1/* Implementation of the internal dcigettext function.
2 Copyright (C) 1995-2024 Free Software Foundation, Inc.
3
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU Lesser General Public License as published by
6 the Free Software Foundation; either version 2.1 of the License, or
7 (at your option) any later version.
8
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU Lesser General Public License for more details.
13
14 You should have received a copy of the GNU Lesser General Public License
15 along with this program. If not, see <https://www.gnu.org/licenses/>. */
16
17/* Tell glibc's <string.h> to provide a prototype for mempcpy().
18 This must come before <config.h> because <config.h> may include
19 <features.h>, and once <features.h> has been included, it's too late. */
20#ifndef _GNU_SOURCE
21# define _GNU_SOURCE 1
22#endif
23
24#ifdef HAVE_CONFIG_H
25# include <config.h>
26#endif
27
28#include <sys/types.h>
29
30#ifdef __GNUC__
31# define alloca __builtin_alloca
32# define HAVE_ALLOCA 1
33#else
34# ifdef _MSC_VER
35# include <malloc.h>
36# define alloca _alloca
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
45char *alloca ();
46# endif
47# endif
48# endif
49# endif
50#endif
51
52#include <errno.h>
53#ifndef errno
54extern int errno;
55#endif
56#ifndef __set_errno
57# define __set_errno(val) errno = (val)
58#endif
59
60#include <stddef.h>
61#include <stdlib.h>
62#include <stdio.h>
63#include <string.h>
64
65#if defined HAVE_UNISTD_H || defined _LIBC
66# include <unistd.h>
67#endif
68
69#include <locale.h>
70
71#ifdef _LIBC
72 /* Guess whether integer division by zero raises signal SIGFPE.
73 Set to 1 only if you know for sure. In case of doubt, set to 0. */
74# if defined __alpha__ || defined __arm__ || defined __i386__ \
75 || defined __m68k__ || defined __s390__
76# define INTDIV0_RAISES_SIGFPE 1
77# else
78# define INTDIV0_RAISES_SIGFPE 0
79# endif
80#endif
81#if !INTDIV0_RAISES_SIGFPE
82# include <signal.h>
83#endif
84
85#if defined HAVE_SYS_PARAM_H || defined _LIBC
86# include <sys/param.h>
87#endif
88
89#if !defined _LIBC
90# include "localcharset.h"
91#endif
92
93#include "gettextP.h"
94#include "plural-exp.h"
95#ifdef _LIBC
96# include <libintl.h>
97#else
98# ifdef IN_LIBGLOCALE
99# include <libintl.h>
100# endif
101# include "libgnuintl.h"
102#endif
103#include "hash-string.h"
104
105/* Handle multi-threaded applications. */
106#ifdef _LIBC
107# include <libc-lock.h>
108# define gl_rwlock_define_initialized __libc_rwlock_define_initialized
109# define gl_rwlock_rdlock __libc_rwlock_rdlock
110# define gl_rwlock_wrlock __libc_rwlock_wrlock
111# define gl_rwlock_unlock __libc_rwlock_unlock
112#else
113# include "lock.h"
114#endif
115
116/* Alignment of types. */
117#if defined __GNUC__ && __GNUC__ >= 2
118# define alignof(TYPE) __alignof__ (TYPE)
119#else
120# define alignof(TYPE) \
121 ((int) &((struct { char dummy1; TYPE dummy2; } *) 0)->dummy2)
122#endif
123
124/* Some compilers, like SunOS4 cc, don't have offsetof in <stddef.h>. */
125#ifndef offsetof
126# define offsetof(type,ident) ((size_t)&(((type*)0)->ident))
127#endif
128
129/* @@ end of prolog @@ */
130
131#ifdef _LIBC
132/* Rename the non ANSI C functions. This is required by the standard
133 because some ANSI C functions will require linking with this object
134 file and the name space must not be polluted. */
135# define strdup __strdup
136# define getcwd __getcwd
137# ifndef stpcpy
138# define stpcpy __stpcpy
139# endif
140# define tfind __tfind
141#else
142# if !defined HAVE_GETCWD
143char *getwd ();
144# define getcwd(buf, max) getwd (buf)
145# else
146# if VMS
147# define getcwd(buf, max) (getcwd) (buf, max, 0)
148# else
149char *getcwd ();
150# endif
151# endif
152# ifndef HAVE_STPCPY
153static char *stpcpy (char *dest, const char *src);
154# endif
155# ifndef HAVE_MEMPCPY
156static void *mempcpy (void *dest, const void *src, size_t n);
157# endif
158#endif
159
160/* Use a replacement if the system does not provide the `tsearch' function
161 family. */
162#if defined HAVE_TSEARCH || defined _LIBC
163# include <search.h>
164#else
165# define tsearch libintl_tsearch
166# define tfind libintl_tfind
167# define tdelete libintl_tdelete
168# define twalk libintl_twalk
169# include "tsearch.h"
170#endif
171
172#ifdef _LIBC
173# define tsearch __tsearch
174#endif
175
176/* Amount to increase buffer size by in each try. */
177#define PATH_INCR 32
178
179/* The following is from pathmax.h. */
180/* Non-POSIX BSD systems might have gcc's limits.h, which doesn't define
181 PATH_MAX but might cause redefinition warnings when sys/param.h is
182 later included (as on MORE/BSD 4.3). */
183#if defined _POSIX_VERSION || (defined HAVE_LIMITS_H && !defined __GNUC__)
184# include <limits.h>
185#endif
186
187#ifndef _POSIX_PATH_MAX
188# define _POSIX_PATH_MAX 255
189#endif
190
191#if !defined PATH_MAX && defined _PC_PATH_MAX
192# define PATH_MAX (__pathconf ("/", _PC_PATH_MAX) < 1 ? 1024 : __pathconf ("/", _PC_PATH_MAX))
193#endif
194
195/* Don't include sys/param.h if it already has been. */
196#if defined HAVE_SYS_PARAM_H && !defined PATH_MAX && !defined MAXPATHLEN
197# include <sys/param.h>
198#endif
199
200#if !defined PATH_MAX && defined MAXPATHLEN
201# define PATH_MAX MAXPATHLEN
202#endif
203
204#ifndef PATH_MAX
205# define PATH_MAX _POSIX_PATH_MAX
206#endif
207
208/* Pathname support.
209 ISSLASH(C) tests whether C is a directory separator character.
210 IS_ABSOLUTE_PATH(P) tests whether P is an absolute path. If it is not,
211 it may be concatenated to a directory pathname.
212 IS_PATH_WITH_DIR(P) tests whether P contains a directory specification.
213 */
214#if defined _WIN32 || defined __WIN32__ || defined __CYGWIN__ || defined __EMX__ || defined __DJGPP__
215 /* Win32, Cygwin, OS/2, DOS */
216# define ISSLASH(C) ((C) == '/' || (C) == '\\')
217# define HAS_DEVICE(P) \
218 ((((P)[0] >= 'A' && (P)[0] <= 'Z') || ((P)[0] >= 'a' && (P)[0] <= 'z')) \
219 && (P)[1] == ':')
220# define IS_ABSOLUTE_PATH(P) (ISSLASH ((P)[0]) || HAS_DEVICE (P))
221# define IS_PATH_WITH_DIR(P) \
222 (strchr (P, '/') != NULL || strchr (P, '\\') != NULL || HAS_DEVICE (P))
223#else
224 /* Unix */
225# define ISSLASH(C) ((C) == '/')
226# define IS_ABSOLUTE_PATH(P) ISSLASH ((P)[0])
227# define IS_PATH_WITH_DIR(P) (strchr (P, '/') != NULL)
228#endif
229
230/* Whether to support different locales in different threads. */
231#if defined _LIBC || HAVE_USELOCALE || defined IN_LIBGLOCALE
232# define HAVE_PER_THREAD_LOCALE
233#endif
234
235/* This is the type used for the search tree where known translations
236 are stored. */
237struct known_translation_t
238{
239 /* Domain in which to search. */
240 const char *domainname;
241
242 /* The category. */
243 int category;
244
245#ifdef HAVE_PER_THREAD_LOCALE
246 /* Name of the relevant locale category, or "" for the global locale. */
247 const char *localename;
248#endif
249
250#ifdef IN_LIBGLOCALE
251 /* The character encoding. */
252 const char *encoding;
253#endif
254
255 /* State of the catalog counter at the point the string was found. */
256 int counter;
257
258 /* Catalog where the string was found. */
259 struct loaded_l10nfile *domain;
260
261 /* And finally the translation. */
262 const char *translation;
263 size_t translation_length;
264
265 /* Pointer to the string in question. */
266 union
267 {
268 char appended[ZERO]; /* used if domain != NULL */
269 const char *ptr; /* used if domain == NULL */
270 }
271 msgid;
272};
273
274gl_rwlock_define_initialized (static, tree_lock)
275
276/* Root of the search tree with known translations. */
277static void *root;
278
279/* Function to compare two entries in the table of known translations. */
280static int
281transcmp (const void *p1, const void *p2)
282{
283 const struct known_translation_t *s1;
284 const struct known_translation_t *s2;
285 int result;
286
287 s1 = (const struct known_translation_t *) p1;
288 s2 = (const struct known_translation_t *) p2;
289
290 result = strcmp (s1->domain != NULL ? s1->msgid.appended : s1->msgid.ptr,
291 s2->domain != NULL ? s2->msgid.appended : s2->msgid.ptr);
292 if (result == 0)
293 {
294 result = strcmp (s1->domainname, s2->domainname);
295 if (result == 0)
296 {
297#ifdef HAVE_PER_THREAD_LOCALE
298 result = strcmp (s1->localename, s2->localename);
299 if (result == 0)
300#endif
301 {
302#ifdef IN_LIBGLOCALE
303 result = strcmp (s1->encoding, s2->encoding);
304 if (result == 0)
305#endif
306 /* We compare the category last (though this is the cheapest
307 operation) since it is hopefully always the same (namely
308 LC_MESSAGES). */
309 result = s1->category - s2->category;
310 }
311 }
312 }
313
314 return result;
315}
316
317/* Name of the default domain used for gettext(3) prior any call to
318 textdomain(3). The default value for this is "messages". */
319const char _nl_default_default_domain[] attribute_hidden = "messages";
320
321#ifndef IN_LIBGLOCALE
322/* Value used as the default domain for gettext(3). */
323const char *_nl_current_default_domain attribute_hidden
324 = _nl_default_default_domain;
325#endif
326
327/* Contains the default location of the message catalogs. */
328#if defined __EMX__
329extern const char _nl_default_dirname[];
330#else
331# ifdef _LIBC
332extern const char _nl_default_dirname[];
333libc_hidden_proto (_nl_default_dirname)
334# endif
335const char _nl_default_dirname[] = LOCALEDIR;
336# ifdef _LIBC
337libc_hidden_data_def (_nl_default_dirname)
338# endif
339#endif
340
341#ifndef IN_LIBGLOCALE
342/* List with bindings of specific domains created by bindtextdomain()
343 calls. */
344struct binding *_nl_domain_bindings;
345#endif
346
347/* Prototypes for local functions. */
348static char *plural_lookup (struct loaded_l10nfile *domain,
349 unsigned long int n,
350 const char *translation, size_t translation_len);
351
352#ifdef IN_LIBGLOCALE
353static const char *guess_category_value (int category,
354 const char *categoryname,
355 const char *localename);
356#else
357static const char *guess_category_value (int category,
358 const char *categoryname);
359#endif
360
361#ifdef _LIBC
362# include "../locale/localeinfo.h"
363# define category_to_name(category) _nl_category_names_get (category)
364#else
365static const char *category_to_name (int category);
366#endif
367#if (defined _LIBC || HAVE_ICONV) && !defined IN_LIBGLOCALE
368static const char *get_output_charset (struct binding *domainbinding);
369#endif
370
371
372/* For those losing systems which don't have `alloca' we have to add
373 some additional code emulating it. */
374#ifdef HAVE_ALLOCA
375/* Nothing has to be done. */
376# define freea(p) /* nothing */
377# define ADD_BLOCK(list, address) /* nothing */
378# define FREE_BLOCKS(list) /* nothing */
379#else
380struct block_list
381{
382 void *address;
383 struct block_list *next;
384};
385# define ADD_BLOCK(list, addr) \
386 do { \
387 struct block_list *newp = (struct block_list *) malloc (sizeof (*newp)); \
388 /* If we cannot get a free block we cannot add the new element to \
389 the list. */ \
390 if (newp != NULL) { \
391 newp->address = (addr); \
392 newp->next = (list); \
393 (list) = newp; \
394 } \
395 } while (0)
396# define FREE_BLOCKS(list) \
397 do { \
398 while (list != NULL) { \
399 struct block_list *old = list; \
400 list = list->next; \
401 free (old->address); \
402 free (old); \
403 } \
404 } while (0)
405# undef alloca
406# define alloca(size) (malloc (size))
407# define freea(p) free (p)
408#endif /* have alloca */
409
410
411#ifdef _LIBC
412/* List of blocks allocated for translations. */
413typedef struct transmem_list
414{
415 struct transmem_list *next;
416 char data[ZERO];
417} transmem_block_t;
418static struct transmem_list *transmem_list;
419#else
420typedef unsigned char transmem_block_t;
421#endif
422
423
424/* Names for the libintl functions are a problem. They must not clash
425 with existing names and they should follow ANSI C. But this source
426 code is also used in GNU C Library where the names have a __
427 prefix. So we have to make a difference here. */
428#ifdef _LIBC
429# define DCIGETTEXT __dcigettext
430#else
431# define DCIGETTEXT libintl_dcigettext
432#endif
433
434/* Lock variable to protect the global data in the gettext implementation. */
435gl_rwlock_define_initialized (, _nl_state_lock attribute_hidden)
436
437/* Checking whether the binaries runs SUID must be done and glibc provides
438 easier methods therefore we make a difference here. */
439#ifdef _LIBC
440# define ENABLE_SECURE __libc_enable_secure
441# define DETERMINE_SECURE
442#else
443# ifndef HAVE_GETUID
444# define getuid() 0
445# endif
446# ifndef HAVE_GETGID
447# define getgid() 0
448# endif
449# ifndef HAVE_GETEUID
450# define geteuid() getuid()
451# endif
452# ifndef HAVE_GETEGID
453# define getegid() getgid()
454# endif
455static int enable_secure;
456# define ENABLE_SECURE (enable_secure == 1)
457# define DETERMINE_SECURE \
458 if (enable_secure == 0) \
459 { \
460 if (getuid () != geteuid () || getgid () != getegid ()) \
461 enable_secure = 1; \
462 else \
463 enable_secure = -1; \
464 }
465#endif
466
467/* Get the function to evaluate the plural expression. */
468#include "eval-plural.h"
469
470/* Look up MSGID in the DOMAINNAME message catalog for the current
471 CATEGORY locale and, if PLURAL is nonzero, search over string
472 depending on the plural form determined by N. */
473#ifdef IN_LIBGLOCALE
474char *
475gl_dcigettext (const char *domainname,
476 const char *msgid1, const char *msgid2,
477 int plural, unsigned long int n,
478 int category,
479 const char *localename, const char *encoding)
480#else
481char *
482DCIGETTEXT (const char *domainname, const char *msgid1, const char *msgid2,
483 int plural, unsigned long int n, int category)
484#endif
485{
486#ifndef HAVE_ALLOCA
487 struct block_list *block_list = NULL;
488#endif
489 struct loaded_l10nfile *domain;
490 struct binding *binding;
491 const char *categoryname;
492 const char *categoryvalue;
493 const char *dirname;
494 char *xdirname = NULL;
495 char *xdomainname;
496 char *single_locale;
497 char *retval;
498 size_t retlen;
499 int saved_errno;
500 struct known_translation_t search;
501 struct known_translation_t **foundp = NULL;
502#if defined HAVE_PER_THREAD_LOCALE && !defined IN_LIBGLOCALE
503 const char *localename;
504#endif
505 size_t domainname_len;
506
507 /* If no real MSGID is given return NULL. */
508 if (msgid1 == NULL)
509 return NULL;
510
511#ifdef _LIBC
512 if (category < 0 || category >= __LC_LAST || category == LC_ALL)
513 /* Bogus. */
514 return (plural == 0
515 ? (char *) msgid1
516 /* Use the Germanic plural rule. */
517 : n == 1 ? (char *) msgid1 : (char *) msgid2);
518#endif
519
520 /* Preserve the `errno' value. */
521 saved_errno = errno;
522
523#ifdef _LIBC
524 __libc_rwlock_define (extern, __libc_setlocale_lock attribute_hidden)
525 __libc_rwlock_rdlock (__libc_setlocale_lock);
526#endif
527
528 gl_rwlock_rdlock (_nl_state_lock);
529
530 /* If DOMAINNAME is NULL, we are interested in the default domain. If
531 CATEGORY is not LC_MESSAGES this might not make much sense but the
532 definition left this undefined. */
533 if (domainname == NULL)
534 domainname = _nl_current_default_domain;
535
536 /* OS/2 specific: backward compatibility with older libintl versions */
537#ifdef LC_MESSAGES_COMPAT
538 if (category == LC_MESSAGES_COMPAT)
539 category = LC_MESSAGES;
540#endif
541
542 /* Try to find the translation among those which we found at
543 some time. */
544 search.domain = NULL;
545 search.msgid.ptr = msgid1;
546 search.domainname = domainname;
547 search.category = category;
548#ifdef HAVE_PER_THREAD_LOCALE
549# ifndef IN_LIBGLOCALE
550# ifdef _LIBC
551 localename = __current_locale_name (category);
552# else
553 categoryname = category_to_name (category);
554# define CATEGORYNAME_INITIALIZED
555 localename = _nl_locale_name_thread_unsafe (category, categoryname);
556 if (localename == NULL)
557 localename = "";
558# endif
559# endif
560 search.localename = localename;
561# ifdef IN_LIBGLOCALE
562 search.encoding = encoding;
563# endif
564
565 /* Since tfind/tsearch manage a balanced tree, concurrent tfind and
566 tsearch calls can be fatal. */
567 gl_rwlock_rdlock (tree_lock);
568
569 foundp = (struct known_translation_t **) tfind (&search, &root, transcmp);
570
571 gl_rwlock_unlock (tree_lock);
572
573 if (foundp != NULL && (*foundp)->counter == _nl_msg_cat_cntr)
574 {
575 /* Now deal with plural. */
576 if (plural)
577 retval = plural_lookup (domain: (*foundp)->domain, n, translation: (*foundp)->translation,
578 translation_len: (*foundp)->translation_length);
579 else
580 retval = (char *) (*foundp)->translation;
581
582 gl_rwlock_unlock (_nl_state_lock);
583# ifdef _LIBC
584 __libc_rwlock_unlock (__libc_setlocale_lock);
585# endif
586 __set_errno (saved_errno);
587 return retval;
588 }
589#endif
590
591 /* See whether this is a SUID binary or not. */
592 DETERMINE_SECURE;
593
594 /* First find matching binding. */
595#ifdef IN_LIBGLOCALE
596 /* We can use a trivial binding, since _nl_find_msg will ignore it anyway,
597 and _nl_load_domain and _nl_find_domain just pass it through. */
598 binding = NULL;
599 dirname = bindtextdomain (domainname, NULL);
600#else
601 for (binding = _nl_domain_bindings; binding != NULL; binding = binding->next)
602 {
603 int compare = strcmp (domainname, binding->domainname);
604 if (compare == 0)
605 /* We found it! */
606 break;
607 if (compare < 0)
608 {
609 /* It is not in the list. */
610 binding = NULL;
611 break;
612 }
613 }
614
615 if (binding == NULL)
616 dirname = _nl_default_dirname;
617 else
618 {
619 dirname = binding->dirname;
620#endif
621 if (!IS_ABSOLUTE_PATH (dirname))
622 {
623 /* We have a relative path. Make it absolute now. */
624 char *cwd = getcwd (NULL, 0);
625 if (cwd == NULL)
626 /* We cannot get the current working directory. Don't
627 signal an error but simply return the default
628 string. */
629 goto return_untranslated;
630 int ret = __asprintf (&xdirname, "%s/%s", cwd, dirname);
631 free (ptr: cwd);
632 if (ret < 0)
633 goto return_untranslated;
634 dirname = xdirname;
635 }
636#ifndef IN_LIBGLOCALE
637 }
638#endif
639
640 /* Now determine the symbolic name of CATEGORY and its value. */
641#ifndef CATEGORYNAME_INITIALIZED
642 categoryname = category_to_name (category);
643#endif
644#ifdef IN_LIBGLOCALE
645 categoryvalue = guess_category_value (category, categoryname, localename);
646#else
647 categoryvalue = guess_category_value (category, categoryname);
648#endif
649
650 domainname_len = strlen (domainname);
651 xdomainname = (char *) alloca (strlen (categoryname)
652 + domainname_len + 5);
653 ADD_BLOCK (block_list, xdomainname);
654
655 stpcpy ((char *) mempcpy (stpcpy (stpcpy (xdomainname, categoryname), "/"),
656 domainname, domainname_len),
657 ".mo");
658
659 /* Creating working area. */
660 single_locale = (char *) alloca (strlen (categoryvalue) + 1);
661 ADD_BLOCK (block_list, single_locale);
662
663
664 /* Search for the given string. This is a loop because we perhaps
665 got an ordered list of languages to consider for the translation. */
666 while (1)
667 {
668 /* Make CATEGORYVALUE point to the next element of the list. */
669 while (categoryvalue[0] != '\0' && categoryvalue[0] == ':')
670 ++categoryvalue;
671 if (categoryvalue[0] == '\0')
672 {
673 /* The whole contents of CATEGORYVALUE has been searched but
674 no valid entry has been found. We solve this situation
675 by implicitly appending a "C" entry, i.e. no translation
676 will take place. */
677 single_locale[0] = 'C';
678 single_locale[1] = '\0';
679 }
680 else
681 {
682 char *cp = single_locale;
683 while (categoryvalue[0] != '\0' && categoryvalue[0] != ':')
684 *cp++ = *categoryvalue++;
685 *cp = '\0';
686
687 /* When this is a SUID binary we must not allow accessing files
688 outside the dedicated directories. */
689 if (ENABLE_SECURE && IS_PATH_WITH_DIR (single_locale))
690 /* Ingore this entry. */
691 continue;
692 }
693
694 /* If the current locale value is "C" or "C.<encoding>" or "POSIX",
695 we don't load a domain. Return the MSGID. */
696 if ((single_locale[0] == 'C'
697 && (single_locale[1] == '\0' || single_locale[1] == '.'))
698 || strcmp (single_locale, "POSIX") == 0)
699 break;
700
701 /* Find structure describing the message catalog matching the
702 DOMAINNAME and CATEGORY. */
703 domain = _nl_find_domain (dirname: dirname, locale: single_locale, domainname: xdomainname, domainbinding: binding);
704
705 if (domain != NULL)
706 {
707#if defined IN_LIBGLOCALE
708 retval = _nl_find_msg (domain, binding, encoding, msgid1, &retlen);
709#else
710 retval = _nl_find_msg (domain_file: domain, domainbinding: binding, msgid: msgid1, convert: 1, lengthp: &retlen);
711#endif
712
713 if (retval == NULL)
714 {
715 int cnt;
716
717 for (cnt = 0; domain->successor[cnt] != NULL; ++cnt)
718 {
719#if defined IN_LIBGLOCALE
720 retval = _nl_find_msg (domain->successor[cnt], binding,
721 encoding, msgid1, &retlen);
722#else
723 retval = _nl_find_msg (domain_file: domain->successor[cnt], domainbinding: binding,
724 msgid: msgid1, convert: 1, lengthp: &retlen);
725#endif
726
727 /* Resource problems are not fatal, instead we return no
728 translation. */
729 if (__builtin_expect (retval == (char *) -1, 0))
730 goto return_untranslated;
731
732 if (retval != NULL)
733 {
734 domain = domain->successor[cnt];
735 break;
736 }
737 }
738 }
739
740 /* Returning -1 means that some resource problem exists
741 (likely memory) and that the strings could not be
742 converted. Return the original strings. */
743 if (__builtin_expect (retval == (char *) -1, 0))
744 break;
745
746 if (retval != NULL)
747 {
748 /* Found the translation of MSGID1 in domain DOMAIN:
749 starting at RETVAL, RETLEN bytes. */
750 free (ptr: xdirname);
751 FREE_BLOCKS (block_list);
752 if (foundp == NULL)
753 {
754 /* Create a new entry and add it to the search tree. */
755 size_t msgid_len;
756 size_t size;
757 struct known_translation_t *newp;
758
759 msgid_len = strlen (msgid1) + 1;
760 size = offsetof (struct known_translation_t, msgid)
761 + msgid_len + domainname_len + 1;
762#ifdef HAVE_PER_THREAD_LOCALE
763 size += strlen (localename) + 1;
764#endif
765 newp = (struct known_translation_t *) malloc (size: size);
766 if (newp != NULL)
767 {
768 char *new_domainname;
769#ifdef HAVE_PER_THREAD_LOCALE
770 char *new_localename;
771#endif
772
773 new_domainname =
774 (char *) mempcpy (newp->msgid.appended, msgid1,
775 msgid_len);
776 memcpy (new_domainname, domainname, domainname_len + 1);
777#ifdef HAVE_PER_THREAD_LOCALE
778 new_localename = new_domainname + domainname_len + 1;
779 strcpy (new_localename, localename);
780#endif
781 newp->domainname = new_domainname;
782 newp->category = category;
783#ifdef HAVE_PER_THREAD_LOCALE
784 newp->localename = new_localename;
785#endif
786#ifdef IN_LIBGLOCALE
787 newp->encoding = encoding;
788#endif
789 newp->counter = _nl_msg_cat_cntr;
790 newp->domain = domain;
791 newp->translation = retval;
792 newp->translation_length = retlen;
793
794 gl_rwlock_wrlock (tree_lock);
795
796 /* Insert the entry in the search tree. */
797 foundp = (struct known_translation_t **)
798 tsearch (newp, &root, transcmp);
799
800 gl_rwlock_unlock (tree_lock);
801
802 if (foundp == NULL
803 || __builtin_expect (*foundp != newp, 0))
804 /* The insert failed. */
805 free (ptr: newp);
806 }
807 }
808 else
809 {
810 /* We can update the existing entry. */
811 (*foundp)->counter = _nl_msg_cat_cntr;
812 (*foundp)->domain = domain;
813 (*foundp)->translation = retval;
814 (*foundp)->translation_length = retlen;
815 }
816
817 __set_errno (saved_errno);
818
819 /* Now deal with plural. */
820 if (plural)
821 retval = plural_lookup (domain, n, translation: retval, translation_len: retlen);
822
823 gl_rwlock_unlock (_nl_state_lock);
824#ifdef _LIBC
825 __libc_rwlock_unlock (__libc_setlocale_lock);
826#endif
827 return retval;
828 }
829 }
830 }
831
832 return_untranslated:
833 /* Return the untranslated MSGID. */
834 free (ptr: xdirname);
835 FREE_BLOCKS (block_list);
836 gl_rwlock_unlock (_nl_state_lock);
837#ifdef _LIBC
838 __libc_rwlock_unlock (__libc_setlocale_lock);
839#endif
840#ifndef _LIBC
841 if (!ENABLE_SECURE)
842 {
843 extern void _nl_log_untranslated (const char *logfilename,
844 const char *domainname,
845 const char *msgid1, const char *msgid2,
846 int plural);
847 const char *logfilename = getenv ("GETTEXT_LOG_UNTRANSLATED");
848
849 if (logfilename != NULL && logfilename[0] != '\0')
850 _nl_log_untranslated (logfilename, domainname, msgid1, msgid2, plural);
851 }
852#endif
853 __set_errno (saved_errno);
854 return (plural == 0
855 ? (char *) msgid1
856 /* Use the Germanic plural rule. */
857 : n == 1 ? (char *) msgid1 : (char *) msgid2);
858}
859
860
861/* Look up the translation of msgid within DOMAIN_FILE and DOMAINBINDING.
862 Return it if found. Return NULL if not found or in case of a conversion
863 failure (problem in the particular message catalog). Return (char *) -1
864 in case of a memory allocation failure during conversion (only if
865 ENCODING != NULL resp. CONVERT == true). */
866char *
867#ifdef IN_LIBGLOCALE
868_nl_find_msg (struct loaded_l10nfile *domain_file,
869 struct binding *domainbinding, const char *encoding,
870 const char *msgid,
871 size_t *lengthp)
872#else
873_nl_find_msg (struct loaded_l10nfile *domain_file,
874 struct binding *domainbinding,
875 const char *msgid, int convert,
876 size_t *lengthp)
877#endif
878{
879 struct loaded_domain *domain;
880 nls_uint32 nstrings;
881 size_t act;
882 char *result;
883 size_t resultlen;
884
885 if (domain_file->decided <= 0)
886 _nl_load_domain (domain: domain_file, domainbinding: domainbinding);
887
888 if (domain_file->data == NULL)
889 return NULL;
890
891 domain = (struct loaded_domain *) domain_file->data;
892
893 nstrings = domain->nstrings;
894
895 /* Locate the MSGID and its translation. */
896 if (domain->hash_tab != NULL)
897 {
898 /* Use the hashing table. */
899 nls_uint32 len = strlen (msgid);
900 nls_uint32 hash_val = __hash_string (str_param: msgid);
901 nls_uint32 idx = hash_val % domain->hash_size;
902 nls_uint32 incr = 1 + (hash_val % (domain->hash_size - 2));
903
904 while (1)
905 {
906 nls_uint32 nstr =
907 W (domain->must_swap_hash_tab, domain->hash_tab[idx]);
908
909 if (nstr == 0)
910 /* Hash table entry is empty. */
911 return NULL;
912
913 nstr--;
914
915 /* Compare msgid with the original string at index nstr.
916 We compare the lengths with >=, not ==, because plural entries
917 are represented by strings with an embedded NUL. */
918 if (nstr < nstrings
919 ? W (domain->must_swap, domain->orig_tab[nstr].length) >= len
920 && (strcmp (msgid,
921 domain->data + W (domain->must_swap,
922 domain->orig_tab[nstr].offset))
923 == 0)
924 : domain->orig_sysdep_tab[nstr - nstrings].length > len
925 && (strcmp (msgid,
926 domain->orig_sysdep_tab[nstr - nstrings].pointer)
927 == 0))
928 {
929 act = nstr;
930 goto found;
931 }
932
933 if (idx >= domain->hash_size - incr)
934 idx -= domain->hash_size - incr;
935 else
936 idx += incr;
937 }
938 /* NOTREACHED */
939 }
940 else
941 {
942 /* Try the default method: binary search in the sorted array of
943 messages. */
944 size_t top, bottom;
945
946 bottom = 0;
947 top = nstrings;
948 while (bottom < top)
949 {
950 int cmp_val;
951
952 act = (bottom + top) / 2;
953 cmp_val = strcmp (msgid, (domain->data
954 + W (domain->must_swap,
955 domain->orig_tab[act].offset)));
956 if (cmp_val < 0)
957 top = act;
958 else if (cmp_val > 0)
959 bottom = act + 1;
960 else
961 goto found;
962 }
963 /* No translation was found. */
964 return NULL;
965 }
966
967 found:
968 /* The translation was found at index ACT. If we have to convert the
969 string to use a different character set, this is the time. */
970 if (act < nstrings)
971 {
972 result = (char *)
973 (domain->data + W (domain->must_swap, domain->trans_tab[act].offset));
974 resultlen = W (domain->must_swap, domain->trans_tab[act].length) + 1;
975 }
976 else
977 {
978 result = (char *) domain->trans_sysdep_tab[act - nstrings].pointer;
979 resultlen = domain->trans_sysdep_tab[act - nstrings].length;
980 }
981
982#if defined _LIBC || HAVE_ICONV
983# ifdef IN_LIBGLOCALE
984 if (encoding != NULL)
985# else
986 if (convert)
987# endif
988 {
989 /* We are supposed to do a conversion. */
990# ifndef IN_LIBGLOCALE
991 const char *encoding = get_output_charset (domainbinding);
992# endif
993 size_t nconversions;
994 struct converted_domain *convd;
995 size_t i;
996
997 /* Protect against reallocation of the table. */
998 gl_rwlock_rdlock (domain->conversions_lock);
999
1000 /* Search whether a table with converted translations for this
1001 encoding has already been allocated. */
1002 nconversions = domain->nconversions;
1003 convd = NULL;
1004
1005 for (i = nconversions; i > 0; )
1006 {
1007 i--;
1008 if (strcmp (domain->conversions[i].encoding, encoding) == 0)
1009 {
1010 convd = &domain->conversions[i];
1011 break;
1012 }
1013 }
1014
1015 gl_rwlock_unlock (domain->conversions_lock);
1016
1017 if (convd == NULL)
1018 {
1019 /* We have to allocate a new conversions table. */
1020 gl_rwlock_wrlock (domain->conversions_lock);
1021 nconversions = domain->nconversions;
1022
1023 /* Maybe in the meantime somebody added the translation.
1024 Recheck. */
1025 for (i = nconversions; i > 0; )
1026 {
1027 i--;
1028 if (strcmp (domain->conversions[i].encoding, encoding) == 0)
1029 {
1030 convd = &domain->conversions[i];
1031 goto found_convd;
1032 }
1033 }
1034
1035 {
1036 /* Allocate a table for the converted translations for this
1037 encoding. */
1038 struct converted_domain *new_conversions =
1039 (struct converted_domain *)
1040 (domain->conversions != NULL
1041 ? realloc (ptr: domain->conversions,
1042 size: (nconversions + 1) * sizeof (struct converted_domain))
1043 : malloc (size: (nconversions + 1) * sizeof (struct converted_domain)));
1044
1045 if (__builtin_expect (new_conversions == NULL, 0))
1046 {
1047 /* Nothing we can do, no more memory. We cannot use the
1048 translation because it might be encoded incorrectly. */
1049 unlock_fail:
1050 gl_rwlock_unlock (domain->conversions_lock);
1051 return (char *) -1;
1052 }
1053
1054 domain->conversions = new_conversions;
1055
1056 /* Copy the 'encoding' string to permanent storage. */
1057 encoding = strdup (encoding);
1058 if (__builtin_expect (encoding == NULL, 0))
1059 /* Nothing we can do, no more memory. We cannot use the
1060 translation because it might be encoded incorrectly. */
1061 goto unlock_fail;
1062
1063 convd = &new_conversions[nconversions];
1064 convd->encoding = encoding;
1065
1066 /* Find out about the character set the file is encoded with.
1067 This can be found (in textual form) in the entry "". If this
1068 entry does not exist or if this does not contain the 'charset='
1069 information, we will assume the charset matches the one the
1070 current locale and we don't have to perform any conversion. */
1071# ifdef _LIBC
1072 convd->conv = (__gconv_t) -1;
1073# else
1074# if HAVE_ICONV
1075 convd->conv = (iconv_t) -1;
1076# endif
1077# endif
1078 {
1079 char *nullentry;
1080 size_t nullentrylen;
1081
1082 /* Get the header entry. This is a recursion, but it doesn't
1083 reallocate domain->conversions because we pass
1084 encoding = NULL or convert = 0, respectively. */
1085 nullentry =
1086# ifdef IN_LIBGLOCALE
1087 _nl_find_msg (domain_file, domainbinding, NULL, "",
1088 &nullentrylen);
1089# else
1090 _nl_find_msg (domain_file, domainbinding, msgid: "", convert: 0, lengthp: &nullentrylen);
1091# endif
1092
1093 /* Resource problems are fatal. If we continue onwards we will
1094 only attempt to calloc a new conv_tab and fail later. */
1095 if (__builtin_expect (nullentry == (char *) -1, 0))
1096 return (char *) -1;
1097
1098 if (nullentry != NULL)
1099 {
1100 const char *charsetstr;
1101
1102 charsetstr = strstr (nullentry, "charset=");
1103 if (charsetstr != NULL)
1104 {
1105 size_t len;
1106 char *charset;
1107 const char *outcharset;
1108
1109 charsetstr += strlen ("charset=");
1110 len = strcspn (charsetstr, " \t\n");
1111
1112 charset = (char *) alloca (len + 1);
1113# if defined _LIBC || HAVE_MEMPCPY
1114 *((char *) mempcpy (charset, charsetstr, len)) = '\0';
1115# else
1116 memcpy (charset, charsetstr, len);
1117 charset[len] = '\0';
1118# endif
1119
1120 outcharset = encoding;
1121
1122# ifdef _LIBC
1123
1124 struct gconv_spec conv_spec;
1125
1126 __gconv_create_spec (&conv_spec, charset, outcharset);
1127
1128 /* We always want to use transliteration. */
1129 conv_spec.translit = true;
1130
1131 int r = __gconv_open (&conv_spec, &convd->conv,
1132 GCONV_AVOID_NOCONV);
1133
1134 __gconv_destroy_spec (&conv_spec);
1135
1136 if (__builtin_expect (r != __GCONV_OK, 0))
1137 {
1138 /* If the output encoding is the same there is
1139 nothing to do. Otherwise do not use the
1140 translation at all. */
1141 if (__builtin_expect (r != __GCONV_NULCONV, 1))
1142 {
1143 gl_rwlock_unlock (domain->conversions_lock);
1144 free (ptr: (char *) encoding);
1145 return NULL;
1146 }
1147
1148 convd->conv = (__gconv_t) -1;
1149 }
1150# else
1151# if HAVE_ICONV
1152 /* When using GNU libc >= 2.2 or GNU libiconv >= 1.5,
1153 we want to use transliteration. */
1154# if (((__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2) \
1155 && !defined __UCLIBC__) \
1156 || _LIBICONV_VERSION >= 0x0105
1157 if (strchr (outcharset, '/') == NULL)
1158 {
1159 char *tmp;
1160
1161 len = strlen (outcharset);
1162 tmp = (char *) alloca (len + 10 + 1);
1163 memcpy (tmp, outcharset, len);
1164 memcpy (tmp + len, "//TRANSLIT", 10 + 1);
1165 outcharset = tmp;
1166
1167 convd->conv = iconv_open (outcharset, charset);
1168
1169 freea (outcharset);
1170 }
1171 else
1172# endif
1173 convd->conv = iconv_open (outcharset, charset);
1174# endif
1175# endif
1176
1177 freea (charset);
1178 }
1179 }
1180 }
1181 convd->conv_tab = NULL;
1182 /* Here domain->conversions is still == new_conversions. */
1183 domain->nconversions++;
1184 }
1185
1186 found_convd:
1187 gl_rwlock_unlock (domain->conversions_lock);
1188 }
1189
1190 if (
1191# ifdef _LIBC
1192 convd->conv != (__gconv_t) -1
1193# else
1194# if HAVE_ICONV
1195 convd->conv != (iconv_t) -1
1196# endif
1197# endif
1198 )
1199 {
1200 /* We are supposed to do a conversion. First allocate an
1201 appropriate table with the same structure as the table
1202 of translations in the file, where we can put the pointers
1203 to the converted strings in.
1204 There is a slight complication with plural entries. They
1205 are represented by consecutive NUL terminated strings. We
1206 handle this case by converting RESULTLEN bytes, including
1207 NULs. */
1208
1209 /* This lock primarily protects the memory management variables
1210 freemem, freemem_size. It also protects write accesses to
1211 convd->conv_tab. It's not worth using a separate lock (such
1212 as domain->conversions_lock) for this purpose, because when
1213 modifying convd->conv_tab, we also need to lock freemem,
1214 freemem_size for most of the time. */
1215 __libc_lock_define_initialized (static, lock)
1216
1217 if (__builtin_expect (convd->conv_tab == NULL, 0))
1218 {
1219 __libc_lock_lock (lock);
1220 if (convd->conv_tab == NULL)
1221 {
1222 convd->conv_tab =
1223 (char **) calloc (nmemb: nstrings + domain->n_sysdep_strings,
1224 size: sizeof (char *));
1225 if (convd->conv_tab != NULL)
1226 goto not_translated_yet;
1227 /* Mark that we didn't succeed allocating a table. */
1228 convd->conv_tab = (char **) -1;
1229 }
1230 __libc_lock_unlock (lock);
1231 }
1232
1233 if (__builtin_expect (convd->conv_tab == (char **) -1, 0))
1234 /* Nothing we can do, no more memory. We cannot use the
1235 translation because it might be encoded incorrectly. */
1236 return (char *) -1;
1237
1238 if (convd->conv_tab[act] == NULL)
1239 {
1240 /* We haven't used this string so far, so it is not
1241 translated yet. Do this now. */
1242 /* We use a bit more efficient memory handling.
1243 We allocate always larger blocks which get used over
1244 time. This is faster than many small allocations. */
1245# define INITIAL_BLOCK_SIZE 4080
1246 static unsigned char *freemem;
1247 static size_t freemem_size;
1248
1249 const unsigned char *inbuf;
1250 unsigned char *outbuf;
1251 int malloc_count;
1252# ifndef _LIBC
1253 transmem_block_t *transmem_list;
1254# endif
1255
1256 __libc_lock_lock (lock);
1257 not_translated_yet:
1258
1259 inbuf = (const unsigned char *) result;
1260 outbuf = freemem + sizeof (size_t);
1261# ifndef _LIBC
1262 transmem_list = NULL;
1263# endif
1264
1265 malloc_count = 0;
1266 while (1)
1267 {
1268 transmem_block_t *newmem;
1269# ifdef _LIBC
1270 size_t non_reversible;
1271 int res;
1272
1273 if (freemem_size < sizeof (size_t))
1274 goto resize_freemem;
1275
1276 res = __gconv (cd: convd->conv,
1277 inbuf: &inbuf, inbufend: inbuf + resultlen,
1278 outbuf: &outbuf,
1279 outbufend: outbuf + freemem_size - sizeof (size_t),
1280 irreversible: &non_reversible);
1281
1282 if (res == __GCONV_OK || res == __GCONV_EMPTY_INPUT)
1283 break;
1284
1285 if (res != __GCONV_FULL_OUTPUT)
1286 {
1287 /* We should not use the translation at all, it
1288 is incorrectly encoded. */
1289 __libc_lock_unlock (lock);
1290 return NULL;
1291 }
1292
1293 inbuf = (const unsigned char *) result;
1294# else
1295# if HAVE_ICONV
1296 const char *inptr = (const char *) inbuf;
1297 size_t inleft = resultlen;
1298 char *outptr = (char *) outbuf;
1299 size_t outleft;
1300
1301 if (freemem_size < sizeof (size_t))
1302 goto resize_freemem;
1303
1304 outleft = freemem_size - sizeof (size_t);
1305 if (iconv (convd->conv,
1306 (ICONV_CONST char **) &inptr, &inleft,
1307 &outptr, &outleft)
1308 != (size_t) (-1))
1309 {
1310 outbuf = (unsigned char *) outptr;
1311 break;
1312 }
1313 if (errno != E2BIG)
1314 {
1315 __libc_lock_unlock (lock);
1316 return NULL;
1317 }
1318# endif
1319# endif
1320
1321 resize_freemem:
1322 /* We must allocate a new buffer or resize the old one. */
1323 if (malloc_count > 0)
1324 {
1325 ++malloc_count;
1326 freemem_size = malloc_count * INITIAL_BLOCK_SIZE;
1327 newmem = (transmem_block_t *) realloc (ptr: transmem_list,
1328 size: freemem_size);
1329# ifdef _LIBC
1330 if (newmem != NULL)
1331 transmem_list = newmem;
1332 else
1333 {
1334 struct transmem_list *old = transmem_list;
1335
1336 transmem_list = transmem_list->next;
1337 free (ptr: old);
1338 }
1339# endif
1340 }
1341 else
1342 {
1343 malloc_count = 1;
1344 freemem_size = INITIAL_BLOCK_SIZE;
1345 newmem = (transmem_block_t *) malloc (size: freemem_size);
1346# ifdef _LIBC
1347 if (newmem != NULL)
1348 {
1349 /* Add the block to the list of blocks we have to free
1350 at some point. */
1351 newmem->next = transmem_list;
1352 transmem_list = newmem;
1353 }
1354 /* Fall through and return -1. */
1355# endif
1356 }
1357 if (__builtin_expect (newmem == NULL, 0))
1358 {
1359 freemem = NULL;
1360 freemem_size = 0;
1361 __libc_lock_unlock (lock);
1362 return (char *) -1;
1363 }
1364
1365# ifdef _LIBC
1366 freemem = (unsigned char *) newmem->data;
1367 freemem_size -= offsetof (struct transmem_list, data);
1368# else
1369 transmem_list = newmem;
1370 freemem = newmem;
1371# endif
1372
1373 outbuf = freemem + sizeof (size_t);
1374 }
1375
1376 /* We have now in our buffer a converted string. Put this
1377 into the table of conversions. */
1378 *(size_t *) freemem = outbuf - freemem - sizeof (size_t);
1379 convd->conv_tab[act] = (char *) freemem;
1380 /* Shrink freemem, but keep it aligned. */
1381 freemem_size -= outbuf - freemem;
1382 freemem = outbuf;
1383 freemem += freemem_size & (alignof (size_t) - 1);
1384 freemem_size = freemem_size & ~ (alignof (size_t) - 1);
1385
1386 __libc_lock_unlock (lock);
1387 }
1388
1389 /* Now convd->conv_tab[act] contains the translation of all
1390 the plural variants. */
1391 result = convd->conv_tab[act] + sizeof (size_t);
1392 resultlen = *(size_t *) convd->conv_tab[act];
1393 }
1394 }
1395
1396 /* The result string is converted. */
1397
1398#endif /* _LIBC || HAVE_ICONV */
1399
1400 *lengthp = resultlen;
1401 return result;
1402}
1403
1404
1405/* Look up a plural variant. */
1406static char *
1407plural_lookup (struct loaded_l10nfile *domain, unsigned long int n,
1408 const char *translation, size_t translation_len)
1409{
1410 struct loaded_domain *domaindata = (struct loaded_domain *) domain->data;
1411 unsigned long int index;
1412 const char *p;
1413
1414 index = plural_eval (pexp: domaindata->plural, n);
1415 if (index >= domaindata->nplurals)
1416 /* This should never happen. It means the plural expression and the
1417 given maximum value do not match. */
1418 index = 0;
1419
1420 /* Skip INDEX strings at TRANSLATION. */
1421 p = translation;
1422 while (index-- > 0)
1423 {
1424 p = strchr (p, '\0');
1425 /* And skip over the NUL byte. */
1426 p++;
1427
1428 if (p >= translation + translation_len)
1429 /* This should never happen. It means the plural expression
1430 evaluated to a value larger than the number of variants
1431 available for MSGID1. */
1432 return (char *) translation;
1433 }
1434 return (char *) p;
1435}
1436
1437#ifndef _LIBC
1438/* Return string representation of locale CATEGORY. */
1439static const char *
1440category_to_name (int category)
1441{
1442 const char *retval;
1443
1444 switch (category)
1445 {
1446#ifdef LC_COLLATE
1447 case LC_COLLATE:
1448 retval = "LC_COLLATE";
1449 break;
1450#endif
1451#ifdef LC_CTYPE
1452 case LC_CTYPE:
1453 retval = "LC_CTYPE";
1454 break;
1455#endif
1456#ifdef LC_MONETARY
1457 case LC_MONETARY:
1458 retval = "LC_MONETARY";
1459 break;
1460#endif
1461#ifdef LC_NUMERIC
1462 case LC_NUMERIC:
1463 retval = "LC_NUMERIC";
1464 break;
1465#endif
1466#ifdef LC_TIME
1467 case LC_TIME:
1468 retval = "LC_TIME";
1469 break;
1470#endif
1471#ifdef LC_MESSAGES
1472 case LC_MESSAGES:
1473 retval = "LC_MESSAGES";
1474 break;
1475#endif
1476#ifdef LC_RESPONSE
1477 case LC_RESPONSE:
1478 retval = "LC_RESPONSE";
1479 break;
1480#endif
1481#ifdef LC_ALL
1482 case LC_ALL:
1483 /* This might not make sense but is perhaps better than any other
1484 value. */
1485 retval = "LC_ALL";
1486 break;
1487#endif
1488 default:
1489 /* If you have a better idea for a default value let me know. */
1490 retval = "LC_XXX";
1491 }
1492
1493 return retval;
1494}
1495#endif
1496
1497/* Guess value of current locale from value of the environment variables
1498 or system-dependent defaults. */
1499static const char *
1500#ifdef IN_LIBGLOCALE
1501guess_category_value (int category, const char *categoryname,
1502 const char *locale)
1503
1504#else
1505guess_category_value (int category, const char *categoryname)
1506#endif
1507{
1508 const char *language;
1509#ifndef IN_LIBGLOCALE
1510 const char *locale;
1511# ifndef _LIBC
1512 const char *language_default;
1513 int locale_defaulted;
1514# endif
1515#endif
1516
1517 /* We use the settings in the following order:
1518 1. The value of the environment variable 'LANGUAGE'. This is a GNU
1519 extension. Its value can be a colon-separated list of locale names.
1520 2. The value of the environment variable 'LC_ALL', 'LC_xxx', or 'LANG'.
1521 More precisely, the first among these that is set to a non-empty value.
1522 This is how POSIX specifies it. The value is a single locale name.
1523 3. A system-dependent preference list of languages. Its value can be a
1524 colon-separated list of locale names.
1525 4. A system-dependent default locale name.
1526 This way:
1527 - System-dependent settings can be overridden by environment variables.
1528 - If the system provides both a list of languages and a default locale,
1529 the former is used. */
1530
1531#ifndef IN_LIBGLOCALE
1532 /* Fetch the locale name, through the POSIX method of looking to `LC_ALL',
1533 `LC_xxx', and `LANG'. On some systems this can be done by the
1534 `setlocale' function itself. */
1535# ifdef _LIBC
1536 locale = __current_locale_name (category);
1537# else
1538 locale_defaulted = 0;
1539# if HAVE_USELOCALE
1540 locale = _nl_locale_name_thread_unsafe (category, categoryname);
1541 if (locale == NULL)
1542# endif
1543 {
1544 locale = _nl_locale_name_posix (category, categoryname);
1545 if (locale == NULL)
1546 {
1547 locale = _nl_locale_name_default ();
1548 locale_defaulted = 1;
1549 }
1550 }
1551# endif
1552#endif
1553
1554 /* Ignore LANGUAGE and its system-dependent analogon if the locale is set
1555 to "C" because
1556 1. "C" locale usually uses the ASCII encoding, and most international
1557 messages use non-ASCII characters. These characters get displayed
1558 as question marks (if using glibc's iconv()) or as invalid 8-bit
1559 characters (because other iconv()s refuse to convert most non-ASCII
1560 characters to ASCII). In any case, the output is ugly.
1561 2. The precise output of some programs in the "C" locale is specified
1562 by POSIX and should not depend on environment variables like
1563 "LANGUAGE" or system-dependent information. We allow such programs
1564 to use gettext().
1565 Ignore LANGUAGE and its system-dependent analogon also if the locale is
1566 set to "C.UTF-8" or, more generally, to "C.<encoding>", because that's
1567 the by-design behaviour for glibc, see
1568 <https://sourceware.org/glibc/wiki/Proposals/C.UTF-8>. */
1569 if (locale[0] == 'C' && (locale[1] == '\0' || locale[1] == '.'))
1570 return locale;
1571
1572 /* The highest priority value is the value of the 'LANGUAGE' environment
1573 variable. */
1574 language = getenv ("LANGUAGE");
1575 if (language != NULL && language[0] != '\0')
1576 return language;
1577#if !defined IN_LIBGLOCALE && !defined _LIBC
1578 /* The next priority value is the locale name, if not defaulted. */
1579 if (locale_defaulted)
1580 {
1581 /* The next priority value is the default language preferences list. */
1582 language_default = _nl_language_preferences_default ();
1583 if (language_default != NULL)
1584 return language_default;
1585 }
1586 /* The least priority value is the locale name, if defaulted. */
1587#endif
1588 return locale;
1589}
1590
1591#if (defined _LIBC || HAVE_ICONV) && !defined IN_LIBGLOCALE
1592/* Returns the output charset. */
1593static const char *
1594get_output_charset (struct binding *domainbinding)
1595{
1596 /* The output charset should normally be determined by the locale. But
1597 sometimes the locale is not used or not correctly set up, so we provide
1598 a possibility for the user to override this: the OUTPUT_CHARSET
1599 environment variable. Moreover, the value specified through
1600 bind_textdomain_codeset overrides both. */
1601 if (domainbinding != NULL && domainbinding->codeset != NULL)
1602 return domainbinding->codeset;
1603 else
1604 {
1605 /* For speed reasons, we look at the value of OUTPUT_CHARSET only
1606 once. This is a user variable that is not supposed to change
1607 during a program run. */
1608 static char *output_charset_cache;
1609 static int output_charset_cached;
1610
1611 if (!output_charset_cached)
1612 {
1613 const char *value = getenv ("OUTPUT_CHARSET");
1614
1615 if (value != NULL && value[0] != '\0')
1616 {
1617 size_t len = strlen (value) + 1;
1618 char *value_copy = (char *) malloc (size: len);
1619
1620 if (value_copy != NULL)
1621 memcpy (value_copy, value, len);
1622 output_charset_cache = value_copy;
1623 }
1624 output_charset_cached = 1;
1625 }
1626
1627 if (output_charset_cache != NULL)
1628 return output_charset_cache;
1629 else
1630 {
1631# ifdef _LIBC
1632 return _NL_CURRENT (LC_CTYPE, CODESET);
1633# else
1634# if HAVE_ICONV
1635 return locale_charset ();
1636# endif
1637# endif
1638 }
1639 }
1640}
1641#endif
1642
1643/* @@ begin of epilog @@ */
1644
1645/* We don't want libintl.a to depend on any other library. So we
1646 avoid the non-standard function stpcpy. In GNU C Library this
1647 function is available, though. Also allow the symbol HAVE_STPCPY
1648 to be defined. */
1649#if !_LIBC && !HAVE_STPCPY
1650static char *
1651stpcpy (char *dest, const char *src)
1652{
1653 while ((*dest++ = *src++) != '\0')
1654 /* Do nothing. */ ;
1655 return dest - 1;
1656}
1657#endif
1658
1659#if !_LIBC && !HAVE_MEMPCPY
1660static void *
1661mempcpy (void *dest, const void *src, size_t n)
1662{
1663 return (void *) ((char *) memcpy (dest, src, n) + n);
1664}
1665#endif
1666
1667#if !_LIBC && !HAVE_TSEARCH
1668# include "tsearch.c"
1669#endif
1670
1671
1672#ifdef _LIBC
1673/* If we want to free all resources we have to do some work at
1674 program's end. */
1675void
1676__intl_freemem (void)
1677{
1678 void *old;
1679
1680 while (_nl_domain_bindings != NULL)
1681 {
1682 struct binding *oldp = _nl_domain_bindings;
1683 _nl_domain_bindings = _nl_domain_bindings->next;
1684 if (oldp->dirname != _nl_default_dirname)
1685 /* Yes, this is a pointer comparison. */
1686 free (ptr: oldp->dirname);
1687 free (ptr: oldp->codeset);
1688 free (ptr: oldp);
1689 }
1690
1691 if (_nl_current_default_domain != _nl_default_default_domain)
1692 /* Yes, again a pointer comparison. */
1693 free (ptr: (char *) _nl_current_default_domain);
1694
1695 /* Remove the search tree with the known translations. */
1696 __tdestroy (root, free);
1697 root = NULL;
1698
1699 while (transmem_list != NULL)
1700 {
1701 old = transmem_list;
1702 transmem_list = transmem_list->next;
1703 free (ptr: old);
1704 }
1705}
1706#endif
1707

source code of glibc/intl/dcigettext.c