3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * A Elbereth Gilthoniel,
13 * silivren penna míriel
14 * o menel aglar elenath!
15 * Na-chaered palan-díriel
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
47 #define PERL_IN_LOCALE_C
48 #include "perl_langinfo.h"
60 /* If the environment says to, we can output debugging information during
61 * initialization. This is done before option parsing, and before any thread
62 * creation, so can be a file-level static */
63 #if ! defined(DEBUGGING)
64 # define debug_initialization 0
65 # define DEBUG_INITIALIZATION_set(v)
67 static bool debug_initialization = FALSE;
68 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
72 /* Returns the Unix errno portion; ignoring any others. This is a macro here
73 * instead of putting it into perl.h, because unclear to khw what should be
75 #define GET_ERRNO saved_errno
77 /* strlen() of a literal string constant. We might want this more general,
78 * but using it in just this file for now. A problem with more generality is
79 * the compiler warnings about comparing unlike signs */
80 #define STRLENs(s) (sizeof("" s "") - 1)
82 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
83 * return of setlocale(), then this is extremely likely to be the C or POSIX
84 * locale. However, the output of setlocale() is documented to be opaque, but
85 * the odds are extremely small that it would return these two strings for some
86 * other locale. Note that VMS in these two locales includes many non-ASCII
87 * characters as controls and punctuation (below are hex bytes):
89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
90 * Oddly, none there are listed as alphas, though some represent alphabetics
91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
92 #define isNAME_C_OR_POSIX(name) \
94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
95 || strEQ((name), "POSIX")))
99 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
100 * looked up. This is in the form of a C string: */
102 #define UTF8NESS_SEP "\v"
103 #define UTF8NESS_PREFIX "\f"
105 /* So, the string looks like:
107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
109 * where the digit 0 after the \a indicates that the locale starting just
110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
112 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
113 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
115 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
118 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
119 * kept there always. The remining portion of the cache is LRU, with the
120 * oldest looked-up locale at the tail end */
123 S_stdize_locale(pTHX_ char *locs)
125 /* Standardize the locale name from a string returned by 'setlocale',
126 * possibly modifying that string.
128 * The typical return value of setlocale() is either
129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
131 * (the space-separated values represent the various sublocales,
132 * in some unspecified order). This is not handled by this function.
134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
135 * which is harmful for further use of the string in setlocale(). This
136 * function removes the trailing new line and everything up through the '='
139 const char * const s = strchr(locs, '=');
142 PERL_ARGS_ASSERT_STDIZE_LOCALE;
145 const char * const t = strchr(s, '.');
148 const char * const u = strchr(t, '\n');
149 if (u && (u[1] == 0)) {
150 const STRLEN len = u - s;
151 Move(s + 1, locs, len, char);
159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
164 /* Two parallel arrays; first the locale categories Perl uses on this system;
165 * the second array is their names. These arrays are in mostly arbitrary
168 const int categories[] = {
170 # ifdef USE_LOCALE_NUMERIC
173 # ifdef USE_LOCALE_CTYPE
176 # ifdef USE_LOCALE_COLLATE
179 # ifdef USE_LOCALE_TIME
182 # ifdef USE_LOCALE_MESSAGES
185 # ifdef USE_LOCALE_MONETARY
188 # ifdef USE_LOCALE_ADDRESS
191 # ifdef USE_LOCALE_IDENTIFICATION
194 # ifdef USE_LOCALE_MEASUREMENT
197 # ifdef USE_LOCALE_PAPER
200 # ifdef USE_LOCALE_TELEPHONE
203 # ifdef USE_LOCALE_SYNTAX
206 # ifdef USE_LOCALE_TOD
212 -1 /* Placeholder because C doesn't allow a
213 trailing comma, and it would get complicated
214 with all the #ifdef's */
217 /* The top-most real element is LC_ALL */
219 const char * const category_names[] = {
221 # ifdef USE_LOCALE_NUMERIC
224 # ifdef USE_LOCALE_CTYPE
227 # ifdef USE_LOCALE_COLLATE
230 # ifdef USE_LOCALE_TIME
233 # ifdef USE_LOCALE_MESSAGES
236 # ifdef USE_LOCALE_MONETARY
239 # ifdef USE_LOCALE_ADDRESS
242 # ifdef USE_LOCALE_IDENTIFICATION
245 # ifdef USE_LOCALE_MEASUREMENT
248 # ifdef USE_LOCALE_PAPER
251 # ifdef USE_LOCALE_TELEPHONE
254 # ifdef USE_LOCALE_SYNTAX
257 # ifdef USE_LOCALE_TOD
263 NULL /* Placeholder */
268 /* On systems with LC_ALL, it is kept in the highest index position. (-2
269 * to account for the final unused placeholder element.) */
270 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
274 /* On systems without LC_ALL, we pretend it is there, one beyond the real
275 * top element, hence in the unused placeholder element. */
276 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
280 /* Pretending there is an LC_ALL element just above allows us to avoid most
281 * special cases. Most loops through these arrays in the code below are
282 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
283 * on either type of system. But the code must be written to not access the
284 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
285 * checked for at compile time by using the #define LC_ALL_INDEX which is only
286 * defined if we do have LC_ALL. */
289 S_category_name(const int category)
295 if (category == LC_ALL) {
301 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
302 if (category == categories[i]) {
303 return category_names[i];
308 const char suffix[] = " (unknown)";
310 Size_t length = sizeof(suffix) + 1;
319 /* Calculate the number of digits */
325 Newx(unknown, length, char);
326 my_snprintf(unknown, length, "%d%s", category, suffix);
332 /* Now create LC_foo_INDEX #defines for just those categories on this system */
333 # ifdef USE_LOCALE_NUMERIC
334 # define LC_NUMERIC_INDEX 0
335 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
337 # define _DUMMY_NUMERIC -1
339 # ifdef USE_LOCALE_CTYPE
340 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
341 # define _DUMMY_CTYPE LC_CTYPE_INDEX
343 # define _DUMMY_CTYPE _DUMMY_NUMERIC
345 # ifdef USE_LOCALE_COLLATE
346 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
347 # define _DUMMY_COLLATE LC_COLLATE_INDEX
349 # define _DUMMY_COLLATE _DUMMY_CTYPE
351 # ifdef USE_LOCALE_TIME
352 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
353 # define _DUMMY_TIME LC_TIME_INDEX
355 # define _DUMMY_TIME _DUMMY_COLLATE
357 # ifdef USE_LOCALE_MESSAGES
358 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
359 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
361 # define _DUMMY_MESSAGES _DUMMY_TIME
363 # ifdef USE_LOCALE_MONETARY
364 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
365 # define _DUMMY_MONETARY LC_MONETARY_INDEX
367 # define _DUMMY_MONETARY _DUMMY_MESSAGES
369 # ifdef USE_LOCALE_ADDRESS
370 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
371 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
373 # define _DUMMY_ADDRESS _DUMMY_MONETARY
375 # ifdef USE_LOCALE_IDENTIFICATION
376 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
377 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
379 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
381 # ifdef USE_LOCALE_MEASUREMENT
382 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
383 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
385 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
387 # ifdef USE_LOCALE_PAPER
388 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
389 # define _DUMMY_PAPER LC_PAPER_INDEX
391 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
393 # ifdef USE_LOCALE_TELEPHONE
394 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
395 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
397 # define _DUMMY_TELEPHONE _DUMMY_PAPER
399 # ifdef USE_LOCALE_SYNTAX
400 # define LC_SYNTAX_INDEX _DUMMY_TELEPHONE + 1
401 # define _DUMMY_SYNTAX LC_SYNTAX_INDEX
403 # define _DUMMY_SYNTAX _DUMMY_TELEPHONE
405 # ifdef USE_LOCALE_TOD
406 # define LC_TOD_INDEX _DUMMY_SYNTAX + 1
407 # define _DUMMY_TOD LC_TOD_INDEX
409 # define _DUMMY_TOD _DUMMY_SYNTAX
412 # define LC_ALL_INDEX _DUMMY_TOD + 1
414 #endif /* ifdef USE_LOCALE */
416 /* Windows requres a customized base-level setlocale() */
418 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
420 # define my_setlocale(cat, locale) setlocale(cat, locale)
423 #ifndef USE_POSIX_2008_LOCALE
425 /* "do_setlocale_c" is intended to be called when the category is a constant
426 * known at compile time; "do_setlocale_r", not known until run time */
427 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
428 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
429 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
431 #else /* Below uses POSIX 2008 */
433 /* We emulate setlocale with our own function. LC_foo is not valid for the
434 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
435 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
436 * proper offset into the array 'category_masks[]'. At runtime, we have to
437 * search through the array (as the actual numbers may not be small contiguous
438 * positive integers which would lend themselves to array lookup). */
439 # define do_setlocale_c(cat, locale) \
440 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
441 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
443 # if ! defined(__GLIBC__) || ! defined(USE_LOCALE_MESSAGES)
445 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
447 # else /* Invalidate glibc cache of loaded translations, see [perl #134264] */
449 # include <libintl.h>
450 # define FIX_GLIBC_LC_MESSAGES_BUG(i) \
452 if ((i) == LC_MESSAGES_INDEX) { \
453 textdomain(textdomain(NULL)); \
459 /* A third array, parallel to the ones above to map from category to its
461 const int category_masks[] = {
462 # ifdef USE_LOCALE_NUMERIC
465 # ifdef USE_LOCALE_CTYPE
468 # ifdef USE_LOCALE_COLLATE
471 # ifdef USE_LOCALE_TIME
474 # ifdef USE_LOCALE_MESSAGES
477 # ifdef USE_LOCALE_MONETARY
480 # ifdef USE_LOCALE_ADDRESS
483 # ifdef USE_LOCALE_IDENTIFICATION
484 LC_IDENTIFICATION_MASK,
486 # ifdef USE_LOCALE_MEASUREMENT
489 # ifdef USE_LOCALE_PAPER
492 # ifdef USE_LOCALE_TELEPHONE
495 # ifdef USE_LOCALE_SYNTAX
498 # ifdef USE_LOCALE_TOD
501 /* LC_ALL can't be turned off by a Configure
502 * option, and in Posix 2008, should always be
503 * here, so compile it in unconditionally.
504 * This could catch some glitches at compile
510 S_emulate_setlocale(const int category,
513 const bool is_index_valid
516 /* This function effectively performs a setlocale() on just the current
517 * thread; thus it is thread-safe. It does this by using the POSIX 2008
518 * locale functions to emulate the behavior of setlocale(). Similar to
519 * regular setlocale(), the return from this function points to memory that
520 * can be overwritten by other system calls, so needs to be copied
521 * immediately if you need to retain it. The difference here is that
522 * system calls besides another setlocale() can overwrite it.
524 * By doing this, most locale-sensitive functions become thread-safe. The
525 * exceptions are mostly those that return a pointer to static memory.
527 * This function takes the same parameters, 'category' and 'locale', that
528 * the regular setlocale() function does, but it also takes two additional
529 * ones. This is because the 2008 functions don't use a category; instead
530 * they use a corresponding mask. Because this function operates in both
531 * worlds, it may need one or the other or both. This function can
532 * calculate the mask from the input category, but to avoid this
533 * calculation, if the caller knows at compile time what the mask is, it
534 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
535 * parameter is ignored.
537 * POSIX 2008, for some sick reason, chose not to provide a method to find
538 * the category name of a locale. Some vendors have created a
539 * querylocale() function to do just that. This function is a lot simpler
540 * to implement on systems that have this. Otherwise, we have to keep
541 * track of what the locale has been set to, so that we can return its
542 * name to emulate setlocale(). It's also possible for C code in some
543 * library to change the locale without us knowing it, though as of
544 * September 2017, there are no occurrences in CPAN of uselocale(). Some
545 * libraries do use setlocale(), but that changes the global locale, and
546 * threads using per-thread locales will just ignore those changes.
547 * Another problem is that without querylocale(), we have to guess at what
548 * was meant by setting a locale of "". We handle this by not actually
549 * ever setting to "" (unless querylocale exists), but to emulate what we
550 * think should happen for "".
560 if (DEBUG_Lv_TEST || debug_initialization) {
561 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
566 /* If the input mask might be incorrect, calculate the correct one */
567 if (! is_index_valid) {
572 if (DEBUG_Lv_TEST || debug_initialization) {
573 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
578 for (i = 0; i <= LC_ALL_INDEX; i++) {
579 if (category == categories[i]) {
585 /* Here, we don't know about this category, so can't handle it.
586 * Fallback to the early POSIX usages */
587 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
588 "Unknown locale category %d; can't set it to %s\n",
596 if (DEBUG_Lv_TEST || debug_initialization) {
597 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
604 mask = category_masks[index];
608 if (DEBUG_Lv_TEST || debug_initialization) {
609 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
614 /* If just querying what the existing locale is ... */
615 if (locale == NULL) {
616 locale_t cur_obj = uselocale((locale_t) 0);
620 if (DEBUG_Lv_TEST || debug_initialization) {
621 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
626 if (cur_obj == LC_GLOBAL_LOCALE) {
627 return my_setlocale(category, NULL);
630 # ifdef HAS_QUERYLOCALE
632 return (char *) querylocale(mask, cur_obj);
636 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
637 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
639 # if defined(_NL_LOCALE_NAME) \
640 && defined(DEBUGGING) \
641 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
642 /* On systems that accept any locale name, the real underlying locale
643 * is often returned by this internal function, so we can't use it */
645 /* Internal glibc for querylocale(), but doesn't handle
646 * empty-string ("") locale properly; who knows what other
647 * glitches. Check for it now, under debug. */
649 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
650 uselocale((locale_t) 0));
652 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
653 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
654 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
656 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
657 if ( strNE(PL_curlocales[index], temp_name)
658 && ! ( isNAME_C_OR_POSIX(temp_name)
659 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
661 # ifdef USE_C_BACKTRACE
663 dump_c_backtrace(Perl_debug_log, 20, 1);
667 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
668 " (%s) and what internal glibc thinks"
669 " (%s)\n", category_names[index],
670 PL_curlocales[index], temp_name);
679 /* Without querylocale(), we have to use our record-keeping we've
682 if (category != LC_ALL) {
686 if (DEBUG_Lv_TEST || debug_initialization) {
687 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
692 return PL_curlocales[index];
694 else { /* For LC_ALL */
696 Size_t names_len = 0;
698 bool are_all_categories_the_same_locale = TRUE;
700 /* If we have a valid LC_ALL value, just return it */
701 if (PL_curlocales[LC_ALL_INDEX]) {
705 if (DEBUG_Lv_TEST || debug_initialization) {
706 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
711 return PL_curlocales[LC_ALL_INDEX];
714 /* Otherwise, we need to construct a string of name=value pairs.
715 * We use the glibc syntax, like
716 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
717 * First calculate the needed size. Along the way, check if all
718 * the locale names are the same */
719 for (i = 0; i < LC_ALL_INDEX; i++) {
723 if (DEBUG_Lv_TEST || debug_initialization) {
724 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
729 names_len += strlen(category_names[i])
731 + strlen(PL_curlocales[i])
734 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
735 are_all_categories_the_same_locale = FALSE;
739 /* If they are the same, we don't actually have to construct the
740 * string; we just make the entry in LC_ALL_INDEX valid, and be
741 * that single name */
742 if (are_all_categories_the_same_locale) {
743 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
744 return PL_curlocales[LC_ALL_INDEX];
747 names_len++; /* Trailing '\0' */
748 SAVEFREEPV(Newx(all_string, names_len, char));
751 /* Then fill in the string */
752 for (i = 0; i < LC_ALL_INDEX; i++) {
756 if (DEBUG_Lv_TEST || debug_initialization) {
757 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
762 my_strlcat(all_string, category_names[i], names_len);
763 my_strlcat(all_string, "=", names_len);
764 my_strlcat(all_string, PL_curlocales[i], names_len);
765 my_strlcat(all_string, ";", names_len);
770 if (DEBUG_L_TEST || debug_initialization) {
771 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
781 SETERRNO(EINVAL, LIB_INVARG);
789 } /* End of this being setlocale(LC_foo, NULL) */
791 /* Here, we are switching locales. */
793 # ifndef HAS_QUERYLOCALE
795 if (strEQ(locale, "")) {
797 /* For non-querylocale() systems, we do the setting of "" ourselves to
798 * be sure that we really know what's going on. We follow the Linux
799 * documented behavior (but if that differs from the actual behavior,
800 * this won't work exactly as the OS implements). We go out and
801 * examine the environment based on our understanding of how the system
802 * works, and use that to figure things out */
804 const char * const lc_all = PerlEnv_getenv("LC_ALL");
806 /* Use any "LC_ALL" environment variable, as it overrides everything
808 if (lc_all && strNE(lc_all, "")) {
813 /* Otherwise, we need to dig deeper. Unless overridden, the
814 * default is the LANG environment variable; if it doesn't exist,
817 const char * default_name;
819 default_name = PerlEnv_getenv("LANG");
821 if (! default_name || strEQ(default_name, "")) {
825 if (category != LC_ALL) {
826 const char * const name = PerlEnv_getenv(category_names[index]);
828 /* Here we are setting a single category. Assume will have the
830 locale = default_name;
832 /* But then look for an overriding environment variable */
833 if (name && strNE(name, "")) {
838 bool did_override = FALSE;
841 /* Here, we are getting LC_ALL. Any categories that don't have
842 * a corresponding environment variable set should be set to
843 * LANG, or to "C" if there is no LANG. If no individual
844 * categories differ from this, we can just set LC_ALL. This
845 * is buggy on systems that have extra categories that we don't
846 * know about. If there is an environment variable that sets
847 * that category, we won't know to look for it, and so our use
848 * of LANG or "C" improperly overrides it. On the other hand,
849 * if we don't do what is done here, and there is no
850 * environment variable, the category's locale should be set to
851 * LANG or "C". So there is no good solution. khw thinks the
852 * best is to look at systems to see what categories they have,
853 * and include them, and then to assume that we know the
856 for (i = 0; i < LC_ALL_INDEX; i++) {
857 const char * const env_override
858 = PerlEnv_getenv(category_names[i]);
859 const char * this_locale = ( env_override
860 && strNE(env_override, ""))
863 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
868 if (strNE(this_locale, default_name)) {
873 /* If all the categories are the same, we can set LC_ALL to
875 if (! did_override) {
876 locale = default_name;
880 /* Here, LC_ALL is no longer valid, as some individual
881 * categories don't match it. We call ourselves
882 * recursively, as that will execute the code that
883 * generates the proper locale string for this situation.
884 * We don't do the remainder of this function, as that is
885 * to update our records, and we've just done that for the
886 * individual categories in the loop above, and doing so
887 * would cause LC_ALL to be done as well */
888 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
892 } /* End of this being setlocale(LC_foo, "") */
893 else if (strchr(locale, ';')) {
895 /* LC_ALL may actually incude a conglomeration of various categories.
896 * Without querylocale, this code uses the glibc (as of this writing)
897 * syntax for representing that, but that is not a stable API, and
898 * other platforms do it differently, so we have to handle all cases
902 const char * s = locale;
903 const char * e = locale + strlen(locale);
905 const char * category_end;
906 const char * name_start;
907 const char * name_end;
909 /* If the string that gives what to set doesn't include all categories,
910 * the omitted ones get set to "C". To get this behavior, first set
911 * all the individual categories to "C", and override the furnished
913 for (i = 0; i < LC_ALL_INDEX; i++) {
914 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
921 /* Parse through the category */
922 while (isWORDCHAR(*p)) {
929 "panic: %s: %d: Unexpected character in locale name '%02X",
930 __FILE__, __LINE__, *(p-1));
933 /* Parse through the locale name */
935 while (p < e && *p != ';') {
938 "panic: %s: %d: Unexpected character in locale name '%02X",
939 __FILE__, __LINE__, *(p-1));
945 /* Space past the semi-colon */
950 /* Find the index of the category name in our lists */
951 for (i = 0; i < LC_ALL_INDEX; i++) {
952 char * individ_locale;
954 /* Keep going if this isn't the index. The strnNE() avoids a
955 * Perl_form(), but would fail if ever a category name could be
956 * a substring of another one, like if there were a
958 if strnNE(s, category_names[i], category_end - s) {
962 /* If this index is for the single category we're changing, we
963 * have found the locale to set it to. */
964 if (category == categories[i]) {
965 locale = Perl_form(aTHX_ "%.*s",
966 (int) (name_end - name_start),
971 assert(category == LC_ALL);
972 individ_locale = Perl_form(aTHX_ "%.*s",
973 (int) (name_end - name_start), name_start);
974 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
983 /* Here we have set all the individual categories by recursive calls.
984 * These collectively should have fixed up LC_ALL, so can just query
985 * what that now is */
986 assert(category == LC_ALL);
988 return do_setlocale_c(LC_ALL, NULL);
989 } /* End of this being setlocale(LC_ALL,
990 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
994 /* Here at the end of having to deal with the absence of querylocale().
995 * Some cases have already been fully handled by recursive calls to this
996 * function. But at this point, we haven't dealt with those, but are now
997 * prepared to, knowing what the locale name to set this category to is.
998 * This would have come for free if this system had had querylocale() */
1000 # endif /* end of ! querylocale */
1002 assert(PL_C_locale_obj);
1004 /* Switching locales generally entails freeing the current one's space (at
1005 * the C library's discretion). We need to stop using that locale before
1006 * the switch. So switch to a known locale object that we don't otherwise
1007 * mess with. This returns the locale object in effect at the time of the
1009 old_obj = uselocale(PL_C_locale_obj);
1013 if (DEBUG_Lv_TEST || debug_initialization) {
1014 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
1023 if (DEBUG_L_TEST || debug_initialization) {
1025 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1036 if (DEBUG_Lv_TEST || debug_initialization) {
1037 PerlIO_printf(Perl_debug_log,
1038 "%s:%d: emulate_setlocale now using %p\n",
1039 __FILE__, __LINE__, PL_C_locale_obj);
1044 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1045 * in fact, we already have switched to it (in preparation for what
1046 * normally is to come). But since we're already there, continue to use
1047 * it instead of trying to create a new locale */
1048 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1052 if (DEBUG_Lv_TEST || debug_initialization) {
1053 PerlIO_printf(Perl_debug_log,
1054 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1059 new_obj = PL_C_locale_obj;
1061 /* We already had switched to the C locale in preparation for freeing
1063 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1064 freelocale(old_obj);
1068 /* If we weren't in a thread safe locale, set so that newlocale() below
1069 * which uses 'old_obj', uses an empty one. Same for our reserved C
1070 * object. The latter is defensive coding, so that, even if there is
1071 * some bug, we will never end up trying to modify either of these, as
1072 * if passed to newlocale(), they can be. */
1073 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1074 old_obj = (locale_t) 0;
1077 /* Ready to create a new locale by modification of the exising one */
1078 new_obj = newlocale(mask, locale, old_obj);
1085 if (DEBUG_L_TEST || debug_initialization) {
1086 PerlIO_printf(Perl_debug_log,
1087 "%s:%d: emulate_setlocale creating new object"
1088 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1093 if (! uselocale(old_obj)) {
1097 if (DEBUG_L_TEST || debug_initialization) {
1098 PerlIO_printf(Perl_debug_log,
1099 "%s:%d: switching back failed: %d\n",
1100 __FILE__, __LINE__, GET_ERRNO);
1112 if (DEBUG_Lv_TEST || debug_initialization) {
1113 PerlIO_printf(Perl_debug_log,
1114 "%s:%d: emulate_setlocale created %p",
1115 __FILE__, __LINE__, new_obj);
1117 PerlIO_printf(Perl_debug_log,
1118 "; should have freed %p", old_obj);
1120 PerlIO_printf(Perl_debug_log, "\n");
1125 /* And switch into it */
1126 if (! uselocale(new_obj)) {
1131 if (DEBUG_L_TEST || debug_initialization) {
1132 PerlIO_printf(Perl_debug_log,
1133 "%s:%d: emulate_setlocale switching to new object"
1134 " failed\n", __FILE__, __LINE__);
1139 if (! uselocale(old_obj)) {
1143 if (DEBUG_L_TEST || debug_initialization) {
1144 PerlIO_printf(Perl_debug_log,
1145 "%s:%d: switching back failed: %d\n",
1146 __FILE__, __LINE__, GET_ERRNO);
1152 freelocale(new_obj);
1160 if (DEBUG_Lv_TEST || debug_initialization) {
1161 PerlIO_printf(Perl_debug_log,
1162 "%s:%d: emulate_setlocale now using %p\n",
1163 __FILE__, __LINE__, new_obj);
1168 /* We are done, except for updating our records (if the system doesn't keep
1169 * them) and in the case of locale "", we don't actually know what the
1170 * locale that got switched to is, as it came from the environment. So
1171 * have to find it */
1173 # ifdef HAS_QUERYLOCALE
1175 if (strEQ(locale, "")) {
1176 locale = querylocale(mask, new_obj);
1181 /* Here, 'locale' is the return value */
1183 /* Without querylocale(), we have to update our records */
1185 if (category == LC_ALL) {
1188 /* For LC_ALL, we change all individual categories to correspond */
1189 /* PL_curlocales is a parallel array, so has same
1190 * length as 'categories' */
1191 for (i = 0; i <= LC_ALL_INDEX; i++) {
1192 Safefree(PL_curlocales[i]);
1193 PL_curlocales[i] = savepv(locale);
1196 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
1200 /* For a single category, if it's not the same as the one in LC_ALL, we
1203 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1204 Safefree(PL_curlocales[LC_ALL_INDEX]);
1205 PL_curlocales[LC_ALL_INDEX] = NULL;
1208 /* Then update the category's record */
1209 Safefree(PL_curlocales[index]);
1210 PL_curlocales[index] = savepv(locale);
1212 FIX_GLIBC_LC_MESSAGES_BUG(index);
1220 #endif /* USE_POSIX_2008_LOCALE */
1222 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1223 didn't natively support them */
1225 /* The way this would work is that we would keep a per-thread list of the
1226 * correct locale for that thread. Any operation that was locale-sensitive
1227 * would have to be changed so that it would look like this:
1230 * setlocale to the correct locale for this operation
1234 * This leaves the global locale in the most recently used operation's, but it
1235 * was locked long enough to get the result. If that result is static, it
1236 * needs to be copied before the unlock.
1238 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1239 * the setup, but are no-ops when not needed, and similarly,
1240 * END_LOCALE_DEPENDENT_OP for the tear-down
1242 * But every call to a locale-sensitive function would have to be changed, and
1243 * if a module didn't cooperate by using the mutex, things would break.
1245 * This code was abandoned before being completed or tested, and is left as-is
1248 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1249 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1252 S_locking_setlocale(pTHX_
1254 const char * locale,
1256 const bool is_index_valid
1259 /* This function kind of performs a setlocale() on just the current thread;
1260 * thus it is kind of thread-safe. It does this by keeping a thread-level
1261 * array of the current locales for each category. Every time a locale is
1262 * switched to, it does the switch globally, but updates the thread's
1263 * array. A query as to what the current locale is just returns the
1264 * appropriate element from the array, and doesn't actually call the system
1265 * setlocale(). The saving into the array is done in an uninterruptible
1266 * section of code, so is unaffected by whatever any other threads might be
1269 * All locale-sensitive operations must work by first starting a critical
1270 * section, then switching to the thread's locale as kept by this function,
1271 * and then doing the operation, then ending the critical section. Thus,
1272 * each gets done in the appropriate locale. simulating thread-safety.
1274 * This function takes the same parameters, 'category' and 'locale', that
1275 * the regular setlocale() function does, but it also takes two additional
1276 * ones. This is because as described earlier. If we know on input the
1277 * index corresponding to the category into the array where we store the
1278 * current locales, we don't have to calculate it. If the caller knows at
1279 * compile time what the index is, it can pass it, setting
1280 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1284 /* If the input index might be incorrect, calculate the correct one */
1285 if (! is_index_valid) {
1288 if (DEBUG_Lv_TEST || debug_initialization) {
1289 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1292 for (i = 0; i <= LC_ALL_INDEX; i++) {
1293 if (category == categories[i]) {
1299 /* Here, we don't know about this category, so can't handle it.
1300 * XXX best we can do is to unsafely set this
1303 return my_setlocale(category, locale);
1307 if (DEBUG_Lv_TEST || debug_initialization) {
1308 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1312 /* For a query, just return what's in our records */
1313 if (new_locale == NULL) {
1314 return curlocales[index];
1318 /* Otherwise, we need to do the switch, and save the result, all in a
1319 * critical section */
1321 Safefree(curlocales[[index]]);
1323 /* It might be that this is called from an already-locked section of code.
1324 * We would have to detect and skip the LOCK/UNLOCK if so */
1327 curlocales[index] = savepv(my_setlocale(category, new_locale));
1329 if (strEQ(new_locale, "")) {
1333 /* The locale values come from the environment, and may not all be the
1334 * same, so for LC_ALL, we have to update all the others, while the
1335 * mutex is still locked */
1337 if (category == LC_ALL) {
1339 for (i = 0; i < LC_ALL_INDEX) {
1340 curlocales[i] = my_setlocale(categories[i], NULL);
1349 return curlocales[index];
1356 S_set_numeric_radix(pTHX_ const bool use_locale)
1358 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1359 * TRUE, use the radix character derived from the current locale */
1361 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1362 || defined(HAS_NL_LANGINFO))
1364 const char * radix = (use_locale)
1365 ? my_nl_langinfo(RADIXCHAR, FALSE)
1366 /* FALSE => already in dest locale */
1369 sv_setpv(PL_numeric_radix_sv, radix);
1371 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1372 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1373 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1374 SvCUR(PL_numeric_radix_sv))
1375 && _is_cur_LC_category_utf8(LC_NUMERIC))
1377 SvUTF8_on(PL_numeric_radix_sv);
1382 if (DEBUG_L_TEST || debug_initialization) {
1383 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1384 SvPVX(PL_numeric_radix_sv),
1385 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1391 PERL_UNUSED_ARG(use_locale);
1393 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1398 S_new_numeric(pTHX_ const char *newnum)
1401 #ifndef USE_LOCALE_NUMERIC
1403 PERL_UNUSED_ARG(newnum);
1407 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1408 * core Perl this and that 'newnum' is the name of the new locale.
1409 * It installs this locale as the current underlying default.
1411 * The default locale and the C locale can be toggled between by use of the
1412 * set_numeric_underlying() and set_numeric_standard() functions, which
1413 * should probably not be called directly, but only via macros like
1414 * SET_NUMERIC_STANDARD() in perl.h.
1416 * The toggling is necessary mainly so that a non-dot radix decimal point
1417 * character can be output, while allowing internal calculations to use a
1420 * This sets several interpreter-level variables:
1421 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1422 * PL_numeric_underlying A boolean indicating if the toggled state is such
1423 * that the current locale is the program's underlying
1425 * PL_numeric_standard An int indicating if the toggled state is such
1426 * that the current locale is the C locale or
1427 * indistinguishable from the C locale. If non-zero, it
1428 * is in C; if > 1, it means it may not be toggled away
1430 * PL_numeric_underlying_is_standard A bool kept by this function
1431 * indicating that the underlying locale and the standard
1432 * C locale are indistinguishable for the purposes of
1433 * LC_NUMERIC. This happens when both of the above two
1434 * variables are true at the same time. (Toggling is a
1435 * no-op under these circumstances.) This variable is
1436 * used to avoid having to recalculate.
1442 Safefree(PL_numeric_name);
1443 PL_numeric_name = NULL;
1444 PL_numeric_standard = TRUE;
1445 PL_numeric_underlying = TRUE;
1446 PL_numeric_underlying_is_standard = TRUE;
1450 save_newnum = stdize_locale(savepv(newnum));
1451 PL_numeric_underlying = TRUE;
1452 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1454 #ifndef TS_W32_BROKEN_LOCALECONV
1456 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1457 * them. But on broken Windows systems calling my_nl_langinfo() for
1458 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1459 * and just always change the locale if not C nor POSIX on those systems */
1460 if (! PL_numeric_standard) {
1461 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1462 FALSE /* Don't toggle locale */ ))
1463 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1468 /* Save the new name if it isn't the same as the previous one, if any */
1469 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1470 Safefree(PL_numeric_name);
1471 PL_numeric_name = save_newnum;
1474 Safefree(save_newnum);
1477 PL_numeric_underlying_is_standard = PL_numeric_standard;
1479 # ifdef HAS_POSIX_2008_LOCALE
1481 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1483 PL_underlying_numeric_obj);
1487 if (DEBUG_L_TEST || debug_initialization) {
1488 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1491 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1492 * have to worry about the radix being a non-dot. (Core operations that
1493 * need the underlying locale change to it temporarily). */
1494 if (PL_numeric_standard) {
1495 set_numeric_radix(0);
1498 set_numeric_standard();
1501 #endif /* USE_LOCALE_NUMERIC */
1506 Perl_set_numeric_standard(pTHX)
1509 #ifdef USE_LOCALE_NUMERIC
1511 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1512 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1513 * macro avoids calling this routine if toggling isn't necessary according
1514 * to our records (which could be wrong if some XS code has changed the
1515 * locale behind our back) */
1519 if (DEBUG_L_TEST || debug_initialization) {
1520 PerlIO_printf(Perl_debug_log,
1521 "Setting LC_NUMERIC locale to standard C\n");
1526 do_setlocale_c(LC_NUMERIC, "C");
1527 PL_numeric_standard = TRUE;
1528 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1529 set_numeric_radix(0);
1531 #endif /* USE_LOCALE_NUMERIC */
1536 Perl_set_numeric_underlying(pTHX)
1539 #ifdef USE_LOCALE_NUMERIC
1541 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1542 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1543 * instead of calling this directly. The macro avoids calling this routine
1544 * if toggling isn't necessary according to our records (which could be
1545 * wrong if some XS code has changed the locale behind our back) */
1549 if (DEBUG_L_TEST || debug_initialization) {
1550 PerlIO_printf(Perl_debug_log,
1551 "Setting LC_NUMERIC locale to %s\n",
1557 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1558 PL_numeric_standard = PL_numeric_underlying_is_standard;
1559 PL_numeric_underlying = TRUE;
1560 set_numeric_radix(! PL_numeric_standard);
1562 #endif /* USE_LOCALE_NUMERIC */
1567 * Set up for a new ctype locale.
1570 S_new_ctype(pTHX_ const char *newctype)
1573 #ifndef USE_LOCALE_CTYPE
1575 PERL_UNUSED_ARG(newctype);
1576 PERL_UNUSED_CONTEXT;
1580 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1581 * core Perl this and that 'newctype' is the name of the new locale.
1583 * This function sets up the folding arrays for all 256 bytes, assuming
1584 * that tofold() is tolc() since fold case is not a concept in POSIX,
1586 * Any code changing the locale (outside this file) should use
1587 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1588 * this function should be called directly only from this file and from
1589 * POSIX::setlocale() */
1593 /* Don't check for problems if we are suppressing the warnings */
1594 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1595 bool maybe_utf8_turkic = FALSE;
1597 PERL_ARGS_ASSERT_NEW_CTYPE;
1599 /* We will replace any bad locale warning with 1) nothing if the new one is
1600 * ok; or 2) a new warning for the bad new locale */
1601 if (PL_warn_locale) {
1602 SvREFCNT_dec_NN(PL_warn_locale);
1603 PL_warn_locale = NULL;
1606 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1608 /* A UTF-8 locale gets standard rules. But note that code still has to
1609 * handle this specially because of the three problematic code points */
1610 if (PL_in_utf8_CTYPE_locale) {
1611 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1613 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1614 * Turkic. Make sure these two are the only anomalies. (We don't use
1615 * towupper and towlower because they aren't in C89.) */
1617 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1619 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1623 if (toupper('i') == 'i' && tolower('I') == 'I') {
1626 check_for_problems = TRUE;
1627 maybe_utf8_turkic = TRUE;
1631 /* We don't populate the other lists if a UTF-8 locale, but do check that
1632 * everything works as expected, unless checking turned off */
1633 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1634 /* Assume enough space for every character being bad. 4 spaces each
1635 * for the 94 printable characters that are output like "'x' "; and 5
1636 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1638 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1639 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1641 unsigned int bad_count = 0; /* Count of bad characters */
1643 for (i = 0; i < 256; i++) {
1644 if (! PL_in_utf8_CTYPE_locale) {
1646 PL_fold_locale[i] = (U8) tolower(i);
1647 else if (islower(i))
1648 PL_fold_locale[i] = (U8) toupper(i);
1650 PL_fold_locale[i] = (U8) i;
1653 /* If checking for locale problems, see if the native ASCII-range
1654 * printables plus \n and \t are in their expected categories in
1655 * the new locale. If not, this could mean big trouble, upending
1656 * Perl's and most programs' assumptions, like having a
1657 * metacharacter with special meaning become a \w. Fortunately,
1658 * it's very rare to find locales that aren't supersets of ASCII
1659 * nowadays. It isn't a problem for most controls to be changed
1660 * into something else; we check only \n and \t, though perhaps \r
1661 * could be an issue as well. */
1662 if ( check_for_problems
1663 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1665 bool is_bad = FALSE;
1666 char name[4] = { '\0' };
1668 /* Convert the name into a string */
1673 else if (i == '\n') {
1674 my_strlcpy(name, "\\n", sizeof(name));
1676 else if (i == '\t') {
1677 my_strlcpy(name, "\\t", sizeof(name));
1681 my_strlcpy(name, "' '", sizeof(name));
1684 /* Check each possibe class */
1685 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1687 DEBUG_L(PerlIO_printf(Perl_debug_log,
1688 "isalnum('%s') unexpectedly is %d\n",
1689 name, cBOOL(isalnum(i))));
1691 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1693 DEBUG_L(PerlIO_printf(Perl_debug_log,
1694 "isalpha('%s') unexpectedly is %d\n",
1695 name, cBOOL(isalpha(i))));
1697 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1699 DEBUG_L(PerlIO_printf(Perl_debug_log,
1700 "isdigit('%s') unexpectedly is %d\n",
1701 name, cBOOL(isdigit(i))));
1703 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1705 DEBUG_L(PerlIO_printf(Perl_debug_log,
1706 "isgraph('%s') unexpectedly is %d\n",
1707 name, cBOOL(isgraph(i))));
1709 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1711 DEBUG_L(PerlIO_printf(Perl_debug_log,
1712 "islower('%s') unexpectedly is %d\n",
1713 name, cBOOL(islower(i))));
1715 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1717 DEBUG_L(PerlIO_printf(Perl_debug_log,
1718 "isprint('%s') unexpectedly is %d\n",
1719 name, cBOOL(isprint(i))));
1721 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1723 DEBUG_L(PerlIO_printf(Perl_debug_log,
1724 "ispunct('%s') unexpectedly is %d\n",
1725 name, cBOOL(ispunct(i))));
1727 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1729 DEBUG_L(PerlIO_printf(Perl_debug_log,
1730 "isspace('%s') unexpectedly is %d\n",
1731 name, cBOOL(isspace(i))));
1733 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1735 DEBUG_L(PerlIO_printf(Perl_debug_log,
1736 "isupper('%s') unexpectedly is %d\n",
1737 name, cBOOL(isupper(i))));
1739 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1741 DEBUG_L(PerlIO_printf(Perl_debug_log,
1742 "isxdigit('%s') unexpectedly is %d\n",
1743 name, cBOOL(isxdigit(i))));
1745 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1747 DEBUG_L(PerlIO_printf(Perl_debug_log,
1748 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1749 name, tolower(i), (int) toLOWER_A(i)));
1751 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1753 DEBUG_L(PerlIO_printf(Perl_debug_log,
1754 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1755 name, toupper(i), (int) toUPPER_A(i)));
1757 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1759 DEBUG_L(PerlIO_printf(Perl_debug_log,
1760 "'\\n' (=%02X) is not a control\n", (int) i));
1763 /* Add to the list; Separate multiple entries with a blank */
1766 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1768 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1774 if (bad_count == 2 && maybe_utf8_turkic) {
1776 *bad_chars_list = '\0';
1777 PL_fold_locale['I'] = 'I';
1778 PL_fold_locale['i'] = 'i';
1779 PL_in_utf8_turkic_locale = TRUE;
1780 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1781 __FILE__, __LINE__, newctype));
1784 PL_in_utf8_turkic_locale = FALSE;
1789 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1790 * this locale requires more than one byte, there are going to be
1792 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1793 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1794 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1796 if ( check_for_problems && MB_CUR_MAX > 1
1797 && ! PL_in_utf8_CTYPE_locale
1799 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1800 * locale. Just assume that the implementation for them (plus
1801 * for POSIX) is correct and the > 1 value is spurious. (Since
1802 * these are specially handled to never be considered UTF-8
1803 * locales, as long as this is the only problem, everything
1804 * should work fine */
1805 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1807 multi_byte_locale = TRUE;
1812 /* If we found problems and we want them output, do so */
1813 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1814 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1816 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1817 PL_warn_locale = Perl_newSVpvf(aTHX_
1818 "Locale '%s' contains (at least) the following characters"
1819 " which have\nunexpected meanings: %s\nThe Perl program"
1820 " will use the expected meanings",
1821 newctype, bad_chars_list);
1824 PL_warn_locale = Perl_newSVpvf(aTHX_
1825 "Locale '%s' may not work well.%s%s%s\n",
1828 ? " Some characters in it are not recognized by"
1832 ? "\nThe following characters (and maybe others)"
1833 " may not have the same meaning as the Perl"
1834 " program expects:\n"
1842 # ifdef HAS_NL_LANGINFO
1844 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1845 /* parameter FALSE is a don't care here */
1846 my_nl_langinfo(CODESET, FALSE));
1850 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1852 /* If we are actually in the scope of the locale or are debugging,
1853 * output the message now. If not in that scope, we save the
1854 * message to be output at the first operation using this locale,
1855 * if that actually happens. Most programs don't use locales, so
1856 * they are immune to bad ones. */
1857 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1859 /* The '0' below suppresses a bogus gcc compiler warning */
1860 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1862 if (IN_LC(LC_CTYPE)) {
1863 SvREFCNT_dec_NN(PL_warn_locale);
1864 PL_warn_locale = NULL;
1870 #endif /* USE_LOCALE_CTYPE */
1875 Perl__warn_problematic_locale()
1878 #ifdef USE_LOCALE_CTYPE
1882 /* Internal-to-core function that outputs the message in PL_warn_locale,
1883 * and then NULLS it. Should be called only through the macro
1884 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1886 if (PL_warn_locale) {
1887 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1888 SvPVX(PL_warn_locale),
1889 0 /* dummy to avoid compiler warning */ );
1890 SvREFCNT_dec_NN(PL_warn_locale);
1891 PL_warn_locale = NULL;
1899 S_new_collate(pTHX_ const char *newcoll)
1902 #ifndef USE_LOCALE_COLLATE
1904 PERL_UNUSED_ARG(newcoll);
1905 PERL_UNUSED_CONTEXT;
1909 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1910 * core Perl this and that 'newcoll' is the name of the new locale.
1912 * The design of locale collation is that every locale change is given an
1913 * index 'PL_collation_ix'. The first time a string particpates in an
1914 * operation that requires collation while locale collation is active, it
1915 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1916 * magic includes the collation index, and the transformation of the string
1917 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1918 * instead of the string itself. If a string changes, the magic is
1919 * cleared. The next time the locale changes, the index is incremented,
1920 * and so we know during a comparison that the transformation is not
1921 * necessarily still valid, and so is recomputed. Note that if the locale
1922 * changes enough times, the index could wrap (a U32), and it is possible
1923 * that a transformation would improperly be considered valid, leading to
1924 * an unlikely bug */
1927 if (PL_collation_name) {
1929 Safefree(PL_collation_name);
1930 PL_collation_name = NULL;
1932 PL_collation_standard = TRUE;
1933 is_standard_collation:
1934 PL_collxfrm_base = 0;
1935 PL_collxfrm_mult = 2;
1936 PL_in_utf8_COLLATE_locale = FALSE;
1937 PL_strxfrm_NUL_replacement = '\0';
1938 PL_strxfrm_max_cp = 0;
1942 /* If this is not the same locale as currently, set the new one up */
1943 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1945 Safefree(PL_collation_name);
1946 PL_collation_name = stdize_locale(savepv(newcoll));
1947 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1948 if (PL_collation_standard) {
1949 goto is_standard_collation;
1952 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1953 PL_strxfrm_NUL_replacement = '\0';
1954 PL_strxfrm_max_cp = 0;
1956 /* A locale collation definition includes primary, secondary, tertiary,
1957 * etc. weights for each character. To sort, the primary weights are
1958 * used, and only if they compare equal, then the secondary weights are
1959 * used, and only if they compare equal, then the tertiary, etc.
1961 * strxfrm() works by taking the input string, say ABC, and creating an
1962 * output transformed string consisting of first the primary weights,
1963 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1964 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1965 * may not have weights at every level. In our example, let's say B
1966 * doesn't have a tertiary weight, and A doesn't have a secondary
1967 * weight. The constructed string is then going to be
1968 * A¹B¹C¹ B²C² A³C³ ....
1969 * This has the desired effect that strcmp() will look at the secondary
1970 * or tertiary weights only if the strings compare equal at all higher
1971 * priority weights. The spaces shown here, like in
1973 * are not just for readability. In the general case, these must
1974 * actually be bytes, which we will call here 'separator weights'; and
1975 * they must be smaller than any other weight value, but since these
1976 * are C strings, only the terminating one can be a NUL (some
1977 * implementations may include a non-NUL separator weight just before
1978 * the NUL). Implementations tend to reserve 01 for the separator
1979 * weights. They are needed so that a shorter string's secondary
1980 * weights won't be misconstrued as primary weights of a longer string,
1981 * etc. By making them smaller than any other weight, the shorter
1982 * string will sort first. (Actually, if all secondary weights are
1983 * smaller than all primary ones, there is no need for a separator
1984 * weight between those two levels, etc.)
1986 * The length of the transformed string is roughly a linear function of
1987 * the input string. It's not exactly linear because some characters
1988 * don't have weights at all levels. When we call strxfrm() we have to
1989 * allocate some memory to hold the transformed string. The
1990 * calculations below try to find coefficients 'm' and 'b' for this
1991 * locale so that m*x + b equals how much space we need, given the size
1992 * of the input string in 'x'. If we calculate too small, we increase
1993 * the size as needed, and call strxfrm() again, but it is better to
1994 * get it right the first time to avoid wasted expensive string
1995 * transformations. */
1998 /* We use the string below to find how long the tranformation of it
1999 * is. Almost all locales are supersets of ASCII, or at least the
2000 * ASCII letters. We use all of them, half upper half lower,
2001 * because if we used fewer, we might hit just the ones that are
2002 * outliers in a particular locale. Most of the strings being
2003 * collated will contain a preponderance of letters, and even if
2004 * they are above-ASCII, they are likely to have the same number of
2005 * weight levels as the ASCII ones. It turns out that digits tend
2006 * to have fewer levels, and some punctuation has more, but those
2007 * are relatively sparse in text, and khw believes this gives a
2008 * reasonable result, but it could be changed if experience so
2010 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
2011 char * x_longer; /* Transformed 'longer' */
2012 Size_t x_len_longer; /* Length of 'x_longer' */
2014 char * x_shorter; /* We also transform a substring of 'longer' */
2015 Size_t x_len_shorter;
2017 /* _mem_collxfrm() is used get the transformation (though here we
2018 * are interested only in its length). It is used because it has
2019 * the intelligence to handle all cases, but to work, it needs some
2020 * values of 'm' and 'b' to get it started. For the purposes of
2021 * this calculation we use a very conservative estimate of 'm' and
2022 * 'b'. This assumes a weight can be multiple bytes, enough to
2023 * hold any UV on the platform, and there are 5 levels, 4 weight
2024 * bytes, and a trailing NUL. */
2025 PL_collxfrm_base = 5;
2026 PL_collxfrm_mult = 5 * sizeof(UV);
2028 /* Find out how long the transformation really is */
2029 x_longer = _mem_collxfrm(longer,
2033 /* We avoid converting to UTF-8 in the
2034 * called function by telling it the
2035 * string is in UTF-8 if the locale is a
2036 * UTF-8 one. Since the string passed
2037 * here is invariant under UTF-8, we can
2038 * claim it's UTF-8 even though it isn't.
2040 PL_in_utf8_COLLATE_locale);
2043 /* Find out how long the transformation of a substring of 'longer'
2044 * is. Together the lengths of these transformations are
2045 * sufficient to calculate 'm' and 'b'. The substring is all of
2046 * 'longer' except the first character. This minimizes the chances
2047 * of being swayed by outliers */
2048 x_shorter = _mem_collxfrm(longer + 1,
2051 PL_in_utf8_COLLATE_locale);
2052 Safefree(x_shorter);
2054 /* If the results are nonsensical for this simple test, the whole
2055 * locale definition is suspect. Mark it so that locale collation
2056 * is not active at all for it. XXX Should we warn? */
2057 if ( x_len_shorter == 0
2058 || x_len_longer == 0
2059 || x_len_shorter >= x_len_longer)
2061 PL_collxfrm_mult = 0;
2062 PL_collxfrm_base = 0;
2065 SSize_t base; /* Temporary */
2067 /* We have both: m * strlen(longer) + b = x_len_longer
2068 * m * strlen(shorter) + b = x_len_shorter;
2069 * subtracting yields:
2070 * m * (strlen(longer) - strlen(shorter))
2071 * = x_len_longer - x_len_shorter
2072 * But we have set things up so that 'shorter' is 1 byte smaller
2073 * than 'longer'. Hence:
2074 * m = x_len_longer - x_len_shorter
2076 * But if something went wrong, make sure the multiplier is at
2079 if (x_len_longer > x_len_shorter) {
2080 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2083 PL_collxfrm_mult = 1;
2088 * but in case something has gone wrong, make sure it is
2090 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2095 /* Add 1 for the trailing NUL */
2096 PL_collxfrm_base = base + 1;
2101 if (DEBUG_L_TEST || debug_initialization) {
2102 PerlIO_printf(Perl_debug_log,
2103 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2105 " collate multipler=%zu, collate base=%zu\n",
2107 PL_in_utf8_COLLATE_locale,
2108 x_len_shorter, x_len_longer,
2109 PL_collxfrm_mult, PL_collxfrm_base);
2116 #endif /* USE_LOCALE_COLLATE */
2124 #define USE_WSETLOCALE
2126 #ifdef USE_WSETLOCALE
2129 S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2136 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2143 Newx(wlocale, req_size, wchar_t);
2144 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2153 wresult = _wsetlocale(category, wlocale);
2157 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2158 Newx(result, req_size, char);
2159 SAVEFREEPV(result); /* is there something better we can do here? */
2160 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2161 result, req_size, NULL, NULL)) {
2176 S_win32_setlocale(pTHX_ int category, const char* locale)
2178 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2179 * difference between the two unless the input locale is "", which normally
2180 * means on Windows to get the machine default, which is set via the
2181 * computer's "Regional and Language Options" (or its current equivalent).
2182 * In POSIX, it instead means to find the locale from the user's
2183 * environment. This routine changes the Windows behavior to first look in
2184 * the environment, and, if anything is found, use that instead of going to
2185 * the machine default. If there is no environment override, the machine
2186 * default is used, by calling the real setlocale() with "".
2188 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2189 * use the particular category's variable if set; otherwise to use the LANG
2192 bool override_LC_ALL = FALSE;
2196 if (locale && strEQ(locale, "")) {
2200 locale = PerlEnv_getenv("LC_ALL");
2202 if (category == LC_ALL) {
2203 override_LC_ALL = TRUE;
2209 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2210 if (category == categories[i]) {
2211 locale = PerlEnv_getenv(category_names[i]);
2216 locale = PerlEnv_getenv("LANG");
2232 #ifdef USE_WSETLOCALE
2233 result = S_wrap_wsetlocale(aTHX_ category, locale);
2235 result = setlocale(category, locale);
2237 DEBUG_L(STMT_START {
2239 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2240 setlocale_debug_string(category, locale, result));
2244 if (! override_LC_ALL) {
2248 /* Here the input category was LC_ALL, and we have set it to what is in the
2249 * LANG variable or the system default if there is no LANG. But these have
2250 * lower priority than the other LC_foo variables, so override it for each
2251 * one that is set. (If they are set to "", it means to use the same thing
2252 * we just set LC_ALL to, so can skip) */
2254 for (i = 0; i < LC_ALL_INDEX; i++) {
2255 result = PerlEnv_getenv(category_names[i]);
2256 if (result && strNE(result, "")) {
2257 #ifdef USE_WSETLOCALE
2258 S_wrap_wsetlocale(aTHX_ categories[i], result);
2260 setlocale(categories[i], result);
2262 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2264 setlocale_debug_string(categories[i], result, "not captured")));
2268 result = setlocale(LC_ALL, NULL);
2269 DEBUG_L(STMT_START {
2271 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2273 setlocale_debug_string(LC_ALL, NULL, result));
2284 =for apidoc_section Locales
2286 =for apidoc Perl_setlocale
2288 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2289 taking the same parameters, and returning the same information, except that it
2290 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2291 instead return C<C> if the underlying locale has a non-dot decimal point
2292 character, or a non-empty thousands separator for displaying floating point
2293 numbers. This is because perl keeps that locale category such that it has a
2294 dot and empty separator, changing the locale briefly during the operations
2295 where the underlying one is required. C<Perl_setlocale> knows about this, and
2296 compensates; regular C<setlocale> doesn't.
2298 Another reason it isn't completely a drop-in replacement is that it is
2299 declared to return S<C<const char *>>, whereas the system setlocale omits the
2300 C<const> (presumably because its API was specified long ago, and can't be
2301 updated; it is illegal to change the information C<setlocale> returns; doing
2302 so leads to segfaults.)
2304 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2305 C<setlocale> can be completely ineffective on some platforms under some
2308 C<Perl_setlocale> should not be used to change the locale except on systems
2309 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2310 the system C<setlocale()> is ineffective, returning the wrong information, and
2311 failing to actually change the locale. C<Perl_setlocale>, however works
2312 properly in all circumstances.
2314 The return points to a per-thread static buffer, which is overwritten the next
2315 time C<Perl_setlocale> is called from the same thread.
2322 Perl_setlocale(const int category, const char * locale)
2324 /* This wraps POSIX::setlocale() */
2328 PERL_UNUSED_ARG(category);
2329 PERL_UNUSED_ARG(locale);
2335 const char * retval;
2336 const char * newlocale;
2339 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2341 #ifdef USE_LOCALE_NUMERIC
2343 /* A NULL locale means only query what the current one is. We have the
2344 * LC_NUMERIC name saved, because we are normally switched into the C
2345 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2346 * the correct results. All other categories don't require special
2348 if (locale == NULL) {
2349 if (category == LC_NUMERIC) {
2351 /* We don't have to copy this return value, as it is a per-thread
2352 * variable, and won't change until a future setlocale */
2353 return PL_numeric_name;
2358 else if (category == LC_ALL) {
2359 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2368 retval = save_to_buffer(do_setlocale_r(category, locale),
2369 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2372 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2374 if (locale == NULL && category == LC_ALL) {
2375 RESTORE_LC_NUMERIC();
2380 DEBUG_L(PerlIO_printf(Perl_debug_log,
2381 "%s:%d: %s\n", __FILE__, __LINE__,
2382 setlocale_debug_string(category, locale, retval)));
2390 /* If locale == NULL, we are just querying the state */
2391 if (locale == NULL) {
2395 /* Now that have switched locales, we have to update our records to
2400 #ifdef USE_LOCALE_CTYPE
2407 #ifdef USE_LOCALE_COLLATE
2410 new_collate(retval);
2414 #ifdef USE_LOCALE_NUMERIC
2417 new_numeric(retval);
2425 /* LC_ALL updates all the things we care about. The values may not
2426 * be the same as 'retval', as the locale "" may have set things
2429 # ifdef USE_LOCALE_CTYPE
2431 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2432 new_ctype(newlocale);
2433 Safefree(newlocale);
2435 # endif /* USE_LOCALE_CTYPE */
2436 # ifdef USE_LOCALE_COLLATE
2438 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2439 new_collate(newlocale);
2440 Safefree(newlocale);
2443 # ifdef USE_LOCALE_NUMERIC
2445 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2446 new_numeric(newlocale);
2447 Safefree(newlocale);
2449 # endif /* USE_LOCALE_NUMERIC */
2462 PERL_STATIC_INLINE const char *
2463 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2465 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2466 * growing it if necessary */
2470 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2476 string_size = strlen(string) + offset + 1;
2478 if (*buf_size == 0) {
2479 Newx(*buf, string_size, char);
2480 *buf_size = string_size;
2482 else if (string_size > *buf_size) {
2483 Renew(*buf, string_size, char);
2484 *buf_size = string_size;
2487 Copy(string, *buf + offset, string_size - offset, char);
2493 =for apidoc Perl_langinfo
2495 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2496 taking the same C<item> parameter values, and returning the same information.
2497 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2498 of Perl's locale handling from your code, and can be used on systems that lack
2499 a native C<nl_langinfo>.
2507 The reason it isn't quite a drop-in replacement is actually an advantage. The
2508 only difference is that it returns S<C<const char *>>, whereas plain
2509 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2510 forbidden to write into the buffer. By declaring this C<const>, the compiler
2511 enforces this restriction, so if it is violated, you know at compilation time,
2512 rather than getting segfaults at runtime.
2516 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2517 without you having to write extra code. The reason for the extra code would be
2518 because these are from the C<LC_NUMERIC> locale category, which is normally
2519 kept set by Perl so that the radix is a dot, and the separator is the empty
2520 string, no matter what the underlying locale is supposed to be, and so to get
2521 the expected results, you have to temporarily toggle into the underlying
2522 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2523 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2524 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2525 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2526 (decimal point) character to be a dot.)
2530 The system function it replaces can have its static return buffer trashed,
2531 not only by a subsequent call to that function, but by a C<freelocale>,
2532 C<setlocale>, or other locale change. The returned buffer of this function is
2533 not changed until the next call to it, so the buffer is never in a trashed
2538 Its return buffer is per-thread, so it also is never overwritten by a call to
2539 this function from another thread; unlike the function it replaces.
2543 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2544 as Windows, hence makes your code more portable. Of the fifty-some possible
2545 items specified by the POSIX 2008 standard,
2546 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2547 only one is completely unimplemented, though on non-Windows platforms, another
2548 significant one is also not implemented). It uses various techniques to
2549 recover the other items, including calling C<L<localeconv(3)>>, and
2550 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2551 available. Later C<strftime()> versions have additional capabilities; C<""> is
2552 returned for those not available on your system.
2554 It is important to note that when called with an item that is recovered by
2555 using C<localeconv>, the buffer from any previous explicit call to
2556 C<localeconv> will be overwritten. This means you must save that buffer's
2557 contents if you need to access them after a call to this function. (But note
2558 that you might not want to be using C<localeconv()> directly anyway, because of
2559 issues like the ones listed in the second item of this list (above) for
2560 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2561 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2564 The details for those items which may deviate from what this emulation returns
2565 and what a native C<nl_langinfo()> would return are specified in
2570 When using C<Perl_langinfo> on systems that don't have a native
2571 C<nl_langinfo()>, you must
2573 #include "perl_langinfo.h"
2575 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2576 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2577 C<langinfo.h> would try to import into the namespace for code that doesn't need
2580 The original impetus for C<Perl_langinfo()> was so that code that needs to
2581 find out the current currency symbol, floating point radix character, or digit
2582 grouping separator can use, on all systems, the simpler and more
2583 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2584 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2585 is better to use the methods given in L<perlcall> to call
2586 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2593 #ifdef HAS_NL_LANGINFO
2594 Perl_langinfo(const nl_item item)
2596 Perl_langinfo(const int item)
2599 return my_nl_langinfo(item, TRUE);
2603 #ifdef HAS_NL_LANGINFO
2604 S_my_nl_langinfo(const nl_item item, bool toggle)
2606 S_my_nl_langinfo(const int item, bool toggle)
2610 const char * retval;
2612 #ifdef USE_LOCALE_NUMERIC
2614 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2615 * two items, and only if not already there */
2616 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2617 || PL_numeric_underlying))
2619 #endif /* No toggling needed if not using LC_NUMERIC */
2623 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2624 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2625 || ! defined(HAS_POSIX_2008_LOCALE) \
2626 || ! defined(DUPLOCALE)
2628 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2629 * for those items dependent on it. This must be copied to a buffer before
2630 * switching back, as some systems destroy the buffer when setlocale() is
2634 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2637 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2640 LOCALE_LOCK; /* Prevent interference from another thread executing
2641 this code section (the only call to nl_langinfo in
2645 /* Copy to a per-thread buffer, which is also one that won't be
2646 * destroyed by a subsequent setlocale(), such as the
2647 * RESTORE_LC_NUMERIC may do just below. */
2648 retval = save_to_buffer(nl_langinfo(item),
2649 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2654 RESTORE_LC_NUMERIC();
2658 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2661 bool do_free = FALSE;
2662 locale_t cur = uselocale((locale_t) 0);
2664 if (cur == LC_GLOBAL_LOCALE) {
2665 cur = duplocale(LC_GLOBAL_LOCALE);
2669 # ifdef USE_LOCALE_NUMERIC
2672 if (PL_underlying_numeric_obj) {
2673 cur = PL_underlying_numeric_obj;
2676 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2683 /* We have to save it to a buffer, because the freelocale() just below
2684 * can invalidate the internal one */
2685 retval = save_to_buffer(nl_langinfo_l(item, cur),
2686 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2695 if (strEQ(retval, "")) {
2696 if (item == YESSTR) {
2699 if (item == NOSTR) {
2706 #else /* Below, emulate nl_langinfo as best we can */
2710 # ifdef HAS_LOCALECONV
2712 const struct lconv* lc;
2714 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2716 # ifdef TS_W32_BROKEN_LOCALECONV
2718 const char * save_global;
2719 const char * save_thread;
2727 # ifdef HAS_STRFTIME
2730 bool return_format = FALSE; /* Return the %format, not the value */
2731 const char * format;
2735 /* We copy the results to a per-thread buffer, even if not
2736 * multi-threaded. This is in part to simplify this code, and partly
2737 * because we need a buffer anyway for strftime(), and partly because a
2738 * call of localeconv() could otherwise wipe out the buffer, and the
2739 * programmer would not be expecting this, as this is a nl_langinfo()
2740 * substitute after all, so s/he might be thinking their localeconv()
2741 * is safe until another localeconv() call. */
2746 /* This is unimplemented */
2747 case ERA: /* For use with strftime() %E modifier */
2752 /* We use only an English set, since we don't know any more */
2753 case YESEXPR: return "^[+1yY]";
2754 case YESSTR: return "yes";
2755 case NOEXPR: return "^[-0nN]";
2756 case NOSTR: return "no";
2762 /* On non-windows, this is unimplemented, in part because of
2763 * inconsistencies between vendors. The Darwin native
2764 * nl_langinfo() implementation simply looks at everything past
2765 * any dot in the name, but that doesn't work for other
2766 * vendors. Many Linux locales that don't have UTF-8 in their
2767 * names really are UTF-8, for example; z/OS locales that do
2768 * have UTF-8 in their names, aren't really UTF-8 */
2773 { /* But on Windows, the name does seem to be consistent, so
2778 const char * name = my_setlocale(LC_CTYPE, NULL);
2780 if (isNAME_C_OR_POSIX(name)) {
2781 return "ANSI_X3.4-1968";
2784 /* Find the dot in the locale name */
2785 first = (const char *) strchr(name, '.');
2791 /* Look at everything past the dot */
2796 if (! isDIGIT(*p)) {
2803 /* Here everything past the dot is a digit. Treat it as a
2805 retval = save_to_buffer("CP", &PL_langinfo_buf,
2806 &PL_langinfo_bufsize, 0);
2807 offset = STRLENs("CP");
2811 retval = save_to_buffer(first, &PL_langinfo_buf,
2812 &PL_langinfo_bufsize, offset);
2818 # ifdef HAS_LOCALECONV
2822 /* We don't bother with localeconv_l() because any system that
2823 * has it is likely to also have nl_langinfo() */
2825 LOCALE_LOCK_V; /* Prevent interference with other threads
2826 using localeconv() */
2828 # ifdef TS_W32_BROKEN_LOCALECONV
2830 /* This is a workaround for a Windows bug prior to VS 15.
2831 * What we do here is, while locked, switch to the global
2832 * locale so localeconv() works; then switch back just before
2833 * the unlock. This can screw things up if some thread is
2834 * already using the global locale while assuming no other is.
2835 * A different workaround would be to call GetCurrencyFormat on
2836 * a known value, and parse it; patches welcome
2838 * We have to use LC_ALL instead of LC_MONETARY because of
2839 * another bug in Windows */
2841 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2842 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2843 save_global= savepv(my_setlocale(LC_ALL, NULL));
2844 my_setlocale(LC_ALL, save_thread);
2850 || ! lc->currency_symbol
2851 || strEQ("", lc->currency_symbol))
2857 /* Leave the first spot empty to be filled in below */
2858 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2859 &PL_langinfo_bufsize, 1);
2860 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2861 { /* khw couldn't figure out how the localedef specifications
2862 would show that the $ should replace the radix; this is
2863 just a guess as to how it might work.*/
2864 PL_langinfo_buf[0] = '.';
2866 else if (lc->p_cs_precedes) {
2867 PL_langinfo_buf[0] = '-';
2870 PL_langinfo_buf[0] = '+';
2873 # ifdef TS_W32_BROKEN_LOCALECONV
2875 my_setlocale(LC_ALL, save_global);
2876 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2877 my_setlocale(LC_ALL, save_thread);
2878 Safefree(save_global);
2879 Safefree(save_thread);
2886 # ifdef TS_W32_BROKEN_LOCALECONV
2890 /* For this, we output a known simple floating point number to
2891 * a buffer, and parse it, looking for the radix */
2894 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2897 if (PL_langinfo_bufsize < 10) {
2898 PL_langinfo_bufsize = 10;
2899 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2902 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2904 if (needed_size >= (int) PL_langinfo_bufsize) {
2905 PL_langinfo_bufsize = needed_size + 1;
2906 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2907 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2909 assert(needed_size < (int) PL_langinfo_bufsize);
2912 ptr = PL_langinfo_buf;
2913 e = PL_langinfo_buf + PL_langinfo_bufsize;
2916 while (ptr < e && *ptr != '1') {
2923 while (ptr < e && *ptr != '5') {
2927 /* Everything in between is the radix string */
2929 PL_langinfo_buf[0] = '?';
2930 PL_langinfo_buf[1] = '\0';
2934 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2938 RESTORE_LC_NUMERIC();
2941 retval = PL_langinfo_buf;
2946 case RADIXCHAR: /* No special handling needed */
2953 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2956 LOCALE_LOCK_V; /* Prevent interference with other threads
2957 using localeconv() */
2959 # ifdef TS_W32_BROKEN_LOCALECONV
2961 /* This should only be for the thousands separator. A
2962 * different work around would be to use GetNumberFormat on a
2963 * known value and parse the result to find the separator */
2964 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2965 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2966 save_global = savepv(my_setlocale(LC_ALL, NULL));
2967 my_setlocale(LC_ALL, save_thread);
2969 /* This is the start of code that for broken Windows replaces
2970 * the above and below code, and instead calls
2971 * GetNumberFormat() and then would parse that to find the
2972 * thousands separator. It needs to handle UTF-16 vs -8
2975 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2976 DEBUG_L(PerlIO_printf(Perl_debug_log,
2977 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2978 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2988 temp = (item == RADIXCHAR)
2990 : lc->thousands_sep;
2996 retval = save_to_buffer(temp, &PL_langinfo_buf,
2997 &PL_langinfo_bufsize, 0);
2999 # ifdef TS_W32_BROKEN_LOCALECONV
3001 my_setlocale(LC_ALL, save_global);
3002 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3003 my_setlocale(LC_ALL, save_thread);
3004 Safefree(save_global);
3005 Safefree(save_thread);
3012 RESTORE_LC_NUMERIC();
3018 # ifdef HAS_STRFTIME
3020 /* These are defined by C89, so we assume that strftime supports
3021 * them, and so are returned unconditionally; they may not be what
3022 * the locale actually says, but should give good enough results
3023 * for someone using them as formats (as opposed to trying to parse
3024 * them to figure out what the locale says). The other format
3025 * items are actually tested to verify they work on the platform */
3026 case D_FMT: return "%x";
3027 case T_FMT: return "%X";
3028 case D_T_FMT: return "%c";
3030 /* These formats are only available in later strfmtime's */
3031 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
3033 /* The rest can be gotten from most versions of strftime(). */
3034 case ABDAY_1: case ABDAY_2: case ABDAY_3:
3035 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
3037 case AM_STR: case PM_STR:
3038 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
3039 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
3040 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
3041 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
3042 case DAY_5: case DAY_6: case DAY_7:
3043 case MON_1: case MON_2: case MON_3: case MON_4:
3044 case MON_5: case MON_6: case MON_7: case MON_8:
3045 case MON_9: case MON_10: case MON_11: case MON_12:
3049 init_tm(&tm); /* Precaution against core dumps */
3053 tm.tm_year = 2017 - 1900;
3060 "panic: %s: %d: switch case: %d problem",
3061 __FILE__, __LINE__, item);
3062 NOT_REACHED; /* NOTREACHED */
3064 case PM_STR: tm.tm_hour = 18;
3069 case ABDAY_7: tm.tm_wday++;
3070 case ABDAY_6: tm.tm_wday++;
3071 case ABDAY_5: tm.tm_wday++;
3072 case ABDAY_4: tm.tm_wday++;
3073 case ABDAY_3: tm.tm_wday++;
3074 case ABDAY_2: tm.tm_wday++;
3079 case DAY_7: tm.tm_wday++;
3080 case DAY_6: tm.tm_wday++;
3081 case DAY_5: tm.tm_wday++;
3082 case DAY_4: tm.tm_wday++;
3083 case DAY_3: tm.tm_wday++;
3084 case DAY_2: tm.tm_wday++;
3089 case ABMON_12: tm.tm_mon++;
3090 case ABMON_11: tm.tm_mon++;
3091 case ABMON_10: tm.tm_mon++;
3092 case ABMON_9: tm.tm_mon++;
3093 case ABMON_8: tm.tm_mon++;
3094 case ABMON_7: tm.tm_mon++;
3095 case ABMON_6: tm.tm_mon++;
3096 case ABMON_5: tm.tm_mon++;
3097 case ABMON_4: tm.tm_mon++;
3098 case ABMON_3: tm.tm_mon++;
3099 case ABMON_2: tm.tm_mon++;
3104 case MON_12: tm.tm_mon++;
3105 case MON_11: tm.tm_mon++;
3106 case MON_10: tm.tm_mon++;
3107 case MON_9: tm.tm_mon++;
3108 case MON_8: tm.tm_mon++;
3109 case MON_7: tm.tm_mon++;
3110 case MON_6: tm.tm_mon++;
3111 case MON_5: tm.tm_mon++;
3112 case MON_4: tm.tm_mon++;
3113 case MON_3: tm.tm_mon++;
3114 case MON_2: tm.tm_mon++;
3121 return_format = TRUE;
3126 return_format = TRUE;
3131 return_format = TRUE;
3136 return_format = TRUE;
3141 format = "%Ow"; /* Find the alternate digit for 0 */
3145 /* We can't use my_strftime() because it doesn't look at
3147 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3150 /* A zero return means one of:
3151 * a) there wasn't enough space in PL_langinfo_buf
3152 * b) the format, like a plain %p, returns empty
3153 * c) it was an illegal format, though some
3154 * implementations of strftime will just return the
3155 * illegal format as a plain character sequence.
3157 * To quickly test for case 'b)', try again but precede
3158 * the format with a plain character. If that result is
3159 * still empty, the problem is either 'a)' or 'c)' */
3161 Size_t format_size = strlen(format) + 1;
3162 Size_t mod_size = format_size + 1;
3166 Newx(mod_format, mod_size, char);
3167 Newx(temp_result, PL_langinfo_bufsize, char);
3169 my_strlcpy(mod_format + 1, format, mod_size);
3170 len = strftime(temp_result,
3171 PL_langinfo_bufsize,
3173 Safefree(mod_format);
3174 Safefree(temp_result);
3176 /* If 'len' is non-zero, it means that we had a case like
3177 * %p which means the current locale doesn't use a.m. or
3178 * p.m., and that is valid */
3181 /* Here, still didn't work. If we get well beyond a
3182 * reasonable size, bail out to prevent an infinite
3185 if (PL_langinfo_bufsize > 100 * format_size) {
3186 *PL_langinfo_buf = '\0';
3189 /* Double the buffer size to retry; Add 1 in case
3190 * original was 0, so we aren't stuck at 0. */
3191 PL_langinfo_bufsize *= 2;
3192 PL_langinfo_bufsize++;
3193 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3201 /* Here, we got a result.
3203 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3204 * alternate format for wday 0. If the value is the same as
3205 * the normal 0, there isn't an alternate, so clear the buffer.
3207 if ( item == ALT_DIGITS
3208 && strEQ(PL_langinfo_buf, "0"))
3210 *PL_langinfo_buf = '\0';
3213 /* ALT_DIGITS is problematic. Experiments on it showed that
3214 * strftime() did not always work properly when going from
3215 * alt-9 to alt-10. Only a few locales have this item defined,
3216 * and in all of them on Linux that khw was able to find,
3217 * nl_langinfo() merely returned the alt-0 character, possibly
3218 * doubled. Most Unicode digits are in blocks of 10
3219 * consecutive code points, so that is sufficient information
3220 * for those scripts, as we can infer alt-1, alt-2, .... But
3221 * for a Japanese locale, a CJK ideographic 0 is returned, and
3222 * the CJK digits are not in code point order, so you can't
3223 * really infer anything. The localedef for this locale did
3224 * specify the succeeding digits, so that strftime() works
3225 * properly on them, without needing to infer anything. But
3226 * the nl_langinfo() return did not give sufficient information
3227 * for the caller to understand what's going on. So until
3228 * there is evidence that it should work differently, this
3229 * returns the alt-0 string for ALT_DIGITS.
3231 * wday was chosen because its range is all a single digit.
3232 * Things like tm_sec have two digits as the minimum: '00' */
3236 retval = PL_langinfo_buf;
3238 /* If to return the format, not the value, overwrite the buffer
3239 * with it. But some strftime()s will keep the original format
3240 * if illegal, so change those to "" */
3241 if (return_format) {
3242 if (strEQ(PL_langinfo_buf, format)) {
3243 *PL_langinfo_buf = '\0';
3246 retval = save_to_buffer(format, &PL_langinfo_buf,
3247 &PL_langinfo_bufsize, 0);
3265 * Initialize locale awareness.
3268 Perl_init_i18nl10n(pTHX_ int printwarn)
3272 * 0 if not to output warning when setup locale is bad
3273 * 1 if to output warning based on value of PERL_BADLANG
3274 * >1 if to output regardless of PERL_BADLANG
3277 * 1 = set ok or not applicable,
3278 * 0 = fallback to a locale of lower priority
3279 * -1 = fallback to all locales failed, not even to the C locale
3281 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3282 * set, debugging information is output.
3284 * This looks more complicated than it is, mainly due to the #ifdefs.
3286 * We try to set LC_ALL to the value determined by the environment. If
3287 * there is no LC_ALL on this platform, we try the individual categories we
3288 * know about. If this works, we are done.
3290 * But if it doesn't work, we have to do something else. We search the
3291 * environment variables ourselves instead of relying on the system to do
3292 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3293 * think there is one), and the ultimate fallback "C". This is all done in
3294 * the same loop as above to avoid duplicating code, but it makes things
3295 * more complex. The 'trial_locales' array is initialized with just one
3296 * element; it causes the behavior described in the paragraph above this to
3297 * happen. If that fails, we add elements to 'trial_locales', and do extra
3298 * loop iterations to cause the behavior described in this paragraph.
3300 * On Ultrix, the locale MUST come from the environment, so there is
3301 * preliminary code to set it. I (khw) am not sure that it is necessary,
3302 * and that this couldn't be folded into the loop, but barring any real
3303 * platforms to test on, it's staying as-is
3305 * A slight complication is that in embedded Perls, the locale may already
3306 * be set-up, and we don't want to get it from the normal environment
3307 * variables. This is handled by having a special environment variable
3308 * indicate we're in this situation. We simply set setlocale's 2nd
3309 * parameter to be a NULL instead of "". That indicates to setlocale that
3310 * it is not to change anything, but to return the current value,
3311 * effectively initializing perl's db to what the locale already is.
3313 * We play the same trick with NULL if a LC_ALL succeeds. We call
3314 * setlocale() on the individual categores with NULL to get their existing
3315 * values for our db, instead of trying to change them.
3323 PERL_UNUSED_ARG(printwarn);
3325 #else /* USE_LOCALE */
3328 const char * const language = PerlEnv_getenv("LANGUAGE");
3332 /* NULL uses the existing already set up locale */
3333 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3336 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3337 unsigned int trial_locales_count;
3338 const char * const lc_all = PerlEnv_getenv("LC_ALL");
3339 const char * const lang = PerlEnv_getenv("LANG");
3340 bool setlocale_failure = FALSE;
3343 /* A later getenv() could zap this, so only use here */
3344 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3346 const bool locwarn = (printwarn > 1
3348 && ( ! bad_lang_use_once
3350 /* disallow with "" or "0" */
3352 && strNE("0", bad_lang_use_once)))));
3354 /* setlocale() return vals; not copied so must be looked at immediately */
3355 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3357 /* current locale for given category; should have been copied so aren't
3359 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3363 /* In some systems you can find out the system default locale
3364 * and use that as the fallback locale. */
3365 # define SYSTEM_DEFAULT_LOCALE
3367 # ifdef SYSTEM_DEFAULT_LOCALE
3369 const char *system_default_locale = NULL;
3374 # define DEBUG_LOCALE_INIT(a,b,c)
3377 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3379 # define DEBUG_LOCALE_INIT(category, locale, result) \
3381 if (debug_initialization) { \
3382 PerlIO_printf(Perl_debug_log, \
3384 __FILE__, __LINE__, \
3385 setlocale_debug_string(category, \
3391 /* Make sure the parallel arrays are properly set up */
3392 # ifdef USE_LOCALE_NUMERIC
3393 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3394 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3395 # ifdef USE_POSIX_2008_LOCALE
3396 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3399 # ifdef USE_LOCALE_CTYPE
3400 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3401 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3402 # ifdef USE_POSIX_2008_LOCALE
3403 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3406 # ifdef USE_LOCALE_COLLATE
3407 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3408 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3409 # ifdef USE_POSIX_2008_LOCALE
3410 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3413 # ifdef USE_LOCALE_TIME
3414 assert(categories[LC_TIME_INDEX] == LC_TIME);
3415 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3416 # ifdef USE_POSIX_2008_LOCALE
3417 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3420 # ifdef USE_LOCALE_MESSAGES
3421 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3422 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3423 # ifdef USE_POSIX_2008_LOCALE
3424 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3427 # ifdef USE_LOCALE_MONETARY
3428 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3429 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3430 # ifdef USE_POSIX_2008_LOCALE
3431 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3434 # ifdef USE_LOCALE_ADDRESS
3435 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3436 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3437 # ifdef USE_POSIX_2008_LOCALE
3438 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3441 # ifdef USE_LOCALE_IDENTIFICATION
3442 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3443 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3444 # ifdef USE_POSIX_2008_LOCALE
3445 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3448 # ifdef USE_LOCALE_MEASUREMENT
3449 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3450 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3451 # ifdef USE_POSIX_2008_LOCALE
3452 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3455 # ifdef USE_LOCALE_PAPER
3456 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3457 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3458 # ifdef USE_POSIX_2008_LOCALE
3459 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3462 # ifdef USE_LOCALE_TELEPHONE
3463 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3464 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3465 # ifdef USE_POSIX_2008_LOCALE
3466 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3469 # ifdef USE_LOCALE_SYNTAX
3470 assert(categories[LC_SYNTAX_INDEX] == LC_SYNTAX);
3471 assert(strEQ(category_names[LC_SYNTAX_INDEX], "LC_SYNTAX"));
3472 # ifdef USE_POSIX_2008_LOCALE
3473 assert(category_masks[LC_SYNTAX_INDEX] == LC_SYNTAX_MASK);
3476 # ifdef USE_LOCALE_TOD
3477 assert(categories[LC_TOD_INDEX] == LC_TOD);
3478 assert(strEQ(category_names[LC_TOD_INDEX], "LC_TOD"));
3479 # ifdef USE_POSIX_2008_LOCALE
3480 assert(category_masks[LC_TOD_INDEX] == LC_TOD_MASK);
3484 assert(categories[LC_ALL_INDEX] == LC_ALL);
3485 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3486 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3487 # ifdef USE_POSIX_2008_LOCALE
3488 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3491 # endif /* DEBUGGING */
3493 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
3494 * why these particular incantations are used. */
3496 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
3499 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
3502 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
3505 /* Initialize the cache of the program's UTF-8ness for the always known
3506 * locales C and POSIX */
3507 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3508 sizeof(PL_locale_utf8ness));
3510 # ifdef USE_THREAD_SAFE_LOCALE
3513 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3517 # ifdef USE_POSIX_2008_LOCALE
3519 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3520 if (! PL_C_locale_obj) {
3521 Perl_croak_nocontext(
3522 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3524 if (DEBUG_Lv_TEST || debug_initialization) {
3525 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3530 # ifdef USE_LOCALE_NUMERIC
3532 PL_numeric_radix_sv = newSVpvs(".");
3536 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3538 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3539 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3542 # ifdef LOCALE_ENVIRON_REQUIRED
3545 * Ultrix setlocale(..., "") fails if there are no environment
3546 * variables from which to get a locale name.
3550 # error Ultrix without LC_ALL not implemented
3556 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3557 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3558 if (sl_result[LC_ALL_INDEX])
3561 setlocale_failure = TRUE;
3563 if (! setlocale_failure) {
3564 const char * locale_param;
3565 for (i = 0; i < LC_ALL_INDEX; i++) {
3566 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3569 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3570 if (! sl_result[i]) {
3571 setlocale_failure = TRUE;
3573 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3578 # endif /* LC_ALL */
3579 # endif /* LOCALE_ENVIRON_REQUIRED */
3581 /* We try each locale in the list until we get one that works, or exhaust
3582 * the list. Normally the loop is executed just once. But if setting the
3583 * locale fails, inside the loop we add fallback trials to the array and so
3584 * will execute the loop multiple times */
3585 trial_locales[0] = setlocale_init;
3586 trial_locales_count = 1;
3588 for (i= 0; i < trial_locales_count; i++) {
3589 const char * trial_locale = trial_locales[i];
3593 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3594 * when i==0, but I (khw) don't think that behavior makes much
3596 setlocale_failure = FALSE;
3598 # ifdef SYSTEM_DEFAULT_LOCALE
3599 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3601 /* On Windows machines, an entry of "" after the 0th means to use
3602 * the system default locale, which we now proceed to get. */
3603 if (strEQ(trial_locale, "")) {
3606 /* Note that this may change the locale, but we are going to do
3607 * that anyway just below */
3608 system_default_locale = do_setlocale_c(LC_ALL, "");
3609 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3611 /* Skip if invalid or if it's already on the list of locales to
3613 if (! system_default_locale) {
3614 goto next_iteration;
3616 for (j = 0; j < trial_locales_count; j++) {
3617 if (strEQ(system_default_locale, trial_locales[j])) {
3618 goto next_iteration;
3622 trial_locale = system_default_locale;
3625 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3627 # endif /* SYSTEM_DEFAULT_LOCALE */
3633 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3634 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3635 if (! sl_result[LC_ALL_INDEX]) {
3636 setlocale_failure = TRUE;
3639 /* Since LC_ALL succeeded, it should have changed all the other
3640 * categories it can to its value; so we massage things so that the
3641 * setlocales below just return their category's current values.
3642 * This adequately handles the case in NetBSD where LC_COLLATE may
3643 * not be defined for a locale, and setting it individually will
3644 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3645 * the POSIX locale. */
3646 trial_locale = NULL;
3649 # endif /* LC_ALL */
3651 if (! setlocale_failure) {
3653 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3655 = savepv(do_setlocale_r(categories[j], trial_locale));
3656 if (! curlocales[j]) {
3657 setlocale_failure = TRUE;
3659 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3662 if (! setlocale_failure) { /* All succeeded */
3663 break; /* Exit trial_locales loop */
3667 /* Here, something failed; will need to try a fallback. */
3673 if (locwarn) { /* Output failure info only on the first one */
3677 PerlIO_printf(Perl_error_log,
3678 "perl: warning: Setting locale failed.\n");
3680 # else /* !LC_ALL */
3682 PerlIO_printf(Perl_error_log,
3683 "perl: warning: Setting locale failed for the categories:\n\t");
3685 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3686 if (! curlocales[j]) {
3687 PerlIO_printf(Perl_error_log, category_names[j]);
3690 Safefree(curlocales[j]);
3694 # endif /* LC_ALL */
3696 PerlIO_printf(Perl_error_log,
3697 "perl: warning: Please check that your locale settings:\n");
3701 PerlIO_printf(Perl_error_log,
3702 "\tLANGUAGE = %c%s%c,\n",
3703 language ? '"' : '(',
3704 language ? language : "unset",
3705 language ? '"' : ')');
3708 PerlIO_printf(Perl_error_log,
3709 "\tLC_ALL = %c%s%c,\n",
3711 lc_all ? lc_all : "unset",
3712 lc_all ? '"' : ')');
3714 # if defined(USE_ENVIRON_ARRAY)
3719 /* Look through the environment for any variables of the
3720 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3721 * already handled above. These are assumed to be locale
3722 * settings. Output them and their values. */
3723 for (e = environ; *e; e++) {
3724 const STRLEN prefix_len = sizeof("LC_") - 1;
3727 if ( strBEGINs(*e, "LC_")
3728 && ! strBEGINs(*e, "LC_ALL=")
3729 && (uppers_len = strspn(*e + prefix_len,
3730 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3731 && ((*e)[prefix_len + uppers_len] == '='))
3733 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3734 (int) (prefix_len + uppers_len), *e,
3735 *e + prefix_len + uppers_len + 1);
3742 PerlIO_printf(Perl_error_log,
3743 "\t(possibly more locale environment variables)\n");
3747 PerlIO_printf(Perl_error_log,
3748 "\tLANG = %c%s%c\n",
3750 lang ? lang : "unset",
3753 PerlIO_printf(Perl_error_log,
3754 " are supported and installed on your system.\n");
3757 /* Calculate what fallback locales to try. We have avoided this
3758 * until we have to, because failure is quite unlikely. This will
3759 * usually change the upper bound of the loop we are in.
3761 * Since the system's default way of setting the locale has not
3762 * found one that works, We use Perl's defined ordering: LC_ALL,
3763 * LANG, and the C locale. We don't try the same locale twice, so
3764 * don't add to the list if already there. (On POSIX systems, the
3765 * LC_ALL element will likely be a repeat of the 0th element "",
3766 * but there's no harm done by doing it explicitly.
3768 * Note that this tries the LC_ALL environment variable even on
3769 * systems which have no LC_ALL locale setting. This may or may
3770 * not have been originally intentional, but there's no real need
3771 * to change the behavior. */
3773 for (j = 0; j < trial_locales_count; j++) {
3774 if (strEQ(lc_all, trial_locales[j])) {
3778 trial_locales[trial_locales_count++] = lc_all;
3783 for (j = 0; j < trial_locales_count; j++) {
3784 if (strEQ(lang, trial_locales[j])) {
3788 trial_locales[trial_locales_count++] = lang;
3792 # if defined(WIN32) && defined(LC_ALL)
3794 /* For Windows, we also try the system default locale before "C".
3795 * (If there exists a Windows without LC_ALL we skip this because
3796 * it gets too complicated. For those, the "C" is the next
3797 * fallback possibility). The "" is the same as the 0th element of
3798 * the array, but the code at the loop above knows to treat it
3799 * differently when not the 0th */
3800 trial_locales[trial_locales_count++] = "";
3804 for (j = 0; j < trial_locales_count; j++) {
3805 if (strEQ("C", trial_locales[j])) {
3809 trial_locales[trial_locales_count++] = "C";
3812 } /* end of first time through the loop */
3820 } /* end of looping through the trial locales */
3822 if (ok < 1) { /* If we tried to fallback */
3824 if (! setlocale_failure) { /* fallback succeeded */
3825 msg = "Falling back to";
3827 else { /* fallback failed */
3830 /* We dropped off the end of the loop, so have to decrement i to
3831 * get back to the value the last time through */
3835 msg = "Failed to fall back to";
3837 /* To continue, we should use whatever values we've got */
3839 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3840 Safefree(curlocales[j]);
3841 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3842 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3847 const char * description;
3848 const char * name = "";
3849 if (strEQ(trial_locales[i], "C")) {
3850 description = "the standard locale";
3854 # ifdef SYSTEM_DEFAULT_LOCALE
3856 else if (strEQ(trial_locales[i], "")) {
3857 description = "the system default locale";
3858 if (system_default_locale) {
3859 name = system_default_locale;
3863 # endif /* SYSTEM_DEFAULT_LOCALE */
3866 description = "a fallback locale";
3867 name = trial_locales[i];
3869 if (name && strNE(name, "")) {
3870 PerlIO_printf(Perl_error_log,
3871 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3874 PerlIO_printf(Perl_error_log,
3875 "perl: warning: %s %s.\n", msg, description);
3878 } /* End of tried to fallback */
3880 /* Done with finding the locales; update our records */
3882 # ifdef USE_LOCALE_CTYPE
3884 new_ctype(curlocales[LC_CTYPE_INDEX]);
3887 # ifdef USE_LOCALE_COLLATE
3889 new_collate(curlocales[LC_COLLATE_INDEX]);
3892 # ifdef USE_LOCALE_NUMERIC
3894 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3898 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3900 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3902 /* This caches whether each category's locale is UTF-8 or not. This
3903 * may involve changing the locale. It is ok to do this at
3904 * initialization time before any threads have started, but not later
3905 * unless thread-safe operations are used.
3906 * Caching means that if the program heeds our dictate not to change
3907 * locales in threaded applications, this data will remain valid, and
3908 * it may get queried without having to change locales. If the
3909 * environment is such that all categories have the same locale, this
3910 * isn't needed, as the code will not change the locale; but this
3911 * handles the uncommon case where the environment has disparate
3912 * locales for the categories */
3913 (void) _is_cur_LC_category_utf8(categories[i]);
3917 Safefree(curlocales[i]);
3920 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3922 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3923 * locale is UTF-8. The call to new_ctype() just above has already
3924 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3925 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3926 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3927 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3928 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3930 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3931 This is an alternative to using the -C command line switch
3932 (the -C if present will override this). */
3934 const char *p = PerlEnv_getenv("PERL_UNICODE");
3935 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3936 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3941 #endif /* USE_LOCALE */
3944 /* So won't continue to output stuff */
3945 DEBUG_INITIALIZATION_set(FALSE);
3952 #ifdef USE_LOCALE_COLLATE
3955 Perl__mem_collxfrm(pTHX_ const char *input_string,
3956 STRLEN len, /* Length of 'input_string' */
3957 STRLEN *xlen, /* Set to length of returned string
3958 (not including the collation index
3960 bool utf8 /* Is the input in UTF-8? */
3964 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3965 * differences. First, it handles embedded NULs. Second, it allocates a bit
3966 * more memory than needed for the transformed data itself. The real
3967 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3968 * the length of that, and doesn't include the collation index size.
3969 * Please see sv_collxfrm() to see how this is used. */
3971 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3973 char * s = (char *) input_string;
3974 STRLEN s_strlen = strlen(input_string);
3976 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3977 STRLEN length_in_chars;
3978 bool first_time = TRUE; /* Cleared after first loop iteration */
3980 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3982 /* Must be NUL-terminated */
3983 assert(*(input_string + len) == '\0');
3985 /* If this locale has defective collation, skip */
3986 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3987 DEBUG_L(PerlIO_printf(Perl_debug_log,
3988 "_mem_collxfrm: locale's collation is defective\n"));
3992 /* Replace any embedded NULs with the control that sorts before any others.
3993 * This will give as good as possible results on strings that don't
3994 * otherwise contain that character, but otherwise there may be
3995 * less-than-perfect results with that character and NUL. This is
3996 * unavoidable unless we replace strxfrm with our own implementation. */
3997 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
4001 STRLEN sans_nuls_len;
4002 int try_non_controls;
4003 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
4004 making sure 2nd byte is NUL.
4006 STRLEN this_replacement_len;
4008 /* If we don't know what non-NUL control character sorts lowest for
4009 * this locale, find it */
4010 if (PL_strxfrm_NUL_replacement == '\0') {
4012 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
4013 includes the collation index
4016 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
4018 /* Unlikely, but it may be that no control will work to replace
4019 * NUL, in which case we instead look for any character. Controls
4020 * are preferred because collation order is, in general, context
4021 * sensitive, with adjoining characters affecting the order, and
4022 * controls are less likely to have such interactions, allowing the
4023 * NUL-replacement to stand on its own. (Another way to look at it
4024 * is to imagine what would happen if the NUL were replaced by a
4025 * combining character; it wouldn't work out all that well.) */
4026 for (try_non_controls = 0;
4027 try_non_controls < 2;
4030 /* Look through all legal code points (NUL isn't) */
4031 for (j = 1; j < 256; j++) {
4032 char * x; /* j's xfrm plus collation index */
4033 STRLEN x_len; /* length of 'x' */
4034 STRLEN trial_len = 1;
4035 char cur_source[] = { '\0', '\0' };
4037 /* Skip non-controls the first time through the loop. The
4038 * controls in a UTF-8 locale are the L1 ones */
4039 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
4046 /* Create a 1-char string of the current code point */
4047 cur_source[0] = (char) j;
4049 /* Then transform it */
4050 x = _mem_collxfrm(cur_source, trial_len, &x_len,
4051 0 /* The string is not in UTF-8 */);
4053 /* Ignore any character that didn't successfully transform.
4059 /* If this character's transformation is lower than
4060 * the current lowest, this one becomes the lowest */
4061 if ( cur_min_x == NULL
4062 || strLT(x + COLLXFRM_HDR_LEN,
4063 cur_min_x + COLLXFRM_HDR_LEN))
4065 PL_strxfrm_NUL_replacement = j;
4066 Safefree(cur_min_x);
4072 } /* end of loop through all 255 characters */
4074 /* Stop looking if found */
4079 /* Unlikely, but possible, if there aren't any controls that
4080 * work in the locale, repeat the loop, looking for any
4081 * character that works */
4082 DEBUG_L(PerlIO_printf(Perl_debug_log,
4083 "_mem_collxfrm: No control worked. Trying non-controls\n"));
4084 } /* End of loop to try first the controls, then any char */
4087 DEBUG_L(PerlIO_printf(Perl_debug_log,
4088 "_mem_collxfrm: Couldn't find any character to replace"
4089 " embedded NULs in locale %s with", PL_collation_name));
4093 DEBUG_L(PerlIO_printf(Perl_debug_log,
4094 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
4095 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
4097 Safefree(cur_min_x);
4098 } /* End of determining the character that is to replace NULs */
4100 /* If the replacement is variant under UTF-8, it must match the
4101 * UTF8-ness of the original */
4102 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
4103 this_replacement_char[0] =
4104 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
4105 this_replacement_char[1] =
4106 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
4107 this_replacement_len = 2;
4110 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
4111 /* this_replacement_char[1] = '\0' was done at initialization */
4112 this_replacement_len = 1;
4115 /* The worst case length for the replaced string would be if every
4116 * character in it is NUL. Multiply that by the length of each
4117 * replacement, and allow for a trailing NUL */
4118 sans_nuls_len = (len * this_replacement_len) + 1;
4119 Newx(sans_nuls, sans_nuls_len, char);
4122 /* Replace each NUL with the lowest collating control. Loop until have
4123 * exhausted all the NULs */
4124 while (s + s_strlen < e) {
4125 my_strlcat(sans_nuls, s, sans_nuls_len);
4127 /* Do the actual replacement */
4128 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
4130 /* Move past the input NUL */
4132 s_strlen = strlen(s);
4135 /* And add anything that trails the final NUL */
4136 my_strlcat(sans_nuls, s, sans_nuls_len);
4138 /* Switch so below we transform this modified string */
4141 } /* End of replacing NULs */
4143 /* Make sure the UTF8ness of the string and locale match */
4144 if (utf8 != PL_in_utf8_COLLATE_locale) {
4145 /* XXX convert above Unicode to 10FFFF? */
4146 const char * const t = s; /* Temporary so we can later find where the
4149 /* Here they don't match. Change the string's to be what the locale is
4152 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
4153 s = (char *) bytes_to_utf8((const U8 *) s, &len);
4156 else { /* locale is not UTF-8; but input is; downgrade the input */
4158 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
4160 /* If the downgrade was successful we are done, but if the input
4161 * contains things that require UTF-8 to represent, have to do
4162 * damage control ... */
4163 if (UNLIKELY(utf8)) {
4165 /* What we do is construct a non-UTF-8 string with
4166 * 1) the characters representable by a single byte converted
4167 * to be so (if necessary);
4168 * 2) and the rest converted to collate the same as the
4169 * highest collating representable character. That makes
4170 * them collate at the end. This is similar to how we
4171 * handle embedded NULs, but we use the highest collating
4172 * code point instead of the smallest. Like the NUL case,
4173 * this isn't perfect, but is the best we can reasonably
4174 * do. Every above-255 code point will sort the same as
4175 * the highest-sorting 0-255 code point. If that code
4176 * point can combine in a sequence with some other code
4177 * points for weight calculations, us changing something to
4178 * be it can adversely affect the results. But in most
4179 * cases, it should work reasonably. And note that this is
4180 * really an illegal situation: using code points above 255
4181 * on a locale where only 0-255 are valid. If two strings
4182 * sort entirely equal, then the sort order for the
4183 * above-255 code points will be in code point order. */
4187 /* If we haven't calculated the code point with the maximum
4188 * collating order for this locale, do so now */
4189 if (! PL_strxfrm_max_cp) {
4192 /* The current transformed string that collates the
4193 * highest (except it also includes the prefixed collation
4195 char * cur_max_x = NULL;
4197 /* Look through all legal code points (NUL isn't) */
4198 for (j = 1; j < 256; j++) {
4201 char cur_source[] = { '\0', '\0' };
4203 /* Create a 1-char string of the current code point */
4204 cur_source[0] = (char) j;
4206 /* Then transform it */
4207 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4209 /* If something went wrong (which it shouldn't), just
4210 * ignore this code point */
4215 /* If this character's transformation is higher than
4216 * the current highest, this one becomes the highest */
4217 if ( cur_max_x == NULL
4218 || strGT(x + COLLXFRM_HDR_LEN,
4219 cur_max_x + COLLXFRM_HDR_LEN))
4221 PL_strxfrm_max_cp = j;
4222 Safefree(cur_max_x);
4231 DEBUG_L(PerlIO_printf(Perl_debug_log,
4232 "_mem_collxfrm: Couldn't find any character to"
4233 " replace above-Latin1 chars in locale %s with",
4234 PL_collation_name));
4238 DEBUG_L(PerlIO_printf(Perl_debug_log,
4239 "_mem_collxfrm: highest 1-byte collating character"
4240 " in locale %s is 0x%02X\n",
4242 PL_strxfrm_max_cp));
4244 Safefree(cur_max_x);
4247 /* Here we know which legal code point collates the highest.
4248 * We are ready to construct the non-UTF-8 string. The length
4249 * will be at least 1 byte smaller than the input string
4250 * (because we changed at least one 2-byte character into a
4251 * single byte), but that is eaten up by the trailing NUL */
4257 char * e = (char *) t + len;
4259 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4261 if (UTF8_IS_INVARIANT(cur_char)) {
4264 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4265 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4267 else { /* Replace illegal cp with highest collating
4269 s[d++] = PL_strxfrm_max_cp;
4273 Renew(s, d, char); /* Free up unused space */
4278 /* Here, we have constructed a modified version of the input. It could
4279 * be that we already had a modified copy before we did this version.
4280 * If so, that copy is no longer needed */
4281 if (t != input_string) {
4286 length_in_chars = (utf8)
4287 ? utf8_length((U8 *) s, (U8 *) s + len)
4290 /* The first element in the output is the collation id, used by
4291 * sv_collxfrm(); then comes the space for the transformed string. The
4292 * equation should give us a good estimate as to how much is needed */
4293 xAlloc = COLLXFRM_HDR_LEN
4295 + (PL_collxfrm_mult * length_in_chars);
4296 Newx(xbuf, xAlloc, char);
4297 if (UNLIKELY(! xbuf)) {
4298 DEBUG_L(PerlIO_printf(Perl_debug_log,
4299 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4303 /* Store the collation id */
4304 *(U32*)xbuf = PL_collation_ix;
4306 /* Then the transformation of the input. We loop until successful, or we
4310 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4312 /* If the transformed string occupies less space than we told strxfrm()
4313 * was available, it means it successfully transformed the whole
4315 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4317 /* Some systems include a trailing NUL in the returned length.
4318 * Ignore it, using a loop in case multiple trailing NULs are
4321 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4326 /* If the first try didn't get it, it means our prediction was low.
4327 * Modify the coefficients so that we predict a larger value in any
4328 * future transformations */
4330 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4331 STRLEN computed_guess = PL_collxfrm_base
4332 + (PL_collxfrm_mult * length_in_chars);
4334 /* On zero-length input, just keep current slope instead of
4336 const STRLEN new_m = (length_in_chars != 0)
4337 ? needed / length_in_chars
4340 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4341 "%s: %d: initial size of %zu bytes for a length "
4342 "%zu string was insufficient, %zu needed\n",
4344 computed_guess, length_in_chars, needed));
4346 /* If slope increased, use it, but discard this result for
4347 * length 1 strings, as we can't be sure that it's a real slope
4349 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4353 STRLEN old_m = PL_collxfrm_mult;
4354 STRLEN old_b = PL_collxfrm_base;
4358 PL_collxfrm_mult = new_m;
4359 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4360 computed_guess = PL_collxfrm_base
4361 + (PL_collxfrm_mult * length_in_chars);
4362 if (computed_guess < needed) {
4363 PL_collxfrm_base += needed - computed_guess;
4366 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4367 "%s: %d: slope is now %zu; was %zu, base "
4368 "is now %zu; was %zu\n",
4370 PL_collxfrm_mult, old_m,
4371 PL_collxfrm_base, old_b));
4373 else { /* Slope didn't change, but 'b' did */
4374 const STRLEN new_b = needed
4377 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4378 "%s: %d: base is now %zu; was %zu\n",
4380 new_b, PL_collxfrm_base));
4381 PL_collxfrm_base = new_b;
4388 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4389 DEBUG_L(PerlIO_printf(Perl_debug_log,
4390 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4391 *xlen, PERL_INT_MAX));
4395 /* A well-behaved strxfrm() returns exactly how much space it needs
4396 * (usually not including the trailing NUL) when it fails due to not
4397 * enough space being provided. Assume that this is the case unless
4398 * it's been proven otherwise */
4399 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4400 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4402 else { /* Here, either:
4403 * 1) The strxfrm() has previously shown bad behavior; or
4404 * 2) It isn't the first time through the loop, which means
4405 * that the strxfrm() is now showing bad behavior, because
4406 * we gave it what it said was needed in the previous
4407 * iteration, and it came back saying it needed still more.
4408 * (Many versions of cygwin fit this. When the buffer size
4409 * isn't sufficient, they return the input size instead of
4410 * how much is needed.)
4411 * Increase the buffer size by a fixed percentage and try again.
4413 xAlloc += (xAlloc / 4) + 1;
4414 PL_strxfrm_is_behaved = FALSE;
4418 if (DEBUG_Lv_TEST || debug_initialization) {
4419 PerlIO_printf(Perl_debug_log,
4420 "_mem_collxfrm required more space than previously calculated"
4421 " for locale %s, trying again with new guess=%d+%zu\n",
4422 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4423 xAlloc - COLLXFRM_HDR_LEN);
4430 Renew(xbuf, xAlloc, char);
4431 if (UNLIKELY(! xbuf)) {
4432 DEBUG_L(PerlIO_printf(Perl_debug_log,
4433 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4443 if (DEBUG_Lv_TEST || debug_initialization) {
4445 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4446 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4447 PerlIO_printf(Perl_debug_log, "%s\n",
4448 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4454 /* Free up unneeded space; retain ehough for trailing NUL */
4455 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4457 if (s != input_string) {
4467 if (DEBUG_Lv_TEST || debug_initialization) {
4468 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4474 if (s != input_string) {
4485 S_print_collxfrm_input_and_return(pTHX_
4486 const char * const s,
4487 const char * const e,
4488 const STRLEN * const xlen,
4492 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4494 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4495 (UV)PL_collation_ix);
4497 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4500 PerlIO_printf(Perl_debug_log, "NULL");
4502 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4504 print_bytes_for_locale(s, e, is_utf8);
4506 PerlIO_printf(Perl_debug_log, "'\n");
4509 # endif /* DEBUGGING */
4510 #endif /* USE_LOCALE_COLLATE */
4515 S_print_bytes_for_locale(pTHX_
4516 const char * const s,
4517 const char * const e,
4521 bool prev_was_printable = TRUE;
4522 bool first_time = TRUE;
4524 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4528 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4531 if (! prev_was_printable) {
4532 PerlIO_printf(Perl_debug_log, " ");
4534 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4535 prev_was_printable = TRUE;
4539 PerlIO_printf(Perl_debug_log, " ");
4541 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4542 prev_was_printable = FALSE;
4544 t += (is_utf8) ? UTF8SKIP(t) : 1;
4549 # endif /* #ifdef DEBUGGING */
4552 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4554 /* Changes the locale for LC_'switch_category" to that of
4555 * LC_'template_category', if they aren't already the same. If not NULL,
4556 * 'template_locale' is the locale that 'template_category' is in.
4558 * Returns a copy of the name of the original locale for 'switch_category'
4559 * so can be switched back to with the companion function
4560 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4562 char * restore_to_locale = NULL;
4564 if (switch_category == template_category) { /* No changes needed */
4568 /* Find the original locale of the category we may need to change, so that
4569 * it can be restored to later */
4570 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4572 if (! restore_to_locale) {
4574 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4575 __FILE__, __LINE__, category_name(switch_category), errno);
4578 /* If the locale of the template category wasn't passed in, find it now */
4579 if (template_locale == NULL) {
4580 template_locale = do_setlocale_r(template_category, NULL);
4581 if (! template_locale) {
4583 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4584 __FILE__, __LINE__, category_name(template_category), errno);
4588 /* It the locales are the same, there's nothing to do */
4589 if (strEQ(restore_to_locale, template_locale)) {
4590 Safefree(restore_to_locale);
4592 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4593 category_name(switch_category), template_locale));
4598 /* Finally, change the locale to the template one */
4599 if (! do_setlocale_r(switch_category, template_locale)) {
4601 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4602 __FILE__, __LINE__, category_name(switch_category),
4603 template_locale, errno);
4606 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4607 category_name(switch_category), template_locale));
4609 return restore_to_locale;
4613 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4615 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4616 * copy that will be freed by this function), or do nothing if the latter
4617 * parameter is NULL */
4619 if (original_locale == NULL) {
4623 if (! do_setlocale_r(category, original_locale)) {
4625 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4627 category_name(category), original_locale, errno);
4630 Safefree(original_locale);
4633 /* is_cur_LC_category_utf8 uses a small char buffer to avoid malloc/free */
4634 #define CUR_LC_BUFFER_SIZE 64
4637 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4639 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4640 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4641 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4642 * could give the wrong result. The result will very likely be correct for
4643 * languages that have commonly used non-ASCII characters, but for notably
4644 * English, it comes down to if the locale's name ends in something like
4645 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4647 * If the platform is early C89, not containing mbtowc(), or we are
4648 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4649 * These work very well for non-Latin locales or those whose currency
4650 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4651 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4652 * on the name containing UTF-8 or not. */
4654 /* Name of current locale corresponding to the input category */
4655 const char *save_input_locale = NULL;
4657 bool is_utf8 = FALSE; /* The return value */
4659 /* The variables below are for the cache of previous lookups using this
4660 * function. The cache is a C string, described at the definition for
4661 * 'C_and_POSIX_utf8ness'.
4663 * The first part of the cache is fixed, for the C and POSIX locales. The
4664 * varying part starts just after them. */
4665 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4667 Size_t utf8ness_cache_size; /* Size of the varying portion */
4668 Size_t input_name_len; /* Length in bytes of save_input_locale */
4669 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4670 the name in the cache */
4671 char * delimited; /* The name plus the delimiters used to store
4673 char buffer[CUR_LC_BUFFER_SIZE]; /* small buffer */
4674 char * name_pos; /* position of 'delimited' in the cache, or 0
4680 assert(category != LC_ALL);
4684 /* Get the desired category's locale */
4685 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4686 if (! save_input_locale) {
4688 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4689 __FILE__, __LINE__, category_name(category), errno);
4692 DEBUG_L(PerlIO_printf(Perl_debug_log,
4693 "Current locale for %s is %s\n",
4694 category_name(category), save_input_locale));
4696 input_name_len = strlen(save_input_locale);
4698 /* In our cache, each name is accompanied by two delimiters and a single
4700 input_name_len_with_overhead = input_name_len + 3;
4702 if ( input_name_len_with_overhead <= CUR_LC_BUFFER_SIZE ) {
4703 /* we can use the buffer, avoid a malloc */
4705 } else { /* need a malloc */
4706 /* Allocate and populate space for a copy of the name surrounded by the
4708 Newx(delimited, input_name_len_with_overhead, char);
4711 delimited[0] = UTF8NESS_SEP[0];
4712 Copy(save_input_locale, delimited + 1, input_name_len, char);
4713 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4714 delimited[input_name_len+2] = '\0';
4716 /* And see if that is in the cache */
4717 name_pos = instr(PL_locale_utf8ness, delimited);
4719 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4723 if (DEBUG_Lv_TEST || debug_initialization) {
4724 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4725 save_input_locale, is_utf8);
4730 /* And, if not already in that position, move it to the beginning of
4731 * the non-constant portion of the list, since it is the most recently
4732 * used. (We don't have to worry about overflow, since just moving
4733 * existing names around) */
4734 if (name_pos > utf8ness_cache) {
4735 Move(utf8ness_cache,
4736 utf8ness_cache + input_name_len_with_overhead,
4737 name_pos - utf8ness_cache, char);
4740 input_name_len_with_overhead - 1, char);
4741 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4744 /* free only when not using the buffer */
4745 if ( delimited != buffer ) Safefree(delimited);
4746 Safefree(save_input_locale);
4750 /* Here we don't have stored the utf8ness for the input locale. We have to
4753 # if defined(USE_LOCALE_CTYPE) \
4754 && ( defined(HAS_NL_LANGINFO) \
4755 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4758 const char *original_ctype_locale
4759 = switch_category_locale_to_template(LC_CTYPE,
4763 /* Here the current LC_CTYPE is set to the locale of the category whose
4764 * information is desired. This means that nl_langinfo() and mbtowc()
4765 * should give the correct results */
4767 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4768 calling the functions if we have this */
4770 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4771 * Unicode code point. */
4773 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4774 __FILE__, __LINE__, (int) MB_CUR_MAX));
4775 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
4777 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4778 goto finish_and_return;
4782 # if defined(HAS_NL_LANGINFO)
4784 { /* The task is easiest if the platform has this POSIX 2001 function.
4785 Except on some platforms it can wrongly return "", so have to have
4786 a fallback. And it can return that it's UTF-8, even if there are
4787 variances from that. For example, Turkish locales may use the
4788 alternate dotted I rules, and sometimes it appears to be a
4789 defective locale definition. XXX We should probably check for
4790 these in the Latin1 range and warn (but on glibc, requires
4791 iswalnum() etc. due to their not handling 80-FF correctly */
4792 const char *codeset = my_nl_langinfo(CODESET, FALSE);
4793 /* FALSE => already in dest locale */
4795 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4796 "\tnllanginfo returned CODESET '%s'\n", codeset));
4798 if (codeset && strNE(codeset, "")) {
4800 /* If the implementation of foldEQ() somehow were
4801 * to change to not go byte-by-byte, this could
4802 * read past end of string, as only one length is
4803 * checked. But currently, a premature NUL will
4804 * compare false, and it will stop there */
4805 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
4806 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
4808 DEBUG_L(PerlIO_printf(Perl_debug_log,
4809 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
4811 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4812 goto finish_and_return;
4817 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
4818 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
4819 * late adder to C89, so very likely to have it. However, testing has
4820 * shown that, like nl_langinfo() above, there are locales that are not
4821 * strictly UTF-8 that this will return that they are */
4828 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4834 /* mbrtowc() and mbtowc() convert a byte string to a wide
4835 * character. Feed a byte string to one of them and check that the
4836 * result is the expected Unicode code point */
4838 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4839 /* Prefer this function if available, as it's reentrant */
4841 memset(&ps, 0, sizeof(ps));;
4842 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
4845 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
4851 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
4853 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
4860 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4861 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
4862 len, (unsigned int) wc, GET_ERRNO));
4864 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
4865 && wc == (wchar_t) UNICODE_REPLACEMENT);
4870 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4871 goto finish_and_return;
4876 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
4877 * try looking at the currency symbol to see if it disambiguates
4878 * things. Often that will be in the native script, and if the symbol
4879 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
4880 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
4881 * string being valid UTF-8 are quite small */
4883 # ifdef USE_LOCALE_MONETARY
4885 /* If have LC_MONETARY, we can look at the currency symbol. Often that
4886 * will be in the native script. We do this one first because there is
4887 * just one string to examine, so potentially avoids work */
4890 const char *original_monetary_locale
4891 = switch_category_locale_to_template(LC_MONETARY,
4894 bool only_ascii = FALSE;
4895 const U8 * currency_string
4896 = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE);
4897 /* 2nd param not relevant for this item */
4898 const U8 * first_variant;
4900 assert( *currency_string == '-'
4901 || *currency_string == '+'
4902 || *currency_string == '.');
4906 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
4908 DEBUG_L(PerlIO_printf(Perl_debug_log, "Couldn't get currency symbol for %s, or contains only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
4912 is_utf8 = is_strict_utf8_string(first_variant, 0);
4915 restore_switched_locale(LC_MONETARY, original_monetary_locale);
4919 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
4920 * otherwise assume the locale is UTF-8 if and only if the symbol
4921 * is non-ascii UTF-8. */
4922 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
4923 save_input_locale, is_utf8));
4924 goto finish_and_return;
4928 # endif /* USE_LOCALE_MONETARY */
4929 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
4931 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
4932 * the names of the months and weekdays, timezone, and am/pm indicator */
4934 const char *original_time_locale
4935 = switch_category_locale_to_template(LC_TIME,
4939 bool is_dst = FALSE;
4943 char * formatted_time;
4945 /* Here the current LC_TIME is set to the locale of the category
4946 * whose information is desired. Look at all the days of the week and
4947 * month names, and the timezone and am/pm indicator for UTF-8 variant
4948 * characters. The first such a one found will tell us if the locale
4949 * is UTF-8 or not */
4951 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
4952 formatted_time = my_strftime("%A %B %Z %p",
4953 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
4954 if ( ! formatted_time
4955 || is_utf8_invariant_string((U8 *) formatted_time, 0))
4958 /* Here, we didn't find a non-ASCII. Try the next time through
4959 * with the complemented dst and am/pm, and try with the next
4960 * weekday. After we have gotten all weekdays, try the next
4963 hour = (hour + 12) % 24;
4971 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
4972 * false otherwise. But first, restore LC_TIME to its original
4973 * locale if we changed it */
4974 restore_switched_locale(LC_TIME, original_time_locale);
4976 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
4978 is_utf8_string((U8 *) formatted_time, 0)));
4979 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
4980 goto finish_and_return;
4983 /* Falling off the end of the loop indicates all the names were just
4984 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
4985 * to its original locale */
4986 restore_switched_locale(LC_TIME, original_time_locale);
4987 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "All time-related words for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
4992 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
4994 /* This code is ifdefd out because it was found to not be necessary in testing
4995 * on our dromedary test machine, which has over 700 locales. There, this
4996 * added no value to looking at the currency symbol and the time strings. I
4997 * left it in so as to avoid rewriting it if real-world experience indicates
4998 * that dromedary is an outlier. Essentially, instead of returning abpve if we
4999 * haven't found illegal utf8, we continue on and examine all the strerror()
5000 * messages on the platform for utf8ness. If all are ASCII, we still don't
5001 * know the answer; but otherwise we have a pretty good indication of the
5002 * utf8ness. The reason this doesn't help much is that the messages may not
5003 * have been translated into the locale. The currency symbol and time strings
5004 * are much more likely to have been translated. */
5007 bool non_ascii = FALSE;
5008 const char *original_messages_locale
5009 = switch_category_locale_to_template(LC_MESSAGES,
5012 const char * errmsg = NULL;
5014 /* Here the current LC_MESSAGES is set to the locale of the category
5015 * whose information is desired. Look through all the messages. We
5016 * can't use Strerror() here because it may expand to code that
5017 * segfaults in miniperl */
5019 for (e = 0; e <= sys_nerr; e++) {
5021 errmsg = sys_errlist[e];
5022 if (errno || !errmsg) {
5025 errmsg = savepv(errmsg);
5026 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5028 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
5034 restore_switched_locale(LC_MESSAGES, original_messages_locale);
5038 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
5039 * any non-ascii means it is one; otherwise we assume it isn't */
5040 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
5043 goto finish_and_return;
5046 DEBUG_L(PerlIO_printf(Perl_debug_log, "All error messages for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
5050 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
5053 /* As a last resort, look at the locale name to see if it matches
5054 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
5055 * return of setlocale(), is actually defined to be opaque, so we can't
5056 * really rely on the absence of various substrings in the name to indicate
5057 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
5058 * be a UTF-8 locale. Similarly for the other common names */
5061 const Size_t final_pos = strlen(save_input_locale) - 1;
5063 if (final_pos >= 3) {
5064 const char *name = save_input_locale;
5066 /* Find next 'U' or 'u' and look from there */
5067 while ((name += strcspn(name, "Uu") + 1)
5068 <= save_input_locale + final_pos - 2)
5070 if ( isALPHA_FOLD_NE(*name, 't')
5071 || isALPHA_FOLD_NE(*(name + 1), 'f'))
5076 if (*(name) == '-') {
5077 if ((name > save_input_locale + final_pos - 1)) {
5082 if (*(name) == '8') {
5083 DEBUG_L(PerlIO_printf(Perl_debug_log,
5084 "Locale %s ends with UTF-8 in name\n",
5085 save_input_locale));
5087 goto finish_and_return;
5090 DEBUG_L(PerlIO_printf(Perl_debug_log,
5091 "Locale %s doesn't end with UTF-8 in name\n",
5092 save_input_locale));
5097 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
5098 if (memENDs(save_input_locale, final_pos, "65001")) {
5099 DEBUG_L(PerlIO_printf(Perl_debug_log,
5100 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
5101 save_input_locale));
5103 goto finish_and_return;
5110 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
5111 * since we are about to return FALSE anyway, there is no point in doing
5112 * this extra work */
5115 if (instr(save_input_locale, "8859")) {
5116 DEBUG_L(PerlIO_printf(Perl_debug_log,
5117 "Locale %s has 8859 in name, not UTF-8 locale\n",
5118 save_input_locale));
5120 goto finish_and_return;
5124 DEBUG_L(PerlIO_printf(Perl_debug_log,
5125 "Assuming locale %s is not a UTF-8 locale\n",
5126 save_input_locale));
5129 # endif /* the code that is compiled when no modern LC_CTYPE */
5133 /* Cache this result so we don't have to go through all this next time. */
5134 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
5135 - (utf8ness_cache - PL_locale_utf8ness);
5137 /* But we can't save it if it is too large for the total space available */
5138 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
5139 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
5141 /* Here it can fit, but we may need to clear out the oldest cached
5142 * result(s) to do so. Check */
5143 if (utf8ness_cache_len + input_name_len_with_overhead
5144 >= utf8ness_cache_size)
5146 /* Here we have to clear something out to make room for this.
5147 * Start looking at the rightmost place where it could fit and find
5148 * the beginning of the entry that extends past that. */
5149 char * cutoff = (char *) my_memrchr(utf8ness_cache,
5152 - input_name_len_with_overhead);
5155 assert(cutoff >= utf8ness_cache);
5157 /* This and all subsequent entries must be removed */
5159 utf8ness_cache_len = strlen(utf8ness_cache);
5162 /* Make space for the new entry */
5163 Move(utf8ness_cache,
5164 utf8ness_cache + input_name_len_with_overhead,
5165 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
5168 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
5169 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
5171 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1] & ~1) != '0') {
5173 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu,"
5174 " inserted_name=%s, its_len=%zu\n",
5176 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
5177 delimited, input_name_len_with_overhead);
5183 if (DEBUG_Lv_TEST) {
5184 const char * s = PL_locale_utf8ness;
5186 /* Audit the structure */
5187 while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) {
5190 if (*s != UTF8NESS_SEP[0]) {
5192 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5193 " separator %.*s<-- HERE %s\n",
5195 (int) (s - PL_locale_utf8ness), PL_locale_utf8ness,
5199 e = strchr(s, UTF8NESS_PREFIX[0]);
5201 e = PL_locale_utf8ness + strlen(PL_locale_utf8ness);
5203 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5204 " separator %.*s<-- HERE %s\n",
5206 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5210 if (*e != '0' && *e != '1') {
5212 "panic: %s: %d: Corrupt utf8ness_cache: utf8ness"
5213 " must be [01] %.*s<-- HERE %s\n",
5215 (int) (e + 1 - PL_locale_utf8ness),
5216 PL_locale_utf8ness, e + 1);
5218 if (ninstr(PL_locale_utf8ness, s, s-1, e)) {
5220 "panic: %s: %d: Corrupt utf8ness_cache: entry"
5221 " has duplicate %.*s<-- HERE %s\n",
5223 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5230 if (DEBUG_Lv_TEST || debug_initialization) {
5232 PerlIO_printf(Perl_debug_log,
5233 "PL_locale_utf8ness is now %s; returning %d\n",
5234 PL_locale_utf8ness, is_utf8);
5239 /* free only when not using the buffer */
5240 if ( delimited != buffer ) Safefree(delimited);
5241 Safefree(save_input_locale);
5248 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
5250 /* Internal function which returns if we are in the scope of a pragma that
5251 * enables the locale category 'category'. 'compiling' should indicate if
5252 * this is during the compilation phase (TRUE) or not (FALSE). */
5254 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
5256 SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0);
5257 if (! these_categories || these_categories == &PL_sv_placeholder) {
5261 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
5262 * a valid unsigned */
5263 assert(category >= -1);
5264 return cBOOL(SvUV(these_categories) & (1U << (category + 1)));
5268 Perl_my_strerror(pTHX_ const int errnum)
5270 /* Returns a mortalized copy of the text of the error message associated
5271 * with 'errnum'. It uses the current locale's text unless the platform
5272 * doesn't have the LC_MESSAGES category or we are not being called from
5273 * within the scope of 'use locale'. In the former case, it uses whatever
5274 * strerror returns; in the latter case it uses the text from the C locale.
5276 * The function just calls strerror(), but temporarily switches, if needed,
5277 * to the C locale */
5281 #ifndef USE_LOCALE_MESSAGES
5283 /* If platform doesn't have messages category, we don't do any switching to
5284 * the C locale; we just use whatever strerror() returns */
5286 errstr = savepv(Strerror(errnum));
5288 #else /* Has locale messages */
5290 const bool within_locale_scope = IN_LC(LC_MESSAGES);
5292 # ifndef USE_ITHREADS
5294 /* This function is trivial without threads. */
5295 if (within_locale_scope) {
5296 errstr = savepv(strerror(errnum));
5299 const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL));
5301 do_setlocale_c(LC_MESSAGES, "C");
5302 errstr = savepv(strerror(errnum));
5303 do_setlocale_c(LC_MESSAGES, save_locale);
5304 Safefree(save_locale);
5307 # elif defined(USE_POSIX_2008_LOCALE) \
5308 && defined(HAS_STRERROR_L) \
5309 && defined(HAS_DUPLOCALE)
5311 /* This function is also trivial if we don't have to worry about thread
5312 * safety and have strerror_l(), as it handles the switch of locales so we
5313 * don't have to deal with that. We don't have to worry about thread
5314 * safety if strerror_r() is also available. Both it and strerror_l() are
5315 * thread-safe. Plain strerror() isn't thread safe. But on threaded
5316 * builds when strerror_r() is available, the apparent call to strerror()
5317 * below is actually a macro that behind-the-scenes calls strerror_r(). */
5319 # ifdef HAS_STRERROR_R
5321 if (within_locale_scope) {
5322 errstr = savepv(strerror(errnum));
5325 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
5330 /* Here we have strerror_l(), but not strerror_r() and we are on a
5331 * threaded-build. We use strerror_l() for everything, constructing a
5332 * locale to pass to it if necessary */
5334 bool do_free = FALSE;
5335 locale_t locale_to_use;
5337 if (within_locale_scope) {
5338 locale_to_use = uselocale((locale_t) 0);
5339 if (locale_to_use == LC_GLOBAL_LOCALE) {
5340 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
5344 else { /* Use C locale if not within 'use locale' scope */
5345 locale_to_use = PL_C_locale_obj;
5348 errstr = savepv(strerror_l(errnum, locale_to_use));
5351 freelocale(locale_to_use);
5355 # else /* Doesn't have strerror_l() */
5357 const char * save_locale = NULL;
5358 bool locale_is_C = FALSE;
5360 /* We have a critical section to prevent another thread from executing this
5361 * same code at the same time. (On thread-safe perls, the LOCK is a
5362 * no-op.) Since this is the only place in core that changes LC_MESSAGES
5363 * (unless the user has called setlocale(), this works to prevent races. */
5366 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5367 "my_strerror called with errnum %d\n", errnum));
5368 if (! within_locale_scope) {
5369 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
5370 if (! save_locale) {
5372 "panic: %s: %d: Could not find current LC_MESSAGES locale,"
5373 " errno=%d\n", __FILE__, __LINE__, errno);
5376 locale_is_C = isNAME_C_OR_POSIX(save_locale);
5378 /* Switch to the C locale if not already in it */
5379 if (! locale_is_C) {
5381 /* The setlocale() just below likely will zap 'save_locale', so
5383 save_locale = savepv(save_locale);
5384 do_setlocale_c(LC_MESSAGES, "C");
5387 } /* end of ! within_locale_scope */
5389 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
5390 __FILE__, __LINE__));
5393 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5394 "Any locale change has been done; about to call Strerror\n"));
5395 errstr = savepv(Strerror(errnum));
5397 if (! within_locale_scope) {
5398 if (save_locale && ! locale_is_C) {
5399 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
5401 "panic: %s: %d: setlocale restore failed, errno=%d\n",
5402 __FILE__, __LINE__, errno);
5404 Safefree(save_locale);
5410 # endif /* End of doesn't have strerror_l */
5413 if (DEBUG_Lv_TEST) {
5414 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
5415 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
5416 PerlIO_printf(Perl_debug_log, "'\n");
5420 #endif /* End of does have locale messages */
5428 =for apidoc switch_to_global_locale
5430 On systems without locale support, or on typical single-threaded builds, or on
5431 platforms that do not support per-thread locale operations, this function does
5432 nothing. On such systems that do have locale support, only a locale global to
5433 the whole program is available.
5435 On multi-threaded builds on systems that do have per-thread locale operations,
5436 this function converts the thread it is running in to use the global locale.
5437 This is for code that has not yet or cannot be updated to handle multi-threaded
5438 locale operation. As long as only a single thread is so-converted, everything
5439 works fine, as all the other threads continue to ignore the global one, so only
5440 this thread looks at it.
5442 However, on Windows systems this isn't quite true prior to Visual Studio 15,
5443 at which point Microsoft fixed a bug. A race can occur if you use the
5444 following operations on earlier Windows platforms:
5448 =item L<POSIX::localeconv|POSIX/localeconv>
5450 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5452 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5456 The first item is not fixable (except by upgrading to a later Visual Studio
5457 release), but it would be possible to work around the latter two items by using
5458 the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches
5461 Without this function call, threads that use the L<C<setlocale(3)>> system
5462 function will not work properly, as all the locale-sensitive functions will
5463 look at the per-thread locale, and C<setlocale> will have no effect on this
5466 Perl code should convert to either call
5467 L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system
5468 C<setlocale>) or use the methods given in L<perlcall> to call
5469 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly
5470 handle all cases of single- vs multi-thread, POSIX 2008-supported or not.
5472 Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can
5473 continue to work if this function is called before transferring control to the
5476 Upon return from the code that needs to use the global locale,
5477 L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe
5478 multi-thread operation.
5484 Perl_switch_to_global_locale()
5487 #ifdef USE_THREAD_SAFE_LOCALE
5490 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
5493 # ifdef HAS_QUERYLOCALE
5495 setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0)));
5502 for (i = 0; i < LC_ALL_INDEX; i++) {
5503 setlocale(categories[i], do_setlocale_r(categories[i], NULL));
5509 uselocale(LC_GLOBAL_LOCALE);
5518 =for apidoc sync_locale
5520 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or
5521 change the locale (though changing the locale is antisocial and dangerous on
5522 multi-threaded systems that don't have multi-thread safe locale operations.
5523 (See L<perllocale/Multi-threaded operation>). Using the system
5524 L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries
5525 called from XS, such as C<Gtk> do so, and this can't be changed. When the
5526 locale is changed by XS code that didn't use
5527 L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the
5528 locale has changed. Use this function to do so, before returning to Perl.
5530 The return value is a boolean: TRUE if the global locale at the time of call
5531 was in effect; and FALSE if a per-thread locale was in effect. This can be
5532 used by the caller that needs to restore things as-they-were to decide whether
5534 L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>.
5549 const char * newlocale;
5552 # ifdef USE_POSIX_2008_LOCALE
5554 bool was_in_global_locale = FALSE;
5555 locale_t cur_obj = uselocale((locale_t) 0);
5557 /* On Windows, unless the foreign code has turned off the thread-safe
5558 * locale setting, any plain setlocale() will have affected what we see, so
5559 * no need to worry. Otherwise, If the foreign code has done a plain
5560 * setlocale(), it will only affect the global locale on POSIX systems, but
5561 * will affect the */
5562 if (cur_obj == LC_GLOBAL_LOCALE) {
5564 # ifdef HAS_QUERY_LOCALE
5566 do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL));
5572 /* We can't trust that we can read the LC_ALL format on the
5573 * platform, so do them individually */
5574 for (i = 0; i < LC_ALL_INDEX; i++) {
5575 do_setlocale_r(categories[i], setlocale(categories[i], NULL));
5580 was_in_global_locale = TRUE;
5585 bool was_in_global_locale = TRUE;
5588 # ifdef USE_LOCALE_CTYPE
5590 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
5591 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5592 "%s:%d: %s\n", __FILE__, __LINE__,
5593 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
5594 new_ctype(newlocale);
5595 Safefree(newlocale);
5597 # endif /* USE_LOCALE_CTYPE */
5598 # ifdef USE_LOCALE_COLLATE
5600 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
5601 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5602 "%s:%d: %s\n", __FILE__, __LINE__,
5603 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
5604 new_collate(newlocale);
5605 Safefree(newlocale);
5608 # ifdef USE_LOCALE_NUMERIC
5610 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
5611 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5612 "%s:%d: %s\n", __FILE__, __LINE__,
5613 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
5614 new_numeric(newlocale);
5615 Safefree(newlocale);
5617 # endif /* USE_LOCALE_NUMERIC */
5619 return was_in_global_locale;
5625 #if defined(DEBUGGING) && defined(USE_LOCALE)
5628 S_setlocale_debug_string(const int category, /* category number,
5630 const char* const locale, /* locale name */
5632 /* return value from setlocale() when attempting to
5633 * set 'category' to 'locale' */
5634 const char* const retval)
5636 /* Returns a pointer to a NUL-terminated string in static storage with
5637 * added text about the info passed in. This is not thread safe and will
5638 * be overwritten by the next call, so this should be used just to
5639 * formulate a string to immediately print or savepv() on. */
5641 static char ret[256];
5643 my_strlcpy(ret, "setlocale(", sizeof(ret));
5644 my_strlcat(ret, category_name(category), sizeof(ret));
5645 my_strlcat(ret, ", ", sizeof(ret));
5648 my_strlcat(ret, "\"", sizeof(ret));
5649 my_strlcat(ret, locale, sizeof(ret));
5650 my_strlcat(ret, "\"", sizeof(ret));
5653 my_strlcat(ret, "NULL", sizeof(ret));
5656 my_strlcat(ret, ") returned ", sizeof(ret));
5659 my_strlcat(ret, "\"", sizeof(ret));
5660 my_strlcat(ret, retval, sizeof(ret));
5661 my_strlcat(ret, "\"", sizeof(ret));
5664 my_strlcat(ret, "NULL", sizeof(ret));
5667 assert(strlen(ret) < sizeof(ret));
5675 Perl_thread_locale_init()
5677 /* Called from a thread on startup*/
5679 #ifdef USE_THREAD_SAFE_LOCALE
5683 /* C starts the new thread in the global C locale. If we are thread-safe,
5684 * we want to not be in the global locale */
5686 DEBUG_L(PerlIO_printf(Perl_debug_log,
5687 "%s:%d: new thread, initial locale is %s; calling setlocale\n",
5688 __FILE__, __LINE__, setlocale(LC_ALL, NULL)));
5692 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
5696 Perl_setlocale(LC_ALL, "C");
5704 Perl_thread_locale_term()
5706 /* Called from a thread as it gets ready to terminate */
5708 #ifdef USE_THREAD_SAFE_LOCALE
5710 /* C starts the new thread in the global C locale. If we are thread-safe,
5711 * we want to not be in the global locale */
5716 locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE);
5717 if (cur_obj != LC_GLOBAL_LOCALE && cur_obj != PL_C_locale_obj) {
5718 freelocale(cur_obj);
5728 * ex: set ts=8 sts=4 sw=4 et: