlocale_lock);
561 __libc_rwlock_unlock (_nl_state_lock);
562 __set_errno (saved_errno);
563 return (plural == 0
564 ? (char *) msgid1
565 /* Use the Germanic plural rule. */
566 : n == 1 ? (char *) msgid1 : (char *) msgid2);
569 stpcpy (stpcpy (strchr (dirname, '\0'), "/"), binding->dirname);
572 /* Now determine the symbolic name of CATEGORY and its value. */
573 categoryname = category_to_name (category);
574 categoryvalue = guess_category_value (category, categoryname);
576 domainname_len = strlen (domainname);
577 xdomainname = (char *) alloca (strlen (categoryname)
578 + domainname_len + 5);
579 ADD_BLOCK (block_list, xdomainname);
581 stpcpy (mempcpy (stpcpy (stpcpy (xdomainname, categoryname), "/"),
582 domainname, domainname_len),
583 ".mo");
585 /* Creating working area. */
586 single_locale = (char *) alloca (strlen (categoryvalue) + 1);
587 ADD_BLOCK (block_list, single_locale);
590 /* Search for the given string. This is a loop because we perhaps
591 got an ordered list of languages to consider for the translation. */
592 while (1)
594 /* Make CATEGORYVALUE point to the next element of the list. */
595 while (categoryvalue[0] != '\0' && categoryvalue[0] == ':')
596 ++categoryvalue;
597 if (categoryvalue[0] == '\0')
599 /* The whole contents of CATEGORYVALUE has been searched but
600 no valid entry has been found. We solve this situation
601 by implicitly appending a "C" entry, i.e. no translation
602 will take place. */
603 single_locale[0] = 'C';
604 single_locale[1] = '\0';
606 else
608 char *cp = single_locale;
609 while (categoryvalue[0] != '\0' && categoryvalue[0] != ':')
610 *cp++ = *categoryvalue++;
611 *cp = '\0';
613 /* When this is a SUID binary we must not allow accessing files
614 outside the dedicated directories. */
615 if (ENABLE_SECURE && strchr (single_locale, '/') != NULL)
616 /* Ingore this entry. */
617 continue;
620 /* If the current locale value is C (or POSIX) we don't load a
621 domain. Return the MSGID. */
622 if (strcmp (single_locale, "C") == 0
623 || strcmp (single_locale, "POSIX") == 0)
625 no_translation:
626 FREE_BLOCKS (block_list);
627 __libc_rwlock_unlock (__libc_setlocale_lock);
628 __libc_rwlock_unlock (_nl_state_lock);
629 __set_errno (saved_errno);
630 return (plural == 0
631 ? (char *) msgid1
632 /* Use the Germanic plural rule. */
633 : n == 1 ? (char *) msgid1 : (char *) msgid2);
637 /* Find structure describing the message catalog matching the
638 DOMAINNAME and CATEGORY. */
639 domain = _nl_find_domain (dirname, single_locale, xdomainname, binding);
641 if (domain != NULL)
643 retval = _nl_find_msg (domain, binding, msgid1, 1, &retlen);
645 if (retval == NULL)
647 int cnt;
649 for (cnt = 0; domain->successor[cnt] != NULL; ++cnt)
651 retval = _nl_find_msg (domain->successor[cnt], binding,
652 msgid1, 1, &retlen);
654 if (retval != NULL)
656 domain = domain->successor[cnt];
657 break;
662 /* Returning -1 means that some resource problem exists
663 (likely memory) and that the strings could not be
664 converted. Return the original strings. */
665 if (__builtin_expect (retval == (char *) -1, 0))
666 goto no_translation;
668 if (retval != NULL)
670 /* Found the translation of MSGID1 in domain DOMAIN:
671 starting at RETVAL, RETLEN bytes. */
672 FREE_BLOCKS (block_list);
673 #if defined HAVE_TSEARCH || defined _LIBC
674 if (foundp == NULL)
676 /* Create a new entry and add it to the search tree. */
677 size_t msgid_len;
678 size_t size;
679 struct known_translation_t *newp;
681 msgid_len = strlen (msgid1) + 1;
682 size = offsetof (struct known_translation_t, msgid)
683 + msgid_len + domainname_len + 1;
684 # ifdef HAVE_PER_THREAD_LOCALE
685 size += strlen (localename) + 1;
686 # endif
687 newp = (struct known_translation_t *) malloc (size);
688 if (newp != NULL)
690 char *new_domainname;
691 # ifdef HAVE_PER_THREAD_LOCALE
692 char *new_localename;
693 # endif
695 new_domainname =
696 mempcpy (newp->msgid.appended, msgid1, msgid_len);
697 memcpy (new_domainname, domainname, domainname_len + 1);
698 # ifdef HAVE_PER_THREAD_LOCALE
699 new_localename = new_domainname + domainname_len + 1;
700 strcpy (new_localename, localename);
701 # endif
702 newp->domainname = new_domainname;
703 newp->category = category;
704 # ifdef HAVE_PER_THREAD_LOCALE
705 newp->localename = new_localename;
706 # endif
707 newp->counter = _nl_msg_cat_cntr;
708 newp->domain = domain;
709 newp->translation = retval;
710 newp->translation_length = retlen;
712 __libc_rwlock_wrlock (tree_lock);
714 /* Insert the entry in the search tree. */
715 foundp = (struct known_translation_t **)
716 tsearch (newp, &root, transcmp);
718 __libc_rwlock_unlock (tree_lock);
720 if (foundp == NULL
721 || __builtin_expect (*foundp != newp, 0))
722 /* The insert failed. */
723 free (newp);
726 else
728 /* We can update the existing entry. */
729 (*foundp)->counter = _nl_msg_cat_cntr;
730 (*foundp)->domain = domain;
731 (*foundp)->translation = retval;
732 (*foundp)->translation_length = retlen;
734 #endif
735 __set_errno (saved_errno);
737 /* Now deal with plural. */
738 if (plural)
739 retval = plural_lookup (domain, n, retval, retlen);
741 __libc_rwlock_unlock (__libc_setlocale_lock);
742 __libc_rwlock_unlock (_nl_state_lock);
743 return retval;
747 /* NOTREACHED */
751 char *
752 internal_function
753 _nl_find_msg (domain_file, domainbinding, msgid, convert, lengthp)
754 struct loaded_l10nfile *domain_file;
755 struct binding *domainbinding;
756 const char *msgid;
757 int convert;
758 size_t *lengthp;
760 struct loaded_domain *domain;
761 nls_uint32 nstrings;
762 size_t act;
763 char *result;
764 size_t resultlen;
766 if (domain_file->decided <= 0)
767 _nl_load_domain (domain_file, domainbinding);
769 if (domain_file->data == NULL)
770 return NULL;
772 domain = (struct loaded_domain *) domain_file->data;
774 nstrings = domain->nstrings;
776 /* Locate the MSGID and its translation. */
777 if (domain->hash_tab != NULL)
779 /* Use the hashing table. */
780 nls_uint32 len = strlen (msgid);
781 nls_uint32 hash_val = __hash_string (msgid);
782 nls_uint32 idx = hash_val % domain->hash_size;
783 nls_uint32 incr = 1 + (hash_val % (domain->hash_size - 2));
785 while (1)
787 nls_uint32 nstr =
788 W (domain->must_swap_hash_tab, domain->hash_tab[idx]);
790 if (nstr == 0)
791 /* Hash table entry is empty. */
792 return NULL;
794 nstr--;
796 /* Compare msgid with the original string at index nstr.
797 We compare the lengths with >=, not ==, because plural entries
798 are represented by strings with an embedded NUL. */
799 if (nstr < nstrings
800 ? W (domain->must_swap, domain->orig_tab[nstr].length) >= len
801 && (strcmp (msgid,
802 domain->data + W (domain->must_swap,
803 domain->orig_tab[nstr].offset))
804 == 0)
805 : domain->orig_sysdep_tab[nstr - nstrings].length > len
806 && (strcmp (msgid,
807 domain->orig_sysdep_tab[nstr - nstrings].pointer)
808 == 0))
810 act = nstr;
811 goto found;
814 if (idx >= domain->hash_size - incr)
815 idx -= domain->hash_size - incr;
816 else
817 idx += incr;
819 /* NOTREACHED */
821 else
823 /* Try the default method: binary search in the sorted array of
824 messages. */
825 size_t top, bottom;
827 bottom = 0;
828 top = nstrings;
829 while (bottom < top)
831 int cmp_val;
833 act = (bottom + top) / 2;
834 cmp_val = strcmp (msgid, (domain->data
835 + W (domain->must_swap,
836 domain->orig_tab[act].offset)));
837 if (cmp_val < 0)
838 top = act;
839 else if (cmp_val > 0)
840 bottom = act + 1;
841 else
842 goto found;
844 /* No translation was found. */
845 return NULL;
848 found:
849 /* The translation was found at index ACT. If we have to convert the
850 string to use a different character set, this is the time. */
851 if (act < nstrings)
853 result = (char *)
854 (domain->data + W (domain->must_swap, domain->trans_tab[act].offset));
855 resultlen = W (domain->must_swap, domain->trans_tab[act].length) + 1;
857 else
859 result = (char *) domain->trans_sysdep_tab[act - nstrings].pointer;
860 resultlen = domain->trans_sysdep_tab[act - nstrings].length;
863 #if defined _LIBC || HAVE_ICONV
864 if (convert)
866 /* We are supposed to do a conversion. */
867 const char *encoding = get_output_charset (domainbinding);
869 /* Protect against reallocation of the table. */
870 __libc_rwlock_rdlock (domain->conversions_lock);
872 /* Search whether a table with converted translations for this
873 encoding has already been allocated. */
874 size_t nconversions = domain->nconversions;
875 struct converted_domain *convd = NULL;
876 size_t i;
878 for (i = nconversions; i > 0; )
880 i--;
881 if (strcmp (domain->conversions[i].encoding, encoding) == 0)
883 convd = &domain->conversions[i];
884 break;
888 __libc_rwlock_unlock (domain->conversions_lock);
890 if (convd == NULL)
892 /* We have to allocate a new conversions table. */
893 __libc_rwlock_wrlock (domain->conversions_lock);
894 nconversions = domain->nconversions;
896 /* Maybe in the meantime somebody added the translation.
897 Recheck. */
898 for (i = nconversions; i > 0; )
900 i--;
901 if (strcmp (domain->conversions[i].encoding, encoding) == 0)
903 convd = &domain->conversions[i];
904 goto found_convd;
908 /* Allocate a table for the converted translations for this
909 encoding. */
910 struct converted_domain *new_conversions =
911 (struct converted_domain *)
912 realloc (domain->conversions,
913 (nconversions + 1) * sizeof (struct converted_domain));
915 if (__builtin_expect (new_conversions == NULL, 0))
917 /* Nothing we can do, no more memory. We cannot use the
918 translation because it might be encoded incorrectly. */
919 unlock_fail:
920 __libc_rwlock_unlock (domain->conversions_lock);
921 return (char *) -1;
924 domain->conversions = new_conversions;
926 /* Copy the 'encoding' string to permanent storage. */
927 encoding = strdup (encoding);
928 if (__builtin_expect (encoding == NULL, 0))
929 /* Nothing we can do, no more memory. We cannot use the
930 translation because it might be encoded incorrectly. */
931 goto unlock_fail;
933 convd = &new_conversions[nconversions];
934 convd->encoding = encoding;
936 /* Find out about the character set the file is encoded with.
937 This can be found (in textual form) in the entry "". If this
938 entry does not exist or if this does not contain the 'charset='
939 information, we will assume the charset matches the one the
940 current locale and we don't have to perform any conversion. */
941 # ifdef _LIBC
942 convd->conv = (__gconv_t) -1;
943 # else
944 # if HAVE_ICONV
945 convd->conv = (iconv_t) -1;
946 # endif
947 # endif
949 char *nullentry;
950 size_t nullentrylen;
952 /* Get the header entry. This is a recursion, but it doesn't
953 reallocate domain->conversions because we pass convert = 0. */
954 nullentry =
955 _nl_find_msg (domain_file, domainbinding, "", 0, &nullentrylen);
957 if (nullentry != NULL)
959 const char *charsetstr;
961 charsetstr = strstr (nullentry, "charset=");
962 if (charsetstr != NULL)
964 size_t len;
965 char *charset;
966 const char *outcharset;
968 charsetstr += strlen ("charset=");
969 len = strcspn (charsetstr, " \t\n");
971 charset = (char *) alloca (len + 1);
972 # if defined _LIBC || HAVE_MEMPCPY
973 *((char *) mempcpy (charset, charsetstr, len)) = '\0';
974 # else
975 memcpy (charset, charsetstr, len);
976 charset[len] = '\0';
977 # endif
979 outcharset = encoding;
981 # ifdef _LIBC
982 /* We always want to use transliteration. */
983 outcharset = norm_add_slashes (outcharset, "TRANSLIT");
984 charset = norm_add_slashes (charset, "");
985 int r = __gconv_open (outcharset, charset, &convd->conv,
986 GCONV_AVOID_NOCONV);
987 if (__builtin_expect (r != __GCONV_OK, 0))
989 /* If the output encoding is the same there is
990 nothing to do. Otherwise do not use the
991 translation at all. */
992 if (__builtin_expect (r != __GCONV_NULCONV, 1))
994 __libc_rwlock_unlock (domain->conversions_lock);
995 free ((char *) encoding);
996 return NULL;
999 convd->conv = (__gconv_t) -1;
1001 # else
1002 # if HAVE_ICONV
1003 /* When using GNU libc >= 2.2 or GNU libiconv >= 1.5,
1004 we want to use transliteration. */
1005 # if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 \
1006 || _LIBICONV_VERSION >= 0x0105
1007 if (strchr (outcharset, '/') == NULL)
1009 char *tmp;
1011 len = strlen (outcharset);
1012 tmp = (char *) alloca (len + 10 + 1);
1013 memcpy (tmp, outcharset, len);
1014 memcpy (tmp + len, "//TRANSLIT", 10 + 1);
1015 outcharset = tmp;
1017 convd->conv = iconv_open (outcharset, charset);
1019 freea (outcharset);
1021 else
1022 # endif
1023 convd->conv = iconv_open (outcharset, charset);
1024 # endif
1025 # endif
1027 freea (charset);
1031 convd->conv_tab = NULL;
1032 /* Here domain->conversions is still == new_conversions. */
1033 domain->nconversions++;
1035 found_convd:
1036 __libc_rwlock_unlock (domain->conversions_lock);
1039 if (
1040 # ifdef _LIBC
1041 convd->conv != (__gconv_t) -1
1042 # else
1043 # if HAVE_ICONV
1044 convd->conv != (iconv_t) -1
1045 # endif
1046 # endif
1049 __libc_lock_define_initialized (static, lock)
1050 /* We are supposed to do a conversion. First allocate an
1051 appropriate table with the same structure as the table
1052 of translations in the file, where we can put the pointers
1053 to the converted strings in.
1054 There is a slight complication with plural entries. They
1055 are represented by consecutive NUL terminated strings. We
1056 handle this case by converting RESULTLEN bytes, including
1057 NULs. */
1059 if (__builtin_expect (convd->conv_tab == NULL, 0))
1061 __libc_lock_lock (lock);
1062 if (convd->conv_tab == NULL)
1064 convd->conv_tab
1065 = calloc (nstrings + domain->n_sysdep_strings,
1066 sizeof (char *));
1067 if (convd->conv_tab != NULL)
1068 goto not_translated_yet;
1069 /* Mark that we didn't succeed allocating a table. */
1070 convd->conv_tab = (char **) -1;
1072 __libc_lock_unlock (lock);
1075 if (__builtin_expect (convd->conv_tab == (char **) -1, 0))
1076 /* Nothing we can do, no more memory. We cannot use the
1077 translation because it might be encoded incorrectly. */
1078 return (char *) -1;
1080 if (convd->conv_tab[act] == NULL)
1082 __libc_lock_lock (lock);
1083 not_translated_yet:;
1085 /* We haven't used this string so far, so it is not
1086 translated yet. Do this now. */
1087 /* We use a bit more efficient memory handling.
1088 We allocate always larger blocks which get used over
1089 time. This is faster than many small allocations. */
1090 # define INITIAL_BLOCK_SIZE 4080
1091 static unsigned char *freemem;
1092 static size_t freemem_size;
1094 const unsigned char *inbuf;
1095 unsigned char *outbuf;
1096 int malloc_count;
1097 # ifndef _LIBC
1098 transmem_block_t *transmem_list = NULL;
1099 # endif
1101 inbuf = (const unsigned char *) result;
1102 outbuf = freemem + sizeof (size_t);
1104 malloc_count = 0;
1105 while (1)
1107 transmem_block_t *newmem;
1108 # ifdef _LIBC
1109 size_t non_reversible;
1110 int res;
1112 if (freemem_size < sizeof (size_t))
1113 goto resize_freemem;
1115 res = __gconv (convd->conv,
1116 &inbuf, inbuf + resultlen,
1117 &outbuf,
1118 outbuf + freemem_size - sizeof (size_t),
1119 &non_reversible);
1121 if (res == __GCONV_OK || res == __GCONV_EMPTY_INPUT)
1122 break;
1124 if (res != __GCONV_FULL_OUTPUT)
1126 /* We should not use the translation at all, it
1127 is incorrectly encoded. */
1128 __libc_lock_unlock (lock);
1129 return NULL;
1132 inbuf = (const unsigned char *) result;
1133 # else
1134 # if HAVE_ICONV
1135 const char *inptr = (const char *) inbuf;
1136 size_t inleft = resultlen;
1137 char *outptr = (char *) outbuf;
1138 size_t outleft;
1140 if (freemem_size < sizeof (size_t))
1141 goto resize_freemem;
1143 outleft = freemem_size - sizeof (size_t);
1144 if (iconv (convd->conv,
1145 (ICONV_CONST char **) &inptr, &inleft,
1146 &outptr, &outleft)
1147 != (size_t) (-1))
1149 outbuf = (unsigned char *) outptr;
1150 break;
1152 if (errno != E2BIG)
1154 __libc_lock_unlock (lock);
1155 return NULL;
1157 # endif
1158 # endif
1160 resize_freemem:
1161 /* We must allocate a new buffer or resize the old one. */
1162 if (malloc_count > 0)
1164 ++malloc_count;
1165 freemem_size = malloc_count * INITIAL_BLOCK_SIZE;
1166 newmem = (transmem_block_t *) realloc (transmem_list,
1167 freemem_size);
1168 # ifdef _LIBC
1169 if (newmem != NULL)
1170 transmem_list = transmem_list->next;
1171 else
1173 struct transmem_list *old = transmem_list;
1175 transmem_list = transmem_list->next;
1176 free (old);
1178 # endif
1180 else
1182 malloc_count = 1;
1183 freemem_size = INITIAL_BLOCK_SIZE;
1184 newmem = (transmem_block_t *) malloc (freemem_size);
1186 if (__builtin_expect (newmem == NULL, 0))
1188 freemem = NULL;
1189 freemem_size = 0;
1190 __libc_lock_unlock (lock);
1191 return (char *) -1;
1194 # ifdef _LIBC
1195 /* Add the block to the list of blocks we have to free
1196 at some point. */
1197 newmem->next = transmem_list;
1198 transmem_list = newmem;
1200 freemem = (unsigned char *) newmem->data;
1201 freemem_size -= offsetof (struct transmem_list, data);
1202 # else
1203 transmem_list = newmem;
1204 freemem = newmem;
1205 # endif
1207 outbuf = freemem + sizeof (size_t);
1210 /* We have now in our buffer a converted string. Put this
1211 into the table of conversions. */
1212 *(size_t *) freemem = outbuf - freemem - sizeof (size_t);
1213 convd->conv_tab[act] = (char *) freemem;
1214 /* Shrink freemem, but keep it aligned. */
1215 freemem_size -= outbuf - freemem;
1216 freemem = outbuf;
1217 freemem += freemem_size & (alignof (size_t) - 1);
1218 freemem_size = freemem_size & ~ (alignof (size_t) - 1);
1220 __libc_lock_unlock (lock);
1223 /* Now convd->conv_tab[act] contains the translation of all
1224 the plural variants. */
1225 result = convd->conv_tab[act] + sizeof (size_t);
1226 resultlen = *(size_t *) convd->conv_tab[act];
1230 /* The result string is converted. */
1232 #endif /* _LIBC || HAVE_ICONV */
1234 *lengthp = resultlen;
1235 return result;
1239 /* Look up a plural variant. */
1240 static char *
1241 internal_function
1242 plural_lookup (domain, n, translation, translation_len)
1243 struct loaded_l10nfile *domain;
1244 unsigned long int n;
1245 const char *translation;
1246 size_t translation_len;
1248 struct loaded_domain *domaindata = (struct loaded_domain *) domain->data;
1249 unsigned long int index;
1250 const char *p;
1252 index = plural_eval (domaindata->plural, n);
1253 if (index >= domaindata->nplurals)
1254 /* This should never happen. It means the plural expression and the
1255 given maximum value do not match. */
1256 index = 0;
1258 /* Skip INDEX strings at TRANSLATION. */
1259 p = translation;
1260 while (index-- > 0)
1262 #ifdef _LIBC
1263 p = __rawmemchr (p, '\0');
1264 #else
1265 p = strchr (p, '\0');
1266 #endif
1267 /* And skip over the NUL byte. */
1268 p++;
1270 if (p >= translation + translation_len)
1271 /* This should never happen. It means the plural expression
1272 evaluated to a value larger than the number of variants
1273 available for MSGID1. */
1274 return (char *) translation;
1276 return (char *) p;
1279 #ifndef _LIBC
1280 /* Return string representation of locale CATEGORY. */
1281 static const char *
1282 internal_function
1283 category_to_name (category)
1284 int category;
1286 const char *retval;
1288 switch (category)
1290 #ifdef LC_COLLATE
1291 case LC_COLLATE:
1292 retval = "LC_COLLATE";
1293 break;
1294 #endif
1295 #ifdef LC_CTYPE
1296 case LC_CTYPE:
1297 retval = "LC_CTYPE";
1298 break;
1299 #endif
1300 #ifdef LC_MONETARY
1301 case LC_MONETARY:
1302 retval = "LC_MONETARY";
1303 break;
1304 #endif
1305 #ifdef LC_NUMERIC
1306 case LC_NUMERIC:
1307 retval = "LC_NUMERIC";
1308 break;
1309 #endif
1310 #ifdef LC_TIME
1311 case LC_TIME:
1312 retval = "LC_TIME";
1313 break;
1314 #endif
1315 #ifdef LC_MESSAGES
1316 case LC_MESSAGES:
1317 retval = "LC_MESSAGES";
1318 break;
1319 #endif
1320 #ifdef LC_RESPONSE
1321 case LC_RESPONSE:
1322 retval = "LC_RESPONSE";
1323 break;
1324 #endif
1325 #ifdef LC_ALL
1326 case LC_ALL:
1327 /* This might not make sense but is perhaps better than any other
1328 value. */
1329 retval = "LC_ALL";
1330 break;
1331 #endif
1332 default:
1333 /* If you have a better idea for a default value let me know. */
1334 retval = "LC_XXX";
1337 return retval;
1339 #endif
1341 /* Guess value of current locale from value of the environment variables. */
1342 static const char *
1343 internal_function
1344 guess_category_value (category, categoryname)
1345 int category;
1346 const char *categoryname;
1348 const char *language;
1349 const char *retval;
1351 /* The highest priority value is the `LANGUAGE' environment
1352 variable. But we don't use the value if the currently selected
1353 locale is the C locale. This is a GNU extension. */
1354 language = getenv ("LANGUAGE");
1355 if (language != NULL && language[0] == '\0')
1356 language = NULL;
1358 /* We have to proceed with the POSIX methods of looking to `LC_ALL',
1359 `LC_xxx', and `LANG'. On some systems this can be done by the
1360 `setlocale' function itself. */
1361 #ifdef _LIBC
1362 retval = __current_locale_name (category);
1363 #else
1364 retval = _nl_locale_name (category, categoryname);
1365 #endif
1367 return language != NULL && strcmp (retval, "C") != 0 ? language : retval;
1370 #if defined _LIBC || HAVE_ICONV
1371 /* Returns the output charset. */
1372 static const char *
1373 internal_function
1374 get_output_charset (domainbinding)
1375 struct binding *domainbinding;
1377 /* The output charset should normally be determined by the locale. But
1378 sometimes the locale is not used or not correctly set up, so we provide
1379 a possibility for the user to override this: the OUTPUT_CHARSET
1380 environment variable. Moreover, the value specified through
1381 bind_textdomain_codeset overrides both. */
1382 if (domainbinding != NULL && domainbinding->codeset != NULL)
1383 return domainbinding->codeset;
1384 else
1386 /* For speed reasons, we look at the value of OUTPUT_CHARSET only
1387 once. This is a user variable that is not supposed to change
1388 during a program run. */
1389 static char *output_charset_cache;
1390 static int output_charset_cached;
1392 if (!output_charset_cached)
1394 const char *value = getenv ("OUTPUT_CHARSET");
1396 if (value != NULL && value[0] != '\0')
1398 size_t len = strlen (value) + 1;
1399 char *value_copy = (char *) malloc (len);
1401 if (value_copy != NULL)
1402 memcpy (value_copy, value, len);
1403 output_charset_cache = value_copy;
1405 output_charset_cached = 1;
1408 if (output_charset_cache != NULL)
1409 return output_charset_cache;
1410 else
1412 # ifdef _LIBC
1413 return _NL_CURRENT (LC_CTYPE, CODESET);
1414 # else
1415 # if HAVE_ICONV
1416 extern const char *locale_charset PARAMS ((void);
1417 return locale_charset ();
1418 # endif
1419 # endif
1423 #endif
1425 /* @@ begin of epilog @@ */
1427 /* We don't want libintl.a to depend on any other library. So we
1428 avoid the non-standard function stpcpy. In GNU C Library this
1429 function is available, though. Also allow the symbol HAVE_STPCPY
1430 to be defined. */
1431 #if !_LIBC && !HAVE_STPCPY
1432 static char *
1433 stpcpy (dest, src)
1434 char *dest;
1435 const char *src;
1437 while ((*dest++ = *src++) != '\0')
1438 /* Do nothing. */ ;
1439 return dest - 1;
1441 #endif
1443 #if !_LIBC && !HAVE_MEMPCPY
1444 static void *
1445 mempcpy (dest, src, n)
1446 void *dest;
1447 const void *src;
1448 size_t n;
1450 return (void *) ((char *) memcpy (dest, src, n) + n);
1452 #endif
1455 #ifdef _LIBC
1456 /* If we want to free all resources we have to do some work at
1457 program's end. */
1458 libc_freeres_fn (free_mem)
1460 void *old;
1462 while (_nl_domain_bindings != NULL)
1464 struct binding *oldp = _nl_domain_bindings;
1465 _nl_domain_bindings = _nl_domain_bindings->next;
1466 if (oldp->dirname != _nl_default_dirname)
1467 /* Yes, this is a pointer comparison. */
1468 free (oldp->dirname);
1469 free (oldp->codeset);
1470 free (oldp);
1473 if (_nl_current_default_domain != _nl_default_default_domain)
1474 /* Yes, again a pointer comparison. */
1475 free ((char *) _nl_current_default_domain);
1477 /* Remove the search tree with the known translations. */
1478 __tdestroy (root, free);
1479 root = NULL;
1481 while (transmem_list != NULL)
1483 old = transmem_list;
1484 transmem_list = transmem_list->next;
1485 free (old);
1488 #endif