3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * A Elbereth Gilthoniel,
13 * silivren penna míriel
14 * o menel aglar elenath!
15 * Na-chaered palan-díriel
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
47 #define PERL_IN_LOCALE_C
48 #include "perl_langinfo.h"
60 /* If the environment says to, we can output debugging information during
61 * initialization. This is done before option parsing, and before any thread
62 * creation, so can be a file-level static */
63 #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
64 # define debug_initialization 0
65 # define DEBUG_INITIALIZATION_set(v)
67 static bool debug_initialization = FALSE;
68 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
72 /* Returns the Unix errno portion; ignoring any others. This is a macro here
73 * instead of putting it into perl.h, because unclear to khw what should be
75 #define GET_ERRNO saved_errno
77 /* strlen() of a literal string constant. We might want this more general,
78 * but using it in just this file for now. A problem with more generality is
79 * the compiler warnings about comparing unlike signs */
80 #define STRLENs(s) (sizeof("" s "") - 1)
82 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
83 * return of setlocale(), then this is extremely likely to be the C or POSIX
84 * locale. However, the output of setlocale() is documented to be opaque, but
85 * the odds are extremely small that it would return these two strings for some
86 * other locale. Note that VMS in these two locales includes many non-ASCII
87 * characters as controls and punctuation (below are hex bytes):
89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
90 * Oddly, none there are listed as alphas, though some represent alphabetics
91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
92 #define isNAME_C_OR_POSIX(name) \
94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
95 || strEQ((name), "POSIX")))
99 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
100 * looked up. This is in the form of a C string: */
102 #define UTF8NESS_SEP "\v"
103 #define UTF8NESS_PREFIX "\f"
105 /* So, the string looks like:
107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
109 * where the digit 0 after the \a indicates that the locale starting just
110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
112 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
113 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
115 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
118 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
119 * kept there always. The remining portion of the cache is LRU, with the
120 * oldest looked-up locale at the tail end */
123 S_stdize_locale(pTHX_ char *locs)
125 /* Standardize the locale name from a string returned by 'setlocale',
126 * possibly modifying that string.
128 * The typical return value of setlocale() is either
129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
131 * (the space-separated values represent the various sublocales,
132 * in some unspecified order). This is not handled by this function.
134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
135 * which is harmful for further use of the string in setlocale(). This
136 * function removes the trailing new line and everything up through the '='
139 const char * const s = strchr(locs, '=');
142 PERL_ARGS_ASSERT_STDIZE_LOCALE;
145 const char * const t = strchr(s, '.');
148 const char * const u = strchr(t, '\n');
149 if (u && (u[1] == 0)) {
150 const STRLEN len = u - s;
151 Move(s + 1, locs, len, char);
159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
164 /* Two parallel arrays; first the locale categories Perl uses on this system;
165 * the second array is their names. These arrays are in mostly arbitrary
168 const int categories[] = {
170 # ifdef USE_LOCALE_NUMERIC
173 # ifdef USE_LOCALE_CTYPE
176 # ifdef USE_LOCALE_COLLATE
179 # ifdef USE_LOCALE_TIME
182 # ifdef USE_LOCALE_MESSAGES
185 # ifdef USE_LOCALE_MONETARY
188 # ifdef USE_LOCALE_ADDRESS
191 # ifdef USE_LOCALE_IDENTIFICATION
194 # ifdef USE_LOCALE_MEASUREMENT
197 # ifdef USE_LOCALE_PAPER
200 # ifdef USE_LOCALE_TELEPHONE
206 -1 /* Placeholder because C doesn't allow a
207 trailing comma, and it would get complicated
208 with all the #ifdef's */
211 /* The top-most real element is LC_ALL */
213 const char * const category_names[] = {
215 # ifdef USE_LOCALE_NUMERIC
218 # ifdef USE_LOCALE_CTYPE
221 # ifdef USE_LOCALE_COLLATE
224 # ifdef USE_LOCALE_TIME
227 # ifdef USE_LOCALE_MESSAGES
230 # ifdef USE_LOCALE_MONETARY
233 # ifdef USE_LOCALE_ADDRESS
236 # ifdef USE_LOCALE_IDENTIFICATION
239 # ifdef USE_LOCALE_MEASUREMENT
242 # ifdef USE_LOCALE_PAPER
245 # ifdef USE_LOCALE_TELEPHONE
251 NULL /* Placeholder */
256 /* On systems with LC_ALL, it is kept in the highest index position. (-2
257 * to account for the final unused placeholder element.) */
258 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
262 /* On systems without LC_ALL, we pretend it is there, one beyond the real
263 * top element, hence in the unused placeholder element. */
264 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
268 /* Pretending there is an LC_ALL element just above allows us to avoid most
269 * special cases. Most loops through these arrays in the code below are
270 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
271 * on either type of system. But the code must be written to not access the
272 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
273 * checked for at compile time by using the #define LC_ALL_INDEX which is only
274 * defined if we do have LC_ALL. */
277 S_category_name(const int category)
283 if (category == LC_ALL) {
289 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
290 if (category == categories[i]) {
291 return category_names[i];
296 const char suffix[] = " (unknown)";
298 Size_t length = sizeof(suffix) + 1;
307 /* Calculate the number of digits */
313 Newx(unknown, length, char);
314 my_snprintf(unknown, length, "%d%s", category, suffix);
320 /* Now create LC_foo_INDEX #defines for just those categories on this system */
321 # ifdef USE_LOCALE_NUMERIC
322 # define LC_NUMERIC_INDEX 0
323 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
325 # define _DUMMY_NUMERIC -1
327 # ifdef USE_LOCALE_CTYPE
328 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
329 # define _DUMMY_CTYPE LC_CTYPE_INDEX
331 # define _DUMMY_CTYPE _DUMMY_NUMERIC
333 # ifdef USE_LOCALE_COLLATE
334 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
335 # define _DUMMY_COLLATE LC_COLLATE_INDEX
337 # define _DUMMY_COLLATE _DUMMY_CTYPE
339 # ifdef USE_LOCALE_TIME
340 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
341 # define _DUMMY_TIME LC_TIME_INDEX
343 # define _DUMMY_TIME _DUMMY_COLLATE
345 # ifdef USE_LOCALE_MESSAGES
346 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
347 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
349 # define _DUMMY_MESSAGES _DUMMY_TIME
351 # ifdef USE_LOCALE_MONETARY
352 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
353 # define _DUMMY_MONETARY LC_MONETARY_INDEX
355 # define _DUMMY_MONETARY _DUMMY_MESSAGES
357 # ifdef USE_LOCALE_ADDRESS
358 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
359 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
361 # define _DUMMY_ADDRESS _DUMMY_MONETARY
363 # ifdef USE_LOCALE_IDENTIFICATION
364 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
365 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
367 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
369 # ifdef USE_LOCALE_MEASUREMENT
370 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
371 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
373 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
375 # ifdef USE_LOCALE_PAPER
376 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
377 # define _DUMMY_PAPER LC_PAPER_INDEX
379 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
381 # ifdef USE_LOCALE_TELEPHONE
382 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
383 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
385 # define _DUMMY_TELEPHONE _DUMMY_PAPER
388 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
390 #endif /* ifdef USE_LOCALE */
392 /* Windows requres a customized base-level setlocale() */
394 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
396 # define my_setlocale(cat, locale) setlocale(cat, locale)
399 #ifndef USE_POSIX_2008_LOCALE
401 /* "do_setlocale_c" is intended to be called when the category is a constant
402 * known at compile time; "do_setlocale_r", not known until run time */
403 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
404 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
406 #else /* Below uses POSIX 2008 */
408 /* We emulate setlocale with our own function. LC_foo is not valid for the
409 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
410 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
411 * proper offset into the array 'category_masks[]'. At runtime, we have to
412 * search through the array (as the actual numbers may not be small contiguous
413 * positive integers which would lend themselves to array lookup). */
414 # define do_setlocale_c(cat, locale) \
415 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
416 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
418 /* A third array, parallel to the ones above to map from category to its
420 const int category_masks[] = {
421 # ifdef USE_LOCALE_NUMERIC
424 # ifdef USE_LOCALE_CTYPE
427 # ifdef USE_LOCALE_COLLATE
430 # ifdef USE_LOCALE_TIME
433 # ifdef USE_LOCALE_MESSAGES
436 # ifdef USE_LOCALE_MONETARY
439 # ifdef USE_LOCALE_ADDRESS
442 # ifdef USE_LOCALE_IDENTIFICATION
443 LC_IDENTIFICATION_MASK,
445 # ifdef USE_LOCALE_MEASUREMENT
448 # ifdef USE_LOCALE_PAPER
451 # ifdef USE_LOCALE_TELEPHONE
454 /* LC_ALL can't be turned off by a Configure
455 * option, and in Posix 2008, should always be
456 * here, so compile it in unconditionally.
457 * This could catch some glitches at compile
463 S_emulate_setlocale(const int category,
466 const bool is_index_valid
469 /* This function effectively performs a setlocale() on just the current
470 * thread; thus it is thread-safe. It does this by using the POSIX 2008
471 * locale functions to emulate the behavior of setlocale(). Similar to
472 * regular setlocale(), the return from this function points to memory that
473 * can be overwritten by other system calls, so needs to be copied
474 * immediately if you need to retain it. The difference here is that
475 * system calls besides another setlocale() can overwrite it.
477 * By doing this, most locale-sensitive functions become thread-safe. The
478 * exceptions are mostly those that return a pointer to static memory.
480 * This function takes the same parameters, 'category' and 'locale', that
481 * the regular setlocale() function does, but it also takes two additional
482 * ones. This is because the 2008 functions don't use a category; instead
483 * they use a corresponding mask. Because this function operates in both
484 * worlds, it may need one or the other or both. This function can
485 * calculate the mask from the input category, but to avoid this
486 * calculation, if the caller knows at compile time what the mask is, it
487 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
488 * parameter is ignored.
490 * POSIX 2008, for some sick reason, chose not to provide a method to find
491 * the category name of a locale. Some vendors have created a
492 * querylocale() function to do just that. This function is a lot simpler
493 * to implement on systems that have this. Otherwise, we have to keep
494 * track of what the locale has been set to, so that we can return its
495 * name to emulate setlocale(). It's also possible for C code in some
496 * library to change the locale without us knowing it, though as of
497 * September 2017, there are no occurrences in CPAN of uselocale(). Some
498 * libraries do use setlocale(), but that changes the global locale, and
499 * threads using per-thread locales will just ignore those changes.
500 * Another problem is that without querylocale(), we have to guess at what
501 * was meant by setting a locale of "". We handle this by not actually
502 * ever setting to "" (unless querylocale exists), but to emulate what we
503 * think should happen for "".
513 if (DEBUG_Lv_TEST || debug_initialization) {
514 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);
519 /* If the input mask might be incorrect, calculate the correct one */
520 if (! is_index_valid) {
525 if (DEBUG_Lv_TEST || debug_initialization) {
526 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
531 for (i = 0; i <= LC_ALL_INDEX; i++) {
532 if (category == categories[i]) {
538 /* Here, we don't know about this category, so can't handle it.
539 * Fallback to the early POSIX usages */
540 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
541 "Unknown locale category %d; can't set it to %s\n",
549 if (DEBUG_Lv_TEST || debug_initialization) {
550 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
557 mask = category_masks[index];
561 if (DEBUG_Lv_TEST || debug_initialization) {
562 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
567 /* If just querying what the existing locale is ... */
568 if (locale == NULL) {
569 locale_t cur_obj = uselocale((locale_t) 0);
573 if (DEBUG_Lv_TEST || debug_initialization) {
574 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
579 if (cur_obj == LC_GLOBAL_LOCALE) {
580 return my_setlocale(category, NULL);
583 # ifdef HAS_QUERYLOCALE
585 return (char *) querylocale(mask, cur_obj);
589 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
590 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
592 # if defined(_NL_LOCALE_NAME) \
593 && defined(DEBUGGING) \
594 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
595 /* On systems that accept any locale name, the real underlying locale
596 * is often returned by this internal function, so we can't use it */
598 /* Internal glibc for querylocale(), but doesn't handle
599 * empty-string ("") locale properly; who knows what other
600 * glitches. Check for it now, under debug. */
602 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
603 uselocale((locale_t) 0));
605 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
606 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
607 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
609 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
610 if ( strNE(PL_curlocales[index], temp_name)
611 && ! ( isNAME_C_OR_POSIX(temp_name)
612 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
614 # ifdef USE_C_BACKTRACE
616 dump_c_backtrace(Perl_debug_log, 20, 1);
620 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
621 " (%s) and what internal glibc thinks"
622 " (%s)\n", category_names[index],
623 PL_curlocales[index], temp_name);
632 /* Without querylocale(), we have to use our record-keeping we've
635 if (category != LC_ALL) {
639 if (DEBUG_Lv_TEST || debug_initialization) {
640 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
645 return PL_curlocales[index];
647 else { /* For LC_ALL */
649 Size_t names_len = 0;
651 bool are_all_categories_the_same_locale = TRUE;
653 /* If we have a valid LC_ALL value, just return it */
654 if (PL_curlocales[LC_ALL_INDEX]) {
658 if (DEBUG_Lv_TEST || debug_initialization) {
659 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
664 return PL_curlocales[LC_ALL_INDEX];
667 /* Otherwise, we need to construct a string of name=value pairs.
668 * We use the glibc syntax, like
669 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
670 * First calculate the needed size. Along the way, check if all
671 * the locale names are the same */
672 for (i = 0; i < LC_ALL_INDEX; i++) {
676 if (DEBUG_Lv_TEST || debug_initialization) {
677 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]);
682 names_len += strlen(category_names[i])
684 + strlen(PL_curlocales[i])
687 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
688 are_all_categories_the_same_locale = FALSE;
692 /* If they are the same, we don't actually have to construct the
693 * string; we just make the entry in LC_ALL_INDEX valid, and be
694 * that single name */
695 if (are_all_categories_the_same_locale) {
696 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
697 return PL_curlocales[LC_ALL_INDEX];
700 names_len++; /* Trailing '\0' */
701 SAVEFREEPV(Newx(all_string, names_len, char));
704 /* Then fill in the string */
705 for (i = 0; i < LC_ALL_INDEX; i++) {
709 if (DEBUG_Lv_TEST || debug_initialization) {
710 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]);
715 my_strlcat(all_string, category_names[i], names_len);
716 my_strlcat(all_string, "=", names_len);
717 my_strlcat(all_string, PL_curlocales[i], names_len);
718 my_strlcat(all_string, ";", names_len);
723 if (DEBUG_L_TEST || debug_initialization) {
724 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
734 SETERRNO(EINVAL, LIB_INVARG);
744 /* Here, we are switching locales. */
746 # ifndef HAS_QUERYLOCALE
748 if (strEQ(locale, "")) {
750 /* For non-querylocale() systems, we do the setting of "" ourselves to
751 * be sure that we really know what's going on. We follow the Linux
752 * documented behavior (but if that differs from the actual behavior,
753 * this won't work exactly as the OS implements). We go out and
754 * examine the environment based on our understanding of how the system
755 * works, and use that to figure things out */
757 const char * const lc_all = PerlEnv_getenv("LC_ALL");
759 /* Use any "LC_ALL" environment variable, as it overrides everything
761 if (lc_all && strNE(lc_all, "")) {
766 /* Otherwise, we need to dig deeper. Unless overridden, the
767 * default is the LANG environment variable; if it doesn't exist,
770 const char * default_name;
772 default_name = PerlEnv_getenv("LANG");
774 if (! default_name || strEQ(default_name, "")) {
777 else if (PL_scopestack_ix != 0) {
778 /* To minimize other threads messing with the environment,
779 * we copy the variable, making it a temporary. But this
780 * doesn't work upon program initialization before any
781 * scopes are created, and at this time, there's nothing
782 * else going on that would interfere. So skip the copy
784 default_name = savepv(default_name);
785 SAVEFREEPV(default_name);
788 if (category != LC_ALL) {
789 const char * const name = PerlEnv_getenv(category_names[index]);
791 /* Here we are setting a single category. Assume will have the
793 locale = default_name;
795 /* But then look for an overriding environment variable */
796 if (name && strNE(name, "")) {
801 bool did_override = FALSE;
804 /* Here, we are getting LC_ALL. Any categories that don't have
805 * a corresponding environment variable set should be set to
806 * LANG, or to "C" if there is no LANG. If no individual
807 * categories differ from this, we can just set LC_ALL. This
808 * is buggy on systems that have extra categories that we don't
809 * know about. If there is an environment variable that sets
810 * that category, we won't know to look for it, and so our use
811 * of LANG or "C" improperly overrides it. On the other hand,
812 * if we don't do what is done here, and there is no
813 * environment variable, the category's locale should be set to
814 * LANG or "C". So there is no good solution. khw thinks the
815 * best is to look at systems to see what categories they have,
816 * and include them, and then to assume that we know the
819 for (i = 0; i < LC_ALL_INDEX; i++) {
820 const char * const env_override
821 = savepv(PerlEnv_getenv(category_names[i]));
822 const char * this_locale = ( env_override
823 && strNE(env_override, ""))
826 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
828 Safefree(env_override);
832 if (strNE(this_locale, default_name)) {
836 Safefree(env_override);
839 /* If all the categories are the same, we can set LC_ALL to
841 if (! did_override) {
842 locale = default_name;
846 /* Here, LC_ALL is no longer valid, as some individual
847 * categories don't match it. We call ourselves
848 * recursively, as that will execute the code that
849 * generates the proper locale string for this situation.
850 * We don't do the remainder of this function, as that is
851 * to update our records, and we've just done that for the
852 * individual categories in the loop above, and doing so
853 * would cause LC_ALL to be done as well */
854 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
859 else if (strchr(locale, ';')) {
861 /* LC_ALL may actually incude a conglomeration of various categories.
862 * Without querylocale, this code uses the glibc (as of this writing)
863 * syntax for representing that, but that is not a stable API, and
864 * other platforms do it differently, so we have to handle all cases
868 const char * s = locale;
869 const char * e = locale + strlen(locale);
871 const char * category_end;
872 const char * name_start;
873 const char * name_end;
875 /* If the string that gives what to set doesn't include all categories,
876 * the omitted ones get set to "C". To get this behavior, first set
877 * all the individual categories to "C", and override the furnished
879 for (i = 0; i < LC_ALL_INDEX; i++) {
880 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
887 /* Parse through the category */
888 while (isWORDCHAR(*p)) {
895 "panic: %s: %d: Unexpected character in locale name '%02X",
896 __FILE__, __LINE__, *(p-1));
899 /* Parse through the locale name */
901 while (p < e && *p != ';') {
904 "panic: %s: %d: Unexpected character in locale name '%02X",
905 __FILE__, __LINE__, *(p-1));
911 /* Space past the semi-colon */
916 /* Find the index of the category name in our lists */
917 for (i = 0; i < LC_ALL_INDEX; i++) {
918 char * individ_locale;
920 /* Keep going if this isn't the index. The strnNE() avoids a
921 * Perl_form(), but would fail if ever a category name could be
922 * a substring of another one, like if there were a
924 if strnNE(s, category_names[i], category_end - s) {
928 /* If this index is for the single category we're changing, we
929 * have found the locale to set it to. */
930 if (category == categories[i]) {
931 locale = Perl_form(aTHX_ "%.*s",
932 (int) (name_end - name_start),
937 assert(category == LC_ALL);
938 individ_locale = Perl_form(aTHX_ "%.*s",
939 (int) (name_end - name_start), name_start);
940 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
949 /* Here we have set all the individual categories by recursive calls.
950 * These collectively should have fixed up LC_ALL, so can just query
951 * what that now is */
952 assert(category == LC_ALL);
954 return do_setlocale_c(LC_ALL, NULL);
959 /* Here at the end of having to deal with the absence of querylocale().
960 * Some cases have already been fully handled by recursive calls to this
961 * function. But at this point, we haven't dealt with those, but are now
962 * prepared to, knowing what the locale name to set this category to is.
963 * This would have come for free if this system had had querylocale() */
965 # endif /* end of ! querylocale */
967 assert(PL_C_locale_obj);
969 /* Switching locales generally entails freeing the current one's space (at
970 * the C library's discretion). We need to stop using that locale before
971 * the switch. So switch to a known locale object that we don't otherwise
972 * mess with. This returns the locale object in effect at the time of the
974 old_obj = uselocale(PL_C_locale_obj);
978 if (DEBUG_Lv_TEST || debug_initialization) {
979 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
988 if (DEBUG_L_TEST || debug_initialization) {
990 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1001 if (DEBUG_Lv_TEST || debug_initialization) {
1002 PerlIO_printf(Perl_debug_log,
1003 "%s:%d: emulate_setlocale now using %p\n",
1004 __FILE__, __LINE__, PL_C_locale_obj);
1009 /* If we are switching to the LC_ALL C locale, it already exists. Use
1010 * it instead of trying to create a new locale */
1011 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1015 if (DEBUG_Lv_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log,
1017 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1022 new_obj = PL_C_locale_obj;
1024 /* We already had switched to the C locale in preparation for freeing
1026 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1027 freelocale(old_obj);
1031 /* If we weren't in a thread safe locale, set so that newlocale() below
1032 * which uses 'old_obj', uses an empty one. Same for our reserved C
1033 * object. The latter is defensive coding, so that, even if there is
1034 * some bug, we will never end up trying to modify either of these, as
1035 * if passed to newlocale(), they can be. */
1036 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1037 old_obj = (locale_t) 0;
1040 /* Ready to create a new locale by modification of the exising one */
1041 new_obj = newlocale(mask, locale, old_obj);
1048 if (DEBUG_L_TEST || debug_initialization) {
1049 PerlIO_printf(Perl_debug_log,
1050 "%s:%d: emulate_setlocale creating new object"
1051 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1056 if (! uselocale(old_obj)) {
1060 if (DEBUG_L_TEST || debug_initialization) {
1061 PerlIO_printf(Perl_debug_log,
1062 "%s:%d: switching back failed: %d\n",
1063 __FILE__, __LINE__, GET_ERRNO);
1075 if (DEBUG_Lv_TEST || debug_initialization) {
1076 PerlIO_printf(Perl_debug_log,
1077 "%s:%d: emulate_setlocale created %p",
1078 __FILE__, __LINE__, new_obj);
1080 PerlIO_printf(Perl_debug_log,
1081 "; should have freed %p", old_obj);
1083 PerlIO_printf(Perl_debug_log, "\n");
1088 /* And switch into it */
1089 if (! uselocale(new_obj)) {
1094 if (DEBUG_L_TEST || debug_initialization) {
1095 PerlIO_printf(Perl_debug_log,
1096 "%s:%d: emulate_setlocale switching to new object"
1097 " failed\n", __FILE__, __LINE__);
1102 if (! uselocale(old_obj)) {
1106 if (DEBUG_L_TEST || debug_initialization) {
1107 PerlIO_printf(Perl_debug_log,
1108 "%s:%d: switching back failed: %d\n",
1109 __FILE__, __LINE__, GET_ERRNO);
1115 freelocale(new_obj);
1123 if (DEBUG_Lv_TEST || debug_initialization) {
1124 PerlIO_printf(Perl_debug_log,
1125 "%s:%d: emulate_setlocale now using %p\n",
1126 __FILE__, __LINE__, new_obj);
1131 /* We are done, except for updating our records (if the system doesn't keep
1132 * them) and in the case of locale "", we don't actually know what the
1133 * locale that got switched to is, as it came from the environment. So
1134 * have to find it */
1136 # ifdef HAS_QUERYLOCALE
1138 if (strEQ(locale, "")) {
1139 locale = querylocale(mask, new_obj);
1144 /* Here, 'locale' is the return value */
1146 /* Without querylocale(), we have to update our records */
1148 if (category == LC_ALL) {
1151 /* For LC_ALL, we change all individual categories to correspond */
1152 /* PL_curlocales is a parallel array, so has same
1153 * length as 'categories' */
1154 for (i = 0; i <= LC_ALL_INDEX; i++) {
1155 Safefree(PL_curlocales[i]);
1156 PL_curlocales[i] = savepv(locale);
1161 /* For a single category, if it's not the same as the one in LC_ALL, we
1164 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1165 Safefree(PL_curlocales[LC_ALL_INDEX]);
1166 PL_curlocales[LC_ALL_INDEX] = NULL;
1169 /* Then update the category's record */
1170 Safefree(PL_curlocales[index]);
1171 PL_curlocales[index] = savepv(locale);
1179 #endif /* USE_POSIX_2008_LOCALE */
1181 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1182 didn't natively support them */
1184 /* The way this would work is that we would keep a per-thread list of the
1185 * correct locale for that thread. Any operation that was locale-sensitive
1186 * would have to be changed so that it would look like this:
1189 * setlocale to the correct locale for this operation
1193 * This leaves the global locale in the most recently used operation's, but it
1194 * was locked long enough to get the result. If that result is static, it
1195 * needs to be copied before the unlock.
1197 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1198 * the setup, but are no-ops when not needed, and similarly,
1199 * END_LOCALE_DEPENDENT_OP for the tear-down
1201 * But every call to a locale-sensitive function would have to be changed, and
1202 * if a module didn't cooperate by using the mutex, things would break.
1204 * This code was abandoned before being completed or tested, and is left as-is
1207 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1208 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1211 S_locking_setlocale(pTHX_
1213 const char * locale,
1215 const bool is_index_valid
1218 /* This function kind of performs a setlocale() on just the current thread;
1219 * thus it is kind of thread-safe. It does this by keeping a thread-level
1220 * array of the current locales for each category. Every time a locale is
1221 * switched to, it does the switch globally, but updates the thread's
1222 * array. A query as to what the current locale is just returns the
1223 * appropriate element from the array, and doesn't actually call the system
1224 * setlocale(). The saving into the array is done in an uninterruptible
1225 * section of code, so is unaffected by whatever any other threads might be
1228 * All locale-sensitive operations must work by first starting a critical
1229 * section, then switching to the thread's locale as kept by this function,
1230 * and then doing the operation, then ending the critical section. Thus,
1231 * each gets done in the appropriate locale. simulating thread-safety.
1233 * This function takes the same parameters, 'category' and 'locale', that
1234 * the regular setlocale() function does, but it also takes two additional
1235 * ones. This is because as described earlier. If we know on input the
1236 * index corresponding to the category into the array where we store the
1237 * current locales, we don't have to calculate it. If the caller knows at
1238 * compile time what the index is, it it can pass it, setting
1239 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1243 /* If the input index might be incorrect, calculate the correct one */
1244 if (! is_index_valid) {
1247 if (DEBUG_Lv_TEST || debug_initialization) {
1248 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1251 for (i = 0; i <= LC_ALL_INDEX; i++) {
1252 if (category == categories[i]) {
1258 /* Here, we don't know about this category, so can't handle it.
1259 * XXX best we can do is to unsafely set this
1262 return my_setlocale(category, locale);
1266 if (DEBUG_Lv_TEST || debug_initialization) {
1267 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1271 /* For a query, just return what's in our records */
1272 if (new_locale == NULL) {
1273 return curlocales[index];
1277 /* Otherwise, we need to do the switch, and save the result, all in a
1278 * critical section */
1280 Safefree(curlocales[[index]]);
1282 /* It might be that this is called from an already-locked section of code.
1283 * We would have to detect and skip the LOCK/UNLOCK if so */
1286 curlocales[index] = savepv(my_setlocale(category, new_locale));
1288 if (strEQ(new_locale, "")) {
1292 /* The locale values come from the environment, and may not all be the
1293 * same, so for LC_ALL, we have to update all the others, while the
1294 * mutex is still locked */
1296 if (category == LC_ALL) {
1298 for (i = 0; i < LC_ALL_INDEX) {
1299 curlocales[i] = my_setlocale(categories[i], NULL);
1308 return curlocales[index];
1315 S_set_numeric_radix(pTHX_ const bool use_locale)
1317 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1318 * TRUE, use the radix character derived from the current locale */
1320 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1321 || defined(HAS_NL_LANGINFO))
1323 const char * radix = (use_locale)
1324 ? my_nl_langinfo(RADIXCHAR, FALSE)
1325 /* FALSE => already in dest locale */
1328 sv_setpv(PL_numeric_radix_sv, radix);
1330 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1331 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1332 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1333 SvCUR(PL_numeric_radix_sv))
1334 && _is_cur_LC_category_utf8(LC_NUMERIC))
1336 SvUTF8_on(PL_numeric_radix_sv);
1341 if (DEBUG_L_TEST || debug_initialization) {
1342 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1343 SvPVX(PL_numeric_radix_sv),
1344 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1350 PERL_UNUSED_ARG(use_locale);
1352 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1357 S_new_numeric(pTHX_ const char *newnum)
1360 #ifndef USE_LOCALE_NUMERIC
1362 PERL_UNUSED_ARG(newnum);
1366 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1367 * core Perl this and that 'newnum' is the name of the new locale.
1368 * It installs this locale as the current underlying default.
1370 * The default locale and the C locale can be toggled between by use of the
1371 * set_numeric_underlying() and set_numeric_standard() functions, which
1372 * should probably not be called directly, but only via macros like
1373 * SET_NUMERIC_STANDARD() in perl.h.
1375 * The toggling is necessary mainly so that a non-dot radix decimal point
1376 * character can be output, while allowing internal calculations to use a
1379 * This sets several interpreter-level variables:
1380 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1381 * PL_numeric_underlying A boolean indicating if the toggled state is such
1382 * that the current locale is the program's underlying
1384 * PL_numeric_standard An int indicating if the toggled state is such
1385 * that the current locale is the C locale or
1386 * indistinguishable from the C locale. If non-zero, it
1387 * is in C; if > 1, it means it may not be toggled away
1389 * PL_numeric_underlying_is_standard A bool kept by this function
1390 * indicating that the underlying locale and the standard
1391 * C locale are indistinguishable for the purposes of
1392 * LC_NUMERIC. This happens when both of the above two
1393 * variables are true at the same time. (Toggling is a
1394 * no-op under these circumstances.) This variable is
1395 * used to avoid having to recalculate.
1401 Safefree(PL_numeric_name);
1402 PL_numeric_name = NULL;
1403 PL_numeric_standard = TRUE;
1404 PL_numeric_underlying = TRUE;
1405 PL_numeric_underlying_is_standard = TRUE;
1409 save_newnum = stdize_locale(savepv(newnum));
1410 PL_numeric_underlying = TRUE;
1411 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1413 #ifndef TS_W32_BROKEN_LOCALECONV
1415 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1416 * them. But on broken Windows systems calling my_nl_langinfo() for
1417 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1418 * and just always change the locale if not C nor POSIX on those systems */
1419 if (! PL_numeric_standard) {
1420 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1421 FALSE /* Don't toggle locale */ ))
1422 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1427 /* Save the new name if it isn't the same as the previous one, if any */
1428 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1429 Safefree(PL_numeric_name);
1430 PL_numeric_name = save_newnum;
1433 Safefree(save_newnum);
1436 PL_numeric_underlying_is_standard = PL_numeric_standard;
1438 # ifdef HAS_POSIX_2008_LOCALE
1440 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1442 PL_underlying_numeric_obj);
1446 if (DEBUG_L_TEST || debug_initialization) {
1447 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1450 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1451 * have to worry about the radix being a non-dot. (Core operations that
1452 * need the underlying locale change to it temporarily). */
1453 if (PL_numeric_standard) {
1454 set_numeric_radix(0);
1457 set_numeric_standard();
1460 #endif /* USE_LOCALE_NUMERIC */
1465 Perl_set_numeric_standard(pTHX)
1468 #ifdef USE_LOCALE_NUMERIC
1470 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1471 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1472 * macro avoids calling this routine if toggling isn't necessary according
1473 * to our records (which could be wrong if some XS code has changed the
1474 * locale behind our back) */
1478 if (DEBUG_L_TEST || debug_initialization) {
1479 PerlIO_printf(Perl_debug_log,
1480 "Setting LC_NUMERIC locale to standard C\n");
1485 do_setlocale_c(LC_NUMERIC, "C");
1486 PL_numeric_standard = TRUE;
1487 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1488 set_numeric_radix(0);
1490 #endif /* USE_LOCALE_NUMERIC */
1495 Perl_set_numeric_underlying(pTHX)
1498 #ifdef USE_LOCALE_NUMERIC
1500 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1501 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1502 * instead of calling this directly. The macro avoids calling this routine
1503 * if toggling isn't necessary according to our records (which could be
1504 * wrong if some XS code has changed the locale behind our back) */
1508 if (DEBUG_L_TEST || debug_initialization) {
1509 PerlIO_printf(Perl_debug_log,
1510 "Setting LC_NUMERIC locale to %s\n",
1516 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1517 PL_numeric_standard = PL_numeric_underlying_is_standard;
1518 PL_numeric_underlying = TRUE;
1519 set_numeric_radix(! PL_numeric_standard);
1521 #endif /* USE_LOCALE_NUMERIC */
1526 * Set up for a new ctype locale.
1529 S_new_ctype(pTHX_ const char *newctype)
1532 #ifndef USE_LOCALE_CTYPE
1534 PERL_UNUSED_ARG(newctype);
1535 PERL_UNUSED_CONTEXT;
1539 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1540 * core Perl this and that 'newctype' is the name of the new locale.
1542 * This function sets up the folding arrays for all 256 bytes, assuming
1543 * that tofold() is tolc() since fold case is not a concept in POSIX,
1545 * Any code changing the locale (outside this file) should use
1546 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1547 * this function should be called directly only from this file and from
1548 * POSIX::setlocale() */
1553 /* Don't check for problems if we are suppressing the warnings */
1554 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1555 bool maybe_utf8_turkic = FALSE;
1557 PERL_ARGS_ASSERT_NEW_CTYPE;
1559 /* We will replace any bad locale warning with 1) nothing if the new one is
1560 * ok; or 2) a new warning for the bad new locale */
1561 if (PL_warn_locale) {
1562 SvREFCNT_dec_NN(PL_warn_locale);
1563 PL_warn_locale = NULL;
1566 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1568 /* A UTF-8 locale gets standard rules. But note that code still has to
1569 * handle this specially because of the three problematic code points */
1570 if (PL_in_utf8_CTYPE_locale) {
1571 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1573 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1574 * Turkic. Make sure these two are the only anomalies. (We don't use
1575 * towupper and towlower because they aren't in C89.) */
1577 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1579 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1583 if (toupper('i') == 'i' && tolower('I') == 'I') {
1586 check_for_problems = TRUE;
1587 maybe_utf8_turkic = TRUE;
1591 /* We don't populate the other lists if a UTF-8 locale, but do check that
1592 * everything works as expected, unless checking turned off */
1593 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1594 /* Assume enough space for every character being bad. 4 spaces each
1595 * for the 94 printable characters that are output like "'x' "; and 5
1596 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1598 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1599 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1601 unsigned int bad_count = 0; /* Count of bad characters */
1603 for (i = 0; i < 256; i++) {
1604 if (! PL_in_utf8_CTYPE_locale) {
1606 PL_fold_locale[i] = (U8) tolower(i);
1607 else if (islower(i))
1608 PL_fold_locale[i] = (U8) toupper(i);
1610 PL_fold_locale[i] = (U8) i;
1613 /* If checking for locale problems, see if the native ASCII-range
1614 * printables plus \n and \t are in their expected categories in
1615 * the new locale. If not, this could mean big trouble, upending
1616 * Perl's and most programs' assumptions, like having a
1617 * metacharacter with special meaning become a \w. Fortunately,
1618 * it's very rare to find locales that aren't supersets of ASCII
1619 * nowadays. It isn't a problem for most controls to be changed
1620 * into something else; we check only \n and \t, though perhaps \r
1621 * could be an issue as well. */
1622 if ( check_for_problems
1623 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1625 bool is_bad = FALSE;
1626 char name[4] = { '\0' };
1628 /* Convert the name into a string */
1633 else if (i == '\n') {
1634 my_strlcpy(name, "\\n", sizeof(name));
1636 else if (i == '\t') {
1637 my_strlcpy(name, "\\t", sizeof(name));
1641 my_strlcpy(name, "' '", sizeof(name));
1644 /* Check each possibe class */
1645 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1647 DEBUG_L(PerlIO_printf(Perl_debug_log,
1648 "isalnum('%s') unexpectedly is %d\n",
1649 name, cBOOL(isalnum(i))));
1651 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1653 DEBUG_L(PerlIO_printf(Perl_debug_log,
1654 "isalpha('%s') unexpectedly is %d\n",
1655 name, cBOOL(isalpha(i))));
1657 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1659 DEBUG_L(PerlIO_printf(Perl_debug_log,
1660 "isdigit('%s') unexpectedly is %d\n",
1661 name, cBOOL(isdigit(i))));
1663 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1665 DEBUG_L(PerlIO_printf(Perl_debug_log,
1666 "isgraph('%s') unexpectedly is %d\n",
1667 name, cBOOL(isgraph(i))));
1669 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1671 DEBUG_L(PerlIO_printf(Perl_debug_log,
1672 "islower('%s') unexpectedly is %d\n",
1673 name, cBOOL(islower(i))));
1675 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1677 DEBUG_L(PerlIO_printf(Perl_debug_log,
1678 "isprint('%s') unexpectedly is %d\n",
1679 name, cBOOL(isprint(i))));
1681 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1683 DEBUG_L(PerlIO_printf(Perl_debug_log,
1684 "ispunct('%s') unexpectedly is %d\n",
1685 name, cBOOL(ispunct(i))));
1687 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1689 DEBUG_L(PerlIO_printf(Perl_debug_log,
1690 "isspace('%s') unexpectedly is %d\n",
1691 name, cBOOL(isspace(i))));
1693 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1695 DEBUG_L(PerlIO_printf(Perl_debug_log,
1696 "isupper('%s') unexpectedly is %d\n",
1697 name, cBOOL(isupper(i))));
1699 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1701 DEBUG_L(PerlIO_printf(Perl_debug_log,
1702 "isxdigit('%s') unexpectedly is %d\n",
1703 name, cBOOL(isxdigit(i))));
1705 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1707 DEBUG_L(PerlIO_printf(Perl_debug_log,
1708 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1709 name, tolower(i), (int) toLOWER_A(i)));
1711 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1713 DEBUG_L(PerlIO_printf(Perl_debug_log,
1714 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1715 name, toupper(i), (int) toUPPER_A(i)));
1717 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1719 DEBUG_L(PerlIO_printf(Perl_debug_log,
1720 "'\\n' (=%02X) is not a control\n", (int) i));
1723 /* Add to the list; Separate multiple entries with a blank */
1726 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1728 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1734 if (bad_count == 2 && maybe_utf8_turkic) {
1736 *bad_chars_list = '\0';
1737 PL_fold_locale['I'] = 'I';
1738 PL_fold_locale['i'] = 'i';
1739 PL_in_utf8_turkic_locale = TRUE;
1740 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1741 __FILE__, __LINE__, newctype));
1744 PL_in_utf8_turkic_locale = FALSE;
1749 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1750 * this locale requires more than one byte, there are going to be
1752 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1753 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1754 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1756 if ( check_for_problems && MB_CUR_MAX > 1
1757 && ! PL_in_utf8_CTYPE_locale
1759 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1760 * locale. Just assume that the implementation for them (plus
1761 * for POSIX) is correct and the > 1 value is spurious. (Since
1762 * these are specially handled to never be considered UTF-8
1763 * locales, as long as this is the only problem, everything
1764 * should work fine */
1765 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1767 multi_byte_locale = TRUE;
1772 /* If we found problems and we want them output, do so */
1773 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1774 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1776 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1777 PL_warn_locale = Perl_newSVpvf(aTHX_
1778 "Locale '%s' contains (at least) the following characters"
1779 " which have\nunexpected meanings: %s\nThe Perl program"
1780 " will use the expected meanings",
1781 newctype, bad_chars_list);
1784 PL_warn_locale = Perl_newSVpvf(aTHX_
1785 "Locale '%s' may not work well.%s%s%s\n",
1788 ? " Some characters in it are not recognized by"
1792 ? "\nThe following characters (and maybe others)"
1793 " may not have the same meaning as the Perl"
1794 " program expects:\n"
1802 # ifdef HAS_NL_LANGINFO
1804 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1805 /* parameter FALSE is a don't care here */
1806 my_nl_langinfo(CODESET, FALSE));
1810 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1812 /* If we are actually in the scope of the locale or are debugging,
1813 * output the message now. If not in that scope, we save the
1814 * message to be output at the first operation using this locale,
1815 * if that actually happens. Most programs don't use locales, so
1816 * they are immune to bad ones. */
1817 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1819 /* The '0' below suppresses a bogus gcc compiler warning */
1820 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1822 if (IN_LC(LC_CTYPE)) {
1823 SvREFCNT_dec_NN(PL_warn_locale);
1824 PL_warn_locale = NULL;
1830 #endif /* USE_LOCALE_CTYPE */
1835 Perl__warn_problematic_locale()
1838 #ifdef USE_LOCALE_CTYPE
1842 /* Internal-to-core function that outputs the message in PL_warn_locale,
1843 * and then NULLS it. Should be called only through the macro
1844 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1846 if (PL_warn_locale) {
1847 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1848 SvPVX(PL_warn_locale),
1849 0 /* dummy to avoid compiler warning */ );
1850 SvREFCNT_dec_NN(PL_warn_locale);
1851 PL_warn_locale = NULL;
1859 S_new_collate(pTHX_ const char *newcoll)
1862 #ifndef USE_LOCALE_COLLATE
1864 PERL_UNUSED_ARG(newcoll);
1865 PERL_UNUSED_CONTEXT;
1869 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1870 * core Perl this and that 'newcoll' is the name of the new locale.
1872 * The design of locale collation is that every locale change is given an
1873 * index 'PL_collation_ix'. The first time a string particpates in an
1874 * operation that requires collation while locale collation is active, it
1875 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1876 * magic includes the collation index, and the transformation of the string
1877 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1878 * instead of the string itself. If a string changes, the magic is
1879 * cleared. The next time the locale changes, the index is incremented,
1880 * and so we know during a comparison that the transformation is not
1881 * necessarily still valid, and so is recomputed. Note that if the locale
1882 * changes enough times, the index could wrap (a U32), and it is possible
1883 * that a transformation would improperly be considered valid, leading to
1884 * an unlikely bug */
1887 if (PL_collation_name) {
1889 Safefree(PL_collation_name);
1890 PL_collation_name = NULL;
1892 PL_collation_standard = TRUE;
1893 is_standard_collation:
1894 PL_collxfrm_base = 0;
1895 PL_collxfrm_mult = 2;
1896 PL_in_utf8_COLLATE_locale = FALSE;
1897 PL_strxfrm_NUL_replacement = '\0';
1898 PL_strxfrm_max_cp = 0;
1902 /* If this is not the same locale as currently, set the new one up */
1903 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1905 Safefree(PL_collation_name);
1906 PL_collation_name = stdize_locale(savepv(newcoll));
1907 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1908 if (PL_collation_standard) {
1909 goto is_standard_collation;
1912 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1913 PL_strxfrm_NUL_replacement = '\0';
1914 PL_strxfrm_max_cp = 0;
1916 /* A locale collation definition includes primary, secondary, tertiary,
1917 * etc. weights for each character. To sort, the primary weights are
1918 * used, and only if they compare equal, then the secondary weights are
1919 * used, and only if they compare equal, then the tertiary, etc.
1921 * strxfrm() works by taking the input string, say ABC, and creating an
1922 * output transformed string consisting of first the primary weights,
1923 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1924 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1925 * may not have weights at every level. In our example, let's say B
1926 * doesn't have a tertiary weight, and A doesn't have a secondary
1927 * weight. The constructed string is then going to be
1928 * A¹B¹C¹ B²C² A³C³ ....
1929 * This has the desired effect that strcmp() will look at the secondary
1930 * or tertiary weights only if the strings compare equal at all higher
1931 * priority weights. The spaces shown here, like in
1933 * are not just for readability. In the general case, these must
1934 * actually be bytes, which we will call here 'separator weights'; and
1935 * they must be smaller than any other weight value, but since these
1936 * are C strings, only the terminating one can be a NUL (some
1937 * implementations may include a non-NUL separator weight just before
1938 * the NUL). Implementations tend to reserve 01 for the separator
1939 * weights. They are needed so that a shorter string's secondary
1940 * weights won't be misconstrued as primary weights of a longer string,
1941 * etc. By making them smaller than any other weight, the shorter
1942 * string will sort first. (Actually, if all secondary weights are
1943 * smaller than all primary ones, there is no need for a separator
1944 * weight between those two levels, etc.)
1946 * The length of the transformed string is roughly a linear function of
1947 * the input string. It's not exactly linear because some characters
1948 * don't have weights at all levels. When we call strxfrm() we have to
1949 * allocate some memory to hold the transformed string. The
1950 * calculations below try to find coefficients 'm' and 'b' for this
1951 * locale so that m*x + b equals how much space we need, given the size
1952 * of the input string in 'x'. If we calculate too small, we increase
1953 * the size as needed, and call strxfrm() again, but it is better to
1954 * get it right the first time to avoid wasted expensive string
1955 * transformations. */
1958 /* We use the string below to find how long the tranformation of it
1959 * is. Almost all locales are supersets of ASCII, or at least the
1960 * ASCII letters. We use all of them, half upper half lower,
1961 * because if we used fewer, we might hit just the ones that are
1962 * outliers in a particular locale. Most of the strings being
1963 * collated will contain a preponderance of letters, and even if
1964 * they are above-ASCII, they are likely to have the same number of
1965 * weight levels as the ASCII ones. It turns out that digits tend
1966 * to have fewer levels, and some punctuation has more, but those
1967 * are relatively sparse in text, and khw believes this gives a
1968 * reasonable result, but it could be changed if experience so
1970 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1971 char * x_longer; /* Transformed 'longer' */
1972 Size_t x_len_longer; /* Length of 'x_longer' */
1974 char * x_shorter; /* We also transform a substring of 'longer' */
1975 Size_t x_len_shorter;
1977 /* _mem_collxfrm() is used get the transformation (though here we
1978 * are interested only in its length). It is used because it has
1979 * the intelligence to handle all cases, but to work, it needs some
1980 * values of 'm' and 'b' to get it started. For the purposes of
1981 * this calculation we use a very conservative estimate of 'm' and
1982 * 'b'. This assumes a weight can be multiple bytes, enough to
1983 * hold any UV on the platform, and there are 5 levels, 4 weight
1984 * bytes, and a trailing NUL. */
1985 PL_collxfrm_base = 5;
1986 PL_collxfrm_mult = 5 * sizeof(UV);
1988 /* Find out how long the transformation really is */
1989 x_longer = _mem_collxfrm(longer,
1993 /* We avoid converting to UTF-8 in the
1994 * called function by telling it the
1995 * string is in UTF-8 if the locale is a
1996 * UTF-8 one. Since the string passed
1997 * here is invariant under UTF-8, we can
1998 * claim it's UTF-8 even though it isn't.
2000 PL_in_utf8_COLLATE_locale);
2003 /* Find out how long the transformation of a substring of 'longer'
2004 * is. Together the lengths of these transformations are
2005 * sufficient to calculate 'm' and 'b'. The substring is all of
2006 * 'longer' except the first character. This minimizes the chances
2007 * of being swayed by outliers */
2008 x_shorter = _mem_collxfrm(longer + 1,
2011 PL_in_utf8_COLLATE_locale);
2012 Safefree(x_shorter);
2014 /* If the results are nonsensical for this simple test, the whole
2015 * locale definition is suspect. Mark it so that locale collation
2016 * is not active at all for it. XXX Should we warn? */
2017 if ( x_len_shorter == 0
2018 || x_len_longer == 0
2019 || x_len_shorter >= x_len_longer)
2021 PL_collxfrm_mult = 0;
2022 PL_collxfrm_base = 0;
2025 SSize_t base; /* Temporary */
2027 /* We have both: m * strlen(longer) + b = x_len_longer
2028 * m * strlen(shorter) + b = x_len_shorter;
2029 * subtracting yields:
2030 * m * (strlen(longer) - strlen(shorter))
2031 * = x_len_longer - x_len_shorter
2032 * But we have set things up so that 'shorter' is 1 byte smaller
2033 * than 'longer'. Hence:
2034 * m = x_len_longer - x_len_shorter
2036 * But if something went wrong, make sure the multiplier is at
2039 if (x_len_longer > x_len_shorter) {
2040 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2043 PL_collxfrm_mult = 1;
2048 * but in case something has gone wrong, make sure it is
2050 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2055 /* Add 1 for the trailing NUL */
2056 PL_collxfrm_base = base + 1;
2061 if (DEBUG_L_TEST || debug_initialization) {
2062 PerlIO_printf(Perl_debug_log,
2063 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2065 " collate multipler=%zu, collate base=%zu\n",
2067 PL_in_utf8_COLLATE_locale,
2068 x_len_shorter, x_len_longer,
2069 PL_collxfrm_mult, PL_collxfrm_base);
2076 #endif /* USE_LOCALE_COLLATE */
2085 S_win32_setlocale(pTHX_ int category, const char* locale)
2087 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2088 * difference between the two unless the input locale is "", which normally
2089 * means on Windows to get the machine default, which is set via the
2090 * computer's "Regional and Language Options" (or its current equivalent).
2091 * In POSIX, it instead means to find the locale from the user's
2092 * environment. This routine changes the Windows behavior to first look in
2093 * the environment, and, if anything is found, use that instead of going to
2094 * the machine default. If there is no environment override, the machine
2095 * default is used, by calling the real setlocale() with "".
2097 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2098 * use the particular category's variable if set; otherwise to use the LANG
2101 bool override_LC_ALL = FALSE;
2105 if (locale && strEQ(locale, "")) {
2109 locale = PerlEnv_getenv("LC_ALL");
2111 if (category == LC_ALL) {
2112 override_LC_ALL = TRUE;
2118 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2119 if (category == categories[i]) {
2120 locale = PerlEnv_getenv(category_names[i]);
2125 locale = PerlEnv_getenv("LANG");
2141 result = setlocale(category, locale);
2142 DEBUG_L(STMT_START {
2144 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2145 setlocale_debug_string(category, locale, result));
2149 if (! override_LC_ALL) {
2153 /* Here the input category was LC_ALL, and we have set it to what is in the
2154 * LANG variable or the system default if there is no LANG. But these have
2155 * lower priority than the other LC_foo variables, so override it for each
2156 * one that is set. (If they are set to "", it means to use the same thing
2157 * we just set LC_ALL to, so can skip) */
2159 for (i = 0; i < LC_ALL_INDEX; i++) {
2160 result = PerlEnv_getenv(category_names[i]);
2161 if (result && strNE(result, "")) {
2162 setlocale(categories[i], result);
2163 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2165 setlocale_debug_string(categories[i], result, "not captured")));
2169 result = setlocale(LC_ALL, NULL);
2170 DEBUG_L(STMT_START {
2172 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2174 setlocale_debug_string(LC_ALL, NULL, result));
2185 =head1 Locale-related functions and macros
2187 =for apidoc Perl_setlocale
2189 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2190 taking the same parameters, and returning the same information, except that it
2191 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2192 instead return C<C> if the underlying locale has a non-dot decimal point
2193 character, or a non-empty thousands separator for displaying floating point
2194 numbers. This is because perl keeps that locale category such that it has a
2195 dot and empty separator, changing the locale briefly during the operations
2196 where the underlying one is required. C<Perl_setlocale> knows about this, and
2197 compensates; regular C<setlocale> doesn't.
2199 Another reason it isn't completely a drop-in replacement is that it is
2200 declared to return S<C<const char *>>, whereas the system setlocale omits the
2201 C<const> (presumably because its API was specified long ago, and can't be
2202 updated; it is illegal to change the information C<setlocale> returns; doing
2203 so leads to segfaults.)
2205 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2206 C<setlocale> can be completely ineffective on some platforms under some
2209 C<Perl_setlocale> should not be used to change the locale except on systems
2210 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2211 the system C<setlocale()> is ineffective, returning the wrong information, and
2212 failing to actually change the locale. C<Perl_setlocale>, however works
2213 properly in all circumstances.
2215 The return points to a per-thread static buffer, which is overwritten the next
2216 time C<Perl_setlocale> is called from the same thread.
2223 Perl_setlocale(const int category, const char * locale)
2225 /* This wraps POSIX::setlocale() */
2229 PERL_UNUSED_ARG(category);
2230 PERL_UNUSED_ARG(locale);
2236 const char * retval;
2237 const char * newlocale;
2240 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2242 #ifdef USE_LOCALE_NUMERIC
2244 /* A NULL locale means only query what the current one is. We have the
2245 * LC_NUMERIC name saved, because we are normally switched into the C
2246 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2247 * the correct results. All other categories don't require special
2249 if (locale == NULL) {
2250 if (category == LC_NUMERIC) {
2252 /* We don't have to copy this return value, as it is a per-thread
2253 * variable, and won't change until a future setlocale */
2254 return PL_numeric_name;
2259 else if (category == LC_ALL) {
2260 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2269 retval = save_to_buffer(do_setlocale_r(category, locale),
2270 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2273 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2275 if (locale == NULL && category == LC_ALL) {
2276 RESTORE_LC_NUMERIC();
2281 DEBUG_L(PerlIO_printf(Perl_debug_log,
2282 "%s:%d: %s\n", __FILE__, __LINE__,
2283 setlocale_debug_string(category, locale, retval)));
2291 /* If locale == NULL, we are just querying the state */
2292 if (locale == NULL) {
2296 /* Now that have switched locales, we have to update our records to
2301 #ifdef USE_LOCALE_CTYPE
2308 #ifdef USE_LOCALE_COLLATE
2311 new_collate(retval);
2315 #ifdef USE_LOCALE_NUMERIC
2318 new_numeric(retval);
2326 /* LC_ALL updates all the things we care about. The values may not
2327 * be the same as 'retval', as the locale "" may have set things
2330 # ifdef USE_LOCALE_CTYPE
2332 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2333 new_ctype(newlocale);
2334 Safefree(newlocale);
2336 # endif /* USE_LOCALE_CTYPE */
2337 # ifdef USE_LOCALE_COLLATE
2339 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2340 new_collate(newlocale);
2341 Safefree(newlocale);
2344 # ifdef USE_LOCALE_NUMERIC
2346 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2347 new_numeric(newlocale);
2348 Safefree(newlocale);
2350 # endif /* USE_LOCALE_NUMERIC */
2363 PERL_STATIC_INLINE const char *
2364 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2366 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2367 * growing it if necessary */
2371 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2377 string_size = strlen(string) + offset + 1;
2379 if (*buf_size == 0) {
2380 Newx(*buf, string_size, char);
2381 *buf_size = string_size;
2383 else if (string_size > *buf_size) {
2384 Renew(*buf, string_size, char);
2385 *buf_size = string_size;
2388 Copy(string, *buf + offset, string_size - offset, char);
2394 =for apidoc Perl_langinfo
2396 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2397 taking the same C<item> parameter values, and returning the same information.
2398 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2399 of Perl's locale handling from your code, and can be used on systems that lack
2400 a native C<nl_langinfo>.
2408 The reason it isn't quite a drop-in replacement is actually an advantage. The
2409 only difference is that it returns S<C<const char *>>, whereas plain
2410 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2411 forbidden to write into the buffer. By declaring this C<const>, the compiler
2412 enforces this restriction, so if it is violated, you know at compilation time,
2413 rather than getting segfaults at runtime.
2417 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2418 without you having to write extra code. The reason for the extra code would be
2419 because these are from the C<LC_NUMERIC> locale category, which is normally
2420 kept set by Perl so that the radix is a dot, and the separator is the empty
2421 string, no matter what the underlying locale is supposed to be, and so to get
2422 the expected results, you have to temporarily toggle into the underlying
2423 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2424 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2425 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2426 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2427 (decimal point) character to be a dot.)
2431 The system function it replaces can have its static return buffer trashed,
2432 not only by a subesequent call to that function, but by a C<freelocale>,
2433 C<setlocale>, or other locale change. The returned buffer of this function is
2434 not changed until the next call to it, so the buffer is never in a trashed
2439 Its return buffer is per-thread, so it also is never overwritten by a call to
2440 this function from another thread; unlike the function it replaces.
2444 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2445 as Windows, hence makes your code more portable. Of the fifty-some possible
2446 items specified by the POSIX 2008 standard,
2447 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2448 only one is completely unimplemented, though on non-Windows platforms, another
2449 significant one is also not implemented). It uses various techniques to
2450 recover the other items, including calling C<L<localeconv(3)>>, and
2451 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2452 available. Later C<strftime()> versions have additional capabilities; C<""> is
2453 returned for those not available on your system.
2455 It is important to note that when called with an item that is recovered by
2456 using C<localeconv>, the buffer from any previous explicit call to
2457 C<localeconv> will be overwritten. This means you must save that buffer's
2458 contents if you need to access them after a call to this function. (But note
2459 that you might not want to be using C<localeconv()> directly anyway, because of
2460 issues like the ones listed in the second item of this list (above) for
2461 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2462 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2465 The details for those items which may deviate from what this emulation returns
2466 and what a native C<nl_langinfo()> would return are specified in
2471 When using C<Perl_langinfo> on systems that don't have a native
2472 C<nl_langinfo()>, you must
2474 #include "perl_langinfo.h"
2476 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2477 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2478 C<langinfo.h> would try to import into the namespace for code that doesn't need
2481 The original impetus for C<Perl_langinfo()> was so that code that needs to
2482 find out the current currency symbol, floating point radix character, or digit
2483 grouping separator can use, on all systems, the simpler and more
2484 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2485 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2486 is better to use the methods given in L<perlcall> to call
2487 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2494 #ifdef HAS_NL_LANGINFO
2495 Perl_langinfo(const nl_item item)
2497 Perl_langinfo(const int item)
2500 return my_nl_langinfo(item, TRUE);
2504 #ifdef HAS_NL_LANGINFO
2505 S_my_nl_langinfo(const nl_item item, bool toggle)
2507 S_my_nl_langinfo(const int item, bool toggle)
2511 const char * retval;
2513 #ifdef USE_LOCALE_NUMERIC
2515 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2516 * two items, and only if not already there */
2517 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2518 || PL_numeric_underlying))
2520 #endif /* No toggling needed if not using LC_NUMERIC */
2524 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2525 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2526 || ! defined(HAS_POSIX_2008_LOCALE) \
2527 || ! defined(DUPLOCALE)
2529 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2530 * for those items dependent on it. This must be copied to a buffer before
2531 * switching back, as some systems destroy the buffer when setlocale() is
2535 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2538 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2541 LOCALE_LOCK; /* Prevent interference from another thread executing
2542 this code section (the only call to nl_langinfo in
2546 /* Copy to a per-thread buffer, which is also one that won't be
2547 * destroyed by a subsequent setlocale(), such as the
2548 * RESTORE_LC_NUMERIC may do just below. */
2549 retval = save_to_buffer(nl_langinfo(item),
2550 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2555 RESTORE_LC_NUMERIC();
2559 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2562 bool do_free = FALSE;
2563 locale_t cur = uselocale((locale_t) 0);
2565 if (cur == LC_GLOBAL_LOCALE) {
2566 cur = duplocale(LC_GLOBAL_LOCALE);
2570 # ifdef USE_LOCALE_NUMERIC
2573 if (PL_underlying_numeric_obj) {
2574 cur = PL_underlying_numeric_obj;
2577 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2584 /* We have to save it to a buffer, because the freelocale() just below
2585 * can invalidate the internal one */
2586 retval = save_to_buffer(nl_langinfo_l(item, cur),
2587 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2596 if (strEQ(retval, "")) {
2597 if (item == YESSTR) {
2600 if (item == NOSTR) {
2607 #else /* Below, emulate nl_langinfo as best we can */
2611 # ifdef HAS_LOCALECONV
2613 const struct lconv* lc;
2615 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2617 # ifdef TS_W32_BROKEN_LOCALECONV
2619 const char * save_global;
2620 const char * save_thread;
2628 # ifdef HAS_STRFTIME
2631 bool return_format = FALSE; /* Return the %format, not the value */
2632 const char * format;
2636 /* We copy the results to a per-thread buffer, even if not
2637 * multi-threaded. This is in part to simplify this code, and partly
2638 * because we need a buffer anyway for strftime(), and partly because a
2639 * call of localeconv() could otherwise wipe out the buffer, and the
2640 * programmer would not be expecting this, as this is a nl_langinfo()
2641 * substitute after all, so s/he might be thinking their localeconv()
2642 * is safe until another localeconv() call. */
2647 /* This is unimplemented */
2648 case ERA: /* For use with strftime() %E modifier */
2653 /* We use only an English set, since we don't know any more */
2654 case YESEXPR: return "^[+1yY]";
2655 case YESSTR: return "yes";
2656 case NOEXPR: return "^[-0nN]";
2657 case NOSTR: return "no";
2663 /* On non-windows, this is unimplemented, in part because of
2664 * inconsistencies between vendors. The Darwin native
2665 * nl_langinfo() implementation simply looks at everything past
2666 * any dot in the name, but that doesn't work for other
2667 * vendors. Many Linux locales that don't have UTF-8 in their
2668 * names really are UTF-8, for example; z/OS locales that do
2669 * have UTF-8 in their names, aren't really UTF-8 */
2674 { /* But on Windows, the name does seem to be consistent, so
2679 const char * name = my_setlocale(LC_CTYPE, NULL);
2681 if (isNAME_C_OR_POSIX(name)) {
2682 return "ANSI_X3.4-1968";
2685 /* Find the dot in the locale name */
2686 first = (const char *) strchr(name, '.');
2692 /* Look at everything past the dot */
2697 if (! isDIGIT(*p)) {
2704 /* Here everything past the dot is a digit. Treat it as a
2706 retval = save_to_buffer("CP", &PL_langinfo_buf,
2707 &PL_langinfo_bufsize, 0);
2708 offset = STRLENs("CP");
2712 retval = save_to_buffer(first, &PL_langinfo_buf,
2713 &PL_langinfo_bufsize, offset);
2719 # ifdef HAS_LOCALECONV
2723 /* We don't bother with localeconv_l() because any system that
2724 * has it is likely to also have nl_langinfo() */
2726 LOCALE_LOCK_V; /* Prevent interference with other threads
2727 using localeconv() */
2729 # ifdef TS_W32_BROKEN_LOCALECONV
2731 /* This is a workaround for a Windows bug prior to VS 15.
2732 * What we do here is, while locked, switch to the global
2733 * locale so localeconv() works; then switch back just before
2734 * the unlock. This can screw things up if some thread is
2735 * already using the global locale while assuming no other is.
2736 * A different workaround would be to call GetCurrencyFormat on
2737 * a known value, and parse it; patches welcome
2739 * We have to use LC_ALL instead of LC_MONETARY because of
2740 * another bug in Windows */
2742 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2743 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2744 save_global= savepv(my_setlocale(LC_ALL, NULL));
2745 my_setlocale(LC_ALL, save_thread);
2751 || ! lc->currency_symbol
2752 || strEQ("", lc->currency_symbol))
2758 /* Leave the first spot empty to be filled in below */
2759 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2760 &PL_langinfo_bufsize, 1);
2761 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2762 { /* khw couldn't figure out how the localedef specifications
2763 would show that the $ should replace the radix; this is
2764 just a guess as to how it might work.*/
2765 PL_langinfo_buf[0] = '.';
2767 else if (lc->p_cs_precedes) {
2768 PL_langinfo_buf[0] = '-';
2771 PL_langinfo_buf[0] = '+';
2774 # ifdef TS_W32_BROKEN_LOCALECONV
2776 my_setlocale(LC_ALL, save_global);
2777 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2778 my_setlocale(LC_ALL, save_thread);
2779 Safefree(save_global);
2780 Safefree(save_thread);
2787 # ifdef TS_W32_BROKEN_LOCALECONV
2791 /* For this, we output a known simple floating point number to
2792 * a buffer, and parse it, looking for the radix */
2795 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2798 if (PL_langinfo_bufsize < 10) {
2799 PL_langinfo_bufsize = 10;
2800 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2803 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2805 if (needed_size >= (int) PL_langinfo_bufsize) {
2806 PL_langinfo_bufsize = needed_size + 1;
2807 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2808 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2810 assert(needed_size < (int) PL_langinfo_bufsize);
2813 ptr = PL_langinfo_buf;
2814 e = PL_langinfo_buf + PL_langinfo_bufsize;
2817 while (ptr < e && *ptr != '1') {
2824 while (ptr < e && *ptr != '5') {
2828 /* Everything in between is the radix string */
2830 PL_langinfo_buf[0] = '?';
2831 PL_langinfo_buf[1] = '\0';
2835 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2839 RESTORE_LC_NUMERIC();
2842 retval = PL_langinfo_buf;
2847 case RADIXCHAR: /* No special handling needed */
2854 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2857 LOCALE_LOCK_V; /* Prevent interference with other threads
2858 using localeconv() */
2860 # ifdef TS_W32_BROKEN_LOCALECONV
2862 /* This should only be for the thousands separator. A
2863 * different work around would be to use GetNumberFormat on a
2864 * known value and parse the result to find the separator */
2865 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2866 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2867 save_global = savepv(my_setlocale(LC_ALL, NULL));
2868 my_setlocale(LC_ALL, save_thread);
2870 /* This is the start of code that for broken Windows replaces
2871 * the above and below code, and instead calls
2872 * GetNumberFormat() and then would parse that to find the
2873 * thousands separator. It needs to handle UTF-16 vs -8
2876 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2877 DEBUG_L(PerlIO_printf(Perl_debug_log,
2878 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2879 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2889 temp = (item == RADIXCHAR)
2891 : lc->thousands_sep;
2897 retval = save_to_buffer(temp, &PL_langinfo_buf,
2898 &PL_langinfo_bufsize, 0);
2900 # ifdef TS_W32_BROKEN_LOCALECONV
2902 my_setlocale(LC_ALL, save_global);
2903 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2904 my_setlocale(LC_ALL, save_thread);
2905 Safefree(save_global);
2906 Safefree(save_thread);
2913 RESTORE_LC_NUMERIC();
2919 # ifdef HAS_STRFTIME
2921 /* These are defined by C89, so we assume that strftime supports
2922 * them, and so are returned unconditionally; they may not be what
2923 * the locale actually says, but should give good enough results
2924 * for someone using them as formats (as opposed to trying to parse
2925 * them to figure out what the locale says). The other format
2926 * items are actually tested to verify they work on the platform */
2927 case D_FMT: return "%x";
2928 case T_FMT: return "%X";
2929 case D_T_FMT: return "%c";
2931 /* These formats are only available in later strfmtime's */
2932 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2934 /* The rest can be gotten from most versions of strftime(). */
2935 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2936 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2938 case AM_STR: case PM_STR:
2939 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2940 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2941 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2942 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2943 case DAY_5: case DAY_6: case DAY_7:
2944 case MON_1: case MON_2: case MON_3: case MON_4:
2945 case MON_5: case MON_6: case MON_7: case MON_8:
2946 case MON_9: case MON_10: case MON_11: case MON_12:
2950 init_tm(&tm); /* Precaution against core dumps */
2954 tm.tm_year = 2017 - 1900;
2961 "panic: %s: %d: switch case: %d problem",
2962 __FILE__, __LINE__, item);
2963 NOT_REACHED; /* NOTREACHED */
2965 case PM_STR: tm.tm_hour = 18;
2970 case ABDAY_7: tm.tm_wday++;
2971 case ABDAY_6: tm.tm_wday++;
2972 case ABDAY_5: tm.tm_wday++;
2973 case ABDAY_4: tm.tm_wday++;
2974 case ABDAY_3: tm.tm_wday++;
2975 case ABDAY_2: tm.tm_wday++;
2980 case DAY_7: tm.tm_wday++;
2981 case DAY_6: tm.tm_wday++;
2982 case DAY_5: tm.tm_wday++;
2983 case DAY_4: tm.tm_wday++;
2984 case DAY_3: tm.tm_wday++;
2985 case DAY_2: tm.tm_wday++;
2990 case ABMON_12: tm.tm_mon++;
2991 case ABMON_11: tm.tm_mon++;
2992 case ABMON_10: tm.tm_mon++;
2993 case ABMON_9: tm.tm_mon++;
2994 case ABMON_8: tm.tm_mon++;
2995 case ABMON_7: tm.tm_mon++;
2996 case ABMON_6: tm.tm_mon++;
2997 case ABMON_5: tm.tm_mon++;
2998 case ABMON_4: tm.tm_mon++;
2999 case ABMON_3: tm.tm_mon++;
3000 case ABMON_2: tm.tm_mon++;
3005 case MON_12: tm.tm_mon++;
3006 case MON_11: tm.tm_mon++;
3007 case MON_10: tm.tm_mon++;
3008 case MON_9: tm.tm_mon++;
3009 case MON_8: tm.tm_mon++;
3010 case MON_7: tm.tm_mon++;
3011 case MON_6: tm.tm_mon++;
3012 case MON_5: tm.tm_mon++;
3013 case MON_4: tm.tm_mon++;
3014 case MON_3: tm.tm_mon++;
3015 case MON_2: tm.tm_mon++;
3022 return_format = TRUE;
3027 return_format = TRUE;
3032 return_format = TRUE;
3037 return_format = TRUE;
3042 format = "%Ow"; /* Find the alternate digit for 0 */
3046 /* We can't use my_strftime() because it doesn't look at
3048 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3051 /* A zero return means one of:
3052 * a) there wasn't enough space in PL_langinfo_buf
3053 * b) the format, like a plain %p, returns empty
3054 * c) it was an illegal format, though some
3055 * implementations of strftime will just return the
3056 * illegal format as a plain character sequence.
3058 * To quickly test for case 'b)', try again but precede
3059 * the format with a plain character. If that result is
3060 * still empty, the problem is either 'a)' or 'c)' */
3062 Size_t format_size = strlen(format) + 1;
3063 Size_t mod_size = format_size + 1;
3067 Newx(mod_format, mod_size, char);
3068 Newx(temp_result, PL_langinfo_bufsize, char);
3070 my_strlcpy(mod_format + 1, format, mod_size);
3071 len = strftime(temp_result,
3072 PL_langinfo_bufsize,
3074 Safefree(mod_format);
3075 Safefree(temp_result);
3077 /* If 'len' is non-zero, it means that we had a case like
3078 * %p which means the current locale doesn't use a.m. or
3079 * p.m., and that is valid */
3082 /* Here, still didn't work. If we get well beyond a
3083 * reasonable size, bail out to prevent an infinite
3086 if (PL_langinfo_bufsize > 100 * format_size) {
3087 *PL_langinfo_buf = '\0';
3090 /* Double the buffer size to retry; Add 1 in case
3091 * original was 0, so we aren't stuck at 0. */
3092 PL_langinfo_bufsize *= 2;
3093 PL_langinfo_bufsize++;
3094 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3102 /* Here, we got a result.
3104 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3105 * alternate format for wday 0. If the value is the same as
3106 * the normal 0, there isn't an alternate, so clear the buffer.
3108 if ( item == ALT_DIGITS
3109 && strEQ(PL_langinfo_buf, "0"))
3111 *PL_langinfo_buf = '\0';
3114 /* ALT_DIGITS is problematic. Experiments on it showed that
3115 * strftime() did not always work properly when going from
3116 * alt-9 to alt-10. Only a few locales have this item defined,
3117 * and in all of them on Linux that khw was able to find,
3118 * nl_langinfo() merely returned the alt-0 character, possibly
3119 * doubled. Most Unicode digits are in blocks of 10
3120 * consecutive code points, so that is sufficient information
3121 * for those scripts, as we can infer alt-1, alt-2, .... But
3122 * for a Japanese locale, a CJK ideographic 0 is returned, and
3123 * the CJK digits are not in code point order, so you can't
3124 * really infer anything. The localedef for this locale did
3125 * specify the succeeding digits, so that strftime() works
3126 * properly on them, without needing to infer anything. But
3127 * the nl_langinfo() return did not give sufficient information
3128 * for the caller to understand what's going on. So until
3129 * there is evidence that it should work differently, this
3130 * returns the alt-0 string for ALT_DIGITS.
3132 * wday was chosen because its range is all a single digit.
3133 * Things like tm_sec have two digits as the minimum: '00' */
3137 retval = PL_langinfo_buf;
3139 /* If to return the format, not the value, overwrite the buffer
3140 * with it. But some strftime()s will keep the original format
3141 * if illegal, so change those to "" */
3142 if (return_format) {
3143 if (strEQ(PL_langinfo_buf, format)) {
3144 *PL_langinfo_buf = '\0';
3147 retval = save_to_buffer(format, &PL_langinfo_buf,
3148 &PL_langinfo_bufsize, 0);
3166 * Initialize locale awareness.
3169 Perl_init_i18nl10n(pTHX_ int printwarn)
3173 * 0 if not to output warning when setup locale is bad
3174 * 1 if to output warning based on value of PERL_BADLANG
3175 * >1 if to output regardless of PERL_BADLANG
3178 * 1 = set ok or not applicable,
3179 * 0 = fallback to a locale of lower priority
3180 * -1 = fallback to all locales failed, not even to the C locale
3182 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3183 * set, debugging information is output.
3185 * This looks more complicated than it is, mainly due to the #ifdefs.
3187 * We try to set LC_ALL to the value determined by the environment. If
3188 * there is no LC_ALL on this platform, we try the individual categories we
3189 * know about. If this works, we are done.
3191 * But if it doesn't work, we have to do something else. We search the
3192 * environment variables ourselves instead of relying on the system to do
3193 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3194 * think there is one), and the ultimate fallback "C". This is all done in
3195 * the same loop as above to avoid duplicating code, but it makes things
3196 * more complex. The 'trial_locales' array is initialized with just one
3197 * element; it causes the behavior described in the paragraph above this to
3198 * happen. If that fails, we add elements to 'trial_locales', and do extra
3199 * loop iterations to cause the behavior described in this paragraph.
3201 * On Ultrix, the locale MUST come from the environment, so there is
3202 * preliminary code to set it. I (khw) am not sure that it is necessary,
3203 * and that this couldn't be folded into the loop, but barring any real
3204 * platforms to test on, it's staying as-is
3206 * A slight complication is that in embedded Perls, the locale may already
3207 * be set-up, and we don't want to get it from the normal environment
3208 * variables. This is handled by having a special environment variable
3209 * indicate we're in this situation. We simply set setlocale's 2nd
3210 * parameter to be a NULL instead of "". That indicates to setlocale that
3211 * it is not to change anything, but to return the current value,
3212 * effectively initializing perl's db to what the locale already is.
3214 * We play the same trick with NULL if a LC_ALL succeeds. We call
3215 * setlocale() on the individual categores with NULL to get their existing
3216 * values for our db, instead of trying to change them.
3225 PERL_UNUSED_ARG(printwarn);
3227 #else /* USE_LOCALE */
3230 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3234 /* NULL uses the existing already set up locale */
3235 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3238 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3239 unsigned int trial_locales_count;
3240 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3241 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3242 bool setlocale_failure = FALSE;
3245 /* A later getenv() could zap this, so only use here */
3246 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3248 const bool locwarn = (printwarn > 1
3250 && ( ! bad_lang_use_once
3252 /* disallow with "" or "0" */
3254 && strNE("0", bad_lang_use_once)))));
3256 /* setlocale() return vals; not copied so must be looked at immediately */
3257 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3259 /* current locale for given category; should have been copied so aren't
3261 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3265 /* In some systems you can find out the system default locale
3266 * and use that as the fallback locale. */
3267 # define SYSTEM_DEFAULT_LOCALE
3269 # ifdef SYSTEM_DEFAULT_LOCALE
3271 const char *system_default_locale = NULL;
3276 # define DEBUG_LOCALE_INIT(a,b,c)
3279 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3281 # define DEBUG_LOCALE_INIT(category, locale, result) \
3283 if (debug_initialization) { \
3284 PerlIO_printf(Perl_debug_log, \
3286 __FILE__, __LINE__, \
3287 setlocale_debug_string(category, \
3293 /* Make sure the parallel arrays are properly set up */
3294 # ifdef USE_LOCALE_NUMERIC
3295 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3296 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3297 # ifdef USE_POSIX_2008_LOCALE
3298 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3301 # ifdef USE_LOCALE_CTYPE
3302 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3303 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3304 # ifdef USE_POSIX_2008_LOCALE
3305 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3308 # ifdef USE_LOCALE_COLLATE
3309 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3310 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3311 # ifdef USE_POSIX_2008_LOCALE
3312 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3315 # ifdef USE_LOCALE_TIME
3316 assert(categories[LC_TIME_INDEX] == LC_TIME);
3317 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3318 # ifdef USE_POSIX_2008_LOCALE
3319 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3322 # ifdef USE_LOCALE_MESSAGES
3323 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3324 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3325 # ifdef USE_POSIX_2008_LOCALE
3326 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3329 # ifdef USE_LOCALE_MONETARY
3330 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3331 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3332 # ifdef USE_POSIX_2008_LOCALE
3333 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3336 # ifdef USE_LOCALE_ADDRESS
3337 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3338 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3339 # ifdef USE_POSIX_2008_LOCALE
3340 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3343 # ifdef USE_LOCALE_IDENTIFICATION
3344 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3345 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3346 # ifdef USE_POSIX_2008_LOCALE
3347 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3350 # ifdef USE_LOCALE_MEASUREMENT
3351 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3352 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3353 # ifdef USE_POSIX_2008_LOCALE
3354 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3357 # ifdef USE_LOCALE_PAPER
3358 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3359 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3360 # ifdef USE_POSIX_2008_LOCALE
3361 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3364 # ifdef USE_LOCALE_TELEPHONE
3365 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3366 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3367 # ifdef USE_POSIX_2008_LOCALE
3368 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3372 assert(categories[LC_ALL_INDEX] == LC_ALL);
3373 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3374 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3375 # ifdef USE_POSIX_2008_LOCALE
3376 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3379 # endif /* DEBUGGING */
3381 /* Initialize the cache of the program's UTF-8ness for the always known
3382 * locales C and POSIX */
3383 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3384 sizeof(PL_locale_utf8ness));
3386 # ifdef USE_THREAD_SAFE_LOCALE
3389 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3393 # ifdef USE_POSIX_2008_LOCALE
3395 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3396 if (! PL_C_locale_obj) {
3397 Perl_croak_nocontext(
3398 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3400 if (DEBUG_Lv_TEST || debug_initialization) {
3401 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3406 # ifdef USE_LOCALE_NUMERIC
3408 PL_numeric_radix_sv = newSVpvs(".");
3412 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3414 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3415 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3418 # ifdef LOCALE_ENVIRON_REQUIRED
3421 * Ultrix setlocale(..., "") fails if there are no environment
3422 * variables from which to get a locale name.
3426 # error Ultrix without LC_ALL not implemented
3432 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3433 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3434 if (sl_result[LC_ALL_INDEX])
3437 setlocale_failure = TRUE;
3439 if (! setlocale_failure) {
3440 const char * locale_param;
3441 for (i = 0; i < LC_ALL_INDEX; i++) {
3442 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3445 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3446 if (! sl_result[i]) {
3447 setlocale_failure = TRUE;
3449 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3454 # endif /* LC_ALL */
3455 # endif /* LOCALE_ENVIRON_REQUIRED */
3457 /* We try each locale in the list until we get one that works, or exhaust
3458 * the list. Normally the loop is executed just once. But if setting the
3459 * locale fails, inside the loop we add fallback trials to the array and so
3460 * will execute the loop multiple times */
3461 trial_locales[0] = setlocale_init;
3462 trial_locales_count = 1;
3464 for (i= 0; i < trial_locales_count; i++) {
3465 const char * trial_locale = trial_locales[i];
3469 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3470 * when i==0, but I (khw) don't think that behavior makes much
3472 setlocale_failure = FALSE;
3474 # ifdef SYSTEM_DEFAULT_LOCALE
3475 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3477 /* On Windows machines, an entry of "" after the 0th means to use
3478 * the system default locale, which we now proceed to get. */
3479 if (strEQ(trial_locale, "")) {
3482 /* Note that this may change the locale, but we are going to do
3483 * that anyway just below */
3484 system_default_locale = do_setlocale_c(LC_ALL, "");
3485 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3487 /* Skip if invalid or if it's already on the list of locales to
3489 if (! system_default_locale) {
3490 goto next_iteration;
3492 for (j = 0; j < trial_locales_count; j++) {
3493 if (strEQ(system_default_locale, trial_locales[j])) {
3494 goto next_iteration;
3498 trial_locale = system_default_locale;
3501 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3503 # endif /* SYSTEM_DEFAULT_LOCALE */
3509 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3510 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3511 if (! sl_result[LC_ALL_INDEX]) {
3512 setlocale_failure = TRUE;
3515 /* Since LC_ALL succeeded, it should have changed all the other
3516 * categories it can to its value; so we massage things so that the
3517 * setlocales below just return their category's current values.
3518 * This adequately handles the case in NetBSD where LC_COLLATE may
3519 * not be defined for a locale, and setting it individually will
3520 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3521 * the POSIX locale. */
3522 trial_locale = NULL;
3525 # endif /* LC_ALL */
3527 if (! setlocale_failure) {
3529 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3531 = savepv(do_setlocale_r(categories[j], trial_locale));
3532 if (! curlocales[j]) {
3533 setlocale_failure = TRUE;
3535 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3538 if (! setlocale_failure) { /* All succeeded */
3539 break; /* Exit trial_locales loop */
3543 /* Here, something failed; will need to try a fallback. */
3549 if (locwarn) { /* Output failure info only on the first one */
3553 PerlIO_printf(Perl_error_log,
3554 "perl: warning: Setting locale failed.\n");
3556 # else /* !LC_ALL */
3558 PerlIO_printf(Perl_error_log,
3559 "perl: warning: Setting locale failed for the categories:\n\t");
3561 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3562 if (! curlocales[j]) {
3563 PerlIO_printf(Perl_error_log, category_names[j]);
3566 Safefree(curlocales[j]);
3570 # endif /* LC_ALL */
3572 PerlIO_printf(Perl_error_log,
3573 "perl: warning: Please check that your locale settings:\n");
3577 PerlIO_printf(Perl_error_log,
3578 "\tLANGUAGE = %c%s%c,\n",
3579 language ? '"' : '(',
3580 language ? language : "unset",
3581 language ? '"' : ')');
3584 PerlIO_printf(Perl_error_log,
3585 "\tLC_ALL = %c%s%c,\n",
3587 lc_all ? lc_all : "unset",
3588 lc_all ? '"' : ')');
3590 # if defined(USE_ENVIRON_ARRAY)
3595 /* Look through the environment for any variables of the
3596 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3597 * already handled above. These are assumed to be locale
3598 * settings. Output them and their values. */
3599 for (e = environ; *e; e++) {
3600 const STRLEN prefix_len = sizeof("LC_") - 1;
3603 if ( strBEGINs(*e, "LC_")
3604 && ! strBEGINs(*e, "LC_ALL=")
3605 && (uppers_len = strspn(*e + prefix_len,
3606 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3607 && ((*e)[prefix_len + uppers_len] == '='))
3609 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3610 (int) (prefix_len + uppers_len), *e,
3611 *e + prefix_len + uppers_len + 1);
3618 PerlIO_printf(Perl_error_log,
3619 "\t(possibly more locale environment variables)\n");
3623 PerlIO_printf(Perl_error_log,
3624 "\tLANG = %c%s%c\n",
3626 lang ? lang : "unset",
3629 PerlIO_printf(Perl_error_log,
3630 " are supported and installed on your system.\n");
3633 /* Calculate what fallback locales to try. We have avoided this
3634 * until we have to, because failure is quite unlikely. This will
3635 * usually change the upper bound of the loop we are in.
3637 * Since the system's default way of setting the locale has not
3638 * found one that works, We use Perl's defined ordering: LC_ALL,
3639 * LANG, and the C locale. We don't try the same locale twice, so
3640 * don't add to the list if already there. (On POSIX systems, the
3641 * LC_ALL element will likely be a repeat of the 0th element "",
3642 * but there's no harm done by doing it explicitly.
3644 * Note that this tries the LC_ALL environment variable even on
3645 * systems which have no LC_ALL locale setting. This may or may
3646 * not have been originally intentional, but there's no real need
3647 * to change the behavior. */
3649 for (j = 0; j < trial_locales_count; j++) {
3650 if (strEQ(lc_all, trial_locales[j])) {
3654 trial_locales[trial_locales_count++] = lc_all;
3659 for (j = 0; j < trial_locales_count; j++) {
3660 if (strEQ(lang, trial_locales[j])) {
3664 trial_locales[trial_locales_count++] = lang;
3668 # if defined(WIN32) && defined(LC_ALL)
3670 /* For Windows, we also try the system default locale before "C".
3671 * (If there exists a Windows without LC_ALL we skip this because
3672 * it gets too complicated. For those, the "C" is the next
3673 * fallback possibility). The "" is the same as the 0th element of
3674 * the array, but the code at the loop above knows to treat it
3675 * differently when not the 0th */
3676 trial_locales[trial_locales_count++] = "";
3680 for (j = 0; j < trial_locales_count; j++) {
3681 if (strEQ("C", trial_locales[j])) {
3685 trial_locales[trial_locales_count++] = "C";
3688 } /* end of first time through the loop */
3696 } /* end of looping through the trial locales */
3698 if (ok < 1) { /* If we tried to fallback */
3700 if (! setlocale_failure) { /* fallback succeeded */
3701 msg = "Falling back to";
3703 else { /* fallback failed */
3706 /* We dropped off the end of the loop, so have to decrement i to
3707 * get back to the value the last time through */
3711 msg = "Failed to fall back to";
3713 /* To continue, we should use whatever values we've got */
3715 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3716 Safefree(curlocales[j]);
3717 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3718 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3723 const char * description;
3724 const char * name = "";
3725 if (strEQ(trial_locales[i], "C")) {
3726 description = "the standard locale";
3730 # ifdef SYSTEM_DEFAULT_LOCALE
3732 else if (strEQ(trial_locales[i], "")) {
3733 description = "the system default locale";
3734 if (system_default_locale) {
3735 name = system_default_locale;
3739 # endif /* SYSTEM_DEFAULT_LOCALE */
3742 description = "a fallback locale";
3743 name = trial_locales[i];
3745 if (name && strNE(name, "")) {
3746 PerlIO_printf(Perl_error_log,
3747 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3750 PerlIO_printf(Perl_error_log,
3751 "perl: warning: %s %s.\n", msg, description);
3754 } /* End of tried to fallback */
3756 /* Done with finding the locales; update our records */
3758 # ifdef USE_LOCALE_CTYPE
3760 new_ctype(curlocales[LC_CTYPE_INDEX]);
3763 # ifdef USE_LOCALE_COLLATE
3765 new_collate(curlocales[LC_COLLATE_INDEX]);
3768 # ifdef USE_LOCALE_NUMERIC
3770 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3774 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3776 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3778 /* This caches whether each category's locale is UTF-8 or not. This
3779 * may involve changing the locale. It is ok to do this at
3780 * initialization time before any threads have started, but not later
3781 * unless thread-safe operations are used.
3782 * Caching means that if the program heeds our dictate not to change
3783 * locales in threaded applications, this data will remain valid, and
3784 * it may get queried without having to change locales. If the
3785 * environment is such that all categories have the same locale, this
3786 * isn't needed, as the code will not change the locale; but this
3787 * handles the uncommon case where the environment has disparate
3788 * locales for the categories */
3789 (void) _is_cur_LC_category_utf8(categories[i]);
3793 Safefree(curlocales[i]);
3796 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3798 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3799 * locale is UTF-8. The call to new_ctype() just above has already
3800 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3801 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3802 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3803 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3804 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3806 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3807 This is an alternative to using the -C command line switch
3808 (the -C if present will override this). */
3810 const char *p = PerlEnv_getenv("PERL_UNICODE");
3811 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3812 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3826 #endif /* USE_LOCALE */
3829 /* So won't continue to output stuff */
3830 DEBUG_INITIALIZATION_set(FALSE);
3837 #ifdef USE_LOCALE_COLLATE
3840 Perl__mem_collxfrm(pTHX_ const char *input_string,
3841 STRLEN len, /* Length of 'input_string' */
3842 STRLEN *xlen, /* Set to length of returned string
3843 (not including the collation index
3845 bool utf8 /* Is the input in UTF-8? */
3849 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3850 * differences. First, it handles embedded NULs. Second, it allocates a bit
3851 * more memory than needed for the transformed data itself. The real
3852 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3853 * the length of that, and doesn't include the collation index size.
3854 * Please see sv_collxfrm() to see how this is used. */
3856 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3858 char * s = (char *) input_string;
3859 STRLEN s_strlen = strlen(input_string);
3861 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3862 STRLEN length_in_chars;
3863 bool first_time = TRUE; /* Cleared after first loop iteration */
3865 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3867 /* Must be NUL-terminated */
3868 assert(*(input_string + len) == '\0');
3870 /* If this locale has defective collation, skip */
3871 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3872 DEBUG_L(PerlIO_printf(Perl_debug_log,
3873 "_mem_collxfrm: locale's collation is defective\n"));
3877 /* Replace any embedded NULs with the control that sorts before any others.
3878 * This will give as good as possible results on strings that don't
3879 * otherwise contain that character, but otherwise there may be
3880 * less-than-perfect results with that character and NUL. This is
3881 * unavoidable unless we replace strxfrm with our own implementation. */
3882 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3886 STRLEN sans_nuls_len;
3887 int try_non_controls;
3888 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3889 making sure 2nd byte is NUL.
3891 STRLEN this_replacement_len;
3893 /* If we don't know what non-NUL control character sorts lowest for
3894 * this locale, find it */
3895 if (PL_strxfrm_NUL_replacement == '\0') {
3897 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3898 includes the collation index
3901 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3903 /* Unlikely, but it may be that no control will work to replace
3904 * NUL, in which case we instead look for any character. Controls
3905 * are preferred because collation order is, in general, context
3906 * sensitive, with adjoining characters affecting the order, and
3907 * controls are less likely to have such interactions, allowing the
3908 * NUL-replacement to stand on its own. (Another way to look at it
3909 * is to imagine what would happen if the NUL were replaced by a
3910 * combining character; it wouldn't work out all that well.) */
3911 for (try_non_controls = 0;
3912 try_non_controls < 2;
3915 /* Look through all legal code points (NUL isn't) */
3916 for (j = 1; j < 256; j++) {
3917 char * x; /* j's xfrm plus collation index */
3918 STRLEN x_len; /* length of 'x' */
3919 STRLEN trial_len = 1;
3920 char cur_source[] = { '\0', '\0' };
3922 /* Skip non-controls the first time through the loop. The
3923 * controls in a UTF-8 locale are the L1 ones */
3924 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3931 /* Create a 1-char string of the current code point */
3932 cur_source[0] = (char) j;
3934 /* Then transform it */
3935 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3936 0 /* The string is not in UTF-8 */);
3938 /* Ignore any character that didn't successfully transform.
3944 /* If this character's transformation is lower than
3945 * the current lowest, this one becomes the lowest */
3946 if ( cur_min_x == NULL
3947 || strLT(x + COLLXFRM_HDR_LEN,
3948 cur_min_x + COLLXFRM_HDR_LEN))
3950 PL_strxfrm_NUL_replacement = j;
3951 Safefree(cur_min_x);
3957 } /* end of loop through all 255 characters */
3959 /* Stop looking if found */
3964 /* Unlikely, but possible, if there aren't any controls that
3965 * work in the locale, repeat the loop, looking for any
3966 * character that works */
3967 DEBUG_L(PerlIO_printf(Perl_debug_log,
3968 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3969 } /* End of loop to try first the controls, then any char */
3972 DEBUG_L(PerlIO_printf(Perl_debug_log,
3973 "_mem_collxfrm: Couldn't find any character to replace"
3974 " embedded NULs in locale %s with", PL_collation_name));
3978 DEBUG_L(PerlIO_printf(Perl_debug_log,
3979 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3980 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3982 Safefree(cur_min_x);
3983 } /* End of determining the character that is to replace NULs */
3985 /* If the replacement is variant under UTF-8, it must match the
3986 * UTF8-ness of the original */
3987 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3988 this_replacement_char[0] =
3989 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3990 this_replacement_char[1] =
3991 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3992 this_replacement_len = 2;
3995 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3996 /* this_replacement_char[1] = '\0' was done at initialization */
3997 this_replacement_len = 1;
4000 /* The worst case length for the replaced string would be if every
4001 * character in it is NUL. Multiply that by the length of each
4002 * replacement, and allow for a trailing NUL */
4003 sans_nuls_len = (len * this_replacement_len) + 1;
4004 Newx(sans_nuls, sans_nuls_len, char);
4007 /* Replace each NUL with the lowest collating control. Loop until have
4008 * exhausted all the NULs */
4009 while (s + s_strlen < e) {
4010 my_strlcat(sans_nuls, s, sans_nuls_len);
4012 /* Do the actual replacement */
4013 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
4015 /* Move past the input NUL */
4017 s_strlen = strlen(s);
4020 /* And add anything that trails the final NUL */
4021 my_strlcat(sans_nuls, s, sans_nuls_len);
4023 /* Switch so below we transform this modified string */
4026 } /* End of replacing NULs */
4028 /* Make sure the UTF8ness of the string and locale match */
4029 if (utf8 != PL_in_utf8_COLLATE_locale) {
4030 /* XXX convert above Unicode to 10FFFF? */
4031 const char * const t = s; /* Temporary so we can later find where the
4034 /* Here they don't match. Change the string's to be what the locale is
4037 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
4038 s = (char *) bytes_to_utf8((const U8 *) s, &len);
4041 else { /* locale is not UTF-8; but input is; downgrade the input */
4043 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
4045 /* If the downgrade was successful we are done, but if the input
4046 * contains things that require UTF-8 to represent, have to do
4047 * damage control ... */
4048 if (UNLIKELY(utf8)) {
4050 /* What we do is construct a non-UTF-8 string with
4051 * 1) the characters representable by a single byte converted
4052 * to be so (if necessary);
4053 * 2) and the rest converted to collate the same as the
4054 * highest collating representable character. That makes
4055 * them collate at the end. This is similar to how we
4056 * handle embedded NULs, but we use the highest collating
4057 * code point instead of the smallest. Like the NUL case,
4058 * this isn't perfect, but is the best we can reasonably
4059 * do. Every above-255 code point will sort the same as
4060 * the highest-sorting 0-255 code point. If that code
4061 * point can combine in a sequence with some other code
4062 * points for weight calculations, us changing something to
4063 * be it can adversely affect the results. But in most
4064 * cases, it should work reasonably. And note that this is
4065 * really an illegal situation: using code points above 255
4066 * on a locale where only 0-255 are valid. If two strings
4067 * sort entirely equal, then the sort order for the
4068 * above-255 code points will be in code point order. */
4072 /* If we haven't calculated the code point with the maximum
4073 * collating order for this locale, do so now */
4074 if (! PL_strxfrm_max_cp) {
4077 /* The current transformed string that collates the
4078 * highest (except it also includes the prefixed collation
4080 char * cur_max_x = NULL;
4082 /* Look through all legal code points (NUL isn't) */
4083 for (j = 1; j < 256; j++) {
4086 char cur_source[] = { '\0', '\0' };
4088 /* Create a 1-char string of the current code point */
4089 cur_source[0] = (char) j;
4091 /* Then transform it */
4092 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4094 /* If something went wrong (which it shouldn't), just
4095 * ignore this code point */
4100 /* If this character's transformation is higher than
4101 * the current highest, this one becomes the highest */
4102 if ( cur_max_x == NULL
4103 || strGT(x + COLLXFRM_HDR_LEN,
4104 cur_max_x + COLLXFRM_HDR_LEN))
4106 PL_strxfrm_max_cp = j;
4107 Safefree(cur_max_x);
4116 DEBUG_L(PerlIO_printf(Perl_debug_log,
4117 "_mem_collxfrm: Couldn't find any character to"
4118 " replace above-Latin1 chars in locale %s with",
4119 PL_collation_name));
4123 DEBUG_L(PerlIO_printf(Perl_debug_log,
4124 "_mem_collxfrm: highest 1-byte collating character"
4125 " in locale %s is 0x%02X\n",
4127 PL_strxfrm_max_cp));
4129 Safefree(cur_max_x);
4132 /* Here we know which legal code point collates the highest.
4133 * We are ready to construct the non-UTF-8 string. The length
4134 * will be at least 1 byte smaller than the input string
4135 * (because we changed at least one 2-byte character into a
4136 * single byte), but that is eaten up by the trailing NUL */
4142 char * e = (char *) t + len;
4144 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4146 if (UTF8_IS_INVARIANT(cur_char)) {
4149 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4150 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4152 else { /* Replace illegal cp with highest collating
4154 s[d++] = PL_strxfrm_max_cp;
4158 Renew(s, d, char); /* Free up unused space */
4163 /* Here, we have constructed a modified version of the input. It could
4164 * be that we already had a modified copy before we did this version.
4165 * If so, that copy is no longer needed */
4166 if (t != input_string) {
4171 length_in_chars = (utf8)
4172 ? utf8_length((U8 *) s, (U8 *) s + len)
4175 /* The first element in the output is the collation id, used by
4176 * sv_collxfrm(); then comes the space for the transformed string. The
4177 * equation should give us a good estimate as to how much is needed */
4178 xAlloc = COLLXFRM_HDR_LEN
4180 + (PL_collxfrm_mult * length_in_chars);
4181 Newx(xbuf, xAlloc, char);
4182 if (UNLIKELY(! xbuf)) {
4183 DEBUG_L(PerlIO_printf(Perl_debug_log,
4184 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4188 /* Store the collation id */
4189 *(U32*)xbuf = PL_collation_ix;
4191 /* Then the transformation of the input. We loop until successful, or we
4195 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4197 /* If the transformed string occupies less space than we told strxfrm()
4198 * was available, it means it successfully transformed the whole
4200 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4202 /* Some systems include a trailing NUL in the returned length.
4203 * Ignore it, using a loop in case multiple trailing NULs are
4206 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4211 /* If the first try didn't get it, it means our prediction was low.
4212 * Modify the coefficients so that we predict a larger value in any
4213 * future transformations */
4215 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4216 STRLEN computed_guess = PL_collxfrm_base
4217 + (PL_collxfrm_mult * length_in_chars);
4219 /* On zero-length input, just keep current slope instead of
4221 const STRLEN new_m = (length_in_chars != 0)
4222 ? needed / length_in_chars
4225 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4226 "%s: %d: initial size of %zu bytes for a length "
4227 "%zu string was insufficient, %zu needed\n",
4229 computed_guess, length_in_chars, needed));
4231 /* If slope increased, use it, but discard this result for
4232 * length 1 strings, as we can't be sure that it's a real slope
4234 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4238 STRLEN old_m = PL_collxfrm_mult;
4239 STRLEN old_b = PL_collxfrm_base;
4243 PL_collxfrm_mult = new_m;
4244 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4245 computed_guess = PL_collxfrm_base
4246 + (PL_collxfrm_mult * length_in_chars);
4247 if (computed_guess < needed) {
4248 PL_collxfrm_base += needed - computed_guess;
4251 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4252 "%s: %d: slope is now %zu; was %zu, base "
4253 "is now %zu; was %zu\n",
4255 PL_collxfrm_mult, old_m,
4256 PL_collxfrm_base, old_b));
4258 else { /* Slope didn't change, but 'b' did */
4259 const STRLEN new_b = needed
4262 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4263 "%s: %d: base is now %zu; was %zu\n",
4265 new_b, PL_collxfrm_base));
4266 PL_collxfrm_base = new_b;
4273 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4274 DEBUG_L(PerlIO_printf(Perl_debug_log,
4275 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4276 *xlen, PERL_INT_MAX));
4280 /* A well-behaved strxfrm() returns exactly how much space it needs
4281 * (usually not including the trailing NUL) when it fails due to not
4282 * enough space being provided. Assume that this is the case unless
4283 * it's been proven otherwise */
4284 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4285 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4287 else { /* Here, either:
4288 * 1) The strxfrm() has previously shown bad behavior; or
4289 * 2) It isn't the first time through the loop, which means
4290 * that the strxfrm() is now showing bad behavior, because
4291 * we gave it what it said was needed in the previous
4292 * iteration, and it came back saying it needed still more.
4293 * (Many versions of cygwin fit this. When the buffer size
4294 * isn't sufficient, they return the input size instead of
4295 * how much is needed.)
4296 * Increase the buffer size by a fixed percentage and try again.
4298 xAlloc += (xAlloc / 4) + 1;
4299 PL_strxfrm_is_behaved = FALSE;
4303 if (DEBUG_Lv_TEST || debug_initialization) {
4304 PerlIO_printf(Perl_debug_log,
4305 "_mem_collxfrm required more space than previously calculated"
4306 " for locale %s, trying again with new guess=%d+%zu\n",
4307 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4308 xAlloc - COLLXFRM_HDR_LEN);
4315 Renew(xbuf, xAlloc, char);
4316 if (UNLIKELY(! xbuf)) {
4317 DEBUG_L(PerlIO_printf(Perl_debug_log,
4318 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4328 if (DEBUG_Lv_TEST || debug_initialization) {
4330 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4331 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4332 PerlIO_printf(Perl_debug_log, "%s\n",
4333 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4339 /* Free up unneeded space; retain ehough for trailing NUL */
4340 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4342 if (s != input_string) {
4350 if (s != input_string) {
4357 if (DEBUG_Lv_TEST || debug_initialization) {
4358 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4369 S_print_collxfrm_input_and_return(pTHX_
4370 const char * const s,
4371 const char * const e,
4372 const STRLEN * const xlen,
4376 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4378 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4379 (UV)PL_collation_ix);
4381 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4384 PerlIO_printf(Perl_debug_log, "NULL");
4386 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4388 print_bytes_for_locale(s, e, is_utf8);
4390 PerlIO_printf(Perl_debug_log, "'\n");
4393 # endif /* DEBUGGING */
4394 #endif /* USE_LOCALE_COLLATE */
4399 S_print_bytes_for_locale(pTHX_
4400 const char * const s,
4401 const char * const e,
4405 bool prev_was_printable = TRUE;
4406 bool first_time = TRUE;
4408 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4412 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4415 if (! prev_was_printable) {
4416 PerlIO_printf(Perl_debug_log, " ");
4418 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4419 prev_was_printable = TRUE;
4423 PerlIO_printf(Perl_debug_log, " ");
4425 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4426 prev_was_printable = FALSE;
4428 t += (is_utf8) ? UTF8SKIP(t) : 1;
4433 # endif /* #ifdef DEBUGGING */
4436 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4438 /* Changes the locale for LC_'switch_category" to that of
4439 * LC_'template_category', if they aren't already the same. If not NULL,
4440 * 'template_locale' is the locale that 'template_category' is in.
4442 * Returns a copy of the name of the original locale for 'switch_category'
4443 * so can be switched back to with the companion function
4444 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4446 char * restore_to_locale = NULL;
4448 if (switch_category == template_category) { /* No changes needed */
4452 /* Find the original locale of the category we may need to change, so that
4453 * it can be restored to later */
4454 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4456 if (! restore_to_locale) {
4458 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4459 __FILE__, __LINE__, category_name(switch_category), errno);
4462 /* If the locale of the template category wasn't passed in, find it now */
4463 if (template_locale == NULL) {
4464 template_locale = do_setlocale_r(template_category, NULL);
4465 if (! template_locale) {
4467 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4468 __FILE__, __LINE__, category_name(template_category), errno);
4472 /* It the locales are the same, there's nothing to do */
4473 if (strEQ(restore_to_locale, template_locale)) {
4474 Safefree(restore_to_locale);
4476 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4477 category_name(switch_category), restore_to_locale));
4482 /* Finally, change the locale to the template one */
4483 if (! do_setlocale_r(switch_category, template_locale)) {
4485 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4486 __FILE__, __LINE__, category_name(switch_category),
4487 template_locale, errno);
4490 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4491 category_name(switch_category), template_locale));
4493 return restore_to_locale;
4497 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4499 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4500 * copy that will be freed by this function), or do nothing if the latter
4501 * parameter is NULL */
4503 if (original_locale == NULL) {
4507 if (! do_setlocale_r(category, original_locale)) {
4509 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4511 category_name(category), original_locale, errno);
4514 Safefree(original_locale);
4517 /* is_cur_LC_category_utf8 uses a small char buffer to avoid malloc/free */
4518 #define CUR_LC_BUFFER_SIZE 64
4521 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4523 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4524 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4525 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4526 * could give the wrong result. The result will very likely be correct for
4527 * languages that have commonly used non-ASCII characters, but for notably
4528 * English, it comes down to if the locale's name ends in something like
4529 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4531 * If the platform is early C89, not containing mbtowc(), or we are
4532 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4533 * These work very well for non-Latin locales or those whose currency
4534 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4535 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4536 * on the name containing UTF-8 or not. */
4538 /* Name of current locale corresponding to the input category */
4539 const char *save_input_locale = NULL;
4541 bool is_utf8 = FALSE; /* The return value */
4543 /* The variables below are for the cache of previous lookups using this
4544 * function. The cache is a C string, described at the definition for
4545 * 'C_and_POSIX_utf8ness'.
4547 * The first part of the cache is fixed, for the C and POSIX locales. The
4548 * varying part starts just after them. */
4549 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4551 Size_t utf8ness_cache_size; /* Size of the varying portion */
4552 Size_t input_name_len; /* Length in bytes of save_input_locale */
4553 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4554 the name in the cache */
4555 char * delimited; /* The name plus the delimiters used to store
4557 char buffer[CUR_LC_BUFFER_SIZE]; /* small buffer */
4558 char * name_pos; /* position of 'delimited' in the cache, or 0
4564 assert(category != LC_ALL);
4568 /* Get the desired category's locale */
4569 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4570 if (! save_input_locale) {
4572 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4573 __FILE__, __LINE__, category_name(category), errno);
4576 DEBUG_L(PerlIO_printf(Perl_debug_log,
4577 "Current locale for %s is %s\n",
4578 category_name(category), save_input_locale));
4580 input_name_len = strlen(save_input_locale);
4582 /* In our cache, each name is accompanied by two delimiters and a single
4584 input_name_len_with_overhead = input_name_len + 3;
4586 if ( input_name_len_with_overhead <= CUR_LC_BUFFER_SIZE ) {
4587 /* we can use the buffer, avoid a malloc */
4589 } else { /* need a malloc */
4590 /* Allocate and populate space for a copy of the name surrounded by the
4592 Newx(delimited, input_name_len_with_overhead, char);
4595 delimited[0] = UTF8NESS_SEP[0];
4596 Copy(save_input_locale, delimited + 1, input_name_len, char);
4597 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4598 delimited[input_name_len+2] = '\0';
4600 /* And see if that is in the cache */
4601 name_pos = instr(PL_locale_utf8ness, delimited);
4603 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4607 if (DEBUG_Lv_TEST || debug_initialization) {
4608 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4609 save_input_locale, is_utf8);
4614 /* And, if not already in that position, move it to the beginning of
4615 * the non-constant portion of the list, since it is the most recently
4616 * used. (We don't have to worry about overflow, since just moving
4617 * existing names around) */
4618 if (name_pos > utf8ness_cache) {
4619 Move(utf8ness_cache,
4620 utf8ness_cache + input_name_len_with_overhead,
4621 name_pos - utf8ness_cache, char);
4624 input_name_len_with_overhead - 1, char);
4625 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4628 /* free only when not using the buffer */
4629 if ( delimited != buffer ) Safefree(delimited);
4630 Safefree(save_input_locale);
4634 /* Here we don't have stored the utf8ness for the input locale. We have to
4637 # if defined(USE_LOCALE_CTYPE) \
4638 && ( defined(HAS_NL_LANGINFO) \
4639 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4642 const char *original_ctype_locale
4643 = switch_category_locale_to_template(LC_CTYPE,
4647 /* Here the current LC_CTYPE is set to the locale of the category whose
4648 * information is desired. This means that nl_langinfo() and mbtowc()
4649 * should give the correct results */
4651 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4652 calling the functions if we have this */
4654 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4655 * Unicode code point. */
4657 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4658 __FILE__, __LINE__, (int) MB_CUR_MAX));
4659 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
4661 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4662 goto finish_and_return;
4666 # if defined(HAS_NL_LANGINFO)
4668 { /* The task is easiest if the platform has this POSIX 2001 function.
4669 Except on some platforms it can wrongly return "", so have to have
4670 a fallback. And it can return that it's UTF-8, even if there are
4671 variances from that. For example, Turkish locales may use the
4672 alternate dotted I rules, and sometimes it appears to be a
4673 defective locale definition. XXX We should probably check for
4674 these in the Latin1 range and warn (but on glibc, requires
4675 iswalnum() etc. due to their not handling 80-FF correctly */
4676 const char *codeset = my_nl_langinfo(CODESET, FALSE);
4677 /* FALSE => already in dest locale */
4679 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4680 "\tnllanginfo returned CODESET '%s'\n", codeset));
4682 if (codeset && strNE(codeset, "")) {
4684 /* If the implementation of foldEQ() somehow were
4685 * to change to not go byte-by-byte, this could
4686 * read past end of string, as only one length is
4687 * checked. But currently, a premature NUL will
4688 * compare false, and it will stop there */
4689 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
4690 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
4692 DEBUG_L(PerlIO_printf(Perl_debug_log,
4693 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
4695 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4696 goto finish_and_return;
4701 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
4702 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
4703 * late adder to C89, so very likely to have it. However, testing has
4704 * shown that, like nl_langinfo() above, there are locales that are not
4705 * strictly UTF-8 that this will return that they are */
4712 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4718 /* mbrtowc() and mbtowc() convert a byte string to a wide
4719 * character. Feed a byte string to one of them and check that the
4720 * result is the expected Unicode code point */
4722 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4723 /* Prefer this function if available, as it's reentrant */
4725 memset(&ps, 0, sizeof(ps));;
4726 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
4729 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
4735 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
4737 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
4744 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4745 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
4746 len, (unsigned int) wc, GET_ERRNO));
4748 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
4749 && wc == (wchar_t) UNICODE_REPLACEMENT);
4754 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4755 goto finish_and_return;
4760 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
4761 * try looking at the currency symbol to see if it disambiguates
4762 * things. Often that will be in the native script, and if the symbol
4763 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
4764 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
4765 * string being valid UTF-8 are quite small */
4767 # ifdef USE_LOCALE_MONETARY
4769 /* If have LC_MONETARY, we can look at the currency symbol. Often that
4770 * will be in the native script. We do this one first because there is
4771 * just one string to examine, so potentially avoids work */
4774 const char *original_monetary_locale
4775 = switch_category_locale_to_template(LC_MONETARY,
4778 bool only_ascii = FALSE;
4779 const U8 * currency_string
4780 = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE);
4781 /* 2nd param not relevant for this item */
4782 const U8 * first_variant;
4784 assert( *currency_string == '-'
4785 || *currency_string == '+'
4786 || *currency_string == '.');
4790 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
4792 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));
4796 is_utf8 = is_strict_utf8_string(first_variant, 0);
4799 restore_switched_locale(LC_MONETARY, original_monetary_locale);
4803 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
4804 * otherwise assume the locale is UTF-8 if and only if the symbol
4805 * is non-ascii UTF-8. */
4806 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
4807 save_input_locale, is_utf8));
4808 goto finish_and_return;
4812 # endif /* USE_LOCALE_MONETARY */
4813 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
4815 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
4816 * the names of the months and weekdays, timezone, and am/pm indicator */
4818 const char *original_time_locale
4819 = switch_category_locale_to_template(LC_TIME,
4823 bool is_dst = FALSE;
4827 char * formatted_time;
4829 /* Here the current LC_TIME is set to the locale of the category
4830 * whose information is desired. Look at all the days of the week and
4831 * month names, and the timezone and am/pm indicator for UTF-8 variant
4832 * characters. The first such a one found will tell us if the locale
4833 * is UTF-8 or not */
4835 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
4836 formatted_time = my_strftime("%A %B %Z %p",
4837 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
4838 if ( ! formatted_time
4839 || is_utf8_invariant_string((U8 *) formatted_time, 0))
4842 /* Here, we didn't find a non-ASCII. Try the next time through
4843 * with the complemented dst and am/pm, and try with the next
4844 * weekday. After we have gotten all weekdays, try the next
4847 hour = (hour + 12) % 24;
4855 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
4856 * false otherwise. But first, restore LC_TIME to its original
4857 * locale if we changed it */
4858 restore_switched_locale(LC_TIME, original_time_locale);
4860 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
4862 is_utf8_string((U8 *) formatted_time, 0)));
4863 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
4864 goto finish_and_return;
4867 /* Falling off the end of the loop indicates all the names were just
4868 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
4869 * to its original locale */
4870 restore_switched_locale(LC_TIME, original_time_locale);
4871 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));
4876 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
4878 /* This code is ifdefd out because it was found to not be necessary in testing
4879 * on our dromedary test machine, which has over 700 locales. There, this
4880 * added no value to looking at the currency symbol and the time strings. I
4881 * left it in so as to avoid rewriting it if real-world experience indicates
4882 * that dromedary is an outlier. Essentially, instead of returning abpve if we
4883 * haven't found illegal utf8, we continue on and examine all the strerror()
4884 * messages on the platform for utf8ness. If all are ASCII, we still don't
4885 * know the answer; but otherwise we have a pretty good indication of the
4886 * utf8ness. The reason this doesn't help much is that the messages may not
4887 * have been translated into the locale. The currency symbol and time strings
4888 * are much more likely to have been translated. */
4891 bool non_ascii = FALSE;
4892 const char *original_messages_locale
4893 = switch_category_locale_to_template(LC_MESSAGES,
4896 const char * errmsg = NULL;
4898 /* Here the current LC_MESSAGES is set to the locale of the category
4899 * whose information is desired. Look through all the messages. We
4900 * can't use Strerror() here because it may expand to code that
4901 * segfaults in miniperl */
4903 for (e = 0; e <= sys_nerr; e++) {
4905 errmsg = sys_errlist[e];
4906 if (errno || !errmsg) {
4909 errmsg = savepv(errmsg);
4910 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
4912 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
4918 restore_switched_locale(LC_MESSAGES, original_messages_locale);
4922 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
4923 * any non-ascii means it is one; otherwise we assume it isn't */
4924 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
4927 goto finish_and_return;
4930 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));
4934 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
4937 /* As a last resort, look at the locale name to see if it matches
4938 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
4939 * return of setlocale(), is actually defined to be opaque, so we can't
4940 * really rely on the absence of various substrings in the name to indicate
4941 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
4942 * be a UTF-8 locale. Similarly for the other common names */
4945 const Size_t final_pos = strlen(save_input_locale) - 1;
4947 if (final_pos >= 3) {
4948 const char *name = save_input_locale;
4950 /* Find next 'U' or 'u' and look from there */
4951 while ((name += strcspn(name, "Uu") + 1)
4952 <= save_input_locale + final_pos - 2)
4954 if ( isALPHA_FOLD_NE(*name, 't')
4955 || isALPHA_FOLD_NE(*(name + 1), 'f'))
4960 if (*(name) == '-') {
4961 if ((name > save_input_locale + final_pos - 1)) {
4966 if (*(name) == '8') {
4967 DEBUG_L(PerlIO_printf(Perl_debug_log,
4968 "Locale %s ends with UTF-8 in name\n",
4969 save_input_locale));
4971 goto finish_and_return;
4974 DEBUG_L(PerlIO_printf(Perl_debug_log,
4975 "Locale %s doesn't end with UTF-8 in name\n",
4976 save_input_locale));
4981 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
4982 if (memENDs(save_input_locale, final_pos, "65001")) {
4983 DEBUG_L(PerlIO_printf(Perl_debug_log,
4984 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
4985 save_input_locale));
4987 goto finish_and_return;
4994 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
4995 * since we are about to return FALSE anyway, there is no point in doing
4996 * this extra work */
4999 if (instr(save_input_locale, "8859")) {
5000 DEBUG_L(PerlIO_printf(Perl_debug_log,
5001 "Locale %s has 8859 in name, not UTF-8 locale\n",
5002 save_input_locale));
5004 goto finish_and_return;
5008 DEBUG_L(PerlIO_printf(Perl_debug_log,
5009 "Assuming locale %s is not a UTF-8 locale\n",
5010 save_input_locale));
5013 # endif /* the code that is compiled when no modern LC_CTYPE */
5017 /* Cache this result so we don't have to go through all this next time. */
5018 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
5019 - (utf8ness_cache - PL_locale_utf8ness);
5021 /* But we can't save it if it is too large for the total space available */
5022 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
5023 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
5025 /* Here it can fit, but we may need to clear out the oldest cached
5026 * result(s) to do so. Check */
5027 if (utf8ness_cache_len + input_name_len_with_overhead
5028 >= utf8ness_cache_size)
5030 /* Here we have to clear something out to make room for this.
5031 * Start looking at the rightmost place where it could fit and find
5032 * the beginning of the entry that extends past that. */
5033 char * cutoff = (char *) my_memrchr(utf8ness_cache,
5036 - input_name_len_with_overhead);
5039 assert(cutoff >= utf8ness_cache);
5041 /* This and all subsequent entries must be removed */
5043 utf8ness_cache_len = strlen(utf8ness_cache);
5046 /* Make space for the new entry */
5047 Move(utf8ness_cache,
5048 utf8ness_cache + input_name_len_with_overhead,
5049 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
5052 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
5053 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
5055 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
5056 & (PERL_UINTMAX_T) ~1) != '0')
5059 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu,"
5060 " inserted_name=%s, its_len=%zu\n",
5062 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
5063 delimited, input_name_len_with_overhead);
5069 if (DEBUG_Lv_TEST) {
5070 const char * s = PL_locale_utf8ness;
5072 /* Audit the structure */
5073 while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) {
5076 if (*s != UTF8NESS_SEP[0]) {
5078 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5079 " separator %.*s<-- HERE %s\n",
5081 (int) (s - PL_locale_utf8ness), PL_locale_utf8ness,
5085 e = strchr(s, UTF8NESS_PREFIX[0]);
5088 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5089 " separator %.*s<-- HERE %s\n",
5091 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5095 if (*e != '0' && *e != '1') {
5097 "panic: %s: %d: Corrupt utf8ness_cache: utf8ness"
5098 " must be [01] %.*s<-- HERE %s\n",
5100 (int) (e + 1 - PL_locale_utf8ness),
5101 PL_locale_utf8ness, e + 1);
5103 if (ninstr(PL_locale_utf8ness, s, s-1, e)) {
5105 "panic: %s: %d: Corrupt utf8ness_cache: entry"
5106 " has duplicate %.*s<-- HERE %s\n",
5108 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5115 if (DEBUG_Lv_TEST || debug_initialization) {
5117 PerlIO_printf(Perl_debug_log,
5118 "PL_locale_utf8ness is now %s; returning %d\n",
5119 PL_locale_utf8ness, is_utf8);
5124 /* free only when not using the buffer */
5125 if ( delimited != buffer ) Safefree(delimited);
5126 Safefree(save_input_locale);
5133 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
5136 /* Internal function which returns if we are in the scope of a pragma that
5137 * enables the locale category 'category'. 'compiling' should indicate if
5138 * this is during the compilation phase (TRUE) or not (FALSE). */
5140 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
5142 SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0);
5143 if (! these_categories || these_categories == &PL_sv_placeholder) {
5147 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
5148 * a valid unsigned */
5149 assert(category >= -1);
5150 return cBOOL(SvUV(these_categories) & (1U << (category + 1)));
5154 Perl_my_strerror(pTHX_ const int errnum)
5156 /* Returns a mortalized copy of the text of the error message associated
5157 * with 'errnum'. It uses the current locale's text unless the platform
5158 * doesn't have the LC_MESSAGES category or we are not being called from
5159 * within the scope of 'use locale'. In the former case, it uses whatever
5160 * strerror returns; in the latter case it uses the text from the C locale.
5162 * The function just calls strerror(), but temporarily switches, if needed,
5163 * to the C locale */
5168 #ifndef USE_LOCALE_MESSAGES
5170 /* If platform doesn't have messages category, we don't do any switching to
5171 * the C locale; we just use whatever strerror() returns */
5173 errstr = savepv(Strerror(errnum));
5175 #else /* Has locale messages */
5177 const bool within_locale_scope = IN_LC(LC_MESSAGES);
5179 # ifndef USE_ITHREADS
5181 /* This function is trivial without threads. */
5182 if (within_locale_scope) {
5183 errstr = savepv(strerror(errnum));
5186 const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL));
5188 do_setlocale_c(LC_MESSAGES, "C");
5189 errstr = savepv(strerror(errnum));
5190 do_setlocale_c(LC_MESSAGES, save_locale);
5191 Safefree(save_locale);
5194 # elif defined(HAS_POSIX_2008_LOCALE) \
5195 && defined(HAS_STRERROR_L) \
5196 && defined(HAS_DUPLOCALE)
5198 /* This function is also trivial if we don't have to worry about thread
5199 * safety and have strerror_l(), as it handles the switch of locales so we
5200 * don't have to deal with that. We don't have to worry about thread
5201 * safety if strerror_r() is also available. Both it and strerror_l() are
5202 * thread-safe. Plain strerror() isn't thread safe. But on threaded
5203 * builds when strerror_r() is available, the apparent call to strerror()
5204 * below is actually a macro that behind-the-scenes calls strerror_r(). */
5206 # ifdef HAS_STRERROR_R
5208 if (within_locale_scope) {
5209 errstr = savepv(strerror(errnum));
5212 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
5217 /* Here we have strerror_l(), but not strerror_r() and we are on a
5218 * threaded-build. We use strerror_l() for everything, constructing a
5219 * locale to pass to it if necessary */
5221 bool do_free = FALSE;
5222 locale_t locale_to_use;
5224 if (within_locale_scope) {
5225 locale_to_use = uselocale((locale_t) 0);
5226 if (locale_to_use == LC_GLOBAL_LOCALE) {
5227 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
5231 else { /* Use C locale if not within 'use locale' scope */
5232 locale_to_use = PL_C_locale_obj;
5235 errstr = savepv(strerror_l(errnum, locale_to_use));
5238 freelocale(locale_to_use);
5242 # else /* Doesn't have strerror_l() */
5244 const char * save_locale = NULL;
5245 bool locale_is_C = FALSE;
5247 /* We have a critical section to prevent another thread from executing this
5248 * same code at the same time. (On thread-safe perls, the LOCK is a
5249 * no-op.) Since this is the only place in core that changes LC_MESSAGES
5250 * (unless the user has called setlocale(), this works to prevent races. */
5253 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5254 "my_strerror called with errnum %d\n", errnum));
5255 if (! within_locale_scope) {
5256 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
5257 if (! save_locale) {
5259 "panic: %s: %d: Could not find current LC_MESSAGES locale,"
5260 " errno=%d\n", __FILE__, __LINE__, errno);
5263 locale_is_C = isNAME_C_OR_POSIX(save_locale);
5265 /* Switch to the C locale if not already in it */
5266 if (! locale_is_C) {
5268 /* The setlocale() just below likely will zap 'save_locale', so
5270 save_locale = savepv(save_locale);
5271 do_setlocale_c(LC_MESSAGES, "C");
5274 } /* end of ! within_locale_scope */
5276 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
5277 __FILE__, __LINE__));
5280 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5281 "Any locale change has been done; about to call Strerror\n"));
5282 errstr = savepv(Strerror(errnum));
5284 if (! within_locale_scope) {
5285 if (save_locale && ! locale_is_C) {
5286 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
5288 "panic: %s: %d: setlocale restore failed, errno=%d\n",
5289 __FILE__, __LINE__, errno);
5291 Safefree(save_locale);
5297 # endif /* End of doesn't have strerror_l */
5300 if (DEBUG_Lv_TEST) {
5301 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
5302 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
5303 PerlIO_printf(Perl_debug_log, "'\n");
5307 #endif /* End of does have locale messages */
5315 =for apidoc switch_to_global_locale
5317 On systems without locale support, or on typical single-threaded builds, or on
5318 platforms that do not support per-thread locale operations, this function does
5319 nothing. On such systems that do have locale support, only a locale global to
5320 the whole program is available.
5322 On multi-threaded builds on systems that do have per-thread locale operations,
5323 this function converts the thread it is running in to use the global locale.
5324 This is for code that has not yet or cannot be updated to handle multi-threaded
5325 locale operation. As long as only a single thread is so-converted, everything
5326 works fine, as all the other threads continue to ignore the global one, so only
5327 this thread looks at it.
5329 However, on Windows systems this isn't quite true prior to Visual Studio 15,
5330 at which point Microsoft fixed a bug. A race can occur if you use the
5331 following operations on earlier Windows platforms:
5335 =item L<POSIX::localeconv|POSIX/localeconv>
5337 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5339 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5343 The first item is not fixable (except by upgrading to a later Visual Studio
5344 release), but it would be possible to work around the latter two items by using
5345 the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches
5348 Without this function call, threads that use the L<C<setlocale(3)>> system
5349 function will not work properly, as all the locale-sensitive functions will
5350 look at the per-thread locale, and C<setlocale> will have no effect on this
5353 Perl code should convert to either call
5354 L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system
5355 C<setlocale>) or use the methods given in L<perlcall> to call
5356 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly
5357 handle all cases of single- vs multi-thread, POSIX 2008-supported or not.
5359 Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can
5360 continue to work if this function is called before transferring control to the
5363 Upon return from the code that needs to use the global locale,
5364 L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe
5365 multi-thread operation.
5371 Perl_switch_to_global_locale()
5374 #ifdef USE_THREAD_SAFE_LOCALE
5377 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
5380 # ifdef HAS_QUERYLOCALE
5382 setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0)));
5389 for (i = 0; i < LC_ALL_INDEX; i++) {
5390 setlocale(categories[i], do_setlocale_r(categories[i], NULL));
5396 uselocale(LC_GLOBAL_LOCALE);
5405 =for apidoc sync_locale
5407 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or
5408 change the locale (though changing the locale is antisocial and dangerous on
5409 multi-threaded systems that don't have multi-thread safe locale operations.
5410 (See L<perllocale/Multi-threaded operation>). Using the system
5411 L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries
5412 called from XS, such as C<Gtk> do so, and this can't be changed. When the
5413 locale is changed by XS code that didn't use
5414 L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the
5415 locale has changed. Use this function to do so, before returning to Perl.
5417 The return value is a boolean: TRUE if the global locale at the time of call
5418 was in effect; and FALSE if a per-thread locale was in effect. This can be
5419 used by the caller that needs to restore things as-they-were to decide whether
5421 L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>.
5436 const char * newlocale;
5439 # ifdef USE_POSIX_2008_LOCALE
5441 bool was_in_global_locale = FALSE;
5442 locale_t cur_obj = uselocale((locale_t) 0);
5444 /* On Windows, unless the foreign code has turned off the thread-safe
5445 * locale setting, any plain setlocale() will have affected what we see, so
5446 * no need to worry. Otherwise, If the foreign code has done a plain
5447 * setlocale(), it will only affect the global locale on POSIX systems, but
5448 * will affect the */
5449 if (cur_obj == LC_GLOBAL_LOCALE) {
5451 # ifdef HAS_QUERY_LOCALE
5453 do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL));
5459 /* We can't trust that we can read the LC_ALL format on the
5460 * platform, so do them individually */
5461 for (i = 0; i < LC_ALL_INDEX; i++) {
5462 do_setlocale_r(categories[i], setlocale(categories[i], NULL));
5467 was_in_global_locale = TRUE;
5472 bool was_in_global_locale = TRUE;
5475 # ifdef USE_LOCALE_CTYPE
5477 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
5478 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5479 "%s:%d: %s\n", __FILE__, __LINE__,
5480 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
5481 new_ctype(newlocale);
5482 Safefree(newlocale);
5484 # endif /* USE_LOCALE_CTYPE */
5485 # ifdef USE_LOCALE_COLLATE
5487 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
5488 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5489 "%s:%d: %s\n", __FILE__, __LINE__,
5490 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
5491 new_collate(newlocale);
5492 Safefree(newlocale);
5495 # ifdef USE_LOCALE_NUMERIC
5497 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
5498 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5499 "%s:%d: %s\n", __FILE__, __LINE__,
5500 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
5501 new_numeric(newlocale);
5502 Safefree(newlocale);
5504 # endif /* USE_LOCALE_NUMERIC */
5506 return was_in_global_locale;
5512 #if defined(DEBUGGING) && defined(USE_LOCALE)
5515 S_setlocale_debug_string(const int category, /* category number,
5517 const char* const locale, /* locale name */
5519 /* return value from setlocale() when attempting to
5520 * set 'category' to 'locale' */
5521 const char* const retval)
5523 /* Returns a pointer to a NUL-terminated string in static storage with
5524 * added text about the info passed in. This is not thread safe and will
5525 * be overwritten by the next call, so this should be used just to
5526 * formulate a string to immediately print or savepv() on. */
5528 /* initialise to a non-null value to keep it out of BSS and so keep
5529 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
5530 static char ret[256] = "If you can read this, thank your buggy C"
5531 " library strlcpy(), and change your hints file"
5534 my_strlcpy(ret, "setlocale(", sizeof(ret));
5535 my_strlcat(ret, category_name(category), sizeof(ret));
5536 my_strlcat(ret, ", ", sizeof(ret));
5539 my_strlcat(ret, "\"", sizeof(ret));
5540 my_strlcat(ret, locale, sizeof(ret));
5541 my_strlcat(ret, "\"", sizeof(ret));
5544 my_strlcat(ret, "NULL", sizeof(ret));
5547 my_strlcat(ret, ") returned ", sizeof(ret));
5550 my_strlcat(ret, "\"", sizeof(ret));
5551 my_strlcat(ret, retval, sizeof(ret));
5552 my_strlcat(ret, "\"", sizeof(ret));
5555 my_strlcat(ret, "NULL", sizeof(ret));
5558 assert(strlen(ret) < sizeof(ret));
5566 Perl_thread_locale_init()
5568 /* Called from a thread on startup*/
5570 #ifdef USE_THREAD_SAFE_LOCALE
5574 /* C starts the new thread in the global C locale. If we are thread-safe,
5575 * we want to not be in the global locale */
5577 DEBUG_L(PerlIO_printf(Perl_debug_log,
5578 "%s:%d: new thread, initial locale is %s; calling setlocale\n",
5579 __FILE__, __LINE__, setlocale(LC_ALL, NULL)));
5583 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
5587 Perl_setlocale(LC_ALL, "C");
5595 Perl_thread_locale_term()
5597 /* Called from a thread as it gets ready to terminate */
5599 #ifdef USE_THREAD_SAFE_LOCALE
5601 /* C starts the new thread in the global C locale. If we are thread-safe,
5602 * we want to not be in the global locale */
5608 locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE);
5609 if (cur_obj != LC_GLOBAL_LOCALE && cur_obj != PL_C_locale_obj) {
5610 freelocale(cur_obj);
5620 * ex: set ts=8 sts=4 sw=4 et: