Updated to fedora-glibc-20071212T1051
[glibc.git] / intl / dcigettext.c
blob7004cd49cfd899497dc1763247e12ca03395ef4f
1 /* Implementation of the internal dcigettext function.
2 Copyright (C) 1995-2005, 2006, 2007 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
10 The GNU C Library 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 GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, write to the Free
17 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
18 02111-1307 USA. */
20 /* Tell glibc's <string.h> to provide a prototype for mempcpy().
21 This must come before <config.h> because <config.h> may include
22 <features.h>, and once <features.h> has been included, it's too late. */
23 #ifndef _GNU_SOURCE
24 # define _GNU_SOURCE 1
25 #endif
27 #ifdef HAVE_CONFIG_H
28 # include <config.h>
29 #endif
31 #include <sys/types.h>
33 #ifdef __GNUC__
34 # define alloca __builtin_alloca
35 # define HAVE_ALLOCA 1
36 #else
37 # if defined HAVE_ALLOCA_H || defined _LIBC
38 # include <alloca.h>
39 # else
40 # ifdef _AIX
41 #pragma alloca
42 # else
43 # ifndef alloca
44 char *alloca ();
45 # endif
46 # endif
47 # endif
48 #endif
50 #include <errno.h>
51 #ifndef errno
52 extern int errno;
53 #endif
54 #ifndef __set_errno
55 # define __set_errno(val) errno = (val)
56 #endif
58 #include <stddef.h>
59 #include <stdlib.h>
60 #include <string.h>
62 #if defined HAVE_UNISTD_H || defined _LIBC
63 # include <unistd.h>
64 #endif
66 #include <locale.h>
68 #if defined HAVE_SYS_PARAM_H || defined _LIBC
69 # include <sys/param.h>
70 #endif
72 #include "gettextP.h"
73 #include "plural-exp.h"
74 #ifdef _LIBC
75 # include <libintl.h>
76 #else
77 # include "libgnuintl.h"
78 #endif
79 #include "hash-string.h"
81 /* Thread safetyness. */
82 #ifdef _LIBC
83 # include <bits/libc-lock.h>
84 #else
85 /* Provide dummy implementation if this is outside glibc. */
86 # define __libc_lock_define_initialized(CLASS, NAME)
87 # define __libc_lock_lock(NAME)
88 # define __libc_lock_unlock(NAME)
89 # define __libc_rwlock_define_initialized(CLASS, NAME)
90 # define __libc_rwlock_rdlock(NAME)
91 # define __libc_rwlock_unlock(NAME)
92 #endif
94 /* Alignment of types. */
95 #if defined __GNUC__ && __GNUC__ >= 2
96 # define alignof(TYPE) __alignof__ (TYPE)
97 #else
98 # define alignof(TYPE) \
99 ((int) &((struct { char dummy1; TYPE dummy2; } *) 0)->dummy2)
100 #endif
102 /* The internal variables in the standalone libintl.a must have different
103 names than the internal variables in GNU libc, otherwise programs
104 using libintl.a cannot be linked statically. */
105 #if !defined _LIBC
106 # define _nl_default_default_domain libintl_nl_default_default_domain
107 # define _nl_current_default_domain libintl_nl_current_default_domain
108 # define _nl_default_dirname libintl_nl_default_dirname
109 # define _nl_domain_bindings libintl_nl_domain_bindings
110 #endif
112 /* Some compilers, like SunOS4 cc, don't have offsetof in <stddef.h>. */
113 #ifndef offsetof
114 # define offsetof(type,ident) ((size_t)&(((type*)0)->ident))
115 #endif
117 /* @@ end of prolog @@ */
119 #ifdef _LIBC
120 /* Rename the non ANSI C functions. This is required by the standard
121 because some ANSI C functions will require linking with this object
122 file and the name space must not be polluted. */
123 # define getcwd __getcwd
124 # ifndef stpcpy
125 # define stpcpy __stpcpy
126 # endif
127 # define tfind __tfind
128 #else
129 # if !defined HAVE_GETCWD
130 char *getwd ();
131 # define getcwd(buf, max) getwd (buf)
132 # else
133 char *getcwd ();
134 # endif
135 # ifndef HAVE_STPCPY
136 static char *stpcpy PARAMS ((char *dest, const char *src));
137 # endif
138 # ifndef HAVE_MEMPCPY
139 static void *mempcpy PARAMS ((void *dest, const void *src, size_t n));
140 # endif
141 #endif
143 /* Amount to increase buffer size by in each try. */
144 #define PATH_INCR 32
146 /* The following is from pathmax.h. */
147 /* Non-POSIX BSD systems might have gcc's limits.h, which doesn't define
148 PATH_MAX but might cause redefinition warnings when sys/param.h is
149 later included (as on MORE/BSD 4.3). */
150 #if defined _POSIX_VERSION || (defined HAVE_LIMITS_H && !defined __GNUC__)
151 # include <limits.h>
152 #endif
154 #ifndef _POSIX_PATH_MAX
155 # define _POSIX_PATH_MAX 255
156 #endif
158 #if !defined PATH_MAX && defined _PC_PATH_MAX
159 # define PATH_MAX (pathconf ("/", _PC_PATH_MAX) < 1 ? 1024 : pathconf ("/", _PC_PATH_MAX))
160 #endif
162 /* Don't include sys/param.h if it already has been. */
163 #if defined HAVE_SYS_PARAM_H && !defined PATH_MAX && !defined MAXPATHLEN
164 # include <sys/param.h>
165 #endif
167 #if !defined PATH_MAX && defined MAXPATHLEN
168 # define PATH_MAX MAXPATHLEN
169 #endif
171 #ifndef PATH_MAX
172 # define PATH_MAX _POSIX_PATH_MAX
173 #endif
175 /* Whether to support different locales in different threads. */
176 #if defined _LIBC || HAVE_NL_LOCALE_NAME
177 # define HAVE_PER_THREAD_LOCALE
178 #endif
180 /* This is the type used for the search tree where known translations
181 are stored. */
182 struct known_translation_t
184 /* Domain in which to search. */
185 const char *domainname;
187 /* The category. */
188 int category;
190 #ifdef HAVE_PER_THREAD_LOCALE
191 /* Name of the relevant locale category, or "" for the global locale. */
192 const char *localename;
193 #endif
195 /* State of the catalog counter at the point the string was found. */
196 int counter;
198 /* Catalog where the string was found. */
199 struct loaded_l10nfile *domain;
201 /* And finally the translation. */
202 const char *translation;
203 size_t translation_length;
205 /* Pointer to the string in question. */
206 union
208 char appended[ZERO]; /* used if domain != NULL */
209 const char *ptr; /* used if domain == NULL */
211 msgid;
214 /* Root of the search tree with known translations. We can use this
215 only if the system provides the `tsearch' function family. */
216 #if defined HAVE_TSEARCH || defined _LIBC
217 # include <search.h>
219 static void *root;
221 # ifdef _LIBC
222 # define tsearch __tsearch
223 # endif
225 /* Function to compare two entries in the table of known translations. */
226 static int transcmp PARAMS ((const void *p1, const void *p2));
227 static int
228 transcmp (p1, p2)
229 const void *p1;
230 const void *p2;
232 const struct known_translation_t *s1;
233 const struct known_translation_t *s2;
234 int result;
236 s1 = (const struct known_translation_t *) p1;
237 s2 = (const struct known_translation_t *) p2;
239 result = strcmp (s1->domain != NULL ? s1->msgid.appended : s1->msgid.ptr,
240 s2->domain != NULL ? s2->msgid.appended : s2->msgid.ptr);
241 if (result == 0)
243 result = strcmp (s1->domainname, s2->domainname);
244 if (result == 0)
246 #ifdef HAVE_PER_THREAD_LOCALE
247 result = strcmp (s1->localename, s2->localename);
248 if (result == 0)
249 #endif
250 /* We compare the category last (though this is the cheapest
251 operation) since it is hopefully always the same (namely
252 LC_MESSAGES). */
253 result = s1->category - s2->category;
257 return result;
259 #endif
261 /* Name of the default domain used for gettext(3) prior any call to
262 textdomain(3). The default value for this is "messages". */
263 const char _nl_default_default_domain[] attribute_hidden = "messages";
265 /* Value used as the default domain for gettext(3). */
266 const char *_nl_current_default_domain attribute_hidden
267 = _nl_default_default_domain;
269 /* Contains the default location of the message catalogs. */
271 #ifdef _LIBC
272 extern const char _nl_default_dirname[];
273 libc_hidden_proto (_nl_default_dirname)
274 #endif
275 const char _nl_default_dirname[] = LOCALEDIR;
276 #ifdef _LIBC
277 libc_hidden_data_def (_nl_default_dirname)
278 #endif
280 /* List with bindings of specific domains created by bindtextdomain()
281 calls. */
282 struct binding *_nl_domain_bindings;
284 /* Prototypes for local functions. */
285 static char *plural_lookup PARAMS ((struct loaded_l10nfile *domain,
286 unsigned long int n,
287 const char *translation,
288 size_t translation_len))
289 internal_function;
290 static const char *guess_category_value PARAMS ((int category,
291 const char *categoryname))
292 internal_function;
293 #ifdef _LIBC
294 # include "../locale/localeinfo.h"
295 # define category_to_name(category) \
296 _nl_category_names.str + _nl_category_name_idxs[category]
297 #else
298 static const char *category_to_name PARAMS ((int category)) internal_function;
299 #endif
302 /* For those loosing systems which don't have `alloca' we have to add
303 some additional code emulating it. */
304 #ifdef HAVE_ALLOCA
305 /* Nothing has to be done. */
306 # define freea(p) /* nothing */
307 # define ADD_BLOCK(list, address) /* nothing */
308 # define FREE_BLOCKS(list) /* nothing */
309 #else
310 struct block_list
312 void *address;
313 struct block_list *next;
315 # define ADD_BLOCK(list, addr) \
316 do { \
317 struct block_list *newp = (struct block_list *) malloc (sizeof (*newp)); \
318 /* If we cannot get a free block we cannot add the new element to \
319 the list. */ \
320 if (newp != NULL) { \
321 newp->address = (addr); \
322 newp->next = (list); \
323 (list) = newp; \
325 } while (0)
326 # define FREE_BLOCKS(list) \
327 do { \
328 while (list != NULL) { \
329 struct block_list *old = list; \
330 list = list->next; \
331 free (old->address); \
332 free (old); \
334 } while (0)
335 # undef alloca
336 # define alloca(size) (malloc (size))
337 # define freea(p) free (p)
338 #endif /* have alloca */
341 #ifdef _LIBC
342 /* List of blocks allocated for translations. */
343 typedef struct transmem_list
345 struct transmem_list *next;
346 char data[ZERO];
347 } transmem_block_t;
348 static struct transmem_list *transmem_list;
349 #else
350 typedef unsigned char transmem_block_t;
351 #endif
352 #if defined _LIBC || HAVE_ICONV
353 static const char *get_output_charset PARAMS ((struct binding *domainbinding))
354 internal_function;
355 #endif
358 /* Names for the libintl functions are a problem. They must not clash
359 with existing names and they should follow ANSI C. But this source
360 code is also used in GNU C Library where the names have a __
361 prefix. So we have to make a difference here. */
362 #ifdef _LIBC
363 # define DCIGETTEXT __dcigettext
364 #else
365 # define DCIGETTEXT libintl_dcigettext
366 #endif
368 /* Lock variable to protect the global data in the gettext implementation. */
369 #ifdef _LIBC
370 __libc_rwlock_define_initialized (, _nl_state_lock attribute_hidden)
371 #endif
373 /* Checking whether the binaries runs SUID must be done and glibc provides
374 easier methods therefore we make a difference here. */
375 #ifdef _LIBC
376 # define ENABLE_SECURE __libc_enable_secure
377 # define DETERMINE_SECURE
378 #else
379 # ifndef HAVE_GETUID
380 # define getuid() 0
381 # endif
382 # ifndef HAVE_GETGID
383 # define getgid() 0
384 # endif
385 # ifndef HAVE_GETEUID
386 # define geteuid() getuid()
387 # endif
388 # ifndef HAVE_GETEGID
389 # define getegid() getgid()
390 # endif
391 static int enable_secure;
392 # define ENABLE_SECURE (enable_secure == 1)
393 # define DETERMINE_SECURE \
394 if (enable_secure == 0) \
396 if (getuid () != geteuid () || getgid () != getegid ()) \
397 enable_secure = 1; \
398 else \
399 enable_secure = -1; \
401 #endif
403 /* Get the function to evaluate the plural expression. */
404 #include "plural-eval.c"
406 /* Look up MSGID in the DOMAINNAME message catalog for the current
407 CATEGORY locale and, if PLURAL is nonzero, search over string
408 depending on the plural form determined by N. */
409 char *
410 DCIGETTEXT (domainname, msgid1, msgid2, plural, n, category)
411 const char *domainname;
412 const char *msgid1;
413 const char *msgid2;
414 int plural;
415 unsigned long int n;
416 int category;
418 #ifndef HAVE_ALLOCA
419 struct block_list *block_list = NULL;
420 #endif
421 struct loaded_l10nfile *domain;
422 struct binding *binding;
423 const char *categoryname;
424 const char *categoryvalue;
425 char *dirname, *xdomainname;
426 char *single_locale;
427 char *retval;
428 size_t retlen;
429 int saved_errno;
430 #if defined HAVE_TSEARCH || defined _LIBC
431 struct known_translation_t search;
432 struct known_translation_t **foundp = NULL;
433 # ifdef HAVE_PER_THREAD_LOCALE
434 const char *localename;
435 # endif
436 #endif
437 size_t domainname_len;
439 /* If no real MSGID is given return NULL. */
440 if (msgid1 == NULL)
441 return NULL;
443 #ifdef _LIBC
444 if (category < 0 || category >= __LC_LAST || category == LC_ALL)
445 /* Bogus. */
446 return (plural == 0
447 ? (char *) msgid1
448 /* Use the Germanic plural rule. */
449 : n == 1 ? (char *) msgid1 : (char *) msgid2);
450 #endif
452 __libc_rwlock_rdlock (_nl_state_lock);
454 /* If DOMAINNAME is NULL, we are interested in the default domain. If
455 CATEGORY is not LC_MESSAGES this might not make much sense but the
456 definition left this undefined. */
457 if (domainname == NULL)
458 domainname = _nl_current_default_domain;
460 #if defined HAVE_TSEARCH || defined _LIBC
461 /* Try to find the translation among those which we found at
462 some time. */
463 search.domain = NULL;
464 search.msgid.ptr = msgid1;
465 search.domainname = domainname;
466 search.category = category;
467 # ifdef HAVE_PER_THREAD_LOCALE
468 # ifdef _LIBC
469 localename = __current_locale_name (category);
470 # endif
471 search.localename = localename;
472 # endif
474 /* Since tfind/tsearch manage a balanced tree, concurrent tfind and
475 tsearch calls can be fatal. */
476 __libc_rwlock_define_initialized (static, tree_lock);
477 __libc_rwlock_rdlock (tree_lock);
479 foundp = (struct known_translation_t **) tfind (&search, &root, transcmp);
481 __libc_rwlock_unlock (tree_lock);
483 if (foundp != NULL && (*foundp)->counter == _nl_msg_cat_cntr)
485 /* Now deal with plural. */
486 if (plural)
487 retval = plural_lookup ((*foundp)->domain, n, (*foundp)->translation,
488 (*foundp)->translation_length);
489 else
490 retval = (char *) (*foundp)->translation;
492 __libc_rwlock_unlock (_nl_state_lock);
493 return retval;
495 #endif
497 /* Preserve the `errno' value. */
498 saved_errno = errno;
500 /* See whether this is a SUID binary or not. */
501 DETERMINE_SECURE;
503 /* First find matching binding. */
504 for (binding = _nl_domain_bindings; binding != NULL; binding = binding->next)
506 int compare = strcmp (domainname, binding->domainname);
507 if (compare == 0)
508 /* We found it! */
509 break;
510 if (compare < 0)
512 /* It is not in the list. */
513 binding = NULL;
514 break;
518 if (binding == NULL)
519 dirname = (char *) _nl_default_dirname;
520 else if (binding->dirname[0] == '/')
521 dirname = binding->dirname;
522 else
524 /* We have a relative path. Make it absolute now. */
525 size_t dirname_len = strlen (binding->dirname) + 1;
526 size_t path_max;
527 char *ret;
529 path_max = (unsigned int) PATH_MAX;
530 path_max += 2; /* The getcwd docs say to do this. */
532 for (;;)
534 dirname = (char *) alloca (path_max + dirname_len);
535 ADD_BLOCK (block_list, dirname);
537 __set_errno (0);
538 ret = getcwd (dirname, path_max);
539 if (ret != NULL || errno != ERANGE)
540 break;
542 path_max += path_max / 2;
543 path_max += PATH_INCR;
546 if (ret == NULL)
548 /* We cannot get the current working directory. Don't signal an
549 error but simply return the default string. */
550 FREE_BLOCKS (block_list);
551 __libc_rwlock_unlock (_nl_state_lock);
552 __set_errno (saved_errno);
553 return (plural == 0
554 ? (char *) msgid1
555 /* Use the Germanic plural rule. */
556 : n == 1 ? (char *) msgid1 : (char *) msgid2);
559 stpcpy (stpcpy (strchr (dirname, '\0'), "/"), binding->dirname);
562 /* Now determine the symbolic name of CATEGORY and its value. */
563 categoryname = category_to_name (category);
564 categoryvalue = guess_category_value (category, categoryname);
566 domainname_len = strlen (domainname);
567 xdomainname = (char *) alloca (strlen (categoryname)
568 + domainname_len + 5);
569 ADD_BLOCK (block_list, xdomainname);
571 stpcpy (mempcpy (stpcpy (stpcpy (xdomainname, categoryname), "/"),
572 domainname, domainname_len),
573 ".mo");
575 /* Creating working area. */
576 single_locale = (char *) alloca (strlen (categoryvalue) + 1);
577 ADD_BLOCK (block_list, single_locale);
580 /* Search for the given string. This is a loop because we perhaps
581 got an ordered list of languages to consider for the translation. */
582 while (1)
584 /* Make CATEGORYVALUE point to the next element of the list. */
585 while (categoryvalue[0] != '\0' && categoryvalue[0] == ':')
586 ++categoryvalue;
587 if (categoryvalue[0] == '\0')
589 /* The whole contents of CATEGORYVALUE has been searched but
590 no valid entry has been found. We solve this situation
591 by implicitly appending a "C" entry, i.e. no translation
592 will take place. */
593 single_locale[0] = 'C';
594 single_locale[1] = '\0';
596 else
598 char *cp = single_locale;
599 while (categoryvalue[0] != '\0' && categoryvalue[0] != ':')
600 *cp++ = *categoryvalue++;
601 *cp = '\0';
603 /* When this is a SUID binary we must not allow accessing files
604 outside the dedicated directories. */
605 if (ENABLE_SECURE && strchr (single_locale, '/') != NULL)
606 /* Ingore this entry. */
607 continue;
610 /* If the current locale value is C (or POSIX) we don't load a
611 domain. Return the MSGID. */
612 if (strcmp (single_locale, "C") == 0
613 || strcmp (single_locale, "POSIX") == 0)
615 no_translation:
616 FREE_BLOCKS (block_list);
617 __libc_rwlock_unlock (_nl_state_lock);
618 __set_errno (saved_errno);
619 return (plural == 0
620 ? (char *) msgid1
621 /* Use the Germanic plural rule. */
622 : n == 1 ? (char *) msgid1 : (char *) msgid2);
626 /* Find structure describing the message catalog matching the
627 DOMAINNAME and CATEGORY. */
628 domain = _nl_find_domain (dirname, single_locale, xdomainname, binding);
630 if (domain != NULL)
632 retval = _nl_find_msg (domain, binding, msgid1, 1, &retlen);
634 if (retval == NULL)
636 int cnt;
638 for (cnt = 0; domain->successor[cnt] != NULL; ++cnt)
640 retval = _nl_find_msg (domain->successor[cnt], binding,
641 msgid1, 1, &retlen);
643 if (retval != NULL)
645 domain = domain->successor[cnt];
646 break;
651 /* Returning -1 means that some resource problem exists
652 (likely memory) and that the strings could not be
653 converted. Return the original strings. */
654 if (__builtin_expect (retval == (char *) -1, 0))
655 goto no_translation;
657 if (retval != NULL)
659 /* Found the translation of MSGID1 in domain DOMAIN:
660 starting at RETVAL, RETLEN bytes. */
661 FREE_BLOCKS (block_list);
662 #if defined HAVE_TSEARCH || defined _LIBC
663 if (foundp == NULL)
665 /* Create a new entry and add it to the search tree. */
666 size_t msgid_len;
667 size_t size;
668 struct known_translation_t *newp;
670 msgid_len = strlen (msgid1) + 1;
671 size = offsetof (struct known_translation_t, msgid)
672 + msgid_len + domainname_len + 1;
673 # ifdef HAVE_PER_THREAD_LOCALE
674 size += strlen (localename) + 1;
675 # endif
676 newp = (struct known_translation_t *) malloc (size);
677 if (newp != NULL)
679 char *new_domainname;
680 # ifdef HAVE_PER_THREAD_LOCALE
681 char *new_localename;
682 # endif
684 new_domainname =
685 mempcpy (newp->msgid.appended, msgid1, msgid_len);
686 memcpy (new_domainname, domainname, domainname_len + 1);
687 # ifdef HAVE_PER_THREAD_LOCALE
688 new_localename = new_domainname + domainname_len + 1;
689 strcpy (new_localename, localename);
690 # endif
691 newp->domainname = new_domainname;
692 newp->category = category;
693 # ifdef HAVE_PER_THREAD_LOCALE
694 newp->localename = new_localename;
695 # endif
696 newp->counter = _nl_msg_cat_cntr;
697 newp->domain = domain;
698 newp->translation = retval;
699 newp->translation_length = retlen;
701 __libc_rwlock_wrlock (tree_lock);
703 /* Insert the entry in the search tree. */
704 foundp = (struct known_translation_t **)
705 tsearch (newp, &root, transcmp);
707 __libc_rwlock_unlock (tree_lock);
709 if (foundp == NULL
710 || __builtin_expect (*foundp != newp, 0))
711 /* The insert failed. */
712 free (newp);
715 else
717 /* We can update the existing entry. */
718 (*foundp)->counter = _nl_msg_cat_cntr;
719 (*foundp)->domain = domain;
720 (*foundp)->translation = retval;
721 (*foundp)->translation_length = retlen;
723 #endif
724 __set_errno (saved_errno);
726 /* Now deal with plural. */
727 if (plural)
728 retval = plural_lookup (domain, n, retval, retlen);
730 __libc_rwlock_unlock (_nl_state_lock);
731 return retval;
735 /* NOTREACHED */
739 char *
740 internal_function
741 _nl_find_msg (domain_file, domainbinding, msgid, convert, lengthp)
742 struct loaded_l10nfile *domain_file;
743 struct binding *domainbinding;
744 const char *msgid;
745 int convert;
746 size_t *lengthp;
748 struct loaded_domain *domain;
749 nls_uint32 nstrings;
750 size_t act;
751 char *result;
752 size_t resultlen;
754 if (domain_file->decided <= 0)
755 _nl_load_domain (domain_file, domainbinding);
757 if (domain_file->data == NULL)
758 return NULL;
760 domain = (struct loaded_domain *) domain_file->data;
762 nstrings = domain->nstrings;
764 /* Locate the MSGID and its translation. */
765 if (domain->hash_tab != NULL)
767 /* Use the hashing table. */
768 nls_uint32 len = strlen (msgid);
769 nls_uint32 hash_val = __hash_string (msgid);
770 nls_uint32 idx = hash_val % domain->hash_size;
771 nls_uint32 incr = 1 + (hash_val % (domain->hash_size - 2));
773 while (1)
775 nls_uint32 nstr =
776 W (domain->must_swap_hash_tab, domain->hash_tab[idx]);
778 if (nstr == 0)
779 /* Hash table entry is empty. */
780 return NULL;
782 nstr--;
784 /* Compare msgid with the original string at index nstr.
785 We compare the lengths with >=, not ==, because plural entries
786 are represented by strings with an embedded NUL. */
787 if (nstr < nstrings
788 ? W (domain->must_swap, domain->orig_tab[nstr].length) >= len
789 && (strcmp (msgid,
790 domain->data + W (domain->must_swap,
791 domain->orig_tab[nstr].offset))
792 == 0)
793 : domain->orig_sysdep_tab[nstr - nstrings].length > len
794 && (strcmp (msgid,
795 domain->orig_sysdep_tab[nstr - nstrings].pointer)
796 == 0))
798 act = nstr;
799 goto found;
802 if (idx >= domain->hash_size - incr)
803 idx -= domain->hash_size - incr;
804 else
805 idx += incr;
807 /* NOTREACHED */
809 else
811 /* Try the default method: binary search in the sorted array of
812 messages. */
813 size_t top, bottom;
815 bottom = 0;
816 top = nstrings;
817 while (bottom < top)
819 int cmp_val;
821 act = (bottom + top) / 2;
822 cmp_val = strcmp (msgid, (domain->data
823 + W (domain->must_swap,
824 domain->orig_tab[act].offset)));
825 if (cmp_val < 0)
826 top = act;
827 else if (cmp_val > 0)
828 bottom = act + 1;
829 else
830 goto found;
832 /* No translation was found. */
833 return NULL;
836 found:
837 /* The translation was found at index ACT. If we have to convert the
838 string to use a different character set, this is the time. */
839 if (act < nstrings)
841 result = (char *)
842 (domain->data + W (domain->must_swap, domain->trans_tab[act].offset));
843 resultlen = W (domain->must_swap, domain->trans_tab[act].length) + 1;
845 else
847 result = (char *) domain->trans_sysdep_tab[act - nstrings].pointer;
848 resultlen = domain->trans_sysdep_tab[act - nstrings].length;
851 #if defined _LIBC || HAVE_ICONV
852 if (convert)
854 /* We are supposed to do a conversion. */
855 const char *encoding = get_output_charset (domainbinding);
857 /* Protect against reallocation of the table. */
858 __libc_rwlock_rdlock (domain->conversions_lock);
860 /* Search whether a table with converted translations for this
861 encoding has already been allocated. */
862 size_t nconversions = domain->nconversions;
863 struct converted_domain *convd = NULL;
864 size_t i;
866 for (i = nconversions; i > 0; )
868 i--;
869 if (strcmp (domain->conversions[i].encoding, encoding) == 0)
871 convd = &domain->conversions[i];
872 break;
876 __libc_rwlock_unlock (domain->conversions_lock);
878 if (convd == NULL)
880 /* We have to allocate a new conversions table. */
881 __libc_rwlock_wrlock (domain->conversions_lock);
883 /* Maybe in the meantime somebody added the translation.
884 Recheck. */
885 for (i = nconversions; i > 0; )
887 i--;
888 if (strcmp (domain->conversions[i].encoding, encoding) == 0)
890 convd = &domain->conversions[i];
891 goto found_convd;
895 /* Allocate a table for the converted translations for this
896 encoding. */
897 struct converted_domain *new_conversions =
898 (struct converted_domain *)
899 realloc (domain->conversions,
900 (nconversions + 1) * sizeof (struct converted_domain));
902 if (__builtin_expect (new_conversions == NULL, 0))
904 /* Nothing we can do, no more memory. We cannot use the
905 translation because it might be encoded incorrectly. */
906 unlock_fail:
907 __libc_rwlock_unlock (domain->conversions_lock);
908 return (char *) -1;
911 domain->conversions = new_conversions;
913 /* Copy the 'encoding' string to permanent storage. */
914 encoding = strdup (encoding);
915 if (__builtin_expect (encoding == NULL, 0))
916 /* Nothing we can do, no more memory. We cannot use the
917 translation because it might be encoded incorrectly. */
918 goto unlock_fail;
920 convd = &new_conversions[nconversions];
921 convd->encoding = encoding;
923 /* Find out about the character set the file is encoded with.
924 This can be found (in textual form) in the entry "". If this
925 entry does not exist or if this does not contain the 'charset='
926 information, we will assume the charset matches the one the
927 current locale and we don't have to perform any conversion. */
928 # ifdef _LIBC
929 convd->conv = (__gconv_t) -1;
930 # else
931 # if HAVE_ICONV
932 convd->conv = (iconv_t) -1;
933 # endif
934 # endif
936 char *nullentry;
937 size_t nullentrylen;
939 /* Get the header entry. This is a recursion, but it doesn't
940 reallocate domain->conversions because we pass convert = 0. */
941 nullentry =
942 _nl_find_msg (domain_file, domainbinding, "", 0, &nullentrylen);
944 if (nullentry != NULL)
946 const char *charsetstr;
948 charsetstr = strstr (nullentry, "charset=");
949 if (charsetstr != NULL)
951 size_t len;
952 char *charset;
953 const char *outcharset;
955 charsetstr += strlen ("charset=");
956 len = strcspn (charsetstr, " \t\n");
958 charset = (char *) alloca (len + 1);
959 # if defined _LIBC || HAVE_MEMPCPY
960 *((char *) mempcpy (charset, charsetstr, len)) = '\0';
961 # else
962 memcpy (charset, charsetstr, len);
963 charset[len] = '\0';
964 # endif
966 outcharset = encoding;
968 # ifdef _LIBC
969 /* We always want to use transliteration. */
970 outcharset = norm_add_slashes (outcharset, "TRANSLIT");
971 charset = norm_add_slashes (charset, "");
972 int r = __gconv_open (outcharset, charset, &convd->conv,
973 GCONV_AVOID_NOCONV);
974 if (__builtin_expect (r != __GCONV_OK, 0))
976 /* If the output encoding is the same there is
977 nothing to do. Otherwise do not use the
978 translation at all. */
979 if (__builtin_expect (r != __GCONV_NULCONV, 1))
981 __libc_rwlock_unlock (domain->conversions_lock);
982 free ((char *) encoding);
983 return NULL;
986 convd->conv = (__gconv_t) -1;
988 # else
989 # if HAVE_ICONV
990 /* When using GNU libc >= 2.2 or GNU libiconv >= 1.5,
991 we want to use transliteration. */
992 # if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 \
993 || _LIBICONV_VERSION >= 0x0105
994 if (strchr (outcharset, '/') == NULL)
996 char *tmp;
998 len = strlen (outcharset);
999 tmp = (char *) alloca (len + 10 + 1);
1000 memcpy (tmp, outcharset, len);
1001 memcpy (tmp + len, "//TRANSLIT", 10 + 1);
1002 outcharset = tmp;
1004 convd->conv = iconv_open (outcharset, charset);
1006 freea (outcharset);
1008 else
1009 # endif
1010 convd->conv = iconv_open (outcharset, charset);
1011 # endif
1012 # endif
1014 freea (charset);
1018 convd->conv_tab = NULL;
1019 /* Here domain->conversions is still == new_conversions. */
1020 domain->nconversions++;
1022 found_convd:
1023 __libc_rwlock_unlock (domain->conversions_lock);
1026 if (
1027 # ifdef _LIBC
1028 convd->conv != (__gconv_t) -1
1029 # else
1030 # if HAVE_ICONV
1031 convd->conv != (iconv_t) -1
1032 # endif
1033 # endif
1036 /* We are supposed to do a conversion. First allocate an
1037 appropriate table with the same structure as the table
1038 of translations in the file, where we can put the pointers
1039 to the converted strings in.
1040 There is a slight complication with plural entries. They
1041 are represented by consecutive NUL terminated strings. We
1042 handle this case by converting RESULTLEN bytes, including
1043 NULs. */
1045 if (convd->conv_tab == NULL
1046 && ((convd->conv_tab =
1047 (char **) calloc (nstrings + domain->n_sysdep_strings,
1048 sizeof (char *)))
1049 == NULL))
1050 /* Mark that we didn't succeed allocating a table. */
1051 convd->conv_tab = (char **) -1;
1053 if (__builtin_expect (convd->conv_tab == (char **) -1, 0))
1054 /* Nothing we can do, no more memory. We cannot use the
1055 translation because it might be encoded incorrectly. */
1056 return (char *) -1;
1058 if (convd->conv_tab[act] == NULL)
1060 /* We haven't used this string so far, so it is not
1061 translated yet. Do this now. */
1062 /* We use a bit more efficient memory handling.
1063 We allocate always larger blocks which get used over
1064 time. This is faster than many small allocations. */
1065 __libc_lock_define_initialized (static, lock)
1066 # define INITIAL_BLOCK_SIZE 4080
1067 static unsigned char *freemem;
1068 static size_t freemem_size;
1070 const unsigned char *inbuf;
1071 unsigned char *outbuf;
1072 int malloc_count;
1073 # ifndef _LIBC
1074 transmem_block_t *transmem_list = NULL;
1075 # endif
1077 __libc_lock_lock (lock);
1079 inbuf = (const unsigned char *) result;
1080 outbuf = freemem + sizeof (size_t);
1082 malloc_count = 0;
1083 while (1)
1085 transmem_block_t *newmem;
1086 # ifdef _LIBC
1087 size_t non_reversible;
1088 int res;
1090 if (freemem_size < sizeof (size_t))
1091 goto resize_freemem;
1093 res = __gconv (convd->conv,
1094 &inbuf, inbuf + resultlen,
1095 &outbuf,
1096 outbuf + freemem_size - sizeof (size_t),
1097 &non_reversible);
1099 if (res == __GCONV_OK || res == __GCONV_EMPTY_INPUT)
1100 break;
1102 if (res != __GCONV_FULL_OUTPUT)
1104 /* We should not use the translation at all, it
1105 is incorrectly encoded. */
1106 __libc_lock_unlock (lock);
1107 return NULL;
1110 inbuf = (const unsigned char *) result;
1111 # else
1112 # if HAVE_ICONV
1113 const char *inptr = (const char *) inbuf;
1114 size_t inleft = resultlen;
1115 char *outptr = (char *) outbuf;
1116 size_t outleft;
1118 if (freemem_size < sizeof (size_t))
1119 goto resize_freemem;
1121 outleft = freemem_size - sizeof (size_t);
1122 if (iconv (convd->conv,
1123 (ICONV_CONST char **) &inptr, &inleft,
1124 &outptr, &outleft)
1125 != (size_t) (-1))
1127 outbuf = (unsigned char *) outptr;
1128 break;
1130 if (errno != E2BIG)
1132 __libc_lock_unlock (lock);
1133 return NULL;
1135 # endif
1136 # endif
1138 resize_freemem:
1139 /* We must allocate a new buffer or resize the old one. */
1140 if (malloc_count > 0)
1142 ++malloc_count;
1143 freemem_size = malloc_count * INITIAL_BLOCK_SIZE;
1144 newmem = (transmem_block_t *) realloc (transmem_list,
1145 freemem_size);
1146 # ifdef _LIBC
1147 if (newmem != NULL)
1148 transmem_list = transmem_list->next;
1149 else
1151 struct transmem_list *old = transmem_list;
1153 transmem_list = transmem_list->next;
1154 free (old);
1156 # endif
1158 else
1160 malloc_count = 1;
1161 freemem_size = INITIAL_BLOCK_SIZE;
1162 newmem = (transmem_block_t *) malloc (freemem_size);
1164 if (__builtin_expect (newmem == NULL, 0))
1166 freemem = NULL;
1167 freemem_size = 0;
1168 __libc_lock_unlock (lock);
1169 return (char *) -1;
1172 # ifdef _LIBC
1173 /* Add the block to the list of blocks we have to free
1174 at some point. */
1175 newmem->next = transmem_list;
1176 transmem_list = newmem;
1178 freemem = (unsigned char *) newmem->data;
1179 freemem_size -= offsetof (struct transmem_list, data);
1180 # else
1181 transmem_list = newmem;
1182 freemem = newmem;
1183 # endif
1185 outbuf = freemem + sizeof (size_t);
1188 /* We have now in our buffer a converted string. Put this
1189 into the table of conversions. */
1190 *(size_t *) freemem = outbuf - freemem - sizeof (size_t);
1191 convd->conv_tab[act] = (char *) freemem;
1192 /* Shrink freemem, but keep it aligned. */
1193 freemem_size -= outbuf - freemem;
1194 freemem = outbuf;
1195 freemem += freemem_size & (alignof (size_t) - 1);
1196 freemem_size = freemem_size & ~ (alignof (size_t) - 1);
1198 __libc_lock_unlock (lock);
1201 /* Now convd->conv_tab[act] contains the translation of all
1202 the plural variants. */
1203 result = convd->conv_tab[act] + sizeof (size_t);
1204 resultlen = *(size_t *) convd->conv_tab[act];
1208 /* The result string is converted. */
1210 #endif /* _LIBC || HAVE_ICONV */
1212 *lengthp = resultlen;
1213 return result;
1217 /* Look up a plural variant. */
1218 static char *
1219 internal_function
1220 plural_lookup (domain, n, translation, translation_len)
1221 struct loaded_l10nfile *domain;
1222 unsigned long int n;
1223 const char *translation;
1224 size_t translation_len;
1226 struct loaded_domain *domaindata = (struct loaded_domain *) domain->data;
1227 unsigned long int index;
1228 const char *p;
1230 index = plural_eval (domaindata->plural, n);
1231 if (index >= domaindata->nplurals)
1232 /* This should never happen. It means the plural expression and the
1233 given maximum value do not match. */
1234 index = 0;
1236 /* Skip INDEX strings at TRANSLATION. */
1237 p = translation;
1238 while (index-- > 0)
1240 #ifdef _LIBC
1241 p = __rawmemchr (p, '\0');
1242 #else
1243 p = strchr (p, '\0');
1244 #endif
1245 /* And skip over the NUL byte. */
1246 p++;
1248 if (p >= translation + translation_len)
1249 /* This should never happen. It means the plural expression
1250 evaluated to a value larger than the number of variants
1251 available for MSGID1. */
1252 return (char *) translation;
1254 return (char *) p;
1257 #ifndef _LIBC
1258 /* Return string representation of locale CATEGORY. */
1259 static const char *
1260 internal_function
1261 category_to_name (category)
1262 int category;
1264 const char *retval;
1266 switch (category)
1268 #ifdef LC_COLLATE
1269 case LC_COLLATE:
1270 retval = "LC_COLLATE";
1271 break;
1272 #endif
1273 #ifdef LC_CTYPE
1274 case LC_CTYPE:
1275 retval = "LC_CTYPE";
1276 break;
1277 #endif
1278 #ifdef LC_MONETARY
1279 case LC_MONETARY:
1280 retval = "LC_MONETARY";
1281 break;
1282 #endif
1283 #ifdef LC_NUMERIC
1284 case LC_NUMERIC:
1285 retval = "LC_NUMERIC";
1286 break;
1287 #endif
1288 #ifdef LC_TIME
1289 case LC_TIME:
1290 retval = "LC_TIME";
1291 break;
1292 #endif
1293 #ifdef LC_MESSAGES
1294 case LC_MESSAGES:
1295 retval = "LC_MESSAGES";
1296 break;
1297 #endif
1298 #ifdef LC_RESPONSE
1299 case LC_RESPONSE:
1300 retval = "LC_RESPONSE";
1301 break;
1302 #endif
1303 #ifdef LC_ALL
1304 case LC_ALL:
1305 /* This might not make sense but is perhaps better than any other
1306 value. */
1307 retval = "LC_ALL";
1308 break;
1309 #endif
1310 default:
1311 /* If you have a better idea for a default value let me know. */
1312 retval = "LC_XXX";
1315 return retval;
1317 #endif
1319 /* Guess value of current locale from value of the environment variables. */
1320 static const char *
1321 internal_function
1322 guess_category_value (category, categoryname)
1323 int category;
1324 const char *categoryname;
1326 const char *language;
1327 const char *retval;
1329 /* The highest priority value is the `LANGUAGE' environment
1330 variable. But we don't use the value if the currently selected
1331 locale is the C locale. This is a GNU extension. */
1332 language = getenv ("LANGUAGE");
1333 if (language != NULL && language[0] == '\0')
1334 language = NULL;
1336 /* We have to proceed with the POSIX methods of looking to `LC_ALL',
1337 `LC_xxx', and `LANG'. On some systems this can be done by the
1338 `setlocale' function itself. */
1339 #ifdef _LIBC
1340 retval = __current_locale_name (category);
1341 #else
1342 retval = _nl_locale_name (category, categoryname);
1343 #endif
1345 return language != NULL && strcmp (retval, "C") != 0 ? language : retval;
1348 #if defined _LIBC || HAVE_ICONV
1349 /* Returns the output charset. */
1350 static const char *
1351 internal_function
1352 get_output_charset (domainbinding)
1353 struct binding *domainbinding;
1355 /* The output charset should normally be determined by the locale. But
1356 sometimes the locale is not used or not correctly set up, so we provide
1357 a possibility for the user to override this: the OUTPUT_CHARSET
1358 environment variable. Moreover, the value specified through
1359 bind_textdomain_codeset overrides both. */
1360 if (domainbinding != NULL && domainbinding->codeset != NULL)
1361 return domainbinding->codeset;
1362 else
1364 /* For speed reasons, we look at the value of OUTPUT_CHARSET only
1365 once. This is a user variable that is not supposed to change
1366 during a program run. */
1367 static char *output_charset_cache;
1368 static int output_charset_cached;
1370 if (!output_charset_cached)
1372 const char *value = getenv ("OUTPUT_CHARSET");
1374 if (value != NULL && value[0] != '\0')
1376 size_t len = strlen (value) + 1;
1377 char *value_copy = (char *) malloc (len);
1379 if (value_copy != NULL)
1380 memcpy (value_copy, value, len);
1381 output_charset_cache = value_copy;
1383 output_charset_cached = 1;
1386 if (output_charset_cache != NULL)
1387 return output_charset_cache;
1388 else
1390 # ifdef _LIBC
1391 return _NL_CURRENT (LC_CTYPE, CODESET);
1392 # else
1393 # if HAVE_ICONV
1394 extern const char *locale_charset PARAMS ((void);
1395 return locale_charset ();
1396 # endif
1397 # endif
1401 #endif
1403 /* @@ begin of epilog @@ */
1405 /* We don't want libintl.a to depend on any other library. So we
1406 avoid the non-standard function stpcpy. In GNU C Library this
1407 function is available, though. Also allow the symbol HAVE_STPCPY
1408 to be defined. */
1409 #if !_LIBC && !HAVE_STPCPY
1410 static char *
1411 stpcpy (dest, src)
1412 char *dest;
1413 const char *src;
1415 while ((*dest++ = *src++) != '\0')
1416 /* Do nothing. */ ;
1417 return dest - 1;
1419 #endif
1421 #if !_LIBC && !HAVE_MEMPCPY
1422 static void *
1423 mempcpy (dest, src, n)
1424 void *dest;
1425 const void *src;
1426 size_t n;
1428 return (void *) ((char *) memcpy (dest, src, n) + n);
1430 #endif
1433 #ifdef _LIBC
1434 /* If we want to free all resources we have to do some work at
1435 program's end. */
1436 libc_freeres_fn (free_mem)
1438 void *old;
1440 while (_nl_domain_bindings != NULL)
1442 struct binding *oldp = _nl_domain_bindings;
1443 _nl_domain_bindings = _nl_domain_bindings->next;
1444 if (oldp->dirname != _nl_default_dirname)
1445 /* Yes, this is a pointer comparison. */
1446 free (oldp->dirname);
1447 free (oldp->codeset);
1448 free (oldp);
1451 if (_nl_current_default_domain != _nl_default_default_domain)
1452 /* Yes, again a pointer comparison. */
1453 free ((char *) _nl_current_default_domain);
1455 /* Remove the search tree with the known translations. */
1456 __tdestroy (root, free);
1457 root = NULL;
1459 while (transmem_list != NULL)
1461 old = transmem_list;
1462 transmem_list = transmem_list->next;
1463 free (old);
1466 #endif