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"
57 /* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
60 #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61 # define debug_initialization 0
62 # define DEBUG_INITIALIZATION_set(v)
64 static bool debug_initialization = FALSE;
65 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
69 /* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
72 #define GET_ERRNO saved_errno
74 /* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
77 #define STRLENs(s) (sizeof("" s "") - 1)
79 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89 #define isNAME_C_OR_POSIX(name) \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
96 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
99 #define UTF8NESS_SEP "\v"
100 #define UTF8NESS_PREFIX "\f"
102 /* So, the string looks like:
104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
109 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
112 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
115 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
120 S_stdize_locale(pTHX_ char *locs)
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
136 const char * const s = strchr(locs, '=');
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
142 const char * const t = strchr(s, '.');
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
161 /* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
165 const int categories[] = {
167 # ifdef USE_LOCALE_NUMERIC
170 # ifdef USE_LOCALE_CTYPE
173 # ifdef USE_LOCALE_COLLATE
176 # ifdef USE_LOCALE_TIME
179 # ifdef USE_LOCALE_MESSAGES
182 # ifdef USE_LOCALE_MONETARY
185 # ifdef USE_LOCALE_ADDRESS
188 # ifdef USE_LOCALE_IDENTIFICATION
191 # ifdef USE_LOCALE_MEASUREMENT
194 # ifdef USE_LOCALE_PAPER
197 # ifdef USE_LOCALE_TELEPHONE
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
208 /* The top-most real element is LC_ALL */
210 const char * category_names[] = {
212 # ifdef USE_LOCALE_NUMERIC
215 # ifdef USE_LOCALE_CTYPE
218 # ifdef USE_LOCALE_COLLATE
221 # ifdef USE_LOCALE_TIME
224 # ifdef USE_LOCALE_MESSAGES
227 # ifdef USE_LOCALE_MONETARY
230 # ifdef USE_LOCALE_ADDRESS
233 # ifdef USE_LOCALE_IDENTIFICATION
236 # ifdef USE_LOCALE_MEASUREMENT
239 # ifdef USE_LOCALE_PAPER
242 # ifdef USE_LOCALE_TELEPHONE
248 NULL /* Placeholder */
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
265 /* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
274 S_category_name(const int category)
280 if (category == LC_ALL) {
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
293 const char suffix[] = " (unknown)";
295 Size_t length = sizeof(suffix) + 1;
304 /* Calculate the number of digits */
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
317 /* Now create LC_foo_INDEX #defines for just those categories on this system */
318 # ifdef USE_LOCALE_NUMERIC
319 # define LC_NUMERIC_INDEX 0
320 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
322 # define _DUMMY_NUMERIC -1
324 # ifdef USE_LOCALE_CTYPE
325 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326 # define _DUMMY_CTYPE LC_CTYPE_INDEX
328 # define _DUMMY_CTYPE _DUMMY_NUMERIC
330 # ifdef USE_LOCALE_COLLATE
331 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332 # define _DUMMY_COLLATE LC_COLLATE_INDEX
334 # define _DUMMY_COLLATE _DUMMY_CTYPE
336 # ifdef USE_LOCALE_TIME
337 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
338 # define _DUMMY_TIME LC_TIME_INDEX
340 # define _DUMMY_TIME _DUMMY_COLLATE
342 # ifdef USE_LOCALE_MESSAGES
343 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
346 # define _DUMMY_MESSAGES _DUMMY_TIME
348 # ifdef USE_LOCALE_MONETARY
349 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350 # define _DUMMY_MONETARY LC_MONETARY_INDEX
352 # define _DUMMY_MONETARY _DUMMY_MESSAGES
354 # ifdef USE_LOCALE_ADDRESS
355 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
358 # define _DUMMY_ADDRESS _DUMMY_MONETARY
360 # ifdef USE_LOCALE_IDENTIFICATION
361 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
364 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
366 # ifdef USE_LOCALE_MEASUREMENT
367 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
370 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
372 # ifdef USE_LOCALE_PAPER
373 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374 # define _DUMMY_PAPER LC_PAPER_INDEX
376 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
378 # ifdef USE_LOCALE_TELEPHONE
379 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
382 # define _DUMMY_TELEPHONE _DUMMY_PAPER
385 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
387 #endif /* ifdef USE_LOCALE */
389 /* Windows requres a customized base-level setlocale() */
391 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
393 # define my_setlocale(cat, locale) setlocale(cat, locale)
396 #ifndef USE_POSIX_2008_LOCALE
398 /* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
403 #else /* Below uses POSIX 2008 */
405 /* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411 # define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
415 /* A third array, parallel to the ones above to map from category to its
417 const int category_masks[] = {
418 # ifdef USE_LOCALE_NUMERIC
421 # ifdef USE_LOCALE_CTYPE
424 # ifdef USE_LOCALE_COLLATE
427 # ifdef USE_LOCALE_TIME
430 # ifdef USE_LOCALE_MESSAGES
433 # ifdef USE_LOCALE_MONETARY
436 # ifdef USE_LOCALE_ADDRESS
439 # ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
442 # ifdef USE_LOCALE_MEASUREMENT
445 # ifdef USE_LOCALE_PAPER
448 # ifdef USE_LOCALE_TELEPHONE
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
460 S_emulate_setlocale(const int category,
463 const bool is_index_valid
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 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);
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
554 mask = category_masks[index];
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
580 # ifdef HAS_QUERYLOCALE
582 return (char *) querylocale(mask, cur_obj);
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
589 # if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
608 # ifdef USE_C_BACKTRACE
610 dump_c_backtrace(Perl_debug_log, 20, 1);
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
626 /* Without querylocale(), we have to use our record-keeping we've
629 if (category != LC_ALL) {
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
639 return PL_curlocales[index];
641 else { /* For LC_ALL */
643 Size_t names_len = 0;
645 bool are_all_categories_the_same_locale = TRUE;
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
658 return PL_curlocales[LC_ALL_INDEX];
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
666 for (i = 0; i < LC_ALL_INDEX; i++) {
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 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]);
676 names_len += strlen(category_names[i])
678 + strlen(PL_curlocales[i])
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 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]);
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
728 SETERRNO(EINVAL, LIB_INVARG);
738 /* Here, we are switching locales. */
740 # ifndef HAS_QUERYLOCALE
742 if (strEQ(locale, "")) {
744 /* For non-querylocale() systems, we do the setting of "" ourselves to
745 * be sure that we really know what's going on. We follow the Linux
746 * documented behavior (but if that differs from the actual behavior,
747 * this won't work exactly as the OS implements). We go out and
748 * examine the environment based on our understanding of how the system
749 * works, and use that to figure things out */
751 const char * const lc_all = PerlEnv_getenv("LC_ALL");
753 /* Use any "LC_ALL" environment variable, as it overrides everything
755 if (lc_all && strNE(lc_all, "")) {
760 /* Otherwise, we need to dig deeper. Unless overridden, the
761 * default is the LANG environment variable; if it doesn't exist,
764 const char * default_name;
766 /* To minimize other threads messing with the environment, we copy
767 * the variable, making it a temporary. But this doesn't work upon
768 * program initialization before any scopes are created, and at
769 * this time, there's nothing else going on that would interfere.
770 * So skip the copy in that case */
771 if (PL_scopestack_ix == 0) {
772 default_name = PerlEnv_getenv("LANG");
775 default_name = savepv(PerlEnv_getenv("LANG"));
778 if (! default_name || strEQ(default_name, "")) {
781 else if (PL_scopestack_ix != 0) {
782 SAVEFREEPV(default_name);
785 if (category != LC_ALL) {
786 const char * const name = PerlEnv_getenv(category_names[index]);
788 /* Here we are setting a single category. Assume will have the
790 locale = default_name;
792 /* But then look for an overriding environment variable */
793 if (name && strNE(name, "")) {
798 bool did_override = FALSE;
801 /* Here, we are getting LC_ALL. Any categories that don't have
802 * a corresponding environment variable set should be set to
803 * LANG, or to "C" if there is no LANG. If no individual
804 * categories differ from this, we can just set LC_ALL. This
805 * is buggy on systems that have extra categories that we don't
806 * know about. If there is an environment variable that sets
807 * that category, we won't know to look for it, and so our use
808 * of LANG or "C" improperly overrides it. On the other hand,
809 * if we don't do what is done here, and there is no
810 * environment variable, the category's locale should be set to
811 * LANG or "C". So there is no good solution. khw thinks the
812 * best is to look at systems to see what categories they have,
813 * and include them, and then to assume that we know the
816 for (i = 0; i < LC_ALL_INDEX; i++) {
817 const char * const env_override
818 = savepv(PerlEnv_getenv(category_names[i]));
819 const char * this_locale = ( env_override
820 && strNE(env_override, ""))
823 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
825 Safefree(env_override);
829 if (strNE(this_locale, default_name)) {
833 Safefree(env_override);
836 /* If all the categories are the same, we can set LC_ALL to
838 if (! did_override) {
839 locale = default_name;
843 /* Here, LC_ALL is no longer valid, as some individual
844 * categories don't match it. We call ourselves
845 * recursively, as that will execute the code that
846 * generates the proper locale string for this situation.
847 * We don't do the remainder of this function, as that is
848 * to update our records, and we've just done that for the
849 * individual categories in the loop above, and doing so
850 * would cause LC_ALL to be done as well */
851 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
856 else if (strchr(locale, ';')) {
858 /* LC_ALL may actually incude a conglomeration of various categories.
859 * Without querylocale, this code uses the glibc (as of this writing)
860 * syntax for representing that, but that is not a stable API, and
861 * other platforms do it differently, so we have to handle all cases
865 const char * s = locale;
866 const char * e = locale + strlen(locale);
868 const char * category_end;
869 const char * name_start;
870 const char * name_end;
872 /* If the string that gives what to set doesn't include all categories,
873 * the omitted ones get set to "C". To get this behavior, first set
874 * all the individual categories to "C", and override the furnished
876 for (i = 0; i < LC_ALL_INDEX; i++) {
877 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
884 /* Parse through the category */
885 while (isWORDCHAR(*p)) {
892 "panic: %s: %d: Unexpected character in locale name '%02X",
893 __FILE__, __LINE__, *(p-1));
896 /* Parse through the locale name */
898 while (p < e && *p != ';') {
901 "panic: %s: %d: Unexpected character in locale name '%02X",
902 __FILE__, __LINE__, *(p-1));
908 /* Space past the semi-colon */
913 /* Find the index of the category name in our lists */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
915 char * individ_locale;
917 /* Keep going if this isn't the index. The strnNE() avoids a
918 * Perl_form(), but would fail if ever a category name could be
919 * a substring of another one, like if there were a
921 if strnNE(s, category_names[i], category_end - s) {
925 /* If this index is for the single category we're changing, we
926 * have found the locale to set it to. */
927 if (category == categories[i]) {
928 locale = Perl_form(aTHX_ "%.*s",
929 (int) (name_end - name_start),
934 assert(category == LC_ALL);
935 individ_locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start), name_start);
937 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
946 /* Here we have set all the individual categories by recursive calls.
947 * These collectively should have fixed up LC_ALL, so can just query
948 * what that now is */
949 assert(category == LC_ALL);
951 return do_setlocale_c(LC_ALL, NULL);
956 /* Here at the end of having to deal with the absence of querylocale().
957 * Some cases have already been fully handled by recursive calls to this
958 * function. But at this point, we haven't dealt with those, but are now
959 * prepared to, knowing what the locale name to set this category to is.
960 * This would have come for free if this system had had querylocale() */
962 # endif /* end of ! querylocale */
964 assert(PL_C_locale_obj);
966 /* Switching locales generally entails freeing the current one's space (at
967 * the C library's discretion). We need to stop using that locale before
968 * the switch. So switch to a known locale object that we don't otherwise
969 * mess with. This returns the locale object in effect at the time of the
971 old_obj = uselocale(PL_C_locale_obj);
975 if (DEBUG_Lv_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
985 if (DEBUG_L_TEST || debug_initialization) {
987 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
998 if (DEBUG_Lv_TEST || debug_initialization) {
999 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1004 /* If we weren't in a thread safe locale, set so that newlocale() below
1005 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1006 The latter is defensive coding, so that, even if there is some bug, we
1007 will never end up trying to modify either of these, as if passed to
1008 newlocale(), they can be. */
1009 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1010 old_obj = (locale_t) 0;
1013 /* Ready to create a new locale by modification of the exising one */
1014 new_obj = newlocale(mask, locale, old_obj);
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1027 if (! uselocale(old_obj)) {
1031 if (DEBUG_L_TEST || debug_initialization) {
1032 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1044 if (DEBUG_Lv_TEST || debug_initialization) {
1045 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p; should have freed %p\n", __FILE__, __LINE__, new_obj, old_obj);
1050 /* And switch into it */
1051 if (! uselocale(new_obj)) {
1056 if (DEBUG_L_TEST || debug_initialization) {
1057 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1062 if (! uselocale(old_obj)) {
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1073 freelocale(new_obj);
1080 if (DEBUG_Lv_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1086 /* We are done, except for updating our records (if the system doesn't keep
1087 * them) and in the case of locale "", we don't actually know what the
1088 * locale that got switched to is, as it came from the environment. So
1089 * have to find it */
1091 # ifdef HAS_QUERYLOCALE
1093 if (strEQ(locale, "")) {
1094 locale = querylocale(mask, new_obj);
1099 /* Here, 'locale' is the return value */
1101 /* Without querylocale(), we have to update our records */
1103 if (category == LC_ALL) {
1106 /* For LC_ALL, we change all individual categories to correspond */
1107 /* PL_curlocales is a parallel array, so has same
1108 * length as 'categories' */
1109 for (i = 0; i <= LC_ALL_INDEX; i++) {
1110 Safefree(PL_curlocales[i]);
1111 PL_curlocales[i] = savepv(locale);
1116 /* For a single category, if it's not the same as the one in LC_ALL, we
1119 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1120 Safefree(PL_curlocales[LC_ALL_INDEX]);
1121 PL_curlocales[LC_ALL_INDEX] = NULL;
1124 /* Then update the category's record */
1125 Safefree(PL_curlocales[index]);
1126 PL_curlocales[index] = savepv(locale);
1134 #endif /* USE_POSIX_2008_LOCALE */
1136 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1137 didn't natively support them */
1139 /* The way this would work is that we would keep a per-thread list of the
1140 * correct locale for that thread. Any operation that was locale-sensitive
1141 * would have to be changed so that it would look like this:
1144 * setlocale to the correct locale for this operation
1148 * This leaves the global locale in the most recently used operation's, but it
1149 * was locked long enough to get the result. If that result is static, it
1150 * needs to be copied before the unlock.
1152 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1153 * the setup, but are no-ops when not needed, and similarly,
1154 * END_LOCALE_DEPENDENT_OP for the tear-down
1156 * But every call to a locale-sensitive function would have to be changed, and
1157 * if a module didn't cooperate by using the mutex, things would break.
1159 * This code was abandoned before being completed or tested, and is left as-is
1162 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1163 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1166 S_locking_setlocale(pTHX_
1168 const char * locale,
1170 const bool is_index_valid
1173 /* This function kind of performs a setlocale() on just the current thread;
1174 * thus it is kind of thread-safe. It does this by keeping a thread-level
1175 * array of the current locales for each category. Every time a locale is
1176 * switched to, it does the switch globally, but updates the thread's
1177 * array. A query as to what the current locale is just returns the
1178 * appropriate element from the array, and doesn't actually call the system
1179 * setlocale(). The saving into the array is done in an uninterruptible
1180 * section of code, so is unaffected by whatever any other threads might be
1183 * All locale-sensitive operations must work by first starting a critical
1184 * section, then switching to the thread's locale as kept by this function,
1185 * and then doing the operation, then ending the critical section. Thus,
1186 * each gets done in the appropriate locale. simulating thread-safety.
1188 * This function takes the same parameters, 'category' and 'locale', that
1189 * the regular setlocale() function does, but it also takes two additional
1190 * ones. This is because as described earlier. If we know on input the
1191 * index corresponding to the category into the array where we store the
1192 * current locales, we don't have to calculate it. If the caller knows at
1193 * compile time what the index is, it it can pass it, setting
1194 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1198 /* If the input index might be incorrect, calculate the correct one */
1199 if (! is_index_valid) {
1202 if (DEBUG_Lv_TEST || debug_initialization) {
1203 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1206 for (i = 0; i <= LC_ALL_INDEX; i++) {
1207 if (category == categories[i]) {
1213 /* Here, we don't know about this category, so can't handle it.
1214 * XXX best we can do is to unsafely set this
1217 return my_setlocale(category, locale);
1221 if (DEBUG_Lv_TEST || debug_initialization) {
1222 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1226 /* For a query, just return what's in our records */
1227 if (new_locale == NULL) {
1228 return curlocales[index];
1232 /* Otherwise, we need to do the switch, and save the result, all in a
1233 * critical section */
1235 Safefree(curlocales[[index]]);
1237 /* It might be that this is called from an already-locked section of code.
1238 * We would have to detect and skip the LOCK/UNLOCK if so */
1241 curlocales[index] = savepv(my_setlocale(category, new_locale));
1243 if (strEQ(new_locale, "")) {
1247 /* The locale values come from the environment, and may not all be the
1248 * same, so for LC_ALL, we have to update all the others, while the
1249 * mutex is still locked */
1251 if (category == LC_ALL) {
1253 for (i = 0; i < LC_ALL_INDEX) {
1254 curlocales[i] = my_setlocale(categories[i], NULL);
1263 return curlocales[index];
1270 S_set_numeric_radix(pTHX_ const bool use_locale)
1272 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1273 * TRUE, use the radix character derived from the current locale */
1275 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1276 || defined(HAS_NL_LANGINFO))
1278 const char * radix = (use_locale)
1279 ? my_nl_langinfo(RADIXCHAR, FALSE)
1280 /* FALSE => already in dest locale */
1283 sv_setpv(PL_numeric_radix_sv, radix);
1285 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1286 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1287 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1288 SvCUR(PL_numeric_radix_sv))
1289 && _is_cur_LC_category_utf8(LC_NUMERIC))
1291 SvUTF8_on(PL_numeric_radix_sv);
1296 if (DEBUG_L_TEST || debug_initialization) {
1297 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1298 SvPVX(PL_numeric_radix_sv),
1299 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1305 PERL_UNUSED_ARG(use_locale);
1307 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1312 S_new_numeric(pTHX_ const char *newnum)
1315 #ifndef USE_LOCALE_NUMERIC
1317 PERL_UNUSED_ARG(newnum);
1321 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1322 * core Perl this and that 'newnum' is the name of the new locale.
1323 * It installs this locale as the current underlying default.
1325 * The default locale and the C locale can be toggled between by use of the
1326 * set_numeric_underlying() and set_numeric_standard() functions, which
1327 * should probably not be called directly, but only via macros like
1328 * SET_NUMERIC_STANDARD() in perl.h.
1330 * The toggling is necessary mainly so that a non-dot radix decimal point
1331 * character can be output, while allowing internal calculations to use a
1334 * This sets several interpreter-level variables:
1335 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1336 * PL_numeric_underlying A boolean indicating if the toggled state is such
1337 * that the current locale is the program's underlying
1339 * PL_numeric_standard An int indicating if the toggled state is such
1340 * that the current locale is the C locale or
1341 * indistinguishable from the C locale. If non-zero, it
1342 * is in C; if > 1, it means it may not be toggled away
1344 * PL_numeric_underlying_is_standard A bool kept by this function
1345 * indicating that the underlying locale and the standard
1346 * C locale are indistinguishable for the purposes of
1347 * LC_NUMERIC. This happens when both of the above two
1348 * variables are true at the same time. (Toggling is a
1349 * no-op under these circumstances.) This variable is
1350 * used to avoid having to recalculate.
1356 Safefree(PL_numeric_name);
1357 PL_numeric_name = NULL;
1358 PL_numeric_standard = TRUE;
1359 PL_numeric_underlying = TRUE;
1360 PL_numeric_underlying_is_standard = TRUE;
1364 save_newnum = stdize_locale(savepv(newnum));
1365 PL_numeric_underlying = TRUE;
1366 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1368 #ifndef TS_W32_BROKEN_LOCALECONV
1370 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1371 * them. But on broken Windows systems calling my_nl_langinfo() for
1372 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1373 * and just always change the locale if not C nor POSIX on those systems */
1374 if (! PL_numeric_standard) {
1375 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1376 FALSE /* Don't toggle locale */ ))
1377 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1382 /* Save the new name if it isn't the same as the previous one, if any */
1383 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1384 Safefree(PL_numeric_name);
1385 PL_numeric_name = save_newnum;
1388 Safefree(save_newnum);
1391 PL_numeric_underlying_is_standard = PL_numeric_standard;
1393 # ifdef HAS_POSIX_2008_LOCALE
1395 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1397 PL_underlying_numeric_obj);
1401 if (DEBUG_L_TEST || debug_initialization) {
1402 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1405 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1406 * have to worry about the radix being a non-dot. (Core operations that
1407 * need the underlying locale change to it temporarily). */
1408 if (PL_numeric_standard) {
1409 set_numeric_radix(0);
1412 set_numeric_standard();
1415 #endif /* USE_LOCALE_NUMERIC */
1420 Perl_set_numeric_standard(pTHX)
1423 #ifdef USE_LOCALE_NUMERIC
1425 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1426 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1427 * macro avoids calling this routine if toggling isn't necessary according
1428 * to our records (which could be wrong if some XS code has changed the
1429 * locale behind our back) */
1433 if (DEBUG_L_TEST || debug_initialization) {
1434 PerlIO_printf(Perl_debug_log,
1435 "Setting LC_NUMERIC locale to standard C\n");
1440 do_setlocale_c(LC_NUMERIC, "C");
1441 PL_numeric_standard = TRUE;
1442 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1443 set_numeric_radix(0);
1445 #endif /* USE_LOCALE_NUMERIC */
1450 Perl_set_numeric_underlying(pTHX)
1453 #ifdef USE_LOCALE_NUMERIC
1455 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1456 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1457 * instead of calling this directly. The macro avoids calling this routine
1458 * if toggling isn't necessary according to our records (which could be
1459 * wrong if some XS code has changed the locale behind our back) */
1463 if (DEBUG_L_TEST || debug_initialization) {
1464 PerlIO_printf(Perl_debug_log,
1465 "Setting LC_NUMERIC locale to %s\n",
1471 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1472 PL_numeric_standard = PL_numeric_underlying_is_standard;
1473 PL_numeric_underlying = TRUE;
1474 set_numeric_radix(! PL_numeric_standard);
1476 #endif /* USE_LOCALE_NUMERIC */
1481 * Set up for a new ctype locale.
1484 S_new_ctype(pTHX_ const char *newctype)
1487 #ifndef USE_LOCALE_CTYPE
1489 PERL_UNUSED_ARG(newctype);
1490 PERL_UNUSED_CONTEXT;
1494 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1495 * core Perl this and that 'newctype' is the name of the new locale.
1497 * This function sets up the folding arrays for all 256 bytes, assuming
1498 * that tofold() is tolc() since fold case is not a concept in POSIX,
1500 * Any code changing the locale (outside this file) should use
1501 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1502 * this function should be called directly only from this file and from
1503 * POSIX::setlocale() */
1508 /* Don't check for problems if we are suppressing the warnings */
1509 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1511 PERL_ARGS_ASSERT_NEW_CTYPE;
1513 /* We will replace any bad locale warning with 1) nothing if the new one is
1514 * ok; or 2) a new warning for the bad new locale */
1515 if (PL_warn_locale) {
1516 SvREFCNT_dec_NN(PL_warn_locale);
1517 PL_warn_locale = NULL;
1520 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1522 /* A UTF-8 locale gets standard rules. But note that code still has to
1523 * handle this specially because of the three problematic code points */
1524 if (PL_in_utf8_CTYPE_locale) {
1525 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1528 /* We don't populate the other lists if a UTF-8 locale, but do check that
1529 * everything works as expected, unless checking turned off */
1530 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1531 /* Assume enough space for every character being bad. 4 spaces each
1532 * for the 94 printable characters that are output like "'x' "; and 5
1533 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1535 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1536 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1538 unsigned int bad_count = 0; /* Count of bad characters */
1540 for (i = 0; i < 256; i++) {
1541 if (! PL_in_utf8_CTYPE_locale) {
1543 PL_fold_locale[i] = (U8) tolower(i);
1544 else if (islower(i))
1545 PL_fold_locale[i] = (U8) toupper(i);
1547 PL_fold_locale[i] = (U8) i;
1550 /* If checking for locale problems, see if the native ASCII-range
1551 * printables plus \n and \t are in their expected categories in
1552 * the new locale. If not, this could mean big trouble, upending
1553 * Perl's and most programs' assumptions, like having a
1554 * metacharacter with special meaning become a \w. Fortunately,
1555 * it's very rare to find locales that aren't supersets of ASCII
1556 * nowadays. It isn't a problem for most controls to be changed
1557 * into something else; we check only \n and \t, though perhaps \r
1558 * could be an issue as well. */
1559 if ( check_for_problems
1560 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1562 bool is_bad = FALSE;
1563 char name[4] = { '\0' };
1565 /* Convert the name into a string */
1570 else if (i == '\n') {
1571 my_strlcpy(name, "\\n", sizeof(name));
1573 else if (i == '\t') {
1574 my_strlcpy(name, "\\t", sizeof(name));
1578 my_strlcpy(name, "' '", sizeof(name));
1581 /* Check each possibe class */
1582 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isalnum('%s') unexpectedly is %d\n",
1586 name, cBOOL(isalnum(i))));
1588 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isalpha('%s') unexpectedly is %d\n",
1592 name, cBOOL(isalpha(i))));
1594 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isdigit('%s') unexpectedly is %d\n",
1598 name, cBOOL(isdigit(i))));
1600 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "isgraph('%s') unexpectedly is %d\n",
1604 name, cBOOL(isgraph(i))));
1606 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "islower('%s') unexpectedly is %d\n",
1610 name, cBOOL(islower(i))));
1612 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "isprint('%s') unexpectedly is %d\n",
1616 name, cBOOL(isprint(i))));
1618 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "ispunct('%s') unexpectedly is %d\n",
1622 name, cBOOL(ispunct(i))));
1624 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1626 DEBUG_L(PerlIO_printf(Perl_debug_log,
1627 "isspace('%s') unexpectedly is %d\n",
1628 name, cBOOL(isspace(i))));
1630 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1632 DEBUG_L(PerlIO_printf(Perl_debug_log,
1633 "isupper('%s') unexpectedly is %d\n",
1634 name, cBOOL(isupper(i))));
1636 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1638 DEBUG_L(PerlIO_printf(Perl_debug_log,
1639 "isxdigit('%s') unexpectedly is %d\n",
1640 name, cBOOL(isxdigit(i))));
1642 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1644 DEBUG_L(PerlIO_printf(Perl_debug_log,
1645 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1646 name, tolower(i), (int) toLOWER_A(i)));
1648 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1650 DEBUG_L(PerlIO_printf(Perl_debug_log,
1651 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1652 name, toupper(i), (int) toUPPER_A(i)));
1654 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1656 DEBUG_L(PerlIO_printf(Perl_debug_log,
1657 "'\\n' (=%02X) is not a control\n", (int) i));
1660 /* Add to the list; Separate multiple entries with a blank */
1663 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1665 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1673 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1674 * this locale requires more than one byte, there are going to be
1676 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1677 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1678 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1680 if ( check_for_problems && MB_CUR_MAX > 1
1681 && ! PL_in_utf8_CTYPE_locale
1683 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1684 * locale. Just assume that the implementation for them (plus
1685 * for POSIX) is correct and the > 1 value is spurious. (Since
1686 * these are specially handled to never be considered UTF-8
1687 * locales, as long as this is the only problem, everything
1688 * should work fine */
1689 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1691 multi_byte_locale = TRUE;
1696 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1697 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1698 PL_warn_locale = Perl_newSVpvf(aTHX_
1699 "Locale '%s' contains (at least) the following characters"
1700 " which have\nunexpected meanings: %s\nThe Perl program"
1701 " will use the expected meanings",
1702 newctype, bad_chars_list);
1705 PL_warn_locale = Perl_newSVpvf(aTHX_
1706 "Locale '%s' may not work well.%s%s%s\n",
1709 ? " Some characters in it are not recognized by"
1713 ? "\nThe following characters (and maybe others)"
1714 " may not have the same meaning as the Perl"
1715 " program expects:\n"
1723 # ifdef HAS_NL_LANGINFO
1725 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1726 /* parameter FALSE is a don't care here */
1727 my_nl_langinfo(CODESET, FALSE));
1731 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1733 /* If we are actually in the scope of the locale or are debugging,
1734 * output the message now. If not in that scope, we save the
1735 * message to be output at the first operation using this locale,
1736 * if that actually happens. Most programs don't use locales, so
1737 * they are immune to bad ones. */
1738 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1740 /* The '0' below suppresses a bogus gcc compiler warning */
1741 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1743 if (IN_LC(LC_CTYPE)) {
1744 SvREFCNT_dec_NN(PL_warn_locale);
1745 PL_warn_locale = NULL;
1751 #endif /* USE_LOCALE_CTYPE */
1756 Perl__warn_problematic_locale()
1759 #ifdef USE_LOCALE_CTYPE
1763 /* Internal-to-core function that outputs the message in PL_warn_locale,
1764 * and then NULLS it. Should be called only through the macro
1765 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1767 if (PL_warn_locale) {
1768 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1769 SvPVX(PL_warn_locale),
1770 0 /* dummy to avoid compiler warning */ );
1771 SvREFCNT_dec_NN(PL_warn_locale);
1772 PL_warn_locale = NULL;
1780 S_new_collate(pTHX_ const char *newcoll)
1783 #ifndef USE_LOCALE_COLLATE
1785 PERL_UNUSED_ARG(newcoll);
1786 PERL_UNUSED_CONTEXT;
1790 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1791 * core Perl this and that 'newcoll' is the name of the new locale.
1793 * The design of locale collation is that every locale change is given an
1794 * index 'PL_collation_ix'. The first time a string particpates in an
1795 * operation that requires collation while locale collation is active, it
1796 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1797 * magic includes the collation index, and the transformation of the string
1798 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1799 * instead of the string itself. If a string changes, the magic is
1800 * cleared. The next time the locale changes, the index is incremented,
1801 * and so we know during a comparison that the transformation is not
1802 * necessarily still valid, and so is recomputed. Note that if the locale
1803 * changes enough times, the index could wrap (a U32), and it is possible
1804 * that a transformation would improperly be considered valid, leading to
1805 * an unlikely bug */
1808 if (PL_collation_name) {
1810 Safefree(PL_collation_name);
1811 PL_collation_name = NULL;
1813 PL_collation_standard = TRUE;
1814 is_standard_collation:
1815 PL_collxfrm_base = 0;
1816 PL_collxfrm_mult = 2;
1817 PL_in_utf8_COLLATE_locale = FALSE;
1818 PL_strxfrm_NUL_replacement = '\0';
1819 PL_strxfrm_max_cp = 0;
1823 /* If this is not the same locale as currently, set the new one up */
1824 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1826 Safefree(PL_collation_name);
1827 PL_collation_name = stdize_locale(savepv(newcoll));
1828 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1829 if (PL_collation_standard) {
1830 goto is_standard_collation;
1833 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1834 PL_strxfrm_NUL_replacement = '\0';
1835 PL_strxfrm_max_cp = 0;
1837 /* A locale collation definition includes primary, secondary, tertiary,
1838 * etc. weights for each character. To sort, the primary weights are
1839 * used, and only if they compare equal, then the secondary weights are
1840 * used, and only if they compare equal, then the tertiary, etc.
1842 * strxfrm() works by taking the input string, say ABC, and creating an
1843 * output transformed string consisting of first the primary weights,
1844 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1845 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1846 * may not have weights at every level. In our example, let's say B
1847 * doesn't have a tertiary weight, and A doesn't have a secondary
1848 * weight. The constructed string is then going to be
1849 * A¹B¹C¹ B²C² A³C³ ....
1850 * This has the desired effect that strcmp() will look at the secondary
1851 * or tertiary weights only if the strings compare equal at all higher
1852 * priority weights. The spaces shown here, like in
1854 * are not just for readability. In the general case, these must
1855 * actually be bytes, which we will call here 'separator weights'; and
1856 * they must be smaller than any other weight value, but since these
1857 * are C strings, only the terminating one can be a NUL (some
1858 * implementations may include a non-NUL separator weight just before
1859 * the NUL). Implementations tend to reserve 01 for the separator
1860 * weights. They are needed so that a shorter string's secondary
1861 * weights won't be misconstrued as primary weights of a longer string,
1862 * etc. By making them smaller than any other weight, the shorter
1863 * string will sort first. (Actually, if all secondary weights are
1864 * smaller than all primary ones, there is no need for a separator
1865 * weight between those two levels, etc.)
1867 * The length of the transformed string is roughly a linear function of
1868 * the input string. It's not exactly linear because some characters
1869 * don't have weights at all levels. When we call strxfrm() we have to
1870 * allocate some memory to hold the transformed string. The
1871 * calculations below try to find coefficients 'm' and 'b' for this
1872 * locale so that m*x + b equals how much space we need, given the size
1873 * of the input string in 'x'. If we calculate too small, we increase
1874 * the size as needed, and call strxfrm() again, but it is better to
1875 * get it right the first time to avoid wasted expensive string
1876 * transformations. */
1879 /* We use the string below to find how long the tranformation of it
1880 * is. Almost all locales are supersets of ASCII, or at least the
1881 * ASCII letters. We use all of them, half upper half lower,
1882 * because if we used fewer, we might hit just the ones that are
1883 * outliers in a particular locale. Most of the strings being
1884 * collated will contain a preponderance of letters, and even if
1885 * they are above-ASCII, they are likely to have the same number of
1886 * weight levels as the ASCII ones. It turns out that digits tend
1887 * to have fewer levels, and some punctuation has more, but those
1888 * are relatively sparse in text, and khw believes this gives a
1889 * reasonable result, but it could be changed if experience so
1891 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1892 char * x_longer; /* Transformed 'longer' */
1893 Size_t x_len_longer; /* Length of 'x_longer' */
1895 char * x_shorter; /* We also transform a substring of 'longer' */
1896 Size_t x_len_shorter;
1898 /* _mem_collxfrm() is used get the transformation (though here we
1899 * are interested only in its length). It is used because it has
1900 * the intelligence to handle all cases, but to work, it needs some
1901 * values of 'm' and 'b' to get it started. For the purposes of
1902 * this calculation we use a very conservative estimate of 'm' and
1903 * 'b'. This assumes a weight can be multiple bytes, enough to
1904 * hold any UV on the platform, and there are 5 levels, 4 weight
1905 * bytes, and a trailing NUL. */
1906 PL_collxfrm_base = 5;
1907 PL_collxfrm_mult = 5 * sizeof(UV);
1909 /* Find out how long the transformation really is */
1910 x_longer = _mem_collxfrm(longer,
1914 /* We avoid converting to UTF-8 in the
1915 * called function by telling it the
1916 * string is in UTF-8 if the locale is a
1917 * UTF-8 one. Since the string passed
1918 * here is invariant under UTF-8, we can
1919 * claim it's UTF-8 even though it isn't.
1921 PL_in_utf8_COLLATE_locale);
1924 /* Find out how long the transformation of a substring of 'longer'
1925 * is. Together the lengths of these transformations are
1926 * sufficient to calculate 'm' and 'b'. The substring is all of
1927 * 'longer' except the first character. This minimizes the chances
1928 * of being swayed by outliers */
1929 x_shorter = _mem_collxfrm(longer + 1,
1932 PL_in_utf8_COLLATE_locale);
1933 Safefree(x_shorter);
1935 /* If the results are nonsensical for this simple test, the whole
1936 * locale definition is suspect. Mark it so that locale collation
1937 * is not active at all for it. XXX Should we warn? */
1938 if ( x_len_shorter == 0
1939 || x_len_longer == 0
1940 || x_len_shorter >= x_len_longer)
1942 PL_collxfrm_mult = 0;
1943 PL_collxfrm_base = 0;
1946 SSize_t base; /* Temporary */
1948 /* We have both: m * strlen(longer) + b = x_len_longer
1949 * m * strlen(shorter) + b = x_len_shorter;
1950 * subtracting yields:
1951 * m * (strlen(longer) - strlen(shorter))
1952 * = x_len_longer - x_len_shorter
1953 * But we have set things up so that 'shorter' is 1 byte smaller
1954 * than 'longer'. Hence:
1955 * m = x_len_longer - x_len_shorter
1957 * But if something went wrong, make sure the multiplier is at
1960 if (x_len_longer > x_len_shorter) {
1961 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1964 PL_collxfrm_mult = 1;
1969 * but in case something has gone wrong, make sure it is
1971 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1976 /* Add 1 for the trailing NUL */
1977 PL_collxfrm_base = base + 1;
1982 if (DEBUG_L_TEST || debug_initialization) {
1983 PerlIO_printf(Perl_debug_log,
1984 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1986 " collate multipler=%zu, collate base=%zu\n",
1988 PL_in_utf8_COLLATE_locale,
1989 x_len_shorter, x_len_longer,
1990 PL_collxfrm_mult, PL_collxfrm_base);
1997 #endif /* USE_LOCALE_COLLATE */
2006 S_win32_setlocale(pTHX_ int category, const char* locale)
2008 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2009 * difference between the two unless the input locale is "", which normally
2010 * means on Windows to get the machine default, which is set via the
2011 * computer's "Regional and Language Options" (or its current equivalent).
2012 * In POSIX, it instead means to find the locale from the user's
2013 * environment. This routine changes the Windows behavior to first look in
2014 * the environment, and, if anything is found, use that instead of going to
2015 * the machine default. If there is no environment override, the machine
2016 * default is used, by calling the real setlocale() with "".
2018 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2019 * use the particular category's variable if set; otherwise to use the LANG
2022 bool override_LC_ALL = FALSE;
2026 if (locale && strEQ(locale, "")) {
2030 locale = PerlEnv_getenv("LC_ALL");
2032 if (category == LC_ALL) {
2033 override_LC_ALL = TRUE;
2039 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2040 if (category == categories[i]) {
2041 locale = PerlEnv_getenv(category_names[i]);
2046 locale = PerlEnv_getenv("LANG");
2062 result = setlocale(category, locale);
2063 DEBUG_L(STMT_START {
2065 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2066 setlocale_debug_string(category, locale, result));
2070 if (! override_LC_ALL) {
2074 /* Here the input category was LC_ALL, and we have set it to what is in the
2075 * LANG variable or the system default if there is no LANG. But these have
2076 * lower priority than the other LC_foo variables, so override it for each
2077 * one that is set. (If they are set to "", it means to use the same thing
2078 * we just set LC_ALL to, so can skip) */
2080 for (i = 0; i < LC_ALL_INDEX; i++) {
2081 result = PerlEnv_getenv(category_names[i]);
2082 if (result && strNE(result, "")) {
2083 setlocale(categories[i], result);
2084 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2086 setlocale_debug_string(categories[i], result, "not captured")));
2090 result = setlocale(LC_ALL, NULL);
2091 DEBUG_L(STMT_START {
2093 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2095 setlocale_debug_string(LC_ALL, NULL, result));
2106 =head1 Locale-related functions and macros
2108 =for apidoc Perl_setlocale
2110 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2111 taking the same parameters, and returning the same information, except that it
2112 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2113 instead return C<C> if the underlying locale has a non-dot decimal point
2114 character, or a non-empty thousands separator for displaying floating point
2115 numbers. This is because perl keeps that locale category such that it has a
2116 dot and empty separator, changing the locale briefly during the operations
2117 where the underlying one is required. C<Perl_setlocale> knows about this, and
2118 compensates; regular C<setlocale> doesn't.
2120 Another reason it isn't completely a drop-in replacement is that it is
2121 declared to return S<C<const char *>>, whereas the system setlocale omits the
2122 C<const> (presumably because its API was specified long ago, and can't be
2123 updated; it is illegal to change the information C<setlocale> returns; doing
2124 so leads to segfaults.)
2126 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2127 C<setlocale> can be completely ineffective on some platforms under some
2130 C<Perl_setlocale> should not be used to change the locale except on systems
2131 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2132 the system C<setlocale()> is ineffective, returning the wrong information, and
2133 failing to actually change the locale. C<Perl_setlocale>, however works
2134 properly in all circumstances.
2136 The return points to a per-thread static buffer, which is overwritten the next
2137 time C<Perl_setlocale> is called from the same thread.
2144 Perl_setlocale(const int category, const char * locale)
2146 /* This wraps POSIX::setlocale() */
2150 PERL_UNUSED_ARG(category);
2151 PERL_UNUSED_ARG(locale);
2157 const char * retval;
2158 const char * newlocale;
2161 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2163 #ifdef USE_LOCALE_NUMERIC
2165 /* A NULL locale means only query what the current one is. We have the
2166 * LC_NUMERIC name saved, because we are normally switched into the C
2167 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2168 * the correct results. All other categories don't require special
2170 if (locale == NULL) {
2171 if (category == LC_NUMERIC) {
2173 /* We don't have to copy this return value, as it is a per-thread
2174 * variable, and won't change until a future setlocale */
2175 return PL_numeric_name;
2180 else if (category == LC_ALL) {
2181 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2190 retval = save_to_buffer(do_setlocale_r(category, locale),
2191 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2194 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2196 if (locale == NULL && category == LC_ALL) {
2197 RESTORE_LC_NUMERIC();
2202 DEBUG_L(PerlIO_printf(Perl_debug_log,
2203 "%s:%d: %s\n", __FILE__, __LINE__,
2204 setlocale_debug_string(category, locale, retval)));
2212 /* If locale == NULL, we are just querying the state */
2213 if (locale == NULL) {
2217 /* Now that have switched locales, we have to update our records to
2222 #ifdef USE_LOCALE_CTYPE
2229 #ifdef USE_LOCALE_COLLATE
2232 new_collate(retval);
2236 #ifdef USE_LOCALE_NUMERIC
2239 new_numeric(retval);
2247 /* LC_ALL updates all the things we care about. The values may not
2248 * be the same as 'retval', as the locale "" may have set things
2251 # ifdef USE_LOCALE_CTYPE
2253 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2254 new_ctype(newlocale);
2255 Safefree(newlocale);
2257 # endif /* USE_LOCALE_CTYPE */
2258 # ifdef USE_LOCALE_COLLATE
2260 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2261 new_collate(newlocale);
2262 Safefree(newlocale);
2265 # ifdef USE_LOCALE_NUMERIC
2267 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2268 new_numeric(newlocale);
2269 Safefree(newlocale);
2271 # endif /* USE_LOCALE_NUMERIC */
2284 PERL_STATIC_INLINE const char *
2285 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2287 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2288 * growing it if necessary */
2292 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2298 string_size = strlen(string) + offset + 1;
2300 if (*buf_size == 0) {
2301 Newx(*buf, string_size, char);
2302 *buf_size = string_size;
2304 else if (string_size > *buf_size) {
2305 Renew(*buf, string_size, char);
2306 *buf_size = string_size;
2309 Copy(string, *buf + offset, string_size - offset, char);
2315 =for apidoc Perl_langinfo
2317 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2318 taking the same C<item> parameter values, and returning the same information.
2319 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2320 of Perl's locale handling from your code, and can be used on systems that lack
2321 a native C<nl_langinfo>.
2329 The reason it isn't quite a drop-in replacement is actually an advantage. The
2330 only difference is that it returns S<C<const char *>>, whereas plain
2331 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2332 forbidden to write into the buffer. By declaring this C<const>, the compiler
2333 enforces this restriction, so if it is violated, you know at compilation time,
2334 rather than getting segfaults at runtime.
2338 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2339 without you having to write extra code. The reason for the extra code would be
2340 because these are from the C<LC_NUMERIC> locale category, which is normally
2341 kept set by Perl so that the radix is a dot, and the separator is the empty
2342 string, no matter what the underlying locale is supposed to be, and so to get
2343 the expected results, you have to temporarily toggle into the underlying
2344 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2345 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2346 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2347 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2348 (decimal point) character to be a dot.)
2352 The system function it replaces can have its static return buffer trashed,
2353 not only by a subesequent call to that function, but by a C<freelocale>,
2354 C<setlocale>, or other locale change. The returned buffer of this function is
2355 not changed until the next call to it, so the buffer is never in a trashed
2360 Its return buffer is per-thread, so it also is never overwritten by a call to
2361 this function from another thread; unlike the function it replaces.
2365 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2366 as Windows, hence makes your code more portable. Of the fifty-some possible
2367 items specified by the POSIX 2008 standard,
2368 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2369 only one is completely unimplemented, though on non-Windows platforms, another
2370 significant one is also not implemented). It uses various techniques to
2371 recover the other items, including calling C<L<localeconv(3)>>, and
2372 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2373 available. Later C<strftime()> versions have additional capabilities; C<""> is
2374 returned for those not available on your system.
2376 It is important to note that when called with an item that is recovered by
2377 using C<localeconv>, the buffer from any previous explicit call to
2378 C<localeconv> will be overwritten. This means you must save that buffer's
2379 contents if you need to access them after a call to this function. (But note
2380 that you might not want to be using C<localeconv()> directly anyway, because of
2381 issues like the ones listed in the second item of this list (above) for
2382 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2383 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2386 The details for those items which may deviate from what this emulation returns
2387 and what a native C<nl_langinfo()> would return are specified in
2392 When using C<Perl_langinfo> on systems that don't have a native
2393 C<nl_langinfo()>, you must
2395 #include "perl_langinfo.h"
2397 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2398 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2399 C<langinfo.h> would try to import into the namespace for code that doesn't need
2402 The original impetus for C<Perl_langinfo()> was so that code that needs to
2403 find out the current currency symbol, floating point radix character, or digit
2404 grouping separator can use, on all systems, the simpler and more
2405 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2406 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2407 is better to use the methods given in L<perlcall> to call
2408 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2415 #ifdef HAS_NL_LANGINFO
2416 Perl_langinfo(const nl_item item)
2418 Perl_langinfo(const int item)
2421 return my_nl_langinfo(item, TRUE);
2425 #ifdef HAS_NL_LANGINFO
2426 S_my_nl_langinfo(const nl_item item, bool toggle)
2428 S_my_nl_langinfo(const int item, bool toggle)
2432 const char * retval;
2434 #ifdef USE_LOCALE_NUMERIC
2436 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2437 * two items, and only if not already there */
2438 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2439 || PL_numeric_underlying))
2441 #endif /* No toggling needed if not using LC_NUMERIC */
2445 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2446 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2447 || ! defined(HAS_POSIX_2008_LOCALE) \
2448 || ! defined(DUPLOCALE)
2450 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2451 * for those items dependent on it. This must be copied to a buffer before
2452 * switching back, as some systems destroy the buffer when setlocale() is
2456 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2459 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2462 LOCALE_LOCK; /* Prevent interference from another thread executing
2463 this code section (the only call to nl_langinfo in
2467 /* Copy to a per-thread buffer, which is also one that won't be
2468 * destroyed by a subsequent setlocale(), such as the
2469 * RESTORE_LC_NUMERIC may do just below. */
2470 retval = save_to_buffer(nl_langinfo(item),
2471 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2476 RESTORE_LC_NUMERIC();
2480 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2483 bool do_free = FALSE;
2484 locale_t cur = uselocale((locale_t) 0);
2486 if (cur == LC_GLOBAL_LOCALE) {
2487 cur = duplocale(LC_GLOBAL_LOCALE);
2491 # ifdef USE_LOCALE_NUMERIC
2494 if (PL_underlying_numeric_obj) {
2495 cur = PL_underlying_numeric_obj;
2498 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2505 /* We have to save it to a buffer, because the freelocale() just below
2506 * can invalidate the internal one */
2507 retval = save_to_buffer(nl_langinfo_l(item, cur),
2508 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2517 if (strEQ(retval, "")) {
2518 if (item == YESSTR) {
2521 if (item == NOSTR) {
2528 #else /* Below, emulate nl_langinfo as best we can */
2532 # ifdef HAS_LOCALECONV
2534 const struct lconv* lc;
2536 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2538 # ifdef TS_W32_BROKEN_LOCALECONV
2540 const char * save_global;
2541 const char * save_thread;
2549 # ifdef HAS_STRFTIME
2552 bool return_format = FALSE; /* Return the %format, not the value */
2553 const char * format;
2557 /* We copy the results to a per-thread buffer, even if not
2558 * multi-threaded. This is in part to simplify this code, and partly
2559 * because we need a buffer anyway for strftime(), and partly because a
2560 * call of localeconv() could otherwise wipe out the buffer, and the
2561 * programmer would not be expecting this, as this is a nl_langinfo()
2562 * substitute after all, so s/he might be thinking their localeconv()
2563 * is safe until another localeconv() call. */
2568 /* This is unimplemented */
2569 case ERA: /* For use with strftime() %E modifier */
2574 /* We use only an English set, since we don't know any more */
2575 case YESEXPR: return "^[+1yY]";
2576 case YESSTR: return "yes";
2577 case NOEXPR: return "^[-0nN]";
2578 case NOSTR: return "no";
2584 /* On non-windows, this is unimplemented, in part because of
2585 * inconsistencies between vendors. The Darwin native
2586 * nl_langinfo() implementation simply looks at everything past
2587 * any dot in the name, but that doesn't work for other
2588 * vendors. Many Linux locales that don't have UTF-8 in their
2589 * names really are UTF-8, for example; z/OS locales that do
2590 * have UTF-8 in their names, aren't really UTF-8 */
2595 { /* But on Windows, the name does seem to be consistent, so
2600 const char * name = my_setlocale(LC_CTYPE, NULL);
2602 if (isNAME_C_OR_POSIX(name)) {
2603 return "ANSI_X3.4-1968";
2606 /* Find the dot in the locale name */
2607 first = (const char *) strchr(name, '.');
2613 /* Look at everything past the dot */
2618 if (! isDIGIT(*p)) {
2625 /* Here everything past the dot is a digit. Treat it as a
2627 retval = save_to_buffer("CP", &PL_langinfo_buf,
2628 &PL_langinfo_bufsize, 0);
2629 offset = STRLENs("CP");
2633 retval = save_to_buffer(first, &PL_langinfo_buf,
2634 &PL_langinfo_bufsize, offset);
2640 # ifdef HAS_LOCALECONV
2644 /* We don't bother with localeconv_l() because any system that
2645 * has it is likely to also have nl_langinfo() */
2647 LOCALE_LOCK_V; /* Prevent interference with other threads
2648 using localeconv() */
2650 # ifdef TS_W32_BROKEN_LOCALECONV
2652 /* This is a workaround for a Windows bug prior to VS 15.
2653 * What we do here is, while locked, switch to the global
2654 * locale so localeconv() works; then switch back just before
2655 * the unlock. This can screw things up if some thread is
2656 * already using the global locale while assuming no other is.
2657 * A different workaround would be to call GetCurrencyFormat on
2658 * a known value, and parse it; patches welcome
2660 * We have to use LC_ALL instead of LC_MONETARY because of
2661 * another bug in Windows */
2663 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2664 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2665 save_global= savepv(my_setlocale(LC_ALL, NULL));
2666 my_setlocale(LC_ALL, save_thread);
2672 || ! lc->currency_symbol
2673 || strEQ("", lc->currency_symbol))
2679 /* Leave the first spot empty to be filled in below */
2680 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2681 &PL_langinfo_bufsize, 1);
2682 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2683 { /* khw couldn't figure out how the localedef specifications
2684 would show that the $ should replace the radix; this is
2685 just a guess as to how it might work.*/
2686 PL_langinfo_buf[0] = '.';
2688 else if (lc->p_cs_precedes) {
2689 PL_langinfo_buf[0] = '-';
2692 PL_langinfo_buf[0] = '+';
2695 # ifdef TS_W32_BROKEN_LOCALECONV
2697 my_setlocale(LC_ALL, save_global);
2698 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2699 my_setlocale(LC_ALL, save_thread);
2700 Safefree(save_global);
2701 Safefree(save_thread);
2708 # ifdef TS_W32_BROKEN_LOCALECONV
2712 /* For this, we output a known simple floating point number to
2713 * a buffer, and parse it, looking for the radix */
2716 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2719 if (PL_langinfo_bufsize < 10) {
2720 PL_langinfo_bufsize = 10;
2721 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2724 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2726 if (needed_size >= (int) PL_langinfo_bufsize) {
2727 PL_langinfo_bufsize = needed_size + 1;
2728 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2729 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2731 assert(needed_size < (int) PL_langinfo_bufsize);
2734 ptr = PL_langinfo_buf;
2735 e = PL_langinfo_buf + PL_langinfo_bufsize;
2738 while (ptr < e && *ptr != '1') {
2745 while (ptr < e && *ptr != '5') {
2749 /* Everything in between is the radix string */
2751 PL_langinfo_buf[0] = '?';
2752 PL_langinfo_buf[1] = '\0';
2756 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2760 RESTORE_LC_NUMERIC();
2763 retval = PL_langinfo_buf;
2768 case RADIXCHAR: /* No special handling needed */
2775 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2778 LOCALE_LOCK_V; /* Prevent interference with other threads
2779 using localeconv() */
2781 # ifdef TS_W32_BROKEN_LOCALECONV
2783 /* This should only be for the thousands separator. A
2784 * different work around would be to use GetNumberFormat on a
2785 * known value and parse the result to find the separator */
2786 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2787 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2788 save_global = savepv(my_setlocale(LC_ALL, NULL));
2789 my_setlocale(LC_ALL, save_thread);
2791 /* This is the start of code that for broken Windows replaces
2792 * the above and below code, and instead calls
2793 * GetNumberFormat() and then would parse that to find the
2794 * thousands separator. It needs to handle UTF-16 vs -8
2797 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2798 DEBUG_L(PerlIO_printf(Perl_debug_log,
2799 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2800 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2810 temp = (item == RADIXCHAR)
2812 : lc->thousands_sep;
2818 retval = save_to_buffer(temp, &PL_langinfo_buf,
2819 &PL_langinfo_bufsize, 0);
2821 # ifdef TS_W32_BROKEN_LOCALECONV
2823 my_setlocale(LC_ALL, save_global);
2824 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2825 my_setlocale(LC_ALL, save_thread);
2826 Safefree(save_global);
2827 Safefree(save_thread);
2834 RESTORE_LC_NUMERIC();
2840 # ifdef HAS_STRFTIME
2842 /* These are defined by C89, so we assume that strftime supports
2843 * them, and so are returned unconditionally; they may not be what
2844 * the locale actually says, but should give good enough results
2845 * for someone using them as formats (as opposed to trying to parse
2846 * them to figure out what the locale says). The other format
2847 * items are actually tested to verify they work on the platform */
2848 case D_FMT: return "%x";
2849 case T_FMT: return "%X";
2850 case D_T_FMT: return "%c";
2852 /* These formats are only available in later strfmtime's */
2853 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2855 /* The rest can be gotten from most versions of strftime(). */
2856 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2857 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2859 case AM_STR: case PM_STR:
2860 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2861 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2862 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2863 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2864 case DAY_5: case DAY_6: case DAY_7:
2865 case MON_1: case MON_2: case MON_3: case MON_4:
2866 case MON_5: case MON_6: case MON_7: case MON_8:
2867 case MON_9: case MON_10: case MON_11: case MON_12:
2871 init_tm(&tm); /* Precaution against core dumps */
2875 tm.tm_year = 2017 - 1900;
2882 "panic: %s: %d: switch case: %d problem",
2883 __FILE__, __LINE__, item);
2884 NOT_REACHED; /* NOTREACHED */
2886 case PM_STR: tm.tm_hour = 18;
2891 case ABDAY_7: tm.tm_wday++;
2892 case ABDAY_6: tm.tm_wday++;
2893 case ABDAY_5: tm.tm_wday++;
2894 case ABDAY_4: tm.tm_wday++;
2895 case ABDAY_3: tm.tm_wday++;
2896 case ABDAY_2: tm.tm_wday++;
2901 case DAY_7: tm.tm_wday++;
2902 case DAY_6: tm.tm_wday++;
2903 case DAY_5: tm.tm_wday++;
2904 case DAY_4: tm.tm_wday++;
2905 case DAY_3: tm.tm_wday++;
2906 case DAY_2: tm.tm_wday++;
2911 case ABMON_12: tm.tm_mon++;
2912 case ABMON_11: tm.tm_mon++;
2913 case ABMON_10: tm.tm_mon++;
2914 case ABMON_9: tm.tm_mon++;
2915 case ABMON_8: tm.tm_mon++;
2916 case ABMON_7: tm.tm_mon++;
2917 case ABMON_6: tm.tm_mon++;
2918 case ABMON_5: tm.tm_mon++;
2919 case ABMON_4: tm.tm_mon++;
2920 case ABMON_3: tm.tm_mon++;
2921 case ABMON_2: tm.tm_mon++;
2926 case MON_12: tm.tm_mon++;
2927 case MON_11: tm.tm_mon++;
2928 case MON_10: tm.tm_mon++;
2929 case MON_9: tm.tm_mon++;
2930 case MON_8: tm.tm_mon++;
2931 case MON_7: tm.tm_mon++;
2932 case MON_6: tm.tm_mon++;
2933 case MON_5: tm.tm_mon++;
2934 case MON_4: tm.tm_mon++;
2935 case MON_3: tm.tm_mon++;
2936 case MON_2: tm.tm_mon++;
2943 return_format = TRUE;
2948 return_format = TRUE;
2953 return_format = TRUE;
2958 return_format = TRUE;
2963 format = "%Ow"; /* Find the alternate digit for 0 */
2967 /* We can't use my_strftime() because it doesn't look at
2969 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2972 /* A zero return means one of:
2973 * a) there wasn't enough space in PL_langinfo_buf
2974 * b) the format, like a plain %p, returns empty
2975 * c) it was an illegal format, though some
2976 * implementations of strftime will just return the
2977 * illegal format as a plain character sequence.
2979 * To quickly test for case 'b)', try again but precede
2980 * the format with a plain character. If that result is
2981 * still empty, the problem is either 'a)' or 'c)' */
2983 Size_t format_size = strlen(format) + 1;
2984 Size_t mod_size = format_size + 1;
2988 Newx(mod_format, mod_size, char);
2989 Newx(temp_result, PL_langinfo_bufsize, char);
2991 my_strlcpy(mod_format + 1, format, mod_size);
2992 len = strftime(temp_result,
2993 PL_langinfo_bufsize,
2995 Safefree(mod_format);
2996 Safefree(temp_result);
2998 /* If 'len' is non-zero, it means that we had a case like
2999 * %p which means the current locale doesn't use a.m. or
3000 * p.m., and that is valid */
3003 /* Here, still didn't work. If we get well beyond a
3004 * reasonable size, bail out to prevent an infinite
3007 if (PL_langinfo_bufsize > 100 * format_size) {
3008 *PL_langinfo_buf = '\0';
3011 /* Double the buffer size to retry; Add 1 in case
3012 * original was 0, so we aren't stuck at 0. */
3013 PL_langinfo_bufsize *= 2;
3014 PL_langinfo_bufsize++;
3015 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3023 /* Here, we got a result.
3025 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3026 * alternate format for wday 0. If the value is the same as
3027 * the normal 0, there isn't an alternate, so clear the buffer.
3029 if ( item == ALT_DIGITS
3030 && strEQ(PL_langinfo_buf, "0"))
3032 *PL_langinfo_buf = '\0';
3035 /* ALT_DIGITS is problematic. Experiments on it showed that
3036 * strftime() did not always work properly when going from
3037 * alt-9 to alt-10. Only a few locales have this item defined,
3038 * and in all of them on Linux that khw was able to find,
3039 * nl_langinfo() merely returned the alt-0 character, possibly
3040 * doubled. Most Unicode digits are in blocks of 10
3041 * consecutive code points, so that is sufficient information
3042 * for those scripts, as we can infer alt-1, alt-2, .... But
3043 * for a Japanese locale, a CJK ideographic 0 is returned, and
3044 * the CJK digits are not in code point order, so you can't
3045 * really infer anything. The localedef for this locale did
3046 * specify the succeeding digits, so that strftime() works
3047 * properly on them, without needing to infer anything. But
3048 * the nl_langinfo() return did not give sufficient information
3049 * for the caller to understand what's going on. So until
3050 * there is evidence that it should work differently, this
3051 * returns the alt-0 string for ALT_DIGITS.
3053 * wday was chosen because its range is all a single digit.
3054 * Things like tm_sec have two digits as the minimum: '00' */
3058 retval = PL_langinfo_buf;
3060 /* If to return the format, not the value, overwrite the buffer
3061 * with it. But some strftime()s will keep the original format
3062 * if illegal, so change those to "" */
3063 if (return_format) {
3064 if (strEQ(PL_langinfo_buf, format)) {
3065 *PL_langinfo_buf = '\0';
3068 retval = save_to_buffer(format, &PL_langinfo_buf,
3069 &PL_langinfo_bufsize, 0);
3087 * Initialize locale awareness.
3090 Perl_init_i18nl10n(pTHX_ int printwarn)
3094 * 0 if not to output warning when setup locale is bad
3095 * 1 if to output warning based on value of PERL_BADLANG
3096 * >1 if to output regardless of PERL_BADLANG
3099 * 1 = set ok or not applicable,
3100 * 0 = fallback to a locale of lower priority
3101 * -1 = fallback to all locales failed, not even to the C locale
3103 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3104 * set, debugging information is output.
3106 * This looks more complicated than it is, mainly due to the #ifdefs.
3108 * We try to set LC_ALL to the value determined by the environment. If
3109 * there is no LC_ALL on this platform, we try the individual categories we
3110 * know about. If this works, we are done.
3112 * But if it doesn't work, we have to do something else. We search the
3113 * environment variables ourselves instead of relying on the system to do
3114 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3115 * think there is one), and the ultimate fallback "C". This is all done in
3116 * the same loop as above to avoid duplicating code, but it makes things
3117 * more complex. The 'trial_locales' array is initialized with just one
3118 * element; it causes the behavior described in the paragraph above this to
3119 * happen. If that fails, we add elements to 'trial_locales', and do extra
3120 * loop iterations to cause the behavior described in this paragraph.
3122 * On Ultrix, the locale MUST come from the environment, so there is
3123 * preliminary code to set it. I (khw) am not sure that it is necessary,
3124 * and that this couldn't be folded into the loop, but barring any real
3125 * platforms to test on, it's staying as-is
3127 * A slight complication is that in embedded Perls, the locale may already
3128 * be set-up, and we don't want to get it from the normal environment
3129 * variables. This is handled by having a special environment variable
3130 * indicate we're in this situation. We simply set setlocale's 2nd
3131 * parameter to be a NULL instead of "". That indicates to setlocale that
3132 * it is not to change anything, but to return the current value,
3133 * effectively initializing perl's db to what the locale already is.
3135 * We play the same trick with NULL if a LC_ALL succeeds. We call
3136 * setlocale() on the individual categores with NULL to get their existing
3137 * values for our db, instead of trying to change them.
3144 PERL_UNUSED_ARG(printwarn);
3146 #else /* USE_LOCALE */
3149 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3153 /* NULL uses the existing already set up locale */
3154 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3157 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3158 unsigned int trial_locales_count;
3159 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3160 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3161 bool setlocale_failure = FALSE;
3164 /* A later getenv() could zap this, so only use here */
3165 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3167 const bool locwarn = (printwarn > 1
3169 && ( ! bad_lang_use_once
3171 /* disallow with "" or "0" */
3173 && strNE("0", bad_lang_use_once)))));
3175 /* setlocale() return vals; not copied so must be looked at immediately */
3176 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3178 /* current locale for given category; should have been copied so aren't
3180 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3184 /* In some systems you can find out the system default locale
3185 * and use that as the fallback locale. */
3186 # define SYSTEM_DEFAULT_LOCALE
3188 # ifdef SYSTEM_DEFAULT_LOCALE
3190 const char *system_default_locale = NULL;
3195 # define DEBUG_LOCALE_INIT(a,b,c)
3198 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3200 # define DEBUG_LOCALE_INIT(category, locale, result) \
3202 if (debug_initialization) { \
3203 PerlIO_printf(Perl_debug_log, \
3205 __FILE__, __LINE__, \
3206 setlocale_debug_string(category, \
3212 /* Make sure the parallel arrays are properly set up */
3213 # ifdef USE_LOCALE_NUMERIC
3214 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3215 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3216 # ifdef USE_POSIX_2008_LOCALE
3217 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3220 # ifdef USE_LOCALE_CTYPE
3221 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3222 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3223 # ifdef USE_POSIX_2008_LOCALE
3224 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3227 # ifdef USE_LOCALE_COLLATE
3228 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3229 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3230 # ifdef USE_POSIX_2008_LOCALE
3231 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3234 # ifdef USE_LOCALE_TIME
3235 assert(categories[LC_TIME_INDEX] == LC_TIME);
3236 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3237 # ifdef USE_POSIX_2008_LOCALE
3238 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3241 # ifdef USE_LOCALE_MESSAGES
3242 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3243 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3244 # ifdef USE_POSIX_2008_LOCALE
3245 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3248 # ifdef USE_LOCALE_MONETARY
3249 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3250 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3251 # ifdef USE_POSIX_2008_LOCALE
3252 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3255 # ifdef USE_LOCALE_ADDRESS
3256 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3257 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3258 # ifdef USE_POSIX_2008_LOCALE
3259 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3262 # ifdef USE_LOCALE_IDENTIFICATION
3263 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3264 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3265 # ifdef USE_POSIX_2008_LOCALE
3266 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3269 # ifdef USE_LOCALE_MEASUREMENT
3270 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3271 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3272 # ifdef USE_POSIX_2008_LOCALE
3273 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3276 # ifdef USE_LOCALE_PAPER
3277 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3278 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3279 # ifdef USE_POSIX_2008_LOCALE
3280 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3283 # ifdef USE_LOCALE_TELEPHONE
3284 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3285 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3286 # ifdef USE_POSIX_2008_LOCALE
3287 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3291 assert(categories[LC_ALL_INDEX] == LC_ALL);
3292 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3293 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3294 # ifdef USE_POSIX_2008_LOCALE
3295 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3298 # endif /* DEBUGGING */
3300 /* Initialize the cache of the program's UTF-8ness for the always known
3301 * locales C and POSIX */
3302 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3303 sizeof(PL_locale_utf8ness));
3305 # ifdef USE_THREAD_SAFE_LOCALE
3308 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3312 # ifdef USE_POSIX_2008_LOCALE
3314 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3315 if (! PL_C_locale_obj) {
3316 Perl_croak_nocontext(
3317 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3319 if (DEBUG_Lv_TEST || debug_initialization) {
3320 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3325 # ifdef USE_LOCALE_NUMERIC
3327 PL_numeric_radix_sv = newSVpvs(".");
3331 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3333 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3334 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3337 # ifdef LOCALE_ENVIRON_REQUIRED
3340 * Ultrix setlocale(..., "") fails if there are no environment
3341 * variables from which to get a locale name.
3345 # error Ultrix without LC_ALL not implemented
3351 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3352 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3353 if (sl_result[LC_ALL_INDEX])
3356 setlocale_failure = TRUE;
3358 if (! setlocale_failure) {
3359 const char * locale_param;
3360 for (i = 0; i < LC_ALL_INDEX; i++) {
3361 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3364 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3365 if (! sl_result[i]) {
3366 setlocale_failure = TRUE;
3368 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3373 # endif /* LC_ALL */
3374 # endif /* LOCALE_ENVIRON_REQUIRED */
3376 /* We try each locale in the list until we get one that works, or exhaust
3377 * the list. Normally the loop is executed just once. But if setting the
3378 * locale fails, inside the loop we add fallback trials to the array and so
3379 * will execute the loop multiple times */
3380 trial_locales[0] = setlocale_init;
3381 trial_locales_count = 1;
3383 for (i= 0; i < trial_locales_count; i++) {
3384 const char * trial_locale = trial_locales[i];
3388 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3389 * when i==0, but I (khw) don't think that behavior makes much
3391 setlocale_failure = FALSE;
3393 # ifdef SYSTEM_DEFAULT_LOCALE
3394 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3396 /* On Windows machines, an entry of "" after the 0th means to use
3397 * the system default locale, which we now proceed to get. */
3398 if (strEQ(trial_locale, "")) {
3401 /* Note that this may change the locale, but we are going to do
3402 * that anyway just below */
3403 system_default_locale = do_setlocale_c(LC_ALL, "");
3404 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3406 /* Skip if invalid or if it's already on the list of locales to
3408 if (! system_default_locale) {
3409 goto next_iteration;
3411 for (j = 0; j < trial_locales_count; j++) {
3412 if (strEQ(system_default_locale, trial_locales[j])) {
3413 goto next_iteration;
3417 trial_locale = system_default_locale;
3420 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3422 # endif /* SYSTEM_DEFAULT_LOCALE */
3428 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3429 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3430 if (! sl_result[LC_ALL_INDEX]) {
3431 setlocale_failure = TRUE;
3434 /* Since LC_ALL succeeded, it should have changed all the other
3435 * categories it can to its value; so we massage things so that the
3436 * setlocales below just return their category's current values.
3437 * This adequately handles the case in NetBSD where LC_COLLATE may
3438 * not be defined for a locale, and setting it individually will
3439 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3440 * the POSIX locale. */
3441 trial_locale = NULL;
3444 # endif /* LC_ALL */
3446 if (! setlocale_failure) {
3448 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3450 = savepv(do_setlocale_r(categories[j], trial_locale));
3451 if (! curlocales[j]) {
3452 setlocale_failure = TRUE;
3454 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3457 if (! setlocale_failure) { /* All succeeded */
3458 break; /* Exit trial_locales loop */
3462 /* Here, something failed; will need to try a fallback. */
3468 if (locwarn) { /* Output failure info only on the first one */
3472 PerlIO_printf(Perl_error_log,
3473 "perl: warning: Setting locale failed.\n");
3475 # else /* !LC_ALL */
3477 PerlIO_printf(Perl_error_log,
3478 "perl: warning: Setting locale failed for the categories:\n\t");
3480 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3481 if (! curlocales[j]) {
3482 PerlIO_printf(Perl_error_log, category_names[j]);
3485 Safefree(curlocales[j]);
3489 # endif /* LC_ALL */
3491 PerlIO_printf(Perl_error_log,
3492 "perl: warning: Please check that your locale settings:\n");
3496 PerlIO_printf(Perl_error_log,
3497 "\tLANGUAGE = %c%s%c,\n",
3498 language ? '"' : '(',
3499 language ? language : "unset",
3500 language ? '"' : ')');
3503 PerlIO_printf(Perl_error_log,
3504 "\tLC_ALL = %c%s%c,\n",
3506 lc_all ? lc_all : "unset",
3507 lc_all ? '"' : ')');
3509 # if defined(USE_ENVIRON_ARRAY)
3514 /* Look through the environment for any variables of the
3515 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3516 * already handled above. These are assumed to be locale
3517 * settings. Output them and their values. */
3518 for (e = environ; *e; e++) {
3519 const STRLEN prefix_len = sizeof("LC_") - 1;
3522 if ( strBEGINs(*e, "LC_")
3523 && ! strBEGINs(*e, "LC_ALL=")
3524 && (uppers_len = strspn(*e + prefix_len,
3525 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3526 && ((*e)[prefix_len + uppers_len] == '='))
3528 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3529 (int) (prefix_len + uppers_len), *e,
3530 *e + prefix_len + uppers_len + 1);
3537 PerlIO_printf(Perl_error_log,
3538 "\t(possibly more locale environment variables)\n");
3542 PerlIO_printf(Perl_error_log,
3543 "\tLANG = %c%s%c\n",
3545 lang ? lang : "unset",
3548 PerlIO_printf(Perl_error_log,
3549 " are supported and installed on your system.\n");
3552 /* Calculate what fallback locales to try. We have avoided this
3553 * until we have to, because failure is quite unlikely. This will
3554 * usually change the upper bound of the loop we are in.
3556 * Since the system's default way of setting the locale has not
3557 * found one that works, We use Perl's defined ordering: LC_ALL,
3558 * LANG, and the C locale. We don't try the same locale twice, so
3559 * don't add to the list if already there. (On POSIX systems, the
3560 * LC_ALL element will likely be a repeat of the 0th element "",
3561 * but there's no harm done by doing it explicitly.
3563 * Note that this tries the LC_ALL environment variable even on
3564 * systems which have no LC_ALL locale setting. This may or may
3565 * not have been originally intentional, but there's no real need
3566 * to change the behavior. */
3568 for (j = 0; j < trial_locales_count; j++) {
3569 if (strEQ(lc_all, trial_locales[j])) {
3573 trial_locales[trial_locales_count++] = lc_all;
3578 for (j = 0; j < trial_locales_count; j++) {
3579 if (strEQ(lang, trial_locales[j])) {
3583 trial_locales[trial_locales_count++] = lang;
3587 # if defined(WIN32) && defined(LC_ALL)
3589 /* For Windows, we also try the system default locale before "C".
3590 * (If there exists a Windows without LC_ALL we skip this because
3591 * it gets too complicated. For those, the "C" is the next
3592 * fallback possibility). The "" is the same as the 0th element of
3593 * the array, but the code at the loop above knows to treat it
3594 * differently when not the 0th */
3595 trial_locales[trial_locales_count++] = "";
3599 for (j = 0; j < trial_locales_count; j++) {
3600 if (strEQ("C", trial_locales[j])) {
3604 trial_locales[trial_locales_count++] = "C";
3607 } /* end of first time through the loop */
3615 } /* end of looping through the trial locales */
3617 if (ok < 1) { /* If we tried to fallback */
3619 if (! setlocale_failure) { /* fallback succeeded */
3620 msg = "Falling back to";
3622 else { /* fallback failed */
3625 /* We dropped off the end of the loop, so have to decrement i to
3626 * get back to the value the last time through */
3630 msg = "Failed to fall back to";
3632 /* To continue, we should use whatever values we've got */
3634 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3635 Safefree(curlocales[j]);
3636 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3637 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3642 const char * description;
3643 const char * name = "";
3644 if (strEQ(trial_locales[i], "C")) {
3645 description = "the standard locale";
3649 # ifdef SYSTEM_DEFAULT_LOCALE
3651 else if (strEQ(trial_locales[i], "")) {
3652 description = "the system default locale";
3653 if (system_default_locale) {
3654 name = system_default_locale;
3658 # endif /* SYSTEM_DEFAULT_LOCALE */
3661 description = "a fallback locale";
3662 name = trial_locales[i];
3664 if (name && strNE(name, "")) {
3665 PerlIO_printf(Perl_error_log,
3666 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3669 PerlIO_printf(Perl_error_log,
3670 "perl: warning: %s %s.\n", msg, description);
3673 } /* End of tried to fallback */
3675 /* Done with finding the locales; update our records */
3677 # ifdef USE_LOCALE_CTYPE
3679 new_ctype(curlocales[LC_CTYPE_INDEX]);
3682 # ifdef USE_LOCALE_COLLATE
3684 new_collate(curlocales[LC_COLLATE_INDEX]);
3687 # ifdef USE_LOCALE_NUMERIC
3689 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3693 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3695 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3697 /* This caches whether each category's locale is UTF-8 or not. This
3698 * may involve changing the locale. It is ok to do this at
3699 * initialization time before any threads have started, but not later
3700 * unless thread-safe operations are used.
3701 * Caching means that if the program heeds our dictate not to change
3702 * locales in threaded applications, this data will remain valid, and
3703 * it may get queried without having to change locales. If the
3704 * environment is such that all categories have the same locale, this
3705 * isn't needed, as the code will not change the locale; but this
3706 * handles the uncommon case where the environment has disparate
3707 * locales for the categories */
3708 (void) _is_cur_LC_category_utf8(categories[i]);
3712 Safefree(curlocales[i]);
3715 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3717 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3718 * locale is UTF-8. The call to new_ctype() just above has already
3719 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3720 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3721 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3722 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3723 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3725 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3726 This is an alternative to using the -C command line switch
3727 (the -C if present will override this). */
3729 const char *p = PerlEnv_getenv("PERL_UNICODE");
3730 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3731 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3745 #endif /* USE_LOCALE */
3748 /* So won't continue to output stuff */
3749 DEBUG_INITIALIZATION_set(FALSE);
3756 #ifdef USE_LOCALE_COLLATE
3759 Perl__mem_collxfrm(pTHX_ const char *input_string,
3760 STRLEN len, /* Length of 'input_string' */
3761 STRLEN *xlen, /* Set to length of returned string
3762 (not including the collation index
3764 bool utf8 /* Is the input in UTF-8? */
3768 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3769 * differences. First, it handles embedded NULs. Second, it allocates a bit
3770 * more memory than needed for the transformed data itself. The real
3771 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3772 * the length of that, and doesn't include the collation index size.
3773 * Please see sv_collxfrm() to see how this is used. */
3775 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3777 char * s = (char *) input_string;
3778 STRLEN s_strlen = strlen(input_string);
3780 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3781 STRLEN length_in_chars;
3782 bool first_time = TRUE; /* Cleared after first loop iteration */
3784 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3786 /* Must be NUL-terminated */
3787 assert(*(input_string + len) == '\0');
3789 /* If this locale has defective collation, skip */
3790 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3791 DEBUG_L(PerlIO_printf(Perl_debug_log,
3792 "_mem_collxfrm: locale's collation is defective\n"));
3796 /* Replace any embedded NULs with the control that sorts before any others.
3797 * This will give as good as possible results on strings that don't
3798 * otherwise contain that character, but otherwise there may be
3799 * less-than-perfect results with that character and NUL. This is
3800 * unavoidable unless we replace strxfrm with our own implementation. */
3801 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3805 STRLEN sans_nuls_len;
3806 int try_non_controls;
3807 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3808 making sure 2nd byte is NUL.
3810 STRLEN this_replacement_len;
3812 /* If we don't know what non-NUL control character sorts lowest for
3813 * this locale, find it */
3814 if (PL_strxfrm_NUL_replacement == '\0') {
3816 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3817 includes the collation index
3820 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3822 /* Unlikely, but it may be that no control will work to replace
3823 * NUL, in which case we instead look for any character. Controls
3824 * are preferred because collation order is, in general, context
3825 * sensitive, with adjoining characters affecting the order, and
3826 * controls are less likely to have such interactions, allowing the
3827 * NUL-replacement to stand on its own. (Another way to look at it
3828 * is to imagine what would happen if the NUL were replaced by a
3829 * combining character; it wouldn't work out all that well.) */
3830 for (try_non_controls = 0;
3831 try_non_controls < 2;
3834 /* Look through all legal code points (NUL isn't) */
3835 for (j = 1; j < 256; j++) {
3836 char * x; /* j's xfrm plus collation index */
3837 STRLEN x_len; /* length of 'x' */
3838 STRLEN trial_len = 1;
3839 char cur_source[] = { '\0', '\0' };
3841 /* Skip non-controls the first time through the loop. The
3842 * controls in a UTF-8 locale are the L1 ones */
3843 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3850 /* Create a 1-char string of the current code point */
3851 cur_source[0] = (char) j;
3853 /* Then transform it */
3854 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3855 0 /* The string is not in UTF-8 */);
3857 /* Ignore any character that didn't successfully transform.
3863 /* If this character's transformation is lower than
3864 * the current lowest, this one becomes the lowest */
3865 if ( cur_min_x == NULL
3866 || strLT(x + COLLXFRM_HDR_LEN,
3867 cur_min_x + COLLXFRM_HDR_LEN))
3869 PL_strxfrm_NUL_replacement = j;
3875 } /* end of loop through all 255 characters */
3877 /* Stop looking if found */
3882 /* Unlikely, but possible, if there aren't any controls that
3883 * work in the locale, repeat the loop, looking for any
3884 * character that works */
3885 DEBUG_L(PerlIO_printf(Perl_debug_log,
3886 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3887 } /* End of loop to try first the controls, then any char */
3890 DEBUG_L(PerlIO_printf(Perl_debug_log,
3891 "_mem_collxfrm: Couldn't find any character to replace"
3892 " embedded NULs in locale %s with", PL_collation_name));
3896 DEBUG_L(PerlIO_printf(Perl_debug_log,
3897 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3898 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3900 Safefree(cur_min_x);
3901 } /* End of determining the character that is to replace NULs */
3903 /* If the replacement is variant under UTF-8, it must match the
3904 * UTF8-ness of the original */
3905 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3906 this_replacement_char[0] =
3907 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3908 this_replacement_char[1] =
3909 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3910 this_replacement_len = 2;
3913 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3914 /* this_replacement_char[1] = '\0' was done at initialization */
3915 this_replacement_len = 1;
3918 /* The worst case length for the replaced string would be if every
3919 * character in it is NUL. Multiply that by the length of each
3920 * replacement, and allow for a trailing NUL */
3921 sans_nuls_len = (len * this_replacement_len) + 1;
3922 Newx(sans_nuls, sans_nuls_len, char);
3925 /* Replace each NUL with the lowest collating control. Loop until have
3926 * exhausted all the NULs */
3927 while (s + s_strlen < e) {
3928 my_strlcat(sans_nuls, s, sans_nuls_len);
3930 /* Do the actual replacement */
3931 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3933 /* Move past the input NUL */
3935 s_strlen = strlen(s);
3938 /* And add anything that trails the final NUL */
3939 my_strlcat(sans_nuls, s, sans_nuls_len);
3941 /* Switch so below we transform this modified string */
3944 } /* End of replacing NULs */
3946 /* Make sure the UTF8ness of the string and locale match */
3947 if (utf8 != PL_in_utf8_COLLATE_locale) {
3948 /* XXX convert above Unicode to 10FFFF? */
3949 const char * const t = s; /* Temporary so we can later find where the
3952 /* Here they don't match. Change the string's to be what the locale is
3955 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3956 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3959 else { /* locale is not UTF-8; but input is; downgrade the input */
3961 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3963 /* If the downgrade was successful we are done, but if the input
3964 * contains things that require UTF-8 to represent, have to do
3965 * damage control ... */
3966 if (UNLIKELY(utf8)) {
3968 /* What we do is construct a non-UTF-8 string with
3969 * 1) the characters representable by a single byte converted
3970 * to be so (if necessary);
3971 * 2) and the rest converted to collate the same as the
3972 * highest collating representable character. That makes
3973 * them collate at the end. This is similar to how we
3974 * handle embedded NULs, but we use the highest collating
3975 * code point instead of the smallest. Like the NUL case,
3976 * this isn't perfect, but is the best we can reasonably
3977 * do. Every above-255 code point will sort the same as
3978 * the highest-sorting 0-255 code point. If that code
3979 * point can combine in a sequence with some other code
3980 * points for weight calculations, us changing something to
3981 * be it can adversely affect the results. But in most
3982 * cases, it should work reasonably. And note that this is
3983 * really an illegal situation: using code points above 255
3984 * on a locale where only 0-255 are valid. If two strings
3985 * sort entirely equal, then the sort order for the
3986 * above-255 code points will be in code point order. */
3990 /* If we haven't calculated the code point with the maximum
3991 * collating order for this locale, do so now */
3992 if (! PL_strxfrm_max_cp) {
3995 /* The current transformed string that collates the
3996 * highest (except it also includes the prefixed collation
3998 char * cur_max_x = NULL;
4000 /* Look through all legal code points (NUL isn't) */
4001 for (j = 1; j < 256; j++) {
4004 char cur_source[] = { '\0', '\0' };
4006 /* Create a 1-char string of the current code point */
4007 cur_source[0] = (char) j;
4009 /* Then transform it */
4010 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4012 /* If something went wrong (which it shouldn't), just
4013 * ignore this code point */
4018 /* If this character's transformation is higher than
4019 * the current highest, this one becomes the highest */
4020 if ( cur_max_x == NULL
4021 || strGT(x + COLLXFRM_HDR_LEN,
4022 cur_max_x + COLLXFRM_HDR_LEN))
4024 PL_strxfrm_max_cp = j;
4033 DEBUG_L(PerlIO_printf(Perl_debug_log,
4034 "_mem_collxfrm: Couldn't find any character to"
4035 " replace above-Latin1 chars in locale %s with",
4036 PL_collation_name));
4040 DEBUG_L(PerlIO_printf(Perl_debug_log,
4041 "_mem_collxfrm: highest 1-byte collating character"
4042 " in locale %s is 0x%02X\n",
4044 PL_strxfrm_max_cp));
4046 Safefree(cur_max_x);
4049 /* Here we know which legal code point collates the highest.
4050 * We are ready to construct the non-UTF-8 string. The length
4051 * will be at least 1 byte smaller than the input string
4052 * (because we changed at least one 2-byte character into a
4053 * single byte), but that is eaten up by the trailing NUL */
4059 char * e = (char *) t + len;
4061 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4063 if (UTF8_IS_INVARIANT(cur_char)) {
4066 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4067 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4069 else { /* Replace illegal cp with highest collating
4071 s[d++] = PL_strxfrm_max_cp;
4075 Renew(s, d, char); /* Free up unused space */
4080 /* Here, we have constructed a modified version of the input. It could
4081 * be that we already had a modified copy before we did this version.
4082 * If so, that copy is no longer needed */
4083 if (t != input_string) {
4088 length_in_chars = (utf8)
4089 ? utf8_length((U8 *) s, (U8 *) s + len)
4092 /* The first element in the output is the collation id, used by
4093 * sv_collxfrm(); then comes the space for the transformed string. The
4094 * equation should give us a good estimate as to how much is needed */
4095 xAlloc = COLLXFRM_HDR_LEN
4097 + (PL_collxfrm_mult * length_in_chars);
4098 Newx(xbuf, xAlloc, char);
4099 if (UNLIKELY(! xbuf)) {
4100 DEBUG_L(PerlIO_printf(Perl_debug_log,
4101 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4105 /* Store the collation id */
4106 *(U32*)xbuf = PL_collation_ix;
4108 /* Then the transformation of the input. We loop until successful, or we
4112 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4114 /* If the transformed string occupies less space than we told strxfrm()
4115 * was available, it means it successfully transformed the whole
4117 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4119 /* Some systems include a trailing NUL in the returned length.
4120 * Ignore it, using a loop in case multiple trailing NULs are
4123 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4128 /* If the first try didn't get it, it means our prediction was low.
4129 * Modify the coefficients so that we predict a larger value in any
4130 * future transformations */
4132 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4133 STRLEN computed_guess = PL_collxfrm_base
4134 + (PL_collxfrm_mult * length_in_chars);
4136 /* On zero-length input, just keep current slope instead of
4138 const STRLEN new_m = (length_in_chars != 0)
4139 ? needed / length_in_chars
4142 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4143 "%s: %d: initial size of %zu bytes for a length "
4144 "%zu string was insufficient, %zu needed\n",
4146 computed_guess, length_in_chars, needed));
4148 /* If slope increased, use it, but discard this result for
4149 * length 1 strings, as we can't be sure that it's a real slope
4151 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4155 STRLEN old_m = PL_collxfrm_mult;
4156 STRLEN old_b = PL_collxfrm_base;
4160 PL_collxfrm_mult = new_m;
4161 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4162 computed_guess = PL_collxfrm_base
4163 + (PL_collxfrm_mult * length_in_chars);
4164 if (computed_guess < needed) {
4165 PL_collxfrm_base += needed - computed_guess;
4168 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4169 "%s: %d: slope is now %zu; was %zu, base "
4170 "is now %zu; was %zu\n",
4172 PL_collxfrm_mult, old_m,
4173 PL_collxfrm_base, old_b));
4175 else { /* Slope didn't change, but 'b' did */
4176 const STRLEN new_b = needed
4179 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4180 "%s: %d: base is now %zu; was %zu\n",
4182 new_b, PL_collxfrm_base));
4183 PL_collxfrm_base = new_b;
4190 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4191 DEBUG_L(PerlIO_printf(Perl_debug_log,
4192 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4193 *xlen, PERL_INT_MAX));
4197 /* A well-behaved strxfrm() returns exactly how much space it needs
4198 * (usually not including the trailing NUL) when it fails due to not
4199 * enough space being provided. Assume that this is the case unless
4200 * it's been proven otherwise */
4201 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4202 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4204 else { /* Here, either:
4205 * 1) The strxfrm() has previously shown bad behavior; or
4206 * 2) It isn't the first time through the loop, which means
4207 * that the strxfrm() is now showing bad behavior, because
4208 * we gave it what it said was needed in the previous
4209 * iteration, and it came back saying it needed still more.
4210 * (Many versions of cygwin fit this. When the buffer size
4211 * isn't sufficient, they return the input size instead of
4212 * how much is needed.)
4213 * Increase the buffer size by a fixed percentage and try again.
4215 xAlloc += (xAlloc / 4) + 1;
4216 PL_strxfrm_is_behaved = FALSE;
4220 if (DEBUG_Lv_TEST || debug_initialization) {
4221 PerlIO_printf(Perl_debug_log,
4222 "_mem_collxfrm required more space than previously calculated"
4223 " for locale %s, trying again with new guess=%d+%zu\n",
4224 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4225 xAlloc - COLLXFRM_HDR_LEN);
4232 Renew(xbuf, xAlloc, char);
4233 if (UNLIKELY(! xbuf)) {
4234 DEBUG_L(PerlIO_printf(Perl_debug_log,
4235 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4245 if (DEBUG_Lv_TEST || debug_initialization) {
4247 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4248 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4249 PerlIO_printf(Perl_debug_log, "%s\n",
4250 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4256 /* Free up unneeded space; retain ehough for trailing NUL */
4257 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4259 if (s != input_string) {
4267 if (s != input_string) {
4274 if (DEBUG_Lv_TEST || debug_initialization) {
4275 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4286 S_print_collxfrm_input_and_return(pTHX_
4287 const char * const s,
4288 const char * const e,
4289 const STRLEN * const xlen,
4293 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4295 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4296 (UV)PL_collation_ix);
4298 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4301 PerlIO_printf(Perl_debug_log, "NULL");
4303 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4305 print_bytes_for_locale(s, e, is_utf8);
4307 PerlIO_printf(Perl_debug_log, "'\n");
4310 # endif /* DEBUGGING */
4311 #endif /* USE_LOCALE_COLLATE */
4315 S_print_bytes_for_locale(pTHX_
4316 const char * const s,
4317 const char * const e,
4321 bool prev_was_printable = TRUE;
4322 bool first_time = TRUE;
4324 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4328 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4331 if (! prev_was_printable) {
4332 PerlIO_printf(Perl_debug_log, " ");
4334 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4335 prev_was_printable = TRUE;
4339 PerlIO_printf(Perl_debug_log, " ");
4341 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4342 prev_was_printable = FALSE;
4344 t += (is_utf8) ? UTF8SKIP(t) : 1;
4349 #endif /* #ifdef DEBUGGING */
4354 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4356 /* Changes the locale for LC_'switch_category" to that of
4357 * LC_'template_category', if they aren't already the same. If not NULL,
4358 * 'template_locale' is the locale that 'template_category' is in.
4360 * Returns a copy of the name of the original locale for 'switch_category'
4361 * so can be switched back to with the companion function
4362 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4364 char * restore_to_locale = NULL;
4366 if (switch_category == template_category) { /* No changes needed */
4370 /* Find the original locale of the category we may need to change, so that
4371 * it can be restored to later */
4372 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4374 if (! restore_to_locale) {
4376 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4377 __FILE__, __LINE__, category_name(switch_category), errno);
4380 /* If the locale of the template category wasn't passed in, find it now */
4381 if (template_locale == NULL) {
4382 template_locale = do_setlocale_r(template_category, NULL);
4383 if (! template_locale) {
4385 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4386 __FILE__, __LINE__, category_name(template_category), errno);
4390 /* It the locales are the same, there's nothing to do */
4391 if (strEQ(restore_to_locale, template_locale)) {
4392 Safefree(restore_to_locale);
4394 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4395 category_name(switch_category), restore_to_locale));
4400 /* Finally, change the locale to the template one */
4401 if (! do_setlocale_r(switch_category, template_locale)) {
4403 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4404 __FILE__, __LINE__, category_name(switch_category),
4405 template_locale, errno);
4408 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4409 category_name(switch_category), template_locale));
4411 return restore_to_locale;
4415 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4417 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4418 * copy that will be freed by this function), or do nothing if the latter
4419 * parameter is NULL */
4421 if (original_locale == NULL) {
4425 if (! do_setlocale_r(category, original_locale)) {
4427 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4429 category_name(category), original_locale, errno);
4432 Safefree(original_locale);
4435 /* is_cur_LC_category_utf8 uses a small char buffer to avoid malloc/free */
4436 #define CUR_LC_BUFFER_SIZE 64
4439 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4441 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4442 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4443 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4444 * could give the wrong result. The result will very likely be correct for
4445 * languages that have commonly used non-ASCII characters, but for notably
4446 * English, it comes down to if the locale's name ends in something like
4447 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4449 * If the platform is early C89, not containing mbtowc(), or we are
4450 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4451 * These work very well for non-Latin locales or those whose currency
4452 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4453 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4454 * on the name containing UTF-8 or not. */
4456 /* Name of current locale corresponding to the input category */
4457 const char *save_input_locale = NULL;
4459 bool is_utf8 = FALSE; /* The return value */
4461 /* The variables below are for the cache of previous lookups using this
4462 * function. The cache is a C string, described at the definition for
4463 * 'C_and_POSIX_utf8ness'.
4465 * The first part of the cache is fixed, for the C and POSIX locales. The
4466 * varying part starts just after them. */
4467 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4469 Size_t utf8ness_cache_size; /* Size of the varying portion */
4470 Size_t input_name_len; /* Length in bytes of save_input_locale */
4471 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4472 the name in the cache */
4473 char * delimited; /* The name plus the delimiters used to store
4475 char buffer[CUR_LC_BUFFER_SIZE]; /* small buffer */
4476 char * name_pos; /* position of 'delimited' in the cache, or 0
4482 assert(category != LC_ALL);
4486 /* Get the desired category's locale */
4487 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4488 if (! save_input_locale) {
4490 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4491 __FILE__, __LINE__, category_name(category), errno);
4494 DEBUG_L(PerlIO_printf(Perl_debug_log,
4495 "Current locale for %s is %s\n",
4496 category_name(category), save_input_locale));
4498 input_name_len = strlen(save_input_locale);
4500 /* In our cache, each name is accompanied by two delimiters and a single
4502 input_name_len_with_overhead = input_name_len + 3;
4504 if ( input_name_len_with_overhead <= CUR_LC_BUFFER_SIZE ) {
4505 /* we can use the buffer, avoid a malloc */
4507 } else { /* need a malloc */
4508 /* Allocate and populate space for a copy of the name surrounded by the
4510 Newx(delimited, input_name_len_with_overhead, char);
4513 delimited[0] = UTF8NESS_SEP[0];
4514 Copy(save_input_locale, delimited + 1, input_name_len, char);
4515 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4516 delimited[input_name_len+2] = '\0';
4518 /* And see if that is in the cache */
4519 name_pos = instr(PL_locale_utf8ness, delimited);
4521 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4525 if (DEBUG_Lv_TEST || debug_initialization) {
4526 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4527 save_input_locale, is_utf8);
4532 /* And, if not already in that position, move it to the beginning of
4533 * the non-constant portion of the list, since it is the most recently
4534 * used. (We don't have to worry about overflow, since just moving
4535 * existing names around) */
4536 if (name_pos > utf8ness_cache) {
4537 Move(utf8ness_cache,
4538 utf8ness_cache + input_name_len_with_overhead,
4539 name_pos - utf8ness_cache, char);
4542 input_name_len_with_overhead - 1, char);
4543 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4546 /* free only when not using the buffer */
4547 if ( delimited != buffer ) Safefree(delimited);
4548 Safefree(save_input_locale);
4552 /* Here we don't have stored the utf8ness for the input locale. We have to
4555 # if defined(USE_LOCALE_CTYPE) \
4556 && ( defined(HAS_NL_LANGINFO) \
4557 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4560 const char *original_ctype_locale
4561 = switch_category_locale_to_template(LC_CTYPE,
4565 /* Here the current LC_CTYPE is set to the locale of the category whose
4566 * information is desired. This means that nl_langinfo() and mbtowc()
4567 * should give the correct results */
4569 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4570 calling the functions if we have this */
4572 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4573 * Unicode code point. */
4575 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4576 __FILE__, __LINE__, (int) MB_CUR_MAX));
4577 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
4579 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4580 goto finish_and_return;
4584 # if defined(HAS_NL_LANGINFO)
4586 { /* The task is easiest if the platform has this POSIX 2001 function.
4587 Except on some platforms it can wrongly return "", so have to have
4588 a fallback. And it can return that it's UTF-8, even if there are
4589 variances from that. For example, Turkish locales may use the
4590 alternate dotted I rules, and sometimes it appears to be a
4591 defective locale definition. XXX We should probably check for
4592 these in the Latin1 range and warn (but on glibc, requires
4593 iswalnum() etc. due to their not handling 80-FF correctly */
4594 const char *codeset = my_nl_langinfo(CODESET, FALSE);
4595 /* FALSE => already in dest locale */
4597 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4598 "\tnllanginfo returned CODESET '%s'\n", codeset));
4600 if (codeset && strNE(codeset, "")) {
4602 /* If the implementation of foldEQ() somehow were
4603 * to change to not go byte-by-byte, this could
4604 * read past end of string, as only one length is
4605 * checked. But currently, a premature NUL will
4606 * compare false, and it will stop there */
4607 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
4608 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
4610 DEBUG_L(PerlIO_printf(Perl_debug_log,
4611 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
4613 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4614 goto finish_and_return;
4619 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
4620 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
4621 * late adder to C89, so very likely to have it. However, testing has
4622 * shown that, like nl_langinfo() above, there are locales that are not
4623 * strictly UTF-8 that this will return that they are */
4630 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4636 /* mbrtowc() and mbtowc() convert a byte string to a wide
4637 * character. Feed a byte string to one of them and check that the
4638 * result is the expected Unicode code point */
4640 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4641 /* Prefer this function if available, as it's reentrant */
4643 memset(&ps, 0, sizeof(ps));;
4644 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
4647 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
4653 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
4655 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
4662 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4663 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
4664 len, (unsigned int) wc, GET_ERRNO));
4666 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
4667 && wc == (wchar_t) UNICODE_REPLACEMENT);
4672 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4673 goto finish_and_return;
4678 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
4679 * try looking at the currency symbol to see if it disambiguates
4680 * things. Often that will be in the native script, and if the symbol
4681 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
4682 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
4683 * string being valid UTF-8 are quite small */
4685 # ifdef USE_LOCALE_MONETARY
4687 /* If have LC_MONETARY, we can look at the currency symbol. Often that
4688 * will be in the native script. We do this one first because there is
4689 * just one string to examine, so potentially avoids work */
4692 const char *original_monetary_locale
4693 = switch_category_locale_to_template(LC_MONETARY,
4696 bool only_ascii = FALSE;
4697 const U8 * currency_string
4698 = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE);
4699 /* 2nd param not relevant for this item */
4700 const U8 * first_variant;
4702 assert( *currency_string == '-'
4703 || *currency_string == '+'
4704 || *currency_string == '.');
4708 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
4710 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));
4714 is_utf8 = is_strict_utf8_string(first_variant, 0);
4717 restore_switched_locale(LC_MONETARY, original_monetary_locale);
4721 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
4722 * otherwise assume the locale is UTF-8 if and only if the symbol
4723 * is non-ascii UTF-8. */
4724 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
4725 save_input_locale, is_utf8));
4726 goto finish_and_return;
4730 # endif /* USE_LOCALE_MONETARY */
4731 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
4733 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
4734 * the names of the months and weekdays, timezone, and am/pm indicator */
4736 const char *original_time_locale
4737 = switch_category_locale_to_template(LC_TIME,
4741 bool is_dst = FALSE;
4745 char * formatted_time;
4747 /* Here the current LC_TIME is set to the locale of the category
4748 * whose information is desired. Look at all the days of the week and
4749 * month names, and the timezone and am/pm indicator for UTF-8 variant
4750 * characters. The first such a one found will tell us if the locale
4751 * is UTF-8 or not */
4753 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
4754 formatted_time = my_strftime("%A %B %Z %p",
4755 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
4756 if ( ! formatted_time
4757 || is_utf8_invariant_string((U8 *) formatted_time, 0))
4760 /* Here, we didn't find a non-ASCII. Try the next time through
4761 * with the complemented dst and am/pm, and try with the next
4762 * weekday. After we have gotten all weekdays, try the next
4765 hour = (hour + 12) % 24;
4773 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
4774 * false otherwise. But first, restore LC_TIME to its original
4775 * locale if we changed it */
4776 restore_switched_locale(LC_TIME, original_time_locale);
4778 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
4780 is_utf8_string((U8 *) formatted_time, 0)));
4781 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
4782 goto finish_and_return;
4785 /* Falling off the end of the loop indicates all the names were just
4786 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
4787 * to its original locale */
4788 restore_switched_locale(LC_TIME, original_time_locale);
4789 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));
4794 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
4796 /* This code is ifdefd out because it was found to not be necessary in testing
4797 * on our dromedary test machine, which has over 700 locales. There, this
4798 * added no value to looking at the currency symbol and the time strings. I
4799 * left it in so as to avoid rewriting it if real-world experience indicates
4800 * that dromedary is an outlier. Essentially, instead of returning abpve if we
4801 * haven't found illegal utf8, we continue on and examine all the strerror()
4802 * messages on the platform for utf8ness. If all are ASCII, we still don't
4803 * know the answer; but otherwise we have a pretty good indication of the
4804 * utf8ness. The reason this doesn't help much is that the messages may not
4805 * have been translated into the locale. The currency symbol and time strings
4806 * are much more likely to have been translated. */
4809 bool non_ascii = FALSE;
4810 const char *original_messages_locale
4811 = switch_category_locale_to_template(LC_MESSAGES,
4814 const char * errmsg = NULL;
4816 /* Here the current LC_MESSAGES is set to the locale of the category
4817 * whose information is desired. Look through all the messages. We
4818 * can't use Strerror() here because it may expand to code that
4819 * segfaults in miniperl */
4821 for (e = 0; e <= sys_nerr; e++) {
4823 errmsg = sys_errlist[e];
4824 if (errno || !errmsg) {
4827 errmsg = savepv(errmsg);
4828 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
4830 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
4836 restore_switched_locale(LC_MESSAGES, original_messages_locale);
4840 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
4841 * any non-ascii means it is one; otherwise we assume it isn't */
4842 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
4845 goto finish_and_return;
4848 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));
4852 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
4855 /* As a last resort, look at the locale name to see if it matches
4856 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
4857 * return of setlocale(), is actually defined to be opaque, so we can't
4858 * really rely on the absence of various substrings in the name to indicate
4859 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
4860 * be a UTF-8 locale. Similarly for the other common names */
4863 const Size_t final_pos = strlen(save_input_locale) - 1;
4865 if (final_pos >= 3) {
4866 const char *name = save_input_locale;
4868 /* Find next 'U' or 'u' and look from there */
4869 while ((name += strcspn(name, "Uu") + 1)
4870 <= save_input_locale + final_pos - 2)
4872 if ( isALPHA_FOLD_NE(*name, 't')
4873 || isALPHA_FOLD_NE(*(name + 1), 'f'))
4878 if (*(name) == '-') {
4879 if ((name > save_input_locale + final_pos - 1)) {
4884 if (*(name) == '8') {
4885 DEBUG_L(PerlIO_printf(Perl_debug_log,
4886 "Locale %s ends with UTF-8 in name\n",
4887 save_input_locale));
4889 goto finish_and_return;
4892 DEBUG_L(PerlIO_printf(Perl_debug_log,
4893 "Locale %s doesn't end with UTF-8 in name\n",
4894 save_input_locale));
4899 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
4900 if (memENDs(save_input_locale, final_pos, "65001")) {
4901 DEBUG_L(PerlIO_printf(Perl_debug_log,
4902 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
4903 save_input_locale));
4905 goto finish_and_return;
4912 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
4913 * since we are about to return FALSE anyway, there is no point in doing
4914 * this extra work */
4917 if (instr(save_input_locale, "8859")) {
4918 DEBUG_L(PerlIO_printf(Perl_debug_log,
4919 "Locale %s has 8859 in name, not UTF-8 locale\n",
4920 save_input_locale));
4922 goto finish_and_return;
4926 DEBUG_L(PerlIO_printf(Perl_debug_log,
4927 "Assuming locale %s is not a UTF-8 locale\n",
4928 save_input_locale));
4931 # endif /* the code that is compiled when no modern LC_CTYPE */
4935 /* Cache this result so we don't have to go through all this next time. */
4936 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
4937 - (utf8ness_cache - PL_locale_utf8ness);
4939 /* But we can't save it if it is too large for the total space available */
4940 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
4941 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
4943 /* Here it can fit, but we may need to clear out the oldest cached
4944 * result(s) to do so. Check */
4945 if (utf8ness_cache_len + input_name_len_with_overhead
4946 >= utf8ness_cache_size)
4948 /* Here we have to clear something out to make room for this.
4949 * Start looking at the rightmost place where it could fit and find
4950 * the beginning of the entry that extends past that. */
4951 char * cutoff = (char *) my_memrchr(utf8ness_cache,
4954 - input_name_len_with_overhead);
4957 assert(cutoff >= utf8ness_cache);
4959 /* This and all subsequent entries must be removed */
4961 utf8ness_cache_len = strlen(utf8ness_cache);
4964 /* Make space for the new entry */
4965 Move(utf8ness_cache,
4966 utf8ness_cache + input_name_len_with_overhead,
4967 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
4970 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
4971 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4973 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
4974 & (PERL_UINTMAX_T) ~1) != '0')
4977 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu,"
4978 " inserted_name=%s, its_len=%zu\n",
4980 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
4981 delimited, input_name_len_with_overhead);
4987 if (DEBUG_Lv_TEST) {
4988 const char * s = PL_locale_utf8ness;
4990 /* Audit the structure */
4991 while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) {
4994 if (*s != UTF8NESS_SEP[0]) {
4996 "panic: %s: %d: Corrupt utf8ness_cache: missing"
4997 " separator %.*s<-- HERE %s\n",
4999 (int) (s - PL_locale_utf8ness), PL_locale_utf8ness,
5003 e = strchr(s, UTF8NESS_PREFIX[0]);
5006 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5007 " separator %.*s<-- HERE %s\n",
5009 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5013 if (*e != '0' && *e != '1') {
5015 "panic: %s: %d: Corrupt utf8ness_cache: utf8ness"
5016 " must be [01] %.*s<-- HERE %s\n",
5018 (int) (e + 1 - PL_locale_utf8ness),
5019 PL_locale_utf8ness, e + 1);
5021 if (ninstr(PL_locale_utf8ness, s, s-1, e)) {
5023 "panic: %s: %d: Corrupt utf8ness_cache: entry"
5024 " has duplicate %.*s<-- HERE %s\n",
5026 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5033 if (DEBUG_Lv_TEST || debug_initialization) {
5035 PerlIO_printf(Perl_debug_log,
5036 "PL_locale_utf8ness is now %s; returning %d\n",
5037 PL_locale_utf8ness, is_utf8);
5042 /* free only when not using the buffer */
5043 if ( delimited != buffer ) Safefree(delimited);
5044 Safefree(save_input_locale);
5051 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
5054 /* Internal function which returns if we are in the scope of a pragma that
5055 * enables the locale category 'category'. 'compiling' should indicate if
5056 * this is during the compilation phase (TRUE) or not (FALSE). */
5058 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
5060 SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0);
5061 if (! these_categories || these_categories == &PL_sv_placeholder) {
5065 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
5066 * a valid unsigned */
5067 assert(category >= -1);
5068 return cBOOL(SvUV(these_categories) & (1U << (category + 1)));
5072 Perl_my_strerror(pTHX_ const int errnum)
5074 /* Returns a mortalized copy of the text of the error message associated
5075 * with 'errnum'. It uses the current locale's text unless the platform
5076 * doesn't have the LC_MESSAGES category or we are not being called from
5077 * within the scope of 'use locale'. In the former case, it uses whatever
5078 * strerror returns; in the latter case it uses the text from the C locale.
5080 * The function just calls strerror(), but temporarily switches, if needed,
5081 * to the C locale */
5086 #ifndef USE_LOCALE_MESSAGES
5088 /* If platform doesn't have messages category, we don't do any switching to
5089 * the C locale; we just use whatever strerror() returns */
5091 errstr = savepv(Strerror(errnum));
5093 #else /* Has locale messages */
5095 const bool within_locale_scope = IN_LC(LC_MESSAGES);
5097 # ifndef USE_ITHREADS
5099 /* This function is trivial without threads. */
5100 if (within_locale_scope) {
5101 errstr = savepv(strerror(errnum));
5104 const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL));
5106 do_setlocale_c(LC_MESSAGES, "C");
5107 errstr = savepv(strerror(errnum));
5108 do_setlocale_c(LC_MESSAGES, save_locale);
5109 Safefree(save_locale);
5112 # elif defined(HAS_POSIX_2008_LOCALE) \
5113 && defined(HAS_STRERROR_L) \
5114 && defined(HAS_DUPLOCALE)
5116 /* This function is also trivial if we don't have to worry about thread
5117 * safety and have strerror_l(), as it handles the switch of locales so we
5118 * don't have to deal with that. We don't have to worry about thread
5119 * safety if strerror_r() is also available. Both it and strerror_l() are
5120 * thread-safe. Plain strerror() isn't thread safe. But on threaded
5121 * builds when strerror_r() is available, the apparent call to strerror()
5122 * below is actually a macro that behind-the-scenes calls strerror_r(). */
5124 # ifdef HAS_STRERROR_R
5126 if (within_locale_scope) {
5127 errstr = savepv(strerror(errnum));
5130 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
5135 /* Here we have strerror_l(), but not strerror_r() and we are on a
5136 * threaded-build. We use strerror_l() for everything, constructing a
5137 * locale to pass to it if necessary */
5139 bool do_free = FALSE;
5140 locale_t locale_to_use;
5142 if (within_locale_scope) {
5143 locale_to_use = uselocale((locale_t) 0);
5144 if (locale_to_use == LC_GLOBAL_LOCALE) {
5145 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
5149 else { /* Use C locale if not within 'use locale' scope */
5150 locale_to_use = PL_C_locale_obj;
5153 errstr = savepv(strerror_l(errnum, locale_to_use));
5156 freelocale(locale_to_use);
5160 # else /* Doesn't have strerror_l() */
5162 const char * save_locale = NULL;
5163 bool locale_is_C = FALSE;
5165 /* We have a critical section to prevent another thread from executing this
5166 * same code at the same time. (On thread-safe perls, the LOCK is a
5167 * no-op.) Since this is the only place in core that changes LC_MESSAGES
5168 * (unless the user has called setlocale(), this works to prevent races. */
5171 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5172 "my_strerror called with errnum %d\n", errnum));
5173 if (! within_locale_scope) {
5174 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
5175 if (! save_locale) {
5177 "panic: %s: %d: Could not find current LC_MESSAGES locale,"
5178 " errno=%d\n", __FILE__, __LINE__, errno);
5181 locale_is_C = isNAME_C_OR_POSIX(save_locale);
5183 /* Switch to the C locale if not already in it */
5184 if (! locale_is_C) {
5186 /* The setlocale() just below likely will zap 'save_locale', so
5188 save_locale = savepv(save_locale);
5189 do_setlocale_c(LC_MESSAGES, "C");
5192 } /* end of ! within_locale_scope */
5194 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
5195 __FILE__, __LINE__));
5198 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5199 "Any locale change has been done; about to call Strerror\n"));
5200 errstr = savepv(Strerror(errnum));
5202 if (! within_locale_scope) {
5203 if (save_locale && ! locale_is_C) {
5204 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
5206 "panic: %s: %d: setlocale restore failed, errno=%d\n",
5207 __FILE__, __LINE__, errno);
5209 Safefree(save_locale);
5215 # endif /* End of doesn't have strerror_l */
5218 if (DEBUG_Lv_TEST) {
5219 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
5220 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
5221 PerlIO_printf(Perl_debug_log, "'\n");
5225 #endif /* End of does have locale messages */
5233 =for apidoc switch_to_global_locale
5235 On systems without locale support, or on single-threaded builds, or on
5236 platforms that do not support per-thread locale operations, this function does
5237 nothing. On such systems that do have locale support, only a locale global to
5238 the whole program is available.
5240 On multi-threaded builds on systems that do have per-thread locale operations,
5241 this function converts the thread it is running in to use the global locale.
5242 This is for code that has not yet or cannot be updated to handle multi-threaded
5243 locale operation. As long as only a single thread is so-converted, everything
5244 works fine, as all the other threads continue to ignore the global one, so only
5245 this thread looks at it.
5247 However, on Windows systems this isn't quite true prior to Visual Studio 15,
5248 at which point Microsoft fixed a bug. A race can occur if you use the
5249 following operations on earlier Windows platforms:
5253 =item L<POSIX::localeconv|POSIX/localeconv>
5255 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5257 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5261 The first item is not fixable (except by upgrading to a later Visual Studio
5262 release), but it would be possible to work around the latter two items by using
5263 the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches
5266 Without this function call, threads that use the L<C<setlocale(3)>> system
5267 function will not work properly, as all the locale-sensitive functions will
5268 look at the per-thread locale, and C<setlocale> will have no effect on this
5271 Perl code should convert to either call
5272 L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system
5273 C<setlocale>) or use the methods given in L<perlcall> to call
5274 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly
5275 handle all cases of single- vs multi-thread, POSIX 2008-supported or not.
5277 Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can
5278 continue to work if this function is called before transferring control to the
5281 Upon return from the code that needs to use the global locale,
5282 L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe
5283 multi-thread operation.
5289 Perl_switch_to_global_locale()
5292 #ifdef USE_THREAD_SAFE_LOCALE
5295 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
5298 # ifdef HAS_QUERYLOCALE
5300 setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0)));
5307 for (i = 0; i < LC_ALL_INDEX; i++) {
5308 setlocale(categories[i], do_setlocale_r(categories[i], NULL));
5314 uselocale(LC_GLOBAL_LOCALE);
5323 =for apidoc sync_locale
5325 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or
5326 change the locale (though changing the locale is antisocial and dangerous on
5327 multi-threaded systems that don't have multi-thread safe locale operations.
5328 (See L<perllocale/Multi-threaded operation>). Using the system
5329 L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries
5330 called from XS, such as C<Gtk> do so, and this can't be changed. When the
5331 locale is changed by XS code that didn't use
5332 L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the
5333 locale has changed. Use this function to do so, before returning to Perl.
5335 The return value is a boolean: TRUE if the global locale at the time of call
5336 was in effect; and FALSE if a per-thread locale was in effect. This can be
5337 used by the caller that needs to restore things as-they-were to decide whether
5339 L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>.
5354 const char * newlocale;
5357 # ifdef USE_POSIX_2008_LOCALE
5359 bool was_in_global_locale = FALSE;
5360 locale_t cur_obj = uselocale((locale_t) 0);
5362 /* On Windows, unless the foreign code has turned off the thread-safe
5363 * locale setting, any plain setlocale() will have affected what we see, so
5364 * no need to worry. Otherwise, If the foreign code has done a plain
5365 * setlocale(), it will only affect the global locale on POSIX systems, but
5366 * will affect the */
5367 if (cur_obj == LC_GLOBAL_LOCALE) {
5369 # ifdef HAS_QUERY_LOCALE
5371 do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL));
5377 /* We can't trust that we can read the LC_ALL format on the
5378 * platform, so do them individually */
5379 for (i = 0; i < LC_ALL_INDEX; i++) {
5380 do_setlocale_r(categories[i], setlocale(categories[i], NULL));
5385 was_in_global_locale = TRUE;
5390 bool was_in_global_locale = TRUE;
5393 # ifdef USE_LOCALE_CTYPE
5395 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
5396 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5397 "%s:%d: %s\n", __FILE__, __LINE__,
5398 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
5399 new_ctype(newlocale);
5400 Safefree(newlocale);
5402 # endif /* USE_LOCALE_CTYPE */
5403 # ifdef USE_LOCALE_COLLATE
5405 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
5406 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5407 "%s:%d: %s\n", __FILE__, __LINE__,
5408 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
5409 new_collate(newlocale);
5410 Safefree(newlocale);
5413 # ifdef USE_LOCALE_NUMERIC
5415 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
5416 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5417 "%s:%d: %s\n", __FILE__, __LINE__,
5418 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
5419 new_numeric(newlocale);
5420 Safefree(newlocale);
5422 # endif /* USE_LOCALE_NUMERIC */
5424 return was_in_global_locale;
5430 #if defined(DEBUGGING) && defined(USE_LOCALE)
5433 S_setlocale_debug_string(const int category, /* category number,
5435 const char* const locale, /* locale name */
5437 /* return value from setlocale() when attempting to
5438 * set 'category' to 'locale' */
5439 const char* const retval)
5441 /* Returns a pointer to a NUL-terminated string in static storage with
5442 * added text about the info passed in. This is not thread safe and will
5443 * be overwritten by the next call, so this should be used just to
5444 * formulate a string to immediately print or savepv() on. */
5446 /* initialise to a non-null value to keep it out of BSS and so keep
5447 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
5448 static char ret[256] = "If you can read this, thank your buggy C"
5449 " library strlcpy(), and change your hints file"
5452 my_strlcpy(ret, "setlocale(", sizeof(ret));
5453 my_strlcat(ret, category_name(category), sizeof(ret));
5454 my_strlcat(ret, ", ", sizeof(ret));
5457 my_strlcat(ret, "\"", sizeof(ret));
5458 my_strlcat(ret, locale, sizeof(ret));
5459 my_strlcat(ret, "\"", sizeof(ret));
5462 my_strlcat(ret, "NULL", sizeof(ret));
5465 my_strlcat(ret, ") returned ", sizeof(ret));
5468 my_strlcat(ret, "\"", sizeof(ret));
5469 my_strlcat(ret, retval, sizeof(ret));
5470 my_strlcat(ret, "\"", sizeof(ret));
5473 my_strlcat(ret, "NULL", sizeof(ret));
5476 assert(strlen(ret) < sizeof(ret));
5484 Perl_thread_locale_init()
5486 /* Called from a thread on startup*/
5488 #ifdef USE_THREAD_SAFE_LOCALE
5492 /* C starts the new thread in the global C locale. If we are thread-safe,
5493 * we want to not be in the global locale */
5495 DEBUG_L(PerlIO_printf(Perl_debug_log,
5496 "%s:%d: new thread, initial locale is %s; calling setlocale\n",
5497 __FILE__, __LINE__, setlocale(LC_ALL, NULL)));
5501 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
5505 Perl_setlocale(LC_ALL, "C");
5513 Perl_thread_locale_term()
5515 /* Called from a thread as it gets ready to terminate */
5517 #ifdef USE_THREAD_SAFE_LOCALE
5519 /* C starts the new thread in the global C locale. If we are thread-safe,
5520 * we want to not be in the global locale */
5525 locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE);
5526 if (cur_obj != LC_GLOBAL_LOCALE) {
5527 freelocale(cur_obj);
5537 * ex: set ts=8 sts=4 sw=4 et: