1 /* Implementation of the internal dcigettext function.
2 Copyright (C) 1995-2005, 2006, 2007, 2008, 2009, 2011, 2012
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, see
18 <http://www.gnu.org/licenses/>. */
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. */
24 # define _GNU_SOURCE 1
31 #include <sys/types.h>
34 # define alloca __builtin_alloca
35 # define HAVE_ALLOCA 1
37 # if defined HAVE_ALLOCA_H || defined _LIBC
55 # define __set_errno(val) errno = (val)
62 #if defined HAVE_UNISTD_H || defined _LIBC
68 #if defined HAVE_SYS_PARAM_H || defined _LIBC
69 # include <sys/param.h>
73 #include "plural-exp.h"
77 # include "libgnuintl.h"
79 #include "hash-string.h"
81 /* Thread safetyness. */
83 # include <bits/libc-lock.h>
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)
94 /* Alignment of types. */
95 #if defined __GNUC__ && __GNUC__ >= 2
96 # define alignof(TYPE) __alignof__ (TYPE)
98 # define alignof(TYPE) \
99 ((int) &((struct { char dummy1; TYPE dummy2; } *) 0)->dummy2)
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. */
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
112 /* Some compilers, like SunOS4 cc, don't have offsetof in <stddef.h>. */
114 # define offsetof(type,ident) ((size_t)&(((type*)0)->ident))
117 /* @@ end of prolog @@ */
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
125 # define stpcpy __stpcpy
127 # define tfind __tfind
129 # if !defined HAVE_GETCWD
131 # define getcwd(buf, max) getwd (buf)
136 static char *stpcpy
PARAMS ((char *dest
, const char *src
));
138 # ifndef HAVE_MEMPCPY
139 static void *mempcpy
PARAMS ((void *dest
, const void *src
, size_t n
));
143 /* Amount to increase buffer size by in each try. */
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__)
154 #ifndef _POSIX_PATH_MAX
155 # define _POSIX_PATH_MAX 255
158 #if !defined PATH_MAX && defined _PC_PATH_MAX
159 # define PATH_MAX (pathconf ("/", _PC_PATH_MAX) < 1 ? 1024 : pathconf ("/", _PC_PATH_MAX))
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>
167 #if !defined PATH_MAX && defined MAXPATHLEN
168 # define PATH_MAX MAXPATHLEN
172 # define PATH_MAX _POSIX_PATH_MAX
175 /* Whether to support different locales in different threads. */
176 #if defined _LIBC || HAVE_NL_LOCALE_NAME
177 # define HAVE_PER_THREAD_LOCALE
180 /* This is the type used for the search tree where known translations
182 struct known_translation_t
184 /* Domain in which to search. */
185 const char *domainname
;
190 #ifdef HAVE_PER_THREAD_LOCALE
191 /* Name of the relevant locale category, or "" for the global locale. */
192 const char *localename
;
195 /* State of the catalog counter at the point the string was found. */
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. */
208 char appended
[ZERO
]; /* used if domain != NULL */
209 const char *ptr
; /* used if domain == NULL */
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
222 # define tsearch __tsearch
225 /* Function to compare two entries in the table of known translations. */
226 static int transcmp
PARAMS ((const void *p1
, const void *p2
));
232 const struct known_translation_t
*s1
;
233 const struct known_translation_t
*s2
;
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
);
243 result
= strcmp (s1
->domainname
, s2
->domainname
);
246 #ifdef HAVE_PER_THREAD_LOCALE
247 result
= strcmp (s1
->localename
, s2
->localename
);
250 /* We compare the category last (though this is the cheapest
251 operation) since it is hopefully always the same (namely
253 result
= s1
->category
- s2
->category
;
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. */
272 extern const char _nl_default_dirname
[];
273 libc_hidden_proto (_nl_default_dirname
)
275 const char _nl_default_dirname
[] = LOCALEDIR
;
277 libc_hidden_data_def (_nl_default_dirname
)
280 /* List with bindings of specific domains created by bindtextdomain()
282 struct binding
*_nl_domain_bindings
;
284 /* Prototypes for local functions. */
285 static char *plural_lookup
PARAMS ((struct loaded_l10nfile
*domain
,
287 const char *translation
,
288 size_t translation_len
))
290 static const char *guess_category_value
PARAMS ((int category
,
291 const char *categoryname
))
294 # include "../locale/localeinfo.h"
295 # define category_to_name(category) \
296 _nl_category_names.str + _nl_category_name_idxs[category]
298 static const char *category_to_name
PARAMS ((int category
)) internal_function
;
302 /* For those loosing systems which don't have `alloca' we have to add
303 some additional code emulating it. */
305 /* Nothing has to be done. */
306 # define freea(p) /* nothing */
307 # define ADD_BLOCK(list, address) /* nothing */
308 # define FREE_BLOCKS(list) /* nothing */
313 struct block_list
*next
;
315 # define ADD_BLOCK(list, addr) \
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 \
320 if (newp != NULL) { \
321 newp->address = (addr); \
322 newp->next = (list); \
326 # define FREE_BLOCKS(list) \
328 while (list != NULL) { \
329 struct block_list *old = list; \
331 free (old->address); \
336 # define alloca(size) (malloc (size))
337 # define freea(p) free (p)
338 #endif /* have alloca */
342 /* List of blocks allocated for translations. */
343 typedef struct transmem_list
345 struct transmem_list
*next
;
348 static struct transmem_list
*transmem_list
;
350 typedef unsigned char transmem_block_t
;
352 #if defined _LIBC || HAVE_ICONV
353 static const char *get_output_charset
PARAMS ((struct binding
*domainbinding
))
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. */
363 # define DCIGETTEXT __dcigettext
365 # define DCIGETTEXT libintl_dcigettext
368 /* Lock variable to protect the global data in the gettext implementation. */
370 __libc_rwlock_define_initialized (, _nl_state_lock attribute_hidden
)
373 /* Checking whether the binaries runs SUID must be done and glibc provides
374 easier methods therefore we make a difference here. */
376 # define ENABLE_SECURE __libc_enable_secure
377 # define DETERMINE_SECURE
385 # ifndef HAVE_GETEUID
386 # define geteuid() getuid()
388 # ifndef HAVE_GETEGID
389 # define getegid() getgid()
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 ()) \
399 enable_secure = -1; \
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. */
410 DCIGETTEXT (domainname
, msgid1
, msgid2
, plural
, n
, category
)
411 const char *domainname
;
419 struct block_list
*block_list
= NULL
;
421 struct loaded_l10nfile
*domain
;
422 struct binding
*binding
;
423 const char *categoryname
;
424 const char *categoryvalue
;
425 char *dirname
, *xdomainname
;
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
;
437 size_t domainname_len
;
439 /* If no real MSGID is given return NULL. */
444 if (category
< 0 || category
>= __LC_LAST
|| category
== LC_ALL
)
448 /* Use the Germanic plural rule. */
449 : n
== 1 ? (char *) msgid1
: (char *) msgid2
);
453 __libc_rwlock_define (extern, __libc_setlocale_lock attribute_hidden
)
454 __libc_rwlock_rdlock (__libc_setlocale_lock
);
457 __libc_rwlock_rdlock (_nl_state_lock
);
459 /* If DOMAINNAME is NULL, we are interested in the default domain. If
460 CATEGORY is not LC_MESSAGES this might not make much sense but the
461 definition left this undefined. */
462 if (domainname
== NULL
)
463 domainname
= _nl_current_default_domain
;
465 #if defined HAVE_TSEARCH || defined _LIBC
466 /* Try to find the translation among those which we found at
468 search
.domain
= NULL
;
469 search
.msgid
.ptr
= msgid1
;
470 search
.domainname
= domainname
;
471 search
.category
= category
;
472 # ifdef HAVE_PER_THREAD_LOCALE
474 localename
= strdupa (__current_locale_name (category
));
476 search
.localename
= localename
;
479 /* Since tfind/tsearch manage a balanced tree, concurrent tfind and
480 tsearch calls can be fatal. */
481 __libc_rwlock_define_initialized (static, tree_lock
);
482 __libc_rwlock_rdlock (tree_lock
);
484 foundp
= (struct known_translation_t
**) tfind (&search
, &root
, transcmp
);
486 __libc_rwlock_unlock (tree_lock
);
488 if (foundp
!= NULL
&& (*foundp
)->counter
== _nl_msg_cat_cntr
)
490 /* Now deal with plural. */
492 retval
= plural_lookup ((*foundp
)->domain
, n
, (*foundp
)->translation
,
493 (*foundp
)->translation_length
);
495 retval
= (char *) (*foundp
)->translation
;
498 __libc_rwlock_unlock (__libc_setlocale_lock
);
500 __libc_rwlock_unlock (_nl_state_lock
);
505 /* Preserve the `errno' value. */
508 /* See whether this is a SUID binary or not. */
511 /* First find matching binding. */
512 for (binding
= _nl_domain_bindings
; binding
!= NULL
; binding
= binding
->next
)
514 int compare
= strcmp (domainname
, binding
->domainname
);
520 /* It is not in the list. */
527 dirname
= (char *) _nl_default_dirname
;
528 else if (binding
->dirname
[0] == '/')
529 dirname
= binding
->dirname
;
532 /* We have a relative path. Make it absolute now. */
533 size_t dirname_len
= strlen (binding
->dirname
) + 1;
537 path_max
= (unsigned int) PATH_MAX
;
538 path_max
+= 2; /* The getcwd docs say to do this. */
542 dirname
= (char *) alloca (path_max
+ dirname_len
);
543 ADD_BLOCK (block_list
, dirname
);
546 ret
= getcwd (dirname
, path_max
);
547 if (ret
!= NULL
|| errno
!= ERANGE
)
550 path_max
+= path_max
/ 2;
551 path_max
+= PATH_INCR
;
557 stpcpy (stpcpy (strchr (dirname
, '\0'), "/"), binding
->dirname
);
560 /* Now determine the symbolic name of CATEGORY and its value. */
561 categoryname
= category_to_name (category
);
562 categoryvalue
= guess_category_value (category
, categoryname
);
564 domainname_len
= strlen (domainname
);
565 xdomainname
= (char *) alloca (strlen (categoryname
)
566 + domainname_len
+ 5);
567 ADD_BLOCK (block_list
, xdomainname
);
569 stpcpy (mempcpy (stpcpy (stpcpy (xdomainname
, categoryname
), "/"),
570 domainname
, domainname_len
),
573 /* Creating working area. */
574 single_locale
= (char *) alloca (strlen (categoryvalue
) + 1);
575 ADD_BLOCK (block_list
, single_locale
);
578 /* Search for the given string. This is a loop because we perhaps
579 got an ordered list of languages to consider for the translation. */
582 /* Make CATEGORYVALUE point to the next element of the list. */
583 while (categoryvalue
[0] != '\0' && categoryvalue
[0] == ':')
585 if (categoryvalue
[0] == '\0')
587 /* The whole contents of CATEGORYVALUE has been searched but
588 no valid entry has been found. We solve this situation
589 by implicitly appending a "C" entry, i.e. no translation
591 single_locale
[0] = 'C';
592 single_locale
[1] = '\0';
596 char *cp
= single_locale
;
597 while (categoryvalue
[0] != '\0' && categoryvalue
[0] != ':')
598 *cp
++ = *categoryvalue
++;
601 /* When this is a SUID binary we must not allow accessing files
602 outside the dedicated directories. */
603 if (ENABLE_SECURE
&& strchr (single_locale
, '/') != NULL
)
604 /* Ingore this entry. */
608 /* If the current locale value is C (or POSIX) we don't load a
609 domain. Return the MSGID. */
610 if (strcmp (single_locale
, "C") == 0
611 || strcmp (single_locale
, "POSIX") == 0)
614 FREE_BLOCKS (block_list
);
615 __libc_rwlock_unlock (__libc_setlocale_lock
);
616 __libc_rwlock_unlock (_nl_state_lock
);
617 __set_errno (saved_errno
);
620 /* Use the Germanic plural rule. */
621 : n
== 1 ? (char *) msgid1
: (char *) msgid2
);
625 /* Find structure describing the message catalog matching the
626 DOMAINNAME and CATEGORY. */
627 domain
= _nl_find_domain (dirname
, single_locale
, xdomainname
, binding
);
631 retval
= _nl_find_msg (domain
, binding
, msgid1
, 1, &retlen
);
637 for (cnt
= 0; domain
->successor
[cnt
] != NULL
; ++cnt
)
639 retval
= _nl_find_msg (domain
->successor
[cnt
], binding
,
644 domain
= domain
->successor
[cnt
];
650 /* Returning -1 means that some resource problem exists
651 (likely memory) and that the strings could not be
652 converted. Return the original strings. */
653 if (__builtin_expect (retval
== (char *) -1, 0))
658 /* Found the translation of MSGID1 in domain DOMAIN:
659 starting at RETVAL, RETLEN bytes. */
660 FREE_BLOCKS (block_list
);
661 #if defined HAVE_TSEARCH || defined _LIBC
664 /* Create a new entry and add it to the search tree. */
667 struct known_translation_t
*newp
;
669 msgid_len
= strlen (msgid1
) + 1;
670 size
= offsetof (struct known_translation_t
, msgid
)
671 + msgid_len
+ domainname_len
+ 1;
672 # ifdef HAVE_PER_THREAD_LOCALE
673 size
+= strlen (localename
) + 1;
675 newp
= (struct known_translation_t
*) malloc (size
);
678 char *new_domainname
;
679 # ifdef HAVE_PER_THREAD_LOCALE
680 char *new_localename
;
684 mempcpy (newp
->msgid
.appended
, msgid1
, msgid_len
);
685 memcpy (new_domainname
, domainname
, domainname_len
+ 1);
686 # ifdef HAVE_PER_THREAD_LOCALE
687 new_localename
= new_domainname
+ domainname_len
+ 1;
688 strcpy (new_localename
, localename
);
690 newp
->domainname
= new_domainname
;
691 newp
->category
= category
;
692 # ifdef HAVE_PER_THREAD_LOCALE
693 newp
->localename
= new_localename
;
695 newp
->counter
= _nl_msg_cat_cntr
;
696 newp
->domain
= domain
;
697 newp
->translation
= retval
;
698 newp
->translation_length
= retlen
;
700 __libc_rwlock_wrlock (tree_lock
);
702 /* Insert the entry in the search tree. */
703 foundp
= (struct known_translation_t
**)
704 tsearch (newp
, &root
, transcmp
);
706 __libc_rwlock_unlock (tree_lock
);
709 || __builtin_expect (*foundp
!= newp
, 0))
710 /* The insert failed. */
716 /* We can update the existing entry. */
717 (*foundp
)->counter
= _nl_msg_cat_cntr
;
718 (*foundp
)->domain
= domain
;
719 (*foundp
)->translation
= retval
;
720 (*foundp
)->translation_length
= retlen
;
723 __set_errno (saved_errno
);
725 /* Now deal with plural. */
727 retval
= plural_lookup (domain
, n
, retval
, retlen
);
729 __libc_rwlock_unlock (__libc_setlocale_lock
);
730 __libc_rwlock_unlock (_nl_state_lock
);
741 _nl_find_msg (domain_file
, domainbinding
, msgid
, convert
, lengthp
)
742 struct loaded_l10nfile
*domain_file
;
743 struct binding
*domainbinding
;
748 struct loaded_domain
*domain
;
754 if (domain_file
->decided
<= 0)
755 _nl_load_domain (domain_file
, domainbinding
);
757 if (domain_file
->data
== 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));
776 W (domain
->must_swap_hash_tab
, domain
->hash_tab
[idx
]);
779 /* Hash table entry is empty. */
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. */
788 ? W (domain
->must_swap
, domain
->orig_tab
[nstr
].length
) >= len
790 domain
->data
+ W (domain
->must_swap
,
791 domain
->orig_tab
[nstr
].offset
))
793 : domain
->orig_sysdep_tab
[nstr
- nstrings
].length
> len
795 domain
->orig_sysdep_tab
[nstr
- nstrings
].pointer
)
802 if (idx
>= domain
->hash_size
- incr
)
803 idx
-= domain
->hash_size
- incr
;
811 /* Try the default method: binary search in the sorted array of
821 act
= (bottom
+ top
) / 2;
822 cmp_val
= strcmp (msgid
, (domain
->data
823 + W (domain
->must_swap
,
824 domain
->orig_tab
[act
].offset
)));
827 else if (cmp_val
> 0)
832 /* No translation was 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. */
842 (domain
->data
+ W (domain
->must_swap
, domain
->trans_tab
[act
].offset
));
843 resultlen
= W (domain
->must_swap
, domain
->trans_tab
[act
].length
) + 1;
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
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
;
866 for (i
= nconversions
; i
> 0; )
869 if (strcmp (domain
->conversions
[i
].encoding
, encoding
) == 0)
871 convd
= &domain
->conversions
[i
];
876 __libc_rwlock_unlock (domain
->conversions_lock
);
880 /* We have to allocate a new conversions table. */
881 __libc_rwlock_wrlock (domain
->conversions_lock
);
882 nconversions
= domain
->nconversions
;
884 /* Maybe in the meantime somebody added the translation.
886 for (i
= nconversions
; i
> 0; )
889 if (strcmp (domain
->conversions
[i
].encoding
, encoding
) == 0)
891 convd
= &domain
->conversions
[i
];
896 /* Allocate a table for the converted translations for this
898 struct converted_domain
*new_conversions
=
899 (struct converted_domain
*)
900 realloc (domain
->conversions
,
901 (nconversions
+ 1) * sizeof (struct converted_domain
));
903 if (__builtin_expect (new_conversions
== NULL
, 0))
905 /* Nothing we can do, no more memory. We cannot use the
906 translation because it might be encoded incorrectly. */
908 __libc_rwlock_unlock (domain
->conversions_lock
);
912 domain
->conversions
= new_conversions
;
914 /* Copy the 'encoding' string to permanent storage. */
915 encoding
= strdup (encoding
);
916 if (__builtin_expect (encoding
== NULL
, 0))
917 /* Nothing we can do, no more memory. We cannot use the
918 translation because it might be encoded incorrectly. */
921 convd
= &new_conversions
[nconversions
];
922 convd
->encoding
= encoding
;
924 /* Find out about the character set the file is encoded with.
925 This can be found (in textual form) in the entry "". If this
926 entry does not exist or if this does not contain the 'charset='
927 information, we will assume the charset matches the one the
928 current locale and we don't have to perform any conversion. */
930 convd
->conv
= (__gconv_t
) -1;
933 convd
->conv
= (iconv_t
) -1;
940 /* Get the header entry. This is a recursion, but it doesn't
941 reallocate domain->conversions because we pass convert = 0. */
943 _nl_find_msg (domain_file
, domainbinding
, "", 0, &nullentrylen
);
945 if (nullentry
!= NULL
)
947 const char *charsetstr
;
949 charsetstr
= strstr (nullentry
, "charset=");
950 if (charsetstr
!= NULL
)
954 const char *outcharset
;
956 charsetstr
+= strlen ("charset=");
957 len
= strcspn (charsetstr
, " \t\n");
959 charset
= (char *) alloca (len
+ 1);
960 # if defined _LIBC || HAVE_MEMPCPY
961 *((char *) mempcpy (charset
, charsetstr
, len
)) = '\0';
963 memcpy (charset
, charsetstr
, len
);
967 outcharset
= encoding
;
970 /* We always want to use transliteration. */
971 outcharset
= norm_add_slashes (outcharset
, "TRANSLIT");
972 charset
= norm_add_slashes (charset
, "");
973 int r
= __gconv_open (outcharset
, charset
, &convd
->conv
,
975 if (__builtin_expect (r
!= __GCONV_OK
, 0))
977 /* If the output encoding is the same there is
978 nothing to do. Otherwise do not use the
979 translation at all. */
980 if (__builtin_expect (r
!= __GCONV_NULCONV
, 1))
982 __libc_rwlock_unlock (domain
->conversions_lock
);
983 free ((char *) encoding
);
987 convd
->conv
= (__gconv_t
) -1;
991 /* When using GNU libc >= 2.2 or GNU libiconv >= 1.5,
992 we want to use transliteration. */
993 # if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 \
994 || _LIBICONV_VERSION >= 0x0105
995 if (strchr (outcharset
, '/') == NULL
)
999 len
= strlen (outcharset
);
1000 tmp
= (char *) alloca (len
+ 10 + 1);
1001 memcpy (tmp
, outcharset
, len
);
1002 memcpy (tmp
+ len
, "//TRANSLIT", 10 + 1);
1005 convd
->conv
= iconv_open (outcharset
, charset
);
1011 convd
->conv
= iconv_open (outcharset
, charset
);
1019 convd
->conv_tab
= NULL
;
1020 /* Here domain->conversions is still == new_conversions. */
1021 domain
->nconversions
++;
1024 __libc_rwlock_unlock (domain
->conversions_lock
);
1029 convd
->conv
!= (__gconv_t
) -1
1032 convd
->conv
!= (iconv_t
) -1
1037 __libc_lock_define_initialized (static, lock
)
1038 /* We are supposed to do a conversion. First allocate an
1039 appropriate table with the same structure as the table
1040 of translations in the file, where we can put the pointers
1041 to the converted strings in.
1042 There is a slight complication with plural entries. They
1043 are represented by consecutive NUL terminated strings. We
1044 handle this case by converting RESULTLEN bytes, including
1047 if (__builtin_expect (convd
->conv_tab
== NULL
, 0))
1049 __libc_lock_lock (lock
);
1050 if (convd
->conv_tab
== NULL
)
1053 = calloc (nstrings
+ domain
->n_sysdep_strings
,
1055 if (convd
->conv_tab
!= NULL
)
1056 goto not_translated_yet
;
1057 /* Mark that we didn't succeed allocating a table. */
1058 convd
->conv_tab
= (char **) -1;
1060 __libc_lock_unlock (lock
);
1063 if (__builtin_expect (convd
->conv_tab
== (char **) -1, 0))
1064 /* Nothing we can do, no more memory. We cannot use the
1065 translation because it might be encoded incorrectly. */
1068 if (convd
->conv_tab
[act
] == NULL
)
1070 __libc_lock_lock (lock
);
1071 not_translated_yet
:;
1073 /* We haven't used this string so far, so it is not
1074 translated yet. Do this now. */
1075 /* We use a bit more efficient memory handling.
1076 We allocate always larger blocks which get used over
1077 time. This is faster than many small allocations. */
1078 # define INITIAL_BLOCK_SIZE 4080
1079 static unsigned char *freemem
;
1080 static size_t freemem_size
;
1082 const unsigned char *inbuf
;
1083 unsigned char *outbuf
;
1086 transmem_block_t
*transmem_list
= NULL
;
1089 inbuf
= (const unsigned char *) result
;
1090 outbuf
= freemem
+ sizeof (size_t);
1095 transmem_block_t
*newmem
;
1097 size_t non_reversible
;
1100 if (freemem_size
< sizeof (size_t))
1101 goto resize_freemem
;
1103 res
= __gconv (convd
->conv
,
1104 &inbuf
, inbuf
+ resultlen
,
1106 outbuf
+ freemem_size
- sizeof (size_t),
1109 if (res
== __GCONV_OK
|| res
== __GCONV_EMPTY_INPUT
)
1112 if (res
!= __GCONV_FULL_OUTPUT
)
1114 /* We should not use the translation at all, it
1115 is incorrectly encoded. */
1116 __libc_lock_unlock (lock
);
1120 inbuf
= (const unsigned char *) result
;
1123 const char *inptr
= (const char *) inbuf
;
1124 size_t inleft
= resultlen
;
1125 char *outptr
= (char *) outbuf
;
1128 if (freemem_size
< sizeof (size_t))
1129 goto resize_freemem
;
1131 outleft
= freemem_size
- sizeof (size_t);
1132 if (iconv (convd
->conv
,
1133 (ICONV_CONST
char **) &inptr
, &inleft
,
1137 outbuf
= (unsigned char *) outptr
;
1142 __libc_lock_unlock (lock
);
1149 /* We must allocate a new buffer or resize the old one. */
1150 if (malloc_count
> 0)
1153 freemem_size
= malloc_count
* INITIAL_BLOCK_SIZE
;
1154 newmem
= (transmem_block_t
*) realloc (transmem_list
,
1158 transmem_list
= newmem
;
1161 struct transmem_list
*old
= transmem_list
;
1163 transmem_list
= transmem_list
->next
;
1171 freemem_size
= INITIAL_BLOCK_SIZE
;
1172 newmem
= (transmem_block_t
*) malloc (freemem_size
);
1174 /* Add the block to the list of blocks we have to free
1176 newmem
->next
= transmem_list
;
1177 transmem_list
= newmem
;
1180 if (__builtin_expect (newmem
== NULL
, 0))
1184 __libc_lock_unlock (lock
);
1189 freemem
= (unsigned char *) newmem
->data
;
1190 freemem_size
-= offsetof (struct transmem_list
, data
);
1192 transmem_list
= newmem
;
1196 outbuf
= freemem
+ sizeof (size_t);
1199 /* We have now in our buffer a converted string. Put this
1200 into the table of conversions. */
1201 *(size_t *) freemem
= outbuf
- freemem
- sizeof (size_t);
1202 convd
->conv_tab
[act
] = (char *) freemem
;
1203 /* Shrink freemem, but keep it aligned. */
1204 freemem_size
-= outbuf
- freemem
;
1206 freemem
+= freemem_size
& (alignof (size_t) - 1);
1207 freemem_size
= freemem_size
& ~ (alignof (size_t) - 1);
1209 __libc_lock_unlock (lock
);
1212 /* Now convd->conv_tab[act] contains the translation of all
1213 the plural variants. */
1214 result
= convd
->conv_tab
[act
] + sizeof (size_t);
1215 resultlen
= *(size_t *) convd
->conv_tab
[act
];
1219 /* The result string is converted. */
1221 #endif /* _LIBC || HAVE_ICONV */
1223 *lengthp
= resultlen
;
1228 /* Look up a plural variant. */
1231 plural_lookup (domain
, n
, translation
, translation_len
)
1232 struct loaded_l10nfile
*domain
;
1233 unsigned long int n
;
1234 const char *translation
;
1235 size_t translation_len
;
1237 struct loaded_domain
*domaindata
= (struct loaded_domain
*) domain
->data
;
1238 unsigned long int index
;
1241 index
= plural_eval (domaindata
->plural
, n
);
1242 if (index
>= domaindata
->nplurals
)
1243 /* This should never happen. It means the plural expression and the
1244 given maximum value do not match. */
1247 /* Skip INDEX strings at TRANSLATION. */
1252 p
= __rawmemchr (p
, '\0');
1254 p
= strchr (p
, '\0');
1256 /* And skip over the NUL byte. */
1259 if (p
>= translation
+ translation_len
)
1260 /* This should never happen. It means the plural expression
1261 evaluated to a value larger than the number of variants
1262 available for MSGID1. */
1263 return (char *) translation
;
1269 /* Return string representation of locale CATEGORY. */
1272 category_to_name (category
)
1281 retval
= "LC_COLLATE";
1286 retval
= "LC_CTYPE";
1291 retval
= "LC_MONETARY";
1296 retval
= "LC_NUMERIC";
1306 retval
= "LC_MESSAGES";
1311 retval
= "LC_RESPONSE";
1316 /* This might not make sense but is perhaps better than any other
1322 /* If you have a better idea for a default value let me know. */
1330 /* Guess value of current locale from value of the environment variables. */
1333 guess_category_value (category
, categoryname
)
1335 const char *categoryname
;
1337 const char *language
;
1340 /* The highest priority value is the `LANGUAGE' environment
1341 variable. But we don't use the value if the currently selected
1342 locale is the C locale. This is a GNU extension. */
1343 language
= getenv ("LANGUAGE");
1344 if (language
!= NULL
&& language
[0] == '\0')
1347 /* We have to proceed with the POSIX methods of looking to `LC_ALL',
1348 `LC_xxx', and `LANG'. On some systems this can be done by the
1349 `setlocale' function itself. */
1351 retval
= __current_locale_name (category
);
1353 retval
= _nl_locale_name (category
, categoryname
);
1356 return language
!= NULL
&& strcmp (retval
, "C") != 0 ? language
: retval
;
1359 #if defined _LIBC || HAVE_ICONV
1360 /* Returns the output charset. */
1363 get_output_charset (domainbinding
)
1364 struct binding
*domainbinding
;
1366 /* The output charset should normally be determined by the locale. But
1367 sometimes the locale is not used or not correctly set up, so we provide
1368 a possibility for the user to override this: the OUTPUT_CHARSET
1369 environment variable. Moreover, the value specified through
1370 bind_textdomain_codeset overrides both. */
1371 if (domainbinding
!= NULL
&& domainbinding
->codeset
!= NULL
)
1372 return domainbinding
->codeset
;
1375 /* For speed reasons, we look at the value of OUTPUT_CHARSET only
1376 once. This is a user variable that is not supposed to change
1377 during a program run. */
1378 static char *output_charset_cache
;
1379 static int output_charset_cached
;
1381 if (!output_charset_cached
)
1383 const char *value
= getenv ("OUTPUT_CHARSET");
1385 if (value
!= NULL
&& value
[0] != '\0')
1387 size_t len
= strlen (value
) + 1;
1388 char *value_copy
= (char *) malloc (len
);
1390 if (value_copy
!= NULL
)
1391 memcpy (value_copy
, value
, len
);
1392 output_charset_cache
= value_copy
;
1394 output_charset_cached
= 1;
1397 if (output_charset_cache
!= NULL
)
1398 return output_charset_cache
;
1402 return _NL_CURRENT (LC_CTYPE
, CODESET
);
1405 extern const char *locale_charset
PARAMS ((void));
1406 return locale_charset ();
1414 /* @@ begin of epilog @@ */
1416 /* We don't want libintl.a to depend on any other library. So we
1417 avoid the non-standard function stpcpy. In GNU C Library this
1418 function is available, though. Also allow the symbol HAVE_STPCPY
1420 #if !_LIBC && !HAVE_STPCPY
1426 while ((*dest
++ = *src
++) != '\0')
1432 #if !_LIBC && !HAVE_MEMPCPY
1434 mempcpy (dest
, src
, n
)
1439 return (void *) ((char *) memcpy (dest
, src
, n
) + n
);
1445 /* If we want to free all resources we have to do some work at
1447 libc_freeres_fn (free_mem
)
1451 while (_nl_domain_bindings
!= NULL
)
1453 struct binding
*oldp
= _nl_domain_bindings
;
1454 _nl_domain_bindings
= _nl_domain_bindings
->next
;
1455 if (oldp
->dirname
!= _nl_default_dirname
)
1456 /* Yes, this is a pointer comparison. */
1457 free (oldp
->dirname
);
1458 free (oldp
->codeset
);
1462 if (_nl_current_default_domain
!= _nl_default_default_domain
)
1463 /* Yes, again a pointer comparison. */
1464 free ((char *) _nl_current_default_domain
);
1466 /* Remove the search tree with the known translations. */
1467 __tdestroy (root
, free
);
1470 while (transmem_list
!= NULL
)
1472 old
= transmem_list
;
1473 transmem_list
= transmem_list
->next
;