1 /* Implementation of the internal dcigettext function.
2 Copyright (C) 1995-2005, 2006, 2007, 2008, 2009
3 Free Software Foundation, Inc.
4 This file is part of the GNU C Library.
6 The GNU C Library is free software; you can redistribute it and/or
7 modify it under the terms of the GNU Lesser General Public
8 License as published by the Free Software Foundation; either
9 version 2.1 of the License, or (at your option) any later version.
11 The GNU C Library is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public
17 License along with the GNU C Library; if not, write to the Free
18 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA
21 /* Tell glibc's <string.h> to provide a prototype for mempcpy().
22 This must come before <config.h> because <config.h> may include
23 <features.h>, and once <features.h> has been included, it's too late. */
25 # define _GNU_SOURCE 1
32 #include <sys/types.h>
35 # define alloca __builtin_alloca
36 # define HAVE_ALLOCA 1
38 # if defined HAVE_ALLOCA_H || defined _LIBC
56 # define __set_errno(val) errno = (val)
63 #if defined HAVE_UNISTD_H || defined _LIBC
69 #if defined HAVE_SYS_PARAM_H || defined _LIBC
70 # include <sys/param.h>
74 #include "plural-exp.h"
78 # include "libgnuintl.h"
80 #include "hash-string.h"
82 /* Thread safetyness. */
84 # include <bits/libc-lock.h>
86 /* Provide dummy implementation if this is outside glibc. */
87 # define __libc_lock_define_initialized(CLASS, NAME)
88 # define __libc_lock_lock(NAME)
89 # define __libc_lock_unlock(NAME)
90 # define __libc_rwlock_define_initialized(CLASS, NAME)
91 # define __libc_rwlock_rdlock(NAME)
92 # define __libc_rwlock_unlock(NAME)
95 /* Alignment of types. */
96 #if defined __GNUC__ && __GNUC__ >= 2
97 # define alignof(TYPE) __alignof__ (TYPE)
99 # define alignof(TYPE) \
100 ((int) &((struct { char dummy1; TYPE dummy2; } *) 0)->dummy2)
103 /* The internal variables in the standalone libintl.a must have different
104 names than the internal variables in GNU libc, otherwise programs
105 using libintl.a cannot be linked statically. */
107 # define _nl_default_default_domain libintl_nl_default_default_domain
108 # define _nl_current_default_domain libintl_nl_current_default_domain
109 # define _nl_default_dirname libintl_nl_default_dirname
110 # define _nl_domain_bindings libintl_nl_domain_bindings
113 /* Some compilers, like SunOS4 cc, don't have offsetof in <stddef.h>. */
115 # define offsetof(type,ident) ((size_t)&(((type*)0)->ident))
118 /* @@ end of prolog @@ */
121 /* Rename the non ANSI C functions. This is required by the standard
122 because some ANSI C functions will require linking with this object
123 file and the name space must not be polluted. */
124 # define getcwd __getcwd
126 # define stpcpy __stpcpy
128 # define tfind __tfind
130 # if !defined HAVE_GETCWD
132 # define getcwd(buf, max) getwd (buf)
137 static char *stpcpy
PARAMS ((char *dest
, const char *src
));
139 # ifndef HAVE_MEMPCPY
140 static void *mempcpy
PARAMS ((void *dest
, const void *src
, size_t n
));
144 /* Amount to increase buffer size by in each try. */
147 /* The following is from pathmax.h. */
148 /* Non-POSIX BSD systems might have gcc's limits.h, which doesn't define
149 PATH_MAX but might cause redefinition warnings when sys/param.h is
150 later included (as on MORE/BSD 4.3). */
151 #if defined _POSIX_VERSION || (defined HAVE_LIMITS_H && !defined __GNUC__)
155 #ifndef _POSIX_PATH_MAX
156 # define _POSIX_PATH_MAX 255
159 #if !defined PATH_MAX && defined _PC_PATH_MAX
160 # define PATH_MAX (pathconf ("/", _PC_PATH_MAX) < 1 ? 1024 : pathconf ("/", _PC_PATH_MAX))
163 /* Don't include sys/param.h if it already has been. */
164 #if defined HAVE_SYS_PARAM_H && !defined PATH_MAX && !defined MAXPATHLEN
165 # include <sys/param.h>
168 #if !defined PATH_MAX && defined MAXPATHLEN
169 # define PATH_MAX MAXPATHLEN
173 # define PATH_MAX _POSIX_PATH_MAX
176 /* Whether to support different locales in different threads. */
177 #if defined _LIBC || HAVE_NL_LOCALE_NAME
178 # define HAVE_PER_THREAD_LOCALE
181 /* This is the type used for the search tree where known translations
183 struct known_translation_t
185 /* Domain in which to search. */
186 const char *domainname
;
191 #ifdef HAVE_PER_THREAD_LOCALE
192 /* Name of the relevant locale category, or "" for the global locale. */
193 const char *localename
;
196 /* State of the catalog counter at the point the string was found. */
199 /* Catalog where the string was found. */
200 struct loaded_l10nfile
*domain
;
202 /* And finally the translation. */
203 const char *translation
;
204 size_t translation_length
;
206 /* Pointer to the string in question. */
209 char appended
[ZERO
]; /* used if domain != NULL */
210 const char *ptr
; /* used if domain == NULL */
215 /* Root of the search tree with known translations. We can use this
216 only if the system provides the `tsearch' function family. */
217 #if defined HAVE_TSEARCH || defined _LIBC
223 # define tsearch __tsearch
226 /* Function to compare two entries in the table of known translations. */
227 static int transcmp
PARAMS ((const void *p1
, const void *p2
));
233 const struct known_translation_t
*s1
;
234 const struct known_translation_t
*s2
;
237 s1
= (const struct known_translation_t
*) p1
;
238 s2
= (const struct known_translation_t
*) p2
;
240 result
= strcmp (s1
->domain
!= NULL
? s1
->msgid
.appended
: s1
->msgid
.ptr
,
241 s2
->domain
!= NULL
? s2
->msgid
.appended
: s2
->msgid
.ptr
);
244 result
= strcmp (s1
->domainname
, s2
->domainname
);
247 #ifdef HAVE_PER_THREAD_LOCALE
248 result
= strcmp (s1
->localename
, s2
->localename
);
251 /* We compare the category last (though this is the cheapest
252 operation) since it is hopefully always the same (namely
254 result
= s1
->category
- s2
->category
;
262 /* Name of the default domain used for gettext(3) prior any call to
263 textdomain(3). The default value for this is "messages". */
264 const char _nl_default_default_domain
[] attribute_hidden
= "messages";
266 /* Value used as the default domain for gettext(3). */
267 const char *_nl_current_default_domain attribute_hidden
268 = _nl_default_default_domain
;
270 /* Contains the default location of the message catalogs. */
273 extern const char _nl_default_dirname
[];
274 libc_hidden_proto (_nl_default_dirname
)
276 const char _nl_default_dirname
[] = LOCALEDIR
;
278 libc_hidden_data_def (_nl_default_dirname
)
281 /* List with bindings of specific domains created by bindtextdomain()
283 struct binding
*_nl_domain_bindings
;
285 /* Prototypes for local functions. */
286 static char *plural_lookup
PARAMS ((struct loaded_l10nfile
*domain
,
288 const char *translation
,
289 size_t translation_len
))
291 static const char *guess_category_value
PARAMS ((int category
,
292 const char *categoryname
))
295 # include "../locale/localeinfo.h"
296 # define category_to_name(category) \
297 _nl_category_names.str + _nl_category_name_idxs[category]
299 static const char *category_to_name
PARAMS ((int category
)) internal_function
;
303 /* For those loosing systems which don't have `alloca' we have to add
304 some additional code emulating it. */
306 /* Nothing has to be done. */
307 # define freea(p) /* nothing */
308 # define ADD_BLOCK(list, address) /* nothing */
309 # define FREE_BLOCKS(list) /* nothing */
314 struct block_list
*next
;
316 # define ADD_BLOCK(list, addr) \
318 struct block_list *newp = (struct block_list *) malloc (sizeof (*newp)); \
319 /* If we cannot get a free block we cannot add the new element to \
321 if (newp != NULL) { \
322 newp->address = (addr); \
323 newp->next = (list); \
327 # define FREE_BLOCKS(list) \
329 while (list != NULL) { \
330 struct block_list *old = list; \
332 free (old->address); \
337 # define alloca(size) (malloc (size))
338 # define freea(p) free (p)
339 #endif /* have alloca */
343 /* List of blocks allocated for translations. */
344 typedef struct transmem_list
346 struct transmem_list
*next
;
349 static struct transmem_list
*transmem_list
;
351 typedef unsigned char transmem_block_t
;
353 #if defined _LIBC || HAVE_ICONV
354 static const char *get_output_charset
PARAMS ((struct binding
*domainbinding
))
359 /* Names for the libintl functions are a problem. They must not clash
360 with existing names and they should follow ANSI C. But this source
361 code is also used in GNU C Library where the names have a __
362 prefix. So we have to make a difference here. */
364 # define DCIGETTEXT __dcigettext
366 # define DCIGETTEXT libintl_dcigettext
369 /* Lock variable to protect the global data in the gettext implementation. */
371 __libc_rwlock_define_initialized (, _nl_state_lock attribute_hidden
)
374 /* Checking whether the binaries runs SUID must be done and glibc provides
375 easier methods therefore we make a difference here. */
377 # define ENABLE_SECURE __libc_enable_secure
378 # define DETERMINE_SECURE
386 # ifndef HAVE_GETEUID
387 # define geteuid() getuid()
389 # ifndef HAVE_GETEGID
390 # define getegid() getgid()
392 static int enable_secure
;
393 # define ENABLE_SECURE (enable_secure == 1)
394 # define DETERMINE_SECURE \
395 if (enable_secure == 0) \
397 if (getuid () != geteuid () || getgid () != getegid ()) \
400 enable_secure = -1; \
404 /* Get the function to evaluate the plural expression. */
405 #include "plural-eval.c"
407 /* Look up MSGID in the DOMAINNAME message catalog for the current
408 CATEGORY locale and, if PLURAL is nonzero, search over string
409 depending on the plural form determined by N. */
411 DCIGETTEXT (domainname
, msgid1
, msgid2
, plural
, n
, category
)
412 const char *domainname
;
420 struct block_list
*block_list
= NULL
;
422 struct loaded_l10nfile
*domain
;
423 struct binding
*binding
;
424 const char *categoryname
;
425 const char *categoryvalue
;
426 char *dirname
, *xdomainname
;
431 #if defined HAVE_TSEARCH || defined _LIBC
432 struct known_translation_t search
;
433 struct known_translation_t
**foundp
= NULL
;
434 # ifdef HAVE_PER_THREAD_LOCALE
435 const char *localename
;
438 size_t domainname_len
;
440 /* If no real MSGID is given return NULL. */
445 if (category
< 0 || category
>= __LC_LAST
|| category
== LC_ALL
)
449 /* Use the Germanic plural rule. */
450 : n
== 1 ? (char *) msgid1
: (char *) msgid2
);
454 __libc_rwlock_define (extern, __libc_setlocale_lock attribute_hidden
)
455 __libc_rwlock_rdlock (__libc_setlocale_lock
);
458 __libc_rwlock_rdlock (_nl_state_lock
);
460 /* If DOMAINNAME is NULL, we are interested in the default domain. If
461 CATEGORY is not LC_MESSAGES this might not make much sense but the
462 definition left this undefined. */
463 if (domainname
== NULL
)
464 domainname
= _nl_current_default_domain
;
466 #if defined HAVE_TSEARCH || defined _LIBC
467 /* Try to find the translation among those which we found at
469 search
.domain
= NULL
;
470 search
.msgid
.ptr
= msgid1
;
471 search
.domainname
= domainname
;
472 search
.category
= category
;
473 # ifdef HAVE_PER_THREAD_LOCALE
475 localename
= strdupa (__current_locale_name (category
));
477 search
.localename
= localename
;
480 /* Since tfind/tsearch manage a balanced tree, concurrent tfind and
481 tsearch calls can be fatal. */
482 __libc_rwlock_define_initialized (static, tree_lock
);
483 __libc_rwlock_rdlock (tree_lock
);
485 foundp
= (struct known_translation_t
**) tfind (&search
, &root
, transcmp
);
487 __libc_rwlock_unlock (tree_lock
);
489 if (foundp
!= NULL
&& (*foundp
)->counter
== _nl_msg_cat_cntr
)
491 /* Now deal with plural. */
493 retval
= plural_lookup ((*foundp
)->domain
, n
, (*foundp
)->translation
,
494 (*foundp
)->translation_length
);
496 retval
= (char *) (*foundp
)->translation
;
499 __libc_rwlock_unlock (__libc_setlocale_lock
);
501 __libc_rwlock_unlock (_nl_state_lock
);
506 /* Preserve the `errno' value. */
509 /* See whether this is a SUID binary or not. */
512 /* First find matching binding. */
513 for (binding
= _nl_domain_bindings
; binding
!= NULL
; binding
= binding
->next
)
515 int compare
= strcmp (domainname
, binding
->domainname
);
521 /* It is not in the list. */
528 dirname
= (char *) _nl_default_dirname
;
529 else if (binding
->dirname
[0] == '/')
530 dirname
= binding
->dirname
;
533 /* We have a relative path. Make it absolute now. */
534 size_t dirname_len
= strlen (binding
->dirname
) + 1;
538 path_max
= (unsigned int) PATH_MAX
;
539 path_max
+= 2; /* The getcwd docs say to do this. */
543 dirname
= (char *) alloca (path_max
+ dirname_len
);
544 ADD_BLOCK (block_list
, dirname
);
547 ret
= getcwd (dirname
, path_max
);
548 if (ret
!= NULL
|| errno
!= ERANGE
)
551 path_max
+= path_max
/ 2;
552 path_max
+= PATH_INCR
;
558 stpcpy (stpcpy (strchr (dirname
, '\0'), "/"), binding
->dirname
);
561 /* Now determine the symbolic name of CATEGORY and its value. */
562 categoryname
= category_to_name (category
);
563 categoryvalue
= guess_category_value (category
, categoryname
);
565 domainname_len
= strlen (domainname
);
566 xdomainname
= (char *) alloca (strlen (categoryname
)
567 + domainname_len
+ 5);
568 ADD_BLOCK (block_list
, xdomainname
);
570 stpcpy (mempcpy (stpcpy (stpcpy (xdomainname
, categoryname
), "/"),
571 domainname
, domainname_len
),
574 /* Creating working area. */
575 single_locale
= (char *) alloca (strlen (categoryvalue
) + 1);
576 ADD_BLOCK (block_list
, single_locale
);
579 /* Search for the given string. This is a loop because we perhaps
580 got an ordered list of languages to consider for the translation. */
583 /* Make CATEGORYVALUE point to the next element of the list. */
584 while (categoryvalue
[0] != '\0' && categoryvalue
[0] == ':')
586 if (categoryvalue
[0] == '\0')
588 /* The whole contents of CATEGORYVALUE has been searched but
589 no valid entry has been found. We solve this situation
590 by implicitly appending a "C" entry, i.e. no translation
592 single_locale
[0] = 'C';
593 single_locale
[1] = '\0';
597 char *cp
= single_locale
;
598 while (categoryvalue
[0] != '\0' && categoryvalue
[0] != ':')
599 *cp
++ = *categoryvalue
++;
602 /* When this is a SUID binary we must not allow accessing files
603 outside the dedicated directories. */
604 if (ENABLE_SECURE
&& strchr (single_locale
, '/') != NULL
)
605 /* Ingore this entry. */
609 /* If the current locale value is C (or POSIX) we don't load a
610 domain. Return the MSGID. */
611 if (strcmp (single_locale
, "C") == 0
612 || strcmp (single_locale
, "POSIX") == 0)
615 FREE_BLOCKS (block_list
);
616 __libc_rwlock_unlock (__libc_setlocale_lock
);
617 __libc_rwlock_unlock (_nl_state_lock
);
618 __set_errno (saved_errno
);
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
);
632 retval
= _nl_find_msg (domain
, binding
, msgid1
, 1, &retlen
);
638 for (cnt
= 0; domain
->successor
[cnt
] != NULL
; ++cnt
)
640 retval
= _nl_find_msg (domain
->successor
[cnt
], binding
,
645 domain
= domain
->successor
[cnt
];
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))
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
665 /* Create a new entry and add it to the search tree. */
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;
676 newp
= (struct known_translation_t
*) malloc (size
);
679 char *new_domainname
;
680 # ifdef HAVE_PER_THREAD_LOCALE
681 char *new_localename
;
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
);
691 newp
->domainname
= new_domainname
;
692 newp
->category
= category
;
693 # ifdef HAVE_PER_THREAD_LOCALE
694 newp
->localename
= new_localename
;
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
);
710 || __builtin_expect (*foundp
!= newp
, 0))
711 /* The insert failed. */
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
;
724 __set_errno (saved_errno
);
726 /* Now deal with plural. */
728 retval
= plural_lookup (domain
, n
, retval
, retlen
);
730 __libc_rwlock_unlock (__libc_setlocale_lock
);
731 __libc_rwlock_unlock (_nl_state_lock
);
742 _nl_find_msg (domain_file
, domainbinding
, msgid
, convert
, lengthp
)
743 struct loaded_l10nfile
*domain_file
;
744 struct binding
*domainbinding
;
749 struct loaded_domain
*domain
;
755 if (domain_file
->decided
<= 0)
756 _nl_load_domain (domain_file
, domainbinding
);
758 if (domain_file
->data
== NULL
)
761 domain
= (struct loaded_domain
*) domain_file
->data
;
763 nstrings
= domain
->nstrings
;
765 /* Locate the MSGID and its translation. */
766 if (domain
->hash_tab
!= NULL
)
768 /* Use the hashing table. */
769 nls_uint32 len
= strlen (msgid
);
770 nls_uint32 hash_val
= __hash_string (msgid
);
771 nls_uint32 idx
= hash_val
% domain
->hash_size
;
772 nls_uint32 incr
= 1 + (hash_val
% (domain
->hash_size
- 2));
777 W (domain
->must_swap_hash_tab
, domain
->hash_tab
[idx
]);
780 /* Hash table entry is empty. */
785 /* Compare msgid with the original string at index nstr.
786 We compare the lengths with >=, not ==, because plural entries
787 are represented by strings with an embedded NUL. */
789 ? W (domain
->must_swap
, domain
->orig_tab
[nstr
].length
) >= len
791 domain
->data
+ W (domain
->must_swap
,
792 domain
->orig_tab
[nstr
].offset
))
794 : domain
->orig_sysdep_tab
[nstr
- nstrings
].length
> len
796 domain
->orig_sysdep_tab
[nstr
- nstrings
].pointer
)
803 if (idx
>= domain
->hash_size
- incr
)
804 idx
-= domain
->hash_size
- incr
;
812 /* Try the default method: binary search in the sorted array of
822 act
= (bottom
+ top
) / 2;
823 cmp_val
= strcmp (msgid
, (domain
->data
824 + W (domain
->must_swap
,
825 domain
->orig_tab
[act
].offset
)));
828 else if (cmp_val
> 0)
833 /* No translation was found. */
838 /* The translation was found at index ACT. If we have to convert the
839 string to use a different character set, this is the time. */
843 (domain
->data
+ W (domain
->must_swap
, domain
->trans_tab
[act
].offset
));
844 resultlen
= W (domain
->must_swap
, domain
->trans_tab
[act
].length
) + 1;
848 result
= (char *) domain
->trans_sysdep_tab
[act
- nstrings
].pointer
;
849 resultlen
= domain
->trans_sysdep_tab
[act
- nstrings
].length
;
852 #if defined _LIBC || HAVE_ICONV
855 /* We are supposed to do a conversion. */
856 const char *encoding
= get_output_charset (domainbinding
);
858 /* Protect against reallocation of the table. */
859 __libc_rwlock_rdlock (domain
->conversions_lock
);
861 /* Search whether a table with converted translations for this
862 encoding has already been allocated. */
863 size_t nconversions
= domain
->nconversions
;
864 struct converted_domain
*convd
= NULL
;
867 for (i
= nconversions
; i
> 0; )
870 if (strcmp (domain
->conversions
[i
].encoding
, encoding
) == 0)
872 convd
= &domain
->conversions
[i
];
877 __libc_rwlock_unlock (domain
->conversions_lock
);
881 /* We have to allocate a new conversions table. */
882 __libc_rwlock_wrlock (domain
->conversions_lock
);
883 nconversions
= domain
->nconversions
;
885 /* Maybe in the meantime somebody added the translation.
887 for (i
= nconversions
; i
> 0; )
890 if (strcmp (domain
->conversions
[i
].encoding
, encoding
) == 0)
892 convd
= &domain
->conversions
[i
];
897 /* Allocate a table for the converted translations for this
899 struct converted_domain
*new_conversions
=
900 (struct converted_domain
*)
901 realloc (domain
->conversions
,
902 (nconversions
+ 1) * sizeof (struct converted_domain
));
904 if (__builtin_expect (new_conversions
== NULL
, 0))
906 /* Nothing we can do, no more memory. We cannot use the
907 translation because it might be encoded incorrectly. */
909 __libc_rwlock_unlock (domain
->conversions_lock
);
913 domain
->conversions
= new_conversions
;
915 /* Copy the 'encoding' string to permanent storage. */
916 encoding
= strdup (encoding
);
917 if (__builtin_expect (encoding
== NULL
, 0))
918 /* Nothing we can do, no more memory. We cannot use the
919 translation because it might be encoded incorrectly. */
922 convd
= &new_conversions
[nconversions
];
923 convd
->encoding
= encoding
;
925 /* Find out about the character set the file is encoded with.
926 This can be found (in textual form) in the entry "". If this
927 entry does not exist or if this does not contain the 'charset='
928 information, we will assume the charset matches the one the
929 current locale and we don't have to perform any conversion. */
931 convd
->conv
= (__gconv_t
) -1;
934 convd
->conv
= (iconv_t
) -1;
941 /* Get the header entry. This is a recursion, but it doesn't
942 reallocate domain->conversions because we pass convert = 0. */
944 _nl_find_msg (domain_file
, domainbinding
, "", 0, &nullentrylen
);
946 if (nullentry
!= NULL
)
948 const char *charsetstr
;
950 charsetstr
= strstr (nullentry
, "charset=");
951 if (charsetstr
!= NULL
)
955 const char *outcharset
;
957 charsetstr
+= strlen ("charset=");
958 len
= strcspn (charsetstr
, " \t\n");
960 charset
= (char *) alloca (len
+ 1);
961 # if defined _LIBC || HAVE_MEMPCPY
962 *((char *) mempcpy (charset
, charsetstr
, len
)) = '\0';
964 memcpy (charset
, charsetstr
, len
);
968 outcharset
= encoding
;
971 /* We always want to use transliteration. */
972 outcharset
= norm_add_slashes (outcharset
, "TRANSLIT");
973 charset
= norm_add_slashes (charset
, "");
974 int r
= __gconv_open (outcharset
, charset
, &convd
->conv
,
976 if (__builtin_expect (r
!= __GCONV_OK
, 0))
978 /* If the output encoding is the same there is
979 nothing to do. Otherwise do not use the
980 translation at all. */
981 if (__builtin_expect (r
!= __GCONV_NULCONV
, 1))
983 __libc_rwlock_unlock (domain
->conversions_lock
);
984 free ((char *) encoding
);
988 convd
->conv
= (__gconv_t
) -1;
992 /* When using GNU libc >= 2.2 or GNU libiconv >= 1.5,
993 we want to use transliteration. */
994 # if (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 2) || __GLIBC__ > 2 \
995 || _LIBICONV_VERSION >= 0x0105
996 if (strchr (outcharset
, '/') == NULL
)
1000 len
= strlen (outcharset
);
1001 tmp
= (char *) alloca (len
+ 10 + 1);
1002 memcpy (tmp
, outcharset
, len
);
1003 memcpy (tmp
+ len
, "//TRANSLIT", 10 + 1);
1006 convd
->conv
= iconv_open (outcharset
, charset
);
1012 convd
->conv
= iconv_open (outcharset
, charset
);
1020 convd
->conv_tab
= NULL
;
1021 /* Here domain->conversions is still == new_conversions. */
1022 domain
->nconversions
++;
1025 __libc_rwlock_unlock (domain
->conversions_lock
);
1030 convd
->conv
!= (__gconv_t
) -1
1033 convd
->conv
!= (iconv_t
) -1
1038 __libc_lock_define_initialized (static, lock
)
1039 /* We are supposed to do a conversion. First allocate an
1040 appropriate table with the same structure as the table
1041 of translations in the file, where we can put the pointers
1042 to the converted strings in.
1043 There is a slight complication with plural entries. They
1044 are represented by consecutive NUL terminated strings. We
1045 handle this case by converting RESULTLEN bytes, including
1048 if (__builtin_expect (convd
->conv_tab
== NULL
, 0))
1050 __libc_lock_lock (lock
);
1051 if (convd
->conv_tab
== NULL
)
1054 = calloc (nstrings
+ domain
->n_sysdep_strings
,
1056 if (convd
->conv_tab
!= NULL
)
1057 goto not_translated_yet
;
1058 /* Mark that we didn't succeed allocating a table. */
1059 convd
->conv_tab
= (char **) -1;
1061 __libc_lock_unlock (lock
);
1064 if (__builtin_expect (convd
->conv_tab
== (char **) -1, 0))
1065 /* Nothing we can do, no more memory. We cannot use the
1066 translation because it might be encoded incorrectly. */
1069 if (convd
->conv_tab
[act
] == NULL
)
1071 __libc_lock_lock (lock
);
1072 not_translated_yet
:;
1074 /* We haven't used this string so far, so it is not
1075 translated yet. Do this now. */
1076 /* We use a bit more efficient memory handling.
1077 We allocate always larger blocks which get used over
1078 time. This is faster than many small allocations. */
1079 # define INITIAL_BLOCK_SIZE 4080
1080 static unsigned char *freemem
;
1081 static size_t freemem_size
;
1083 const unsigned char *inbuf
;
1084 unsigned char *outbuf
;
1087 transmem_block_t
*transmem_list
= NULL
;
1090 inbuf
= (const unsigned char *) result
;
1091 outbuf
= freemem
+ sizeof (size_t);
1096 transmem_block_t
*newmem
;
1098 size_t non_reversible
;
1101 if (freemem_size
< sizeof (size_t))
1102 goto resize_freemem
;
1104 res
= __gconv (convd
->conv
,
1105 &inbuf
, inbuf
+ resultlen
,
1107 outbuf
+ freemem_size
- sizeof (size_t),
1110 if (res
== __GCONV_OK
|| res
== __GCONV_EMPTY_INPUT
)
1113 if (res
!= __GCONV_FULL_OUTPUT
)
1115 /* We should not use the translation at all, it
1116 is incorrectly encoded. */
1117 __libc_lock_unlock (lock
);
1121 inbuf
= (const unsigned char *) result
;
1124 const char *inptr
= (const char *) inbuf
;
1125 size_t inleft
= resultlen
;
1126 char *outptr
= (char *) outbuf
;
1129 if (freemem_size
< sizeof (size_t))
1130 goto resize_freemem
;
1132 outleft
= freemem_size
- sizeof (size_t);
1133 if (iconv (convd
->conv
,
1134 (ICONV_CONST
char **) &inptr
, &inleft
,
1138 outbuf
= (unsigned char *) outptr
;
1143 __libc_lock_unlock (lock
);
1150 /* We must allocate a new buffer or resize the old one. */
1151 if (malloc_count
> 0)
1154 freemem_size
= malloc_count
* INITIAL_BLOCK_SIZE
;
1155 newmem
= (transmem_block_t
*) realloc (transmem_list
,
1159 transmem_list
= transmem_list
->next
;
1162 struct transmem_list
*old
= transmem_list
;
1164 transmem_list
= transmem_list
->next
;
1172 freemem_size
= INITIAL_BLOCK_SIZE
;
1173 newmem
= (transmem_block_t
*) malloc (freemem_size
);
1175 if (__builtin_expect (newmem
== NULL
, 0))
1179 __libc_lock_unlock (lock
);
1184 /* Add the block to the list of blocks we have to free
1186 newmem
->next
= transmem_list
;
1187 transmem_list
= newmem
;
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
;