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)
405 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
407 #else /* Below uses POSIX 2008 */
409 /* We emulate setlocale with our own function. LC_foo is not valid for the
410 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
411 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
412 * proper offset into the array 'category_masks[]'. At runtime, we have to
413 * search through the array (as the actual numbers may not be small contiguous
414 * positive integers which would lend themselves to array lookup). */
415 # define do_setlocale_c(cat, locale) \
416 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
417 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
419 # if ! defined(__GLIBC__) || ! defined(USE_LOCALE_MESSAGES)
421 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
423 # else /* Invalidate glibc cache of loaded translations, see [perl #134264] */
425 # include <libintl.h>
426 # define FIX_GLIBC_LC_MESSAGES_BUG(i) \
428 if ((i) == LC_MESSAGES_INDEX) { \
429 textdomain(textdomain(NULL)); \
435 /* A third array, parallel to the ones above to map from category to its
437 const int category_masks[] = {
438 # ifdef USE_LOCALE_NUMERIC
441 # ifdef USE_LOCALE_CTYPE
444 # ifdef USE_LOCALE_COLLATE
447 # ifdef USE_LOCALE_TIME
450 # ifdef USE_LOCALE_MESSAGES
453 # ifdef USE_LOCALE_MONETARY
456 # ifdef USE_LOCALE_ADDRESS
459 # ifdef USE_LOCALE_IDENTIFICATION
460 LC_IDENTIFICATION_MASK,
462 # ifdef USE_LOCALE_MEASUREMENT
465 # ifdef USE_LOCALE_PAPER
468 # ifdef USE_LOCALE_TELEPHONE
471 /* LC_ALL can't be turned off by a Configure
472 * option, and in Posix 2008, should always be
473 * here, so compile it in unconditionally.
474 * This could catch some glitches at compile
480 S_emulate_setlocale(const int category,
483 const bool is_index_valid
486 /* This function effectively performs a setlocale() on just the current
487 * thread; thus it is thread-safe. It does this by using the POSIX 2008
488 * locale functions to emulate the behavior of setlocale(). Similar to
489 * regular setlocale(), the return from this function points to memory that
490 * can be overwritten by other system calls, so needs to be copied
491 * immediately if you need to retain it. The difference here is that
492 * system calls besides another setlocale() can overwrite it.
494 * By doing this, most locale-sensitive functions become thread-safe. The
495 * exceptions are mostly those that return a pointer to static memory.
497 * This function takes the same parameters, 'category' and 'locale', that
498 * the regular setlocale() function does, but it also takes two additional
499 * ones. This is because the 2008 functions don't use a category; instead
500 * they use a corresponding mask. Because this function operates in both
501 * worlds, it may need one or the other or both. This function can
502 * calculate the mask from the input category, but to avoid this
503 * calculation, if the caller knows at compile time what the mask is, it
504 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
505 * parameter is ignored.
507 * POSIX 2008, for some sick reason, chose not to provide a method to find
508 * the category name of a locale. Some vendors have created a
509 * querylocale() function to do just that. This function is a lot simpler
510 * to implement on systems that have this. Otherwise, we have to keep
511 * track of what the locale has been set to, so that we can return its
512 * name to emulate setlocale(). It's also possible for C code in some
513 * library to change the locale without us knowing it, though as of
514 * September 2017, there are no occurrences in CPAN of uselocale(). Some
515 * libraries do use setlocale(), but that changes the global locale, and
516 * threads using per-thread locales will just ignore those changes.
517 * Another problem is that without querylocale(), we have to guess at what
518 * was meant by setting a locale of "". We handle this by not actually
519 * ever setting to "" (unless querylocale exists), but to emulate what we
520 * think should happen for "".
530 if (DEBUG_Lv_TEST || debug_initialization) {
531 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);
536 /* If the input mask might be incorrect, calculate the correct one */
537 if (! is_index_valid) {
542 if (DEBUG_Lv_TEST || debug_initialization) {
543 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
548 for (i = 0; i <= LC_ALL_INDEX; i++) {
549 if (category == categories[i]) {
555 /* Here, we don't know about this category, so can't handle it.
556 * Fallback to the early POSIX usages */
557 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
558 "Unknown locale category %d; can't set it to %s\n",
566 if (DEBUG_Lv_TEST || debug_initialization) {
567 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
574 mask = category_masks[index];
578 if (DEBUG_Lv_TEST || debug_initialization) {
579 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
584 /* If just querying what the existing locale is ... */
585 if (locale == NULL) {
586 locale_t cur_obj = uselocale((locale_t) 0);
590 if (DEBUG_Lv_TEST || debug_initialization) {
591 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
596 if (cur_obj == LC_GLOBAL_LOCALE) {
597 return my_setlocale(category, NULL);
600 # ifdef HAS_QUERYLOCALE
602 return (char *) querylocale(mask, cur_obj);
606 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
607 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
609 # if defined(_NL_LOCALE_NAME) \
610 && defined(DEBUGGING) \
611 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
612 /* On systems that accept any locale name, the real underlying locale
613 * is often returned by this internal function, so we can't use it */
615 /* Internal glibc for querylocale(), but doesn't handle
616 * empty-string ("") locale properly; who knows what other
617 * glitches. Check for it now, under debug. */
619 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
620 uselocale((locale_t) 0));
622 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
623 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
624 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
626 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
627 if ( strNE(PL_curlocales[index], temp_name)
628 && ! ( isNAME_C_OR_POSIX(temp_name)
629 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
631 # ifdef USE_C_BACKTRACE
633 dump_c_backtrace(Perl_debug_log, 20, 1);
637 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
638 " (%s) and what internal glibc thinks"
639 " (%s)\n", category_names[index],
640 PL_curlocales[index], temp_name);
649 /* Without querylocale(), we have to use our record-keeping we've
652 if (category != LC_ALL) {
656 if (DEBUG_Lv_TEST || debug_initialization) {
657 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
662 return PL_curlocales[index];
664 else { /* For LC_ALL */
666 Size_t names_len = 0;
668 bool are_all_categories_the_same_locale = TRUE;
670 /* If we have a valid LC_ALL value, just return it */
671 if (PL_curlocales[LC_ALL_INDEX]) {
675 if (DEBUG_Lv_TEST || debug_initialization) {
676 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
681 return PL_curlocales[LC_ALL_INDEX];
684 /* Otherwise, we need to construct a string of name=value pairs.
685 * We use the glibc syntax, like
686 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
687 * First calculate the needed size. Along the way, check if all
688 * the locale names are the same */
689 for (i = 0; i < LC_ALL_INDEX; i++) {
693 if (DEBUG_Lv_TEST || debug_initialization) {
694 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]);
699 names_len += strlen(category_names[i])
701 + strlen(PL_curlocales[i])
704 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
705 are_all_categories_the_same_locale = FALSE;
709 /* If they are the same, we don't actually have to construct the
710 * string; we just make the entry in LC_ALL_INDEX valid, and be
711 * that single name */
712 if (are_all_categories_the_same_locale) {
713 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
714 return PL_curlocales[LC_ALL_INDEX];
717 names_len++; /* Trailing '\0' */
718 SAVEFREEPV(Newx(all_string, names_len, char));
721 /* Then fill in the string */
722 for (i = 0; i < LC_ALL_INDEX; i++) {
726 if (DEBUG_Lv_TEST || debug_initialization) {
727 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]);
732 my_strlcat(all_string, category_names[i], names_len);
733 my_strlcat(all_string, "=", names_len);
734 my_strlcat(all_string, PL_curlocales[i], names_len);
735 my_strlcat(all_string, ";", names_len);
740 if (DEBUG_L_TEST || debug_initialization) {
741 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
751 SETERRNO(EINVAL, LIB_INVARG);
759 } /* End of this being setlocale(LC_foo, NULL) */
761 /* Here, we are switching locales. */
763 # ifndef HAS_QUERYLOCALE
765 if (strEQ(locale, "")) {
767 /* For non-querylocale() systems, we do the setting of "" ourselves to
768 * be sure that we really know what's going on. We follow the Linux
769 * documented behavior (but if that differs from the actual behavior,
770 * this won't work exactly as the OS implements). We go out and
771 * examine the environment based on our understanding of how the system
772 * works, and use that to figure things out */
774 const char * const lc_all = PerlEnv_getenv("LC_ALL");
776 /* Use any "LC_ALL" environment variable, as it overrides everything
778 if (lc_all && strNE(lc_all, "")) {
783 /* Otherwise, we need to dig deeper. Unless overridden, the
784 * default is the LANG environment variable; if it doesn't exist,
787 const char * default_name;
789 default_name = PerlEnv_getenv("LANG");
791 if (! default_name || strEQ(default_name, "")) {
794 else if (PL_scopestack_ix != 0) {
795 /* To minimize other threads messing with the environment,
796 * we copy the variable, making it a temporary. But this
797 * doesn't work upon program initialization before any
798 * scopes are created, and at this time, there's nothing
799 * else going on that would interfere. So skip the copy
801 default_name = savepv(default_name);
802 SAVEFREEPV(default_name);
805 if (category != LC_ALL) {
806 const char * const name = PerlEnv_getenv(category_names[index]);
808 /* Here we are setting a single category. Assume will have the
810 locale = default_name;
812 /* But then look for an overriding environment variable */
813 if (name && strNE(name, "")) {
818 bool did_override = FALSE;
821 /* Here, we are getting LC_ALL. Any categories that don't have
822 * a corresponding environment variable set should be set to
823 * LANG, or to "C" if there is no LANG. If no individual
824 * categories differ from this, we can just set LC_ALL. This
825 * is buggy on systems that have extra categories that we don't
826 * know about. If there is an environment variable that sets
827 * that category, we won't know to look for it, and so our use
828 * of LANG or "C" improperly overrides it. On the other hand,
829 * if we don't do what is done here, and there is no
830 * environment variable, the category's locale should be set to
831 * LANG or "C". So there is no good solution. khw thinks the
832 * best is to look at systems to see what categories they have,
833 * and include them, and then to assume that we know the
836 for (i = 0; i < LC_ALL_INDEX; i++) {
837 const char * const env_override
838 = savepv(PerlEnv_getenv(category_names[i]));
839 const char * this_locale = ( env_override
840 && strNE(env_override, ""))
843 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
845 Safefree(env_override);
849 if (strNE(this_locale, default_name)) {
853 Safefree(env_override);
856 /* If all the categories are the same, we can set LC_ALL to
858 if (! did_override) {
859 locale = default_name;
863 /* Here, LC_ALL is no longer valid, as some individual
864 * categories don't match it. We call ourselves
865 * recursively, as that will execute the code that
866 * generates the proper locale string for this situation.
867 * We don't do the remainder of this function, as that is
868 * to update our records, and we've just done that for the
869 * individual categories in the loop above, and doing so
870 * would cause LC_ALL to be done as well */
871 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
875 } /* End of this being setlocale(LC_foo, "") */
876 else if (strchr(locale, ';')) {
878 /* LC_ALL may actually incude a conglomeration of various categories.
879 * Without querylocale, this code uses the glibc (as of this writing)
880 * syntax for representing that, but that is not a stable API, and
881 * other platforms do it differently, so we have to handle all cases
885 const char * s = locale;
886 const char * e = locale + strlen(locale);
888 const char * category_end;
889 const char * name_start;
890 const char * name_end;
892 /* If the string that gives what to set doesn't include all categories,
893 * the omitted ones get set to "C". To get this behavior, first set
894 * all the individual categories to "C", and override the furnished
896 for (i = 0; i < LC_ALL_INDEX; i++) {
897 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
904 /* Parse through the category */
905 while (isWORDCHAR(*p)) {
912 "panic: %s: %d: Unexpected character in locale name '%02X",
913 __FILE__, __LINE__, *(p-1));
916 /* Parse through the locale name */
918 while (p < e && *p != ';') {
921 "panic: %s: %d: Unexpected character in locale name '%02X",
922 __FILE__, __LINE__, *(p-1));
928 /* Space past the semi-colon */
933 /* Find the index of the category name in our lists */
934 for (i = 0; i < LC_ALL_INDEX; i++) {
935 char * individ_locale;
937 /* Keep going if this isn't the index. The strnNE() avoids a
938 * Perl_form(), but would fail if ever a category name could be
939 * a substring of another one, like if there were a
941 if strnNE(s, category_names[i], category_end - s) {
945 /* If this index is for the single category we're changing, we
946 * have found the locale to set it to. */
947 if (category == categories[i]) {
948 locale = Perl_form(aTHX_ "%.*s",
949 (int) (name_end - name_start),
954 assert(category == LC_ALL);
955 individ_locale = Perl_form(aTHX_ "%.*s",
956 (int) (name_end - name_start), name_start);
957 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
966 /* Here we have set all the individual categories by recursive calls.
967 * These collectively should have fixed up LC_ALL, so can just query
968 * what that now is */
969 assert(category == LC_ALL);
971 return do_setlocale_c(LC_ALL, NULL);
972 } /* End of this being setlocale(LC_ALL,
973 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
977 /* Here at the end of having to deal with the absence of querylocale().
978 * Some cases have already been fully handled by recursive calls to this
979 * function. But at this point, we haven't dealt with those, but are now
980 * prepared to, knowing what the locale name to set this category to is.
981 * This would have come for free if this system had had querylocale() */
983 # endif /* end of ! querylocale */
985 assert(PL_C_locale_obj);
987 /* Switching locales generally entails freeing the current one's space (at
988 * the C library's discretion). We need to stop using that locale before
989 * the switch. So switch to a known locale object that we don't otherwise
990 * mess with. This returns the locale object in effect at the time of the
992 old_obj = uselocale(PL_C_locale_obj);
996 if (DEBUG_Lv_TEST || debug_initialization) {
997 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
1006 if (DEBUG_L_TEST || debug_initialization) {
1008 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1019 if (DEBUG_Lv_TEST || debug_initialization) {
1020 PerlIO_printf(Perl_debug_log,
1021 "%s:%d: emulate_setlocale now using %p\n",
1022 __FILE__, __LINE__, PL_C_locale_obj);
1027 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1028 * in fact, we already have switched to it (in preparation for what
1029 * normally is to come). But since we're already there, continue to use
1030 * it instead of trying to create a new locale */
1031 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1035 if (DEBUG_Lv_TEST || debug_initialization) {
1036 PerlIO_printf(Perl_debug_log,
1037 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1042 new_obj = PL_C_locale_obj;
1044 /* We already had switched to the C locale in preparation for freeing
1046 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1047 freelocale(old_obj);
1051 /* If we weren't in a thread safe locale, set so that newlocale() below
1052 * which uses 'old_obj', uses an empty one. Same for our reserved C
1053 * object. The latter is defensive coding, so that, even if there is
1054 * some bug, we will never end up trying to modify either of these, as
1055 * if passed to newlocale(), they can be. */
1056 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1057 old_obj = (locale_t) 0;
1060 /* Ready to create a new locale by modification of the exising one */
1061 new_obj = newlocale(mask, locale, old_obj);
1068 if (DEBUG_L_TEST || debug_initialization) {
1069 PerlIO_printf(Perl_debug_log,
1070 "%s:%d: emulate_setlocale creating new object"
1071 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1076 if (! uselocale(old_obj)) {
1080 if (DEBUG_L_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log,
1082 "%s:%d: switching back failed: %d\n",
1083 __FILE__, __LINE__, GET_ERRNO);
1095 if (DEBUG_Lv_TEST || debug_initialization) {
1096 PerlIO_printf(Perl_debug_log,
1097 "%s:%d: emulate_setlocale created %p",
1098 __FILE__, __LINE__, new_obj);
1100 PerlIO_printf(Perl_debug_log,
1101 "; should have freed %p", old_obj);
1103 PerlIO_printf(Perl_debug_log, "\n");
1108 /* And switch into it */
1109 if (! uselocale(new_obj)) {
1114 if (DEBUG_L_TEST || debug_initialization) {
1115 PerlIO_printf(Perl_debug_log,
1116 "%s:%d: emulate_setlocale switching to new object"
1117 " failed\n", __FILE__, __LINE__);
1122 if (! uselocale(old_obj)) {
1126 if (DEBUG_L_TEST || debug_initialization) {
1127 PerlIO_printf(Perl_debug_log,
1128 "%s:%d: switching back failed: %d\n",
1129 __FILE__, __LINE__, GET_ERRNO);
1135 freelocale(new_obj);
1143 if (DEBUG_Lv_TEST || debug_initialization) {
1144 PerlIO_printf(Perl_debug_log,
1145 "%s:%d: emulate_setlocale now using %p\n",
1146 __FILE__, __LINE__, new_obj);
1151 /* We are done, except for updating our records (if the system doesn't keep
1152 * them) and in the case of locale "", we don't actually know what the
1153 * locale that got switched to is, as it came from the environment. So
1154 * have to find it */
1156 # ifdef HAS_QUERYLOCALE
1158 if (strEQ(locale, "")) {
1159 locale = querylocale(mask, new_obj);
1164 /* Here, 'locale' is the return value */
1166 /* Without querylocale(), we have to update our records */
1168 if (category == LC_ALL) {
1171 /* For LC_ALL, we change all individual categories to correspond */
1172 /* PL_curlocales is a parallel array, so has same
1173 * length as 'categories' */
1174 for (i = 0; i <= LC_ALL_INDEX; i++) {
1175 Safefree(PL_curlocales[i]);
1176 PL_curlocales[i] = savepv(locale);
1179 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
1183 /* For a single category, if it's not the same as the one in LC_ALL, we
1186 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1187 Safefree(PL_curlocales[LC_ALL_INDEX]);
1188 PL_curlocales[LC_ALL_INDEX] = NULL;
1191 /* Then update the category's record */
1192 Safefree(PL_curlocales[index]);
1193 PL_curlocales[index] = savepv(locale);
1195 FIX_GLIBC_LC_MESSAGES_BUG(index);
1203 #endif /* USE_POSIX_2008_LOCALE */
1205 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1206 didn't natively support them */
1208 /* The way this would work is that we would keep a per-thread list of the
1209 * correct locale for that thread. Any operation that was locale-sensitive
1210 * would have to be changed so that it would look like this:
1213 * setlocale to the correct locale for this operation
1217 * This leaves the global locale in the most recently used operation's, but it
1218 * was locked long enough to get the result. If that result is static, it
1219 * needs to be copied before the unlock.
1221 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1222 * the setup, but are no-ops when not needed, and similarly,
1223 * END_LOCALE_DEPENDENT_OP for the tear-down
1225 * But every call to a locale-sensitive function would have to be changed, and
1226 * if a module didn't cooperate by using the mutex, things would break.
1228 * This code was abandoned before being completed or tested, and is left as-is
1231 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1232 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1235 S_locking_setlocale(pTHX_
1237 const char * locale,
1239 const bool is_index_valid
1242 /* This function kind of performs a setlocale() on just the current thread;
1243 * thus it is kind of thread-safe. It does this by keeping a thread-level
1244 * array of the current locales for each category. Every time a locale is
1245 * switched to, it does the switch globally, but updates the thread's
1246 * array. A query as to what the current locale is just returns the
1247 * appropriate element from the array, and doesn't actually call the system
1248 * setlocale(). The saving into the array is done in an uninterruptible
1249 * section of code, so is unaffected by whatever any other threads might be
1252 * All locale-sensitive operations must work by first starting a critical
1253 * section, then switching to the thread's locale as kept by this function,
1254 * and then doing the operation, then ending the critical section. Thus,
1255 * each gets done in the appropriate locale. simulating thread-safety.
1257 * This function takes the same parameters, 'category' and 'locale', that
1258 * the regular setlocale() function does, but it also takes two additional
1259 * ones. This is because as described earlier. If we know on input the
1260 * index corresponding to the category into the array where we store the
1261 * current locales, we don't have to calculate it. If the caller knows at
1262 * compile time what the index is, it can pass it, setting
1263 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1267 /* If the input index might be incorrect, calculate the correct one */
1268 if (! is_index_valid) {
1271 if (DEBUG_Lv_TEST || debug_initialization) {
1272 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1275 for (i = 0; i <= LC_ALL_INDEX; i++) {
1276 if (category == categories[i]) {
1282 /* Here, we don't know about this category, so can't handle it.
1283 * XXX best we can do is to unsafely set this
1286 return my_setlocale(category, locale);
1290 if (DEBUG_Lv_TEST || debug_initialization) {
1291 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1295 /* For a query, just return what's in our records */
1296 if (new_locale == NULL) {
1297 return curlocales[index];
1301 /* Otherwise, we need to do the switch, and save the result, all in a
1302 * critical section */
1304 Safefree(curlocales[[index]]);
1306 /* It might be that this is called from an already-locked section of code.
1307 * We would have to detect and skip the LOCK/UNLOCK if so */
1310 curlocales[index] = savepv(my_setlocale(category, new_locale));
1312 if (strEQ(new_locale, "")) {
1316 /* The locale values come from the environment, and may not all be the
1317 * same, so for LC_ALL, we have to update all the others, while the
1318 * mutex is still locked */
1320 if (category == LC_ALL) {
1322 for (i = 0; i < LC_ALL_INDEX) {
1323 curlocales[i] = my_setlocale(categories[i], NULL);
1332 return curlocales[index];
1339 S_set_numeric_radix(pTHX_ const bool use_locale)
1341 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1342 * TRUE, use the radix character derived from the current locale */
1344 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1345 || defined(HAS_NL_LANGINFO))
1347 const char * radix = (use_locale)
1348 ? my_nl_langinfo(RADIXCHAR, FALSE)
1349 /* FALSE => already in dest locale */
1352 sv_setpv(PL_numeric_radix_sv, radix);
1354 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1355 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1356 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1357 SvCUR(PL_numeric_radix_sv))
1358 && _is_cur_LC_category_utf8(LC_NUMERIC))
1360 SvUTF8_on(PL_numeric_radix_sv);
1365 if (DEBUG_L_TEST || debug_initialization) {
1366 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1367 SvPVX(PL_numeric_radix_sv),
1368 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1374 PERL_UNUSED_ARG(use_locale);
1376 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1381 S_new_numeric(pTHX_ const char *newnum)
1384 #ifndef USE_LOCALE_NUMERIC
1386 PERL_UNUSED_ARG(newnum);
1390 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1391 * core Perl this and that 'newnum' is the name of the new locale.
1392 * It installs this locale as the current underlying default.
1394 * The default locale and the C locale can be toggled between by use of the
1395 * set_numeric_underlying() and set_numeric_standard() functions, which
1396 * should probably not be called directly, but only via macros like
1397 * SET_NUMERIC_STANDARD() in perl.h.
1399 * The toggling is necessary mainly so that a non-dot radix decimal point
1400 * character can be output, while allowing internal calculations to use a
1403 * This sets several interpreter-level variables:
1404 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1405 * PL_numeric_underlying A boolean indicating if the toggled state is such
1406 * that the current locale is the program's underlying
1408 * PL_numeric_standard An int indicating if the toggled state is such
1409 * that the current locale is the C locale or
1410 * indistinguishable from the C locale. If non-zero, it
1411 * is in C; if > 1, it means it may not be toggled away
1413 * PL_numeric_underlying_is_standard A bool kept by this function
1414 * indicating that the underlying locale and the standard
1415 * C locale are indistinguishable for the purposes of
1416 * LC_NUMERIC. This happens when both of the above two
1417 * variables are true at the same time. (Toggling is a
1418 * no-op under these circumstances.) This variable is
1419 * used to avoid having to recalculate.
1425 Safefree(PL_numeric_name);
1426 PL_numeric_name = NULL;
1427 PL_numeric_standard = TRUE;
1428 PL_numeric_underlying = TRUE;
1429 PL_numeric_underlying_is_standard = TRUE;
1433 save_newnum = stdize_locale(savepv(newnum));
1434 PL_numeric_underlying = TRUE;
1435 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1437 #ifndef TS_W32_BROKEN_LOCALECONV
1439 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1440 * them. But on broken Windows systems calling my_nl_langinfo() for
1441 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1442 * and just always change the locale if not C nor POSIX on those systems */
1443 if (! PL_numeric_standard) {
1444 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1445 FALSE /* Don't toggle locale */ ))
1446 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1451 /* Save the new name if it isn't the same as the previous one, if any */
1452 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1453 Safefree(PL_numeric_name);
1454 PL_numeric_name = save_newnum;
1457 Safefree(save_newnum);
1460 PL_numeric_underlying_is_standard = PL_numeric_standard;
1462 # ifdef HAS_POSIX_2008_LOCALE
1464 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1466 PL_underlying_numeric_obj);
1470 if (DEBUG_L_TEST || debug_initialization) {
1471 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1474 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1475 * have to worry about the radix being a non-dot. (Core operations that
1476 * need the underlying locale change to it temporarily). */
1477 if (PL_numeric_standard) {
1478 set_numeric_radix(0);
1481 set_numeric_standard();
1484 #endif /* USE_LOCALE_NUMERIC */
1489 Perl_set_numeric_standard(pTHX)
1492 #ifdef USE_LOCALE_NUMERIC
1494 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1495 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1496 * macro avoids calling this routine if toggling isn't necessary according
1497 * to our records (which could be wrong if some XS code has changed the
1498 * locale behind our back) */
1502 if (DEBUG_L_TEST || debug_initialization) {
1503 PerlIO_printf(Perl_debug_log,
1504 "Setting LC_NUMERIC locale to standard C\n");
1509 do_setlocale_c(LC_NUMERIC, "C");
1510 PL_numeric_standard = TRUE;
1511 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1512 set_numeric_radix(0);
1514 #endif /* USE_LOCALE_NUMERIC */
1519 Perl_set_numeric_underlying(pTHX)
1522 #ifdef USE_LOCALE_NUMERIC
1524 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1525 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1526 * instead of calling this directly. The macro avoids calling this routine
1527 * if toggling isn't necessary according to our records (which could be
1528 * wrong if some XS code has changed the locale behind our back) */
1532 if (DEBUG_L_TEST || debug_initialization) {
1533 PerlIO_printf(Perl_debug_log,
1534 "Setting LC_NUMERIC locale to %s\n",
1540 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1541 PL_numeric_standard = PL_numeric_underlying_is_standard;
1542 PL_numeric_underlying = TRUE;
1543 set_numeric_radix(! PL_numeric_standard);
1545 #endif /* USE_LOCALE_NUMERIC */
1550 * Set up for a new ctype locale.
1553 S_new_ctype(pTHX_ const char *newctype)
1556 #ifndef USE_LOCALE_CTYPE
1558 PERL_UNUSED_ARG(newctype);
1559 PERL_UNUSED_CONTEXT;
1563 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1564 * core Perl this and that 'newctype' is the name of the new locale.
1566 * This function sets up the folding arrays for all 256 bytes, assuming
1567 * that tofold() is tolc() since fold case is not a concept in POSIX,
1569 * Any code changing the locale (outside this file) should use
1570 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1571 * this function should be called directly only from this file and from
1572 * POSIX::setlocale() */
1577 /* Don't check for problems if we are suppressing the warnings */
1578 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1579 bool maybe_utf8_turkic = FALSE;
1581 PERL_ARGS_ASSERT_NEW_CTYPE;
1583 /* We will replace any bad locale warning with 1) nothing if the new one is
1584 * ok; or 2) a new warning for the bad new locale */
1585 if (PL_warn_locale) {
1586 SvREFCNT_dec_NN(PL_warn_locale);
1587 PL_warn_locale = NULL;
1590 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1592 /* A UTF-8 locale gets standard rules. But note that code still has to
1593 * handle this specially because of the three problematic code points */
1594 if (PL_in_utf8_CTYPE_locale) {
1595 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1597 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1598 * Turkic. Make sure these two are the only anomalies. (We don't use
1599 * towupper and towlower because they aren't in C89.) */
1601 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1603 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1607 if (toupper('i') == 'i' && tolower('I') == 'I') {
1610 check_for_problems = TRUE;
1611 maybe_utf8_turkic = TRUE;
1615 /* We don't populate the other lists if a UTF-8 locale, but do check that
1616 * everything works as expected, unless checking turned off */
1617 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1618 /* Assume enough space for every character being bad. 4 spaces each
1619 * for the 94 printable characters that are output like "'x' "; and 5
1620 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1622 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1623 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1625 unsigned int bad_count = 0; /* Count of bad characters */
1627 for (i = 0; i < 256; i++) {
1628 if (! PL_in_utf8_CTYPE_locale) {
1630 PL_fold_locale[i] = (U8) tolower(i);
1631 else if (islower(i))
1632 PL_fold_locale[i] = (U8) toupper(i);
1634 PL_fold_locale[i] = (U8) i;
1637 /* If checking for locale problems, see if the native ASCII-range
1638 * printables plus \n and \t are in their expected categories in
1639 * the new locale. If not, this could mean big trouble, upending
1640 * Perl's and most programs' assumptions, like having a
1641 * metacharacter with special meaning become a \w. Fortunately,
1642 * it's very rare to find locales that aren't supersets of ASCII
1643 * nowadays. It isn't a problem for most controls to be changed
1644 * into something else; we check only \n and \t, though perhaps \r
1645 * could be an issue as well. */
1646 if ( check_for_problems
1647 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1649 bool is_bad = FALSE;
1650 char name[4] = { '\0' };
1652 /* Convert the name into a string */
1657 else if (i == '\n') {
1658 my_strlcpy(name, "\\n", sizeof(name));
1660 else if (i == '\t') {
1661 my_strlcpy(name, "\\t", sizeof(name));
1665 my_strlcpy(name, "' '", sizeof(name));
1668 /* Check each possibe class */
1669 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1671 DEBUG_L(PerlIO_printf(Perl_debug_log,
1672 "isalnum('%s') unexpectedly is %d\n",
1673 name, cBOOL(isalnum(i))));
1675 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1677 DEBUG_L(PerlIO_printf(Perl_debug_log,
1678 "isalpha('%s') unexpectedly is %d\n",
1679 name, cBOOL(isalpha(i))));
1681 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1683 DEBUG_L(PerlIO_printf(Perl_debug_log,
1684 "isdigit('%s') unexpectedly is %d\n",
1685 name, cBOOL(isdigit(i))));
1687 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1689 DEBUG_L(PerlIO_printf(Perl_debug_log,
1690 "isgraph('%s') unexpectedly is %d\n",
1691 name, cBOOL(isgraph(i))));
1693 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1695 DEBUG_L(PerlIO_printf(Perl_debug_log,
1696 "islower('%s') unexpectedly is %d\n",
1697 name, cBOOL(islower(i))));
1699 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1701 DEBUG_L(PerlIO_printf(Perl_debug_log,
1702 "isprint('%s') unexpectedly is %d\n",
1703 name, cBOOL(isprint(i))));
1705 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1707 DEBUG_L(PerlIO_printf(Perl_debug_log,
1708 "ispunct('%s') unexpectedly is %d\n",
1709 name, cBOOL(ispunct(i))));
1711 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1713 DEBUG_L(PerlIO_printf(Perl_debug_log,
1714 "isspace('%s') unexpectedly is %d\n",
1715 name, cBOOL(isspace(i))));
1717 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1719 DEBUG_L(PerlIO_printf(Perl_debug_log,
1720 "isupper('%s') unexpectedly is %d\n",
1721 name, cBOOL(isupper(i))));
1723 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1725 DEBUG_L(PerlIO_printf(Perl_debug_log,
1726 "isxdigit('%s') unexpectedly is %d\n",
1727 name, cBOOL(isxdigit(i))));
1729 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1731 DEBUG_L(PerlIO_printf(Perl_debug_log,
1732 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1733 name, tolower(i), (int) toLOWER_A(i)));
1735 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1737 DEBUG_L(PerlIO_printf(Perl_debug_log,
1738 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1739 name, toupper(i), (int) toUPPER_A(i)));
1741 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1743 DEBUG_L(PerlIO_printf(Perl_debug_log,
1744 "'\\n' (=%02X) is not a control\n", (int) i));
1747 /* Add to the list; Separate multiple entries with a blank */
1750 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1752 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1758 if (bad_count == 2 && maybe_utf8_turkic) {
1760 *bad_chars_list = '\0';
1761 PL_fold_locale['I'] = 'I';
1762 PL_fold_locale['i'] = 'i';
1763 PL_in_utf8_turkic_locale = TRUE;
1764 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1765 __FILE__, __LINE__, newctype));
1768 PL_in_utf8_turkic_locale = FALSE;
1773 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1774 * this locale requires more than one byte, there are going to be
1776 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1777 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1778 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1780 if ( check_for_problems && MB_CUR_MAX > 1
1781 && ! PL_in_utf8_CTYPE_locale
1783 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1784 * locale. Just assume that the implementation for them (plus
1785 * for POSIX) is correct and the > 1 value is spurious. (Since
1786 * these are specially handled to never be considered UTF-8
1787 * locales, as long as this is the only problem, everything
1788 * should work fine */
1789 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1791 multi_byte_locale = TRUE;
1796 /* If we found problems and we want them output, do so */
1797 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1798 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1800 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1801 PL_warn_locale = Perl_newSVpvf(aTHX_
1802 "Locale '%s' contains (at least) the following characters"
1803 " which have\nunexpected meanings: %s\nThe Perl program"
1804 " will use the expected meanings",
1805 newctype, bad_chars_list);
1808 PL_warn_locale = Perl_newSVpvf(aTHX_
1809 "Locale '%s' may not work well.%s%s%s\n",
1812 ? " Some characters in it are not recognized by"
1816 ? "\nThe following characters (and maybe others)"
1817 " may not have the same meaning as the Perl"
1818 " program expects:\n"
1826 # ifdef HAS_NL_LANGINFO
1828 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1829 /* parameter FALSE is a don't care here */
1830 my_nl_langinfo(CODESET, FALSE));
1834 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1836 /* If we are actually in the scope of the locale or are debugging,
1837 * output the message now. If not in that scope, we save the
1838 * message to be output at the first operation using this locale,
1839 * if that actually happens. Most programs don't use locales, so
1840 * they are immune to bad ones. */
1841 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1843 /* The '0' below suppresses a bogus gcc compiler warning */
1844 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1846 if (IN_LC(LC_CTYPE)) {
1847 SvREFCNT_dec_NN(PL_warn_locale);
1848 PL_warn_locale = NULL;
1854 #endif /* USE_LOCALE_CTYPE */
1859 Perl__warn_problematic_locale()
1862 #ifdef USE_LOCALE_CTYPE
1866 /* Internal-to-core function that outputs the message in PL_warn_locale,
1867 * and then NULLS it. Should be called only through the macro
1868 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1870 if (PL_warn_locale) {
1871 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1872 SvPVX(PL_warn_locale),
1873 0 /* dummy to avoid compiler warning */ );
1874 SvREFCNT_dec_NN(PL_warn_locale);
1875 PL_warn_locale = NULL;
1883 S_new_collate(pTHX_ const char *newcoll)
1886 #ifndef USE_LOCALE_COLLATE
1888 PERL_UNUSED_ARG(newcoll);
1889 PERL_UNUSED_CONTEXT;
1893 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1894 * core Perl this and that 'newcoll' is the name of the new locale.
1896 * The design of locale collation is that every locale change is given an
1897 * index 'PL_collation_ix'. The first time a string particpates in an
1898 * operation that requires collation while locale collation is active, it
1899 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1900 * magic includes the collation index, and the transformation of the string
1901 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1902 * instead of the string itself. If a string changes, the magic is
1903 * cleared. The next time the locale changes, the index is incremented,
1904 * and so we know during a comparison that the transformation is not
1905 * necessarily still valid, and so is recomputed. Note that if the locale
1906 * changes enough times, the index could wrap (a U32), and it is possible
1907 * that a transformation would improperly be considered valid, leading to
1908 * an unlikely bug */
1911 if (PL_collation_name) {
1913 Safefree(PL_collation_name);
1914 PL_collation_name = NULL;
1916 PL_collation_standard = TRUE;
1917 is_standard_collation:
1918 PL_collxfrm_base = 0;
1919 PL_collxfrm_mult = 2;
1920 PL_in_utf8_COLLATE_locale = FALSE;
1921 PL_strxfrm_NUL_replacement = '\0';
1922 PL_strxfrm_max_cp = 0;
1926 /* If this is not the same locale as currently, set the new one up */
1927 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1929 Safefree(PL_collation_name);
1930 PL_collation_name = stdize_locale(savepv(newcoll));
1931 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1932 if (PL_collation_standard) {
1933 goto is_standard_collation;
1936 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1937 PL_strxfrm_NUL_replacement = '\0';
1938 PL_strxfrm_max_cp = 0;
1940 /* A locale collation definition includes primary, secondary, tertiary,
1941 * etc. weights for each character. To sort, the primary weights are
1942 * used, and only if they compare equal, then the secondary weights are
1943 * used, and only if they compare equal, then the tertiary, etc.
1945 * strxfrm() works by taking the input string, say ABC, and creating an
1946 * output transformed string consisting of first the primary weights,
1947 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1948 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1949 * may not have weights at every level. In our example, let's say B
1950 * doesn't have a tertiary weight, and A doesn't have a secondary
1951 * weight. The constructed string is then going to be
1952 * A¹B¹C¹ B²C² A³C³ ....
1953 * This has the desired effect that strcmp() will look at the secondary
1954 * or tertiary weights only if the strings compare equal at all higher
1955 * priority weights. The spaces shown here, like in
1957 * are not just for readability. In the general case, these must
1958 * actually be bytes, which we will call here 'separator weights'; and
1959 * they must be smaller than any other weight value, but since these
1960 * are C strings, only the terminating one can be a NUL (some
1961 * implementations may include a non-NUL separator weight just before
1962 * the NUL). Implementations tend to reserve 01 for the separator
1963 * weights. They are needed so that a shorter string's secondary
1964 * weights won't be misconstrued as primary weights of a longer string,
1965 * etc. By making them smaller than any other weight, the shorter
1966 * string will sort first. (Actually, if all secondary weights are
1967 * smaller than all primary ones, there is no need for a separator
1968 * weight between those two levels, etc.)
1970 * The length of the transformed string is roughly a linear function of
1971 * the input string. It's not exactly linear because some characters
1972 * don't have weights at all levels. When we call strxfrm() we have to
1973 * allocate some memory to hold the transformed string. The
1974 * calculations below try to find coefficients 'm' and 'b' for this
1975 * locale so that m*x + b equals how much space we need, given the size
1976 * of the input string in 'x'. If we calculate too small, we increase
1977 * the size as needed, and call strxfrm() again, but it is better to
1978 * get it right the first time to avoid wasted expensive string
1979 * transformations. */
1982 /* We use the string below to find how long the tranformation of it
1983 * is. Almost all locales are supersets of ASCII, or at least the
1984 * ASCII letters. We use all of them, half upper half lower,
1985 * because if we used fewer, we might hit just the ones that are
1986 * outliers in a particular locale. Most of the strings being
1987 * collated will contain a preponderance of letters, and even if
1988 * they are above-ASCII, they are likely to have the same number of
1989 * weight levels as the ASCII ones. It turns out that digits tend
1990 * to have fewer levels, and some punctuation has more, but those
1991 * are relatively sparse in text, and khw believes this gives a
1992 * reasonable result, but it could be changed if experience so
1994 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1995 char * x_longer; /* Transformed 'longer' */
1996 Size_t x_len_longer; /* Length of 'x_longer' */
1998 char * x_shorter; /* We also transform a substring of 'longer' */
1999 Size_t x_len_shorter;
2001 /* _mem_collxfrm() is used get the transformation (though here we
2002 * are interested only in its length). It is used because it has
2003 * the intelligence to handle all cases, but to work, it needs some
2004 * values of 'm' and 'b' to get it started. For the purposes of
2005 * this calculation we use a very conservative estimate of 'm' and
2006 * 'b'. This assumes a weight can be multiple bytes, enough to
2007 * hold any UV on the platform, and there are 5 levels, 4 weight
2008 * bytes, and a trailing NUL. */
2009 PL_collxfrm_base = 5;
2010 PL_collxfrm_mult = 5 * sizeof(UV);
2012 /* Find out how long the transformation really is */
2013 x_longer = _mem_collxfrm(longer,
2017 /* We avoid converting to UTF-8 in the
2018 * called function by telling it the
2019 * string is in UTF-8 if the locale is a
2020 * UTF-8 one. Since the string passed
2021 * here is invariant under UTF-8, we can
2022 * claim it's UTF-8 even though it isn't.
2024 PL_in_utf8_COLLATE_locale);
2027 /* Find out how long the transformation of a substring of 'longer'
2028 * is. Together the lengths of these transformations are
2029 * sufficient to calculate 'm' and 'b'. The substring is all of
2030 * 'longer' except the first character. This minimizes the chances
2031 * of being swayed by outliers */
2032 x_shorter = _mem_collxfrm(longer + 1,
2035 PL_in_utf8_COLLATE_locale);
2036 Safefree(x_shorter);
2038 /* If the results are nonsensical for this simple test, the whole
2039 * locale definition is suspect. Mark it so that locale collation
2040 * is not active at all for it. XXX Should we warn? */
2041 if ( x_len_shorter == 0
2042 || x_len_longer == 0
2043 || x_len_shorter >= x_len_longer)
2045 PL_collxfrm_mult = 0;
2046 PL_collxfrm_base = 0;
2049 SSize_t base; /* Temporary */
2051 /* We have both: m * strlen(longer) + b = x_len_longer
2052 * m * strlen(shorter) + b = x_len_shorter;
2053 * subtracting yields:
2054 * m * (strlen(longer) - strlen(shorter))
2055 * = x_len_longer - x_len_shorter
2056 * But we have set things up so that 'shorter' is 1 byte smaller
2057 * than 'longer'. Hence:
2058 * m = x_len_longer - x_len_shorter
2060 * But if something went wrong, make sure the multiplier is at
2063 if (x_len_longer > x_len_shorter) {
2064 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2067 PL_collxfrm_mult = 1;
2072 * but in case something has gone wrong, make sure it is
2074 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2079 /* Add 1 for the trailing NUL */
2080 PL_collxfrm_base = base + 1;
2085 if (DEBUG_L_TEST || debug_initialization) {
2086 PerlIO_printf(Perl_debug_log,
2087 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2089 " collate multipler=%zu, collate base=%zu\n",
2091 PL_in_utf8_COLLATE_locale,
2092 x_len_shorter, x_len_longer,
2093 PL_collxfrm_mult, PL_collxfrm_base);
2100 #endif /* USE_LOCALE_COLLATE */
2108 #define USE_WSETLOCALE
2110 #ifdef USE_WSETLOCALE
2113 S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2120 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2127 Newx(wlocale, req_size, wchar_t);
2128 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2137 wresult = _wsetlocale(category, wlocale);
2141 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2142 Newx(result, req_size, char);
2143 SAVEFREEPV(result); /* is there something better we can do here? */
2144 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2145 result, req_size, NULL, NULL)) {
2160 S_win32_setlocale(pTHX_ int category, const char* locale)
2162 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2163 * difference between the two unless the input locale is "", which normally
2164 * means on Windows to get the machine default, which is set via the
2165 * computer's "Regional and Language Options" (or its current equivalent).
2166 * In POSIX, it instead means to find the locale from the user's
2167 * environment. This routine changes the Windows behavior to first look in
2168 * the environment, and, if anything is found, use that instead of going to
2169 * the machine default. If there is no environment override, the machine
2170 * default is used, by calling the real setlocale() with "".
2172 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2173 * use the particular category's variable if set; otherwise to use the LANG
2176 bool override_LC_ALL = FALSE;
2180 if (locale && strEQ(locale, "")) {
2184 locale = PerlEnv_getenv("LC_ALL");
2186 if (category == LC_ALL) {
2187 override_LC_ALL = TRUE;
2193 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2194 if (category == categories[i]) {
2195 locale = PerlEnv_getenv(category_names[i]);
2200 locale = PerlEnv_getenv("LANG");
2216 #ifdef USE_WSETLOCALE
2217 result = S_wrap_wsetlocale(aTHX_ category, locale);
2219 result = setlocale(category, locale);
2221 DEBUG_L(STMT_START {
2223 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2224 setlocale_debug_string(category, locale, result));
2228 if (! override_LC_ALL) {
2232 /* Here the input category was LC_ALL, and we have set it to what is in the
2233 * LANG variable or the system default if there is no LANG. But these have
2234 * lower priority than the other LC_foo variables, so override it for each
2235 * one that is set. (If they are set to "", it means to use the same thing
2236 * we just set LC_ALL to, so can skip) */
2238 for (i = 0; i < LC_ALL_INDEX; i++) {
2239 result = PerlEnv_getenv(category_names[i]);
2240 if (result && strNE(result, "")) {
2241 #ifdef USE_WSETLOCALE
2242 S_wrap_wsetlocale(aTHX_ categories[i], result);
2244 setlocale(categories[i], result);
2246 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2248 setlocale_debug_string(categories[i], result, "not captured")));
2252 result = setlocale(LC_ALL, NULL);
2253 DEBUG_L(STMT_START {
2255 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2257 setlocale_debug_string(LC_ALL, NULL, result));
2268 =head1 Locale-related functions and macros
2270 =for apidoc Perl_setlocale
2272 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2273 taking the same parameters, and returning the same information, except that it
2274 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2275 instead return C<C> if the underlying locale has a non-dot decimal point
2276 character, or a non-empty thousands separator for displaying floating point
2277 numbers. This is because perl keeps that locale category such that it has a
2278 dot and empty separator, changing the locale briefly during the operations
2279 where the underlying one is required. C<Perl_setlocale> knows about this, and
2280 compensates; regular C<setlocale> doesn't.
2282 Another reason it isn't completely a drop-in replacement is that it is
2283 declared to return S<C<const char *>>, whereas the system setlocale omits the
2284 C<const> (presumably because its API was specified long ago, and can't be
2285 updated; it is illegal to change the information C<setlocale> returns; doing
2286 so leads to segfaults.)
2288 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2289 C<setlocale> can be completely ineffective on some platforms under some
2292 C<Perl_setlocale> should not be used to change the locale except on systems
2293 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2294 the system C<setlocale()> is ineffective, returning the wrong information, and
2295 failing to actually change the locale. C<Perl_setlocale>, however works
2296 properly in all circumstances.
2298 The return points to a per-thread static buffer, which is overwritten the next
2299 time C<Perl_setlocale> is called from the same thread.
2306 Perl_setlocale(const int category, const char * locale)
2308 /* This wraps POSIX::setlocale() */
2312 PERL_UNUSED_ARG(category);
2313 PERL_UNUSED_ARG(locale);
2319 const char * retval;
2320 const char * newlocale;
2323 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2325 #ifdef USE_LOCALE_NUMERIC
2327 /* A NULL locale means only query what the current one is. We have the
2328 * LC_NUMERIC name saved, because we are normally switched into the C
2329 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2330 * the correct results. All other categories don't require special
2332 if (locale == NULL) {
2333 if (category == LC_NUMERIC) {
2335 /* We don't have to copy this return value, as it is a per-thread
2336 * variable, and won't change until a future setlocale */
2337 return PL_numeric_name;
2342 else if (category == LC_ALL) {
2343 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2352 retval = save_to_buffer(do_setlocale_r(category, locale),
2353 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2356 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2358 if (locale == NULL && category == LC_ALL) {
2359 RESTORE_LC_NUMERIC();
2364 DEBUG_L(PerlIO_printf(Perl_debug_log,
2365 "%s:%d: %s\n", __FILE__, __LINE__,
2366 setlocale_debug_string(category, locale, retval)));
2374 /* If locale == NULL, we are just querying the state */
2375 if (locale == NULL) {
2379 /* Now that have switched locales, we have to update our records to
2384 #ifdef USE_LOCALE_CTYPE
2391 #ifdef USE_LOCALE_COLLATE
2394 new_collate(retval);
2398 #ifdef USE_LOCALE_NUMERIC
2401 new_numeric(retval);
2409 /* LC_ALL updates all the things we care about. The values may not
2410 * be the same as 'retval', as the locale "" may have set things
2413 # ifdef USE_LOCALE_CTYPE
2415 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2416 new_ctype(newlocale);
2417 Safefree(newlocale);
2419 # endif /* USE_LOCALE_CTYPE */
2420 # ifdef USE_LOCALE_COLLATE
2422 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2423 new_collate(newlocale);
2424 Safefree(newlocale);
2427 # ifdef USE_LOCALE_NUMERIC
2429 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2430 new_numeric(newlocale);
2431 Safefree(newlocale);
2433 # endif /* USE_LOCALE_NUMERIC */
2446 PERL_STATIC_INLINE const char *
2447 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2449 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2450 * growing it if necessary */
2454 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2460 string_size = strlen(string) + offset + 1;
2462 if (*buf_size == 0) {
2463 Newx(*buf, string_size, char);
2464 *buf_size = string_size;
2466 else if (string_size > *buf_size) {
2467 Renew(*buf, string_size, char);
2468 *buf_size = string_size;
2471 Copy(string, *buf + offset, string_size - offset, char);
2477 =for apidoc Perl_langinfo
2479 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2480 taking the same C<item> parameter values, and returning the same information.
2481 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2482 of Perl's locale handling from your code, and can be used on systems that lack
2483 a native C<nl_langinfo>.
2491 The reason it isn't quite a drop-in replacement is actually an advantage. The
2492 only difference is that it returns S<C<const char *>>, whereas plain
2493 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2494 forbidden to write into the buffer. By declaring this C<const>, the compiler
2495 enforces this restriction, so if it is violated, you know at compilation time,
2496 rather than getting segfaults at runtime.
2500 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2501 without you having to write extra code. The reason for the extra code would be
2502 because these are from the C<LC_NUMERIC> locale category, which is normally
2503 kept set by Perl so that the radix is a dot, and the separator is the empty
2504 string, no matter what the underlying locale is supposed to be, and so to get
2505 the expected results, you have to temporarily toggle into the underlying
2506 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2507 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2508 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2509 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2510 (decimal point) character to be a dot.)
2514 The system function it replaces can have its static return buffer trashed,
2515 not only by a subesequent call to that function, but by a C<freelocale>,
2516 C<setlocale>, or other locale change. The returned buffer of this function is
2517 not changed until the next call to it, so the buffer is never in a trashed
2522 Its return buffer is per-thread, so it also is never overwritten by a call to
2523 this function from another thread; unlike the function it replaces.
2527 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2528 as Windows, hence makes your code more portable. Of the fifty-some possible
2529 items specified by the POSIX 2008 standard,
2530 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2531 only one is completely unimplemented, though on non-Windows platforms, another
2532 significant one is also not implemented). It uses various techniques to
2533 recover the other items, including calling C<L<localeconv(3)>>, and
2534 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2535 available. Later C<strftime()> versions have additional capabilities; C<""> is
2536 returned for those not available on your system.
2538 It is important to note that when called with an item that is recovered by
2539 using C<localeconv>, the buffer from any previous explicit call to
2540 C<localeconv> will be overwritten. This means you must save that buffer's
2541 contents if you need to access them after a call to this function. (But note
2542 that you might not want to be using C<localeconv()> directly anyway, because of
2543 issues like the ones listed in the second item of this list (above) for
2544 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2545 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2548 The details for those items which may deviate from what this emulation returns
2549 and what a native C<nl_langinfo()> would return are specified in
2554 When using C<Perl_langinfo> on systems that don't have a native
2555 C<nl_langinfo()>, you must
2557 #include "perl_langinfo.h"
2559 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2560 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2561 C<langinfo.h> would try to import into the namespace for code that doesn't need
2564 The original impetus for C<Perl_langinfo()> was so that code that needs to
2565 find out the current currency symbol, floating point radix character, or digit
2566 grouping separator can use, on all systems, the simpler and more
2567 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2568 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2569 is better to use the methods given in L<perlcall> to call
2570 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2577 #ifdef HAS_NL_LANGINFO
2578 Perl_langinfo(const nl_item item)
2580 Perl_langinfo(const int item)
2583 return my_nl_langinfo(item, TRUE);
2587 #ifdef HAS_NL_LANGINFO
2588 S_my_nl_langinfo(const nl_item item, bool toggle)
2590 S_my_nl_langinfo(const int item, bool toggle)
2594 const char * retval;
2596 #ifdef USE_LOCALE_NUMERIC
2598 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2599 * two items, and only if not already there */
2600 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2601 || PL_numeric_underlying))
2603 #endif /* No toggling needed if not using LC_NUMERIC */
2607 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2608 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2609 || ! defined(HAS_POSIX_2008_LOCALE) \
2610 || ! defined(DUPLOCALE)
2612 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2613 * for those items dependent on it. This must be copied to a buffer before
2614 * switching back, as some systems destroy the buffer when setlocale() is
2618 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2621 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2624 LOCALE_LOCK; /* Prevent interference from another thread executing
2625 this code section (the only call to nl_langinfo in
2629 /* Copy to a per-thread buffer, which is also one that won't be
2630 * destroyed by a subsequent setlocale(), such as the
2631 * RESTORE_LC_NUMERIC may do just below. */
2632 retval = save_to_buffer(nl_langinfo(item),
2633 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2638 RESTORE_LC_NUMERIC();
2642 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2645 bool do_free = FALSE;
2646 locale_t cur = uselocale((locale_t) 0);
2648 if (cur == LC_GLOBAL_LOCALE) {
2649 cur = duplocale(LC_GLOBAL_LOCALE);
2653 # ifdef USE_LOCALE_NUMERIC
2656 if (PL_underlying_numeric_obj) {
2657 cur = PL_underlying_numeric_obj;
2660 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2667 /* We have to save it to a buffer, because the freelocale() just below
2668 * can invalidate the internal one */
2669 retval = save_to_buffer(nl_langinfo_l(item, cur),
2670 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2679 if (strEQ(retval, "")) {
2680 if (item == YESSTR) {
2683 if (item == NOSTR) {
2690 #else /* Below, emulate nl_langinfo as best we can */
2694 # ifdef HAS_LOCALECONV
2696 const struct lconv* lc;
2698 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2700 # ifdef TS_W32_BROKEN_LOCALECONV
2702 const char * save_global;
2703 const char * save_thread;
2711 # ifdef HAS_STRFTIME
2714 bool return_format = FALSE; /* Return the %format, not the value */
2715 const char * format;
2719 /* We copy the results to a per-thread buffer, even if not
2720 * multi-threaded. This is in part to simplify this code, and partly
2721 * because we need a buffer anyway for strftime(), and partly because a
2722 * call of localeconv() could otherwise wipe out the buffer, and the
2723 * programmer would not be expecting this, as this is a nl_langinfo()
2724 * substitute after all, so s/he might be thinking their localeconv()
2725 * is safe until another localeconv() call. */
2730 /* This is unimplemented */
2731 case ERA: /* For use with strftime() %E modifier */
2736 /* We use only an English set, since we don't know any more */
2737 case YESEXPR: return "^[+1yY]";
2738 case YESSTR: return "yes";
2739 case NOEXPR: return "^[-0nN]";
2740 case NOSTR: return "no";
2746 /* On non-windows, this is unimplemented, in part because of
2747 * inconsistencies between vendors. The Darwin native
2748 * nl_langinfo() implementation simply looks at everything past
2749 * any dot in the name, but that doesn't work for other
2750 * vendors. Many Linux locales that don't have UTF-8 in their
2751 * names really are UTF-8, for example; z/OS locales that do
2752 * have UTF-8 in their names, aren't really UTF-8 */
2757 { /* But on Windows, the name does seem to be consistent, so
2762 const char * name = my_setlocale(LC_CTYPE, NULL);
2764 if (isNAME_C_OR_POSIX(name)) {
2765 return "ANSI_X3.4-1968";
2768 /* Find the dot in the locale name */
2769 first = (const char *) strchr(name, '.');
2775 /* Look at everything past the dot */
2780 if (! isDIGIT(*p)) {
2787 /* Here everything past the dot is a digit. Treat it as a
2789 retval = save_to_buffer("CP", &PL_langinfo_buf,
2790 &PL_langinfo_bufsize, 0);
2791 offset = STRLENs("CP");
2795 retval = save_to_buffer(first, &PL_langinfo_buf,
2796 &PL_langinfo_bufsize, offset);
2802 # ifdef HAS_LOCALECONV
2806 /* We don't bother with localeconv_l() because any system that
2807 * has it is likely to also have nl_langinfo() */
2809 LOCALE_LOCK_V; /* Prevent interference with other threads
2810 using localeconv() */
2812 # ifdef TS_W32_BROKEN_LOCALECONV
2814 /* This is a workaround for a Windows bug prior to VS 15.
2815 * What we do here is, while locked, switch to the global
2816 * locale so localeconv() works; then switch back just before
2817 * the unlock. This can screw things up if some thread is
2818 * already using the global locale while assuming no other is.
2819 * A different workaround would be to call GetCurrencyFormat on
2820 * a known value, and parse it; patches welcome
2822 * We have to use LC_ALL instead of LC_MONETARY because of
2823 * another bug in Windows */
2825 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2826 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2827 save_global= savepv(my_setlocale(LC_ALL, NULL));
2828 my_setlocale(LC_ALL, save_thread);
2834 || ! lc->currency_symbol
2835 || strEQ("", lc->currency_symbol))
2841 /* Leave the first spot empty to be filled in below */
2842 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2843 &PL_langinfo_bufsize, 1);
2844 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2845 { /* khw couldn't figure out how the localedef specifications
2846 would show that the $ should replace the radix; this is
2847 just a guess as to how it might work.*/
2848 PL_langinfo_buf[0] = '.';
2850 else if (lc->p_cs_precedes) {
2851 PL_langinfo_buf[0] = '-';
2854 PL_langinfo_buf[0] = '+';
2857 # ifdef TS_W32_BROKEN_LOCALECONV
2859 my_setlocale(LC_ALL, save_global);
2860 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2861 my_setlocale(LC_ALL, save_thread);
2862 Safefree(save_global);
2863 Safefree(save_thread);
2870 # ifdef TS_W32_BROKEN_LOCALECONV
2874 /* For this, we output a known simple floating point number to
2875 * a buffer, and parse it, looking for the radix */
2878 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2881 if (PL_langinfo_bufsize < 10) {
2882 PL_langinfo_bufsize = 10;
2883 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2886 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2888 if (needed_size >= (int) PL_langinfo_bufsize) {
2889 PL_langinfo_bufsize = needed_size + 1;
2890 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2891 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2893 assert(needed_size < (int) PL_langinfo_bufsize);
2896 ptr = PL_langinfo_buf;
2897 e = PL_langinfo_buf + PL_langinfo_bufsize;
2900 while (ptr < e && *ptr != '1') {
2907 while (ptr < e && *ptr != '5') {
2911 /* Everything in between is the radix string */
2913 PL_langinfo_buf[0] = '?';
2914 PL_langinfo_buf[1] = '\0';
2918 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2922 RESTORE_LC_NUMERIC();
2925 retval = PL_langinfo_buf;
2930 case RADIXCHAR: /* No special handling needed */
2937 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2940 LOCALE_LOCK_V; /* Prevent interference with other threads
2941 using localeconv() */
2943 # ifdef TS_W32_BROKEN_LOCALECONV
2945 /* This should only be for the thousands separator. A
2946 * different work around would be to use GetNumberFormat on a
2947 * known value and parse the result to find the separator */
2948 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2949 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2950 save_global = savepv(my_setlocale(LC_ALL, NULL));
2951 my_setlocale(LC_ALL, save_thread);
2953 /* This is the start of code that for broken Windows replaces
2954 * the above and below code, and instead calls
2955 * GetNumberFormat() and then would parse that to find the
2956 * thousands separator. It needs to handle UTF-16 vs -8
2959 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2960 DEBUG_L(PerlIO_printf(Perl_debug_log,
2961 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2962 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2972 temp = (item == RADIXCHAR)
2974 : lc->thousands_sep;
2980 retval = save_to_buffer(temp, &PL_langinfo_buf,
2981 &PL_langinfo_bufsize, 0);
2983 # ifdef TS_W32_BROKEN_LOCALECONV
2985 my_setlocale(LC_ALL, save_global);
2986 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2987 my_setlocale(LC_ALL, save_thread);
2988 Safefree(save_global);
2989 Safefree(save_thread);
2996 RESTORE_LC_NUMERIC();
3002 # ifdef HAS_STRFTIME
3004 /* These are defined by C89, so we assume that strftime supports
3005 * them, and so are returned unconditionally; they may not be what
3006 * the locale actually says, but should give good enough results
3007 * for someone using them as formats (as opposed to trying to parse
3008 * them to figure out what the locale says). The other format
3009 * items are actually tested to verify they work on the platform */
3010 case D_FMT: return "%x";
3011 case T_FMT: return "%X";
3012 case D_T_FMT: return "%c";
3014 /* These formats are only available in later strfmtime's */
3015 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
3017 /* The rest can be gotten from most versions of strftime(). */
3018 case ABDAY_1: case ABDAY_2: case ABDAY_3:
3019 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
3021 case AM_STR: case PM_STR:
3022 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
3023 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
3024 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
3025 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
3026 case DAY_5: case DAY_6: case DAY_7:
3027 case MON_1: case MON_2: case MON_3: case MON_4:
3028 case MON_5: case MON_6: case MON_7: case MON_8:
3029 case MON_9: case MON_10: case MON_11: case MON_12:
3033 init_tm(&tm); /* Precaution against core dumps */
3037 tm.tm_year = 2017 - 1900;
3044 "panic: %s: %d: switch case: %d problem",
3045 __FILE__, __LINE__, item);
3046 NOT_REACHED; /* NOTREACHED */
3048 case PM_STR: tm.tm_hour = 18;
3053 case ABDAY_7: tm.tm_wday++;
3054 case ABDAY_6: tm.tm_wday++;
3055 case ABDAY_5: tm.tm_wday++;
3056 case ABDAY_4: tm.tm_wday++;
3057 case ABDAY_3: tm.tm_wday++;
3058 case ABDAY_2: tm.tm_wday++;
3063 case DAY_7: tm.tm_wday++;
3064 case DAY_6: tm.tm_wday++;
3065 case DAY_5: tm.tm_wday++;
3066 case DAY_4: tm.tm_wday++;
3067 case DAY_3: tm.tm_wday++;
3068 case DAY_2: tm.tm_wday++;
3073 case ABMON_12: tm.tm_mon++;
3074 case ABMON_11: tm.tm_mon++;
3075 case ABMON_10: tm.tm_mon++;
3076 case ABMON_9: tm.tm_mon++;
3077 case ABMON_8: tm.tm_mon++;
3078 case ABMON_7: tm.tm_mon++;
3079 case ABMON_6: tm.tm_mon++;
3080 case ABMON_5: tm.tm_mon++;
3081 case ABMON_4: tm.tm_mon++;
3082 case ABMON_3: tm.tm_mon++;
3083 case ABMON_2: tm.tm_mon++;
3088 case MON_12: tm.tm_mon++;
3089 case MON_11: tm.tm_mon++;
3090 case MON_10: tm.tm_mon++;
3091 case MON_9: tm.tm_mon++;
3092 case MON_8: tm.tm_mon++;
3093 case MON_7: tm.tm_mon++;
3094 case MON_6: tm.tm_mon++;
3095 case MON_5: tm.tm_mon++;
3096 case MON_4: tm.tm_mon++;
3097 case MON_3: tm.tm_mon++;
3098 case MON_2: tm.tm_mon++;
3105 return_format = TRUE;
3110 return_format = TRUE;
3115 return_format = TRUE;
3120 return_format = TRUE;
3125 format = "%Ow"; /* Find the alternate digit for 0 */
3129 /* We can't use my_strftime() because it doesn't look at
3131 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3134 /* A zero return means one of:
3135 * a) there wasn't enough space in PL_langinfo_buf
3136 * b) the format, like a plain %p, returns empty
3137 * c) it was an illegal format, though some
3138 * implementations of strftime will just return the
3139 * illegal format as a plain character sequence.
3141 * To quickly test for case 'b)', try again but precede
3142 * the format with a plain character. If that result is
3143 * still empty, the problem is either 'a)' or 'c)' */
3145 Size_t format_size = strlen(format) + 1;
3146 Size_t mod_size = format_size + 1;
3150 Newx(mod_format, mod_size, char);
3151 Newx(temp_result, PL_langinfo_bufsize, char);
3153 my_strlcpy(mod_format + 1, format, mod_size);
3154 len = strftime(temp_result,
3155 PL_langinfo_bufsize,
3157 Safefree(mod_format);
3158 Safefree(temp_result);
3160 /* If 'len' is non-zero, it means that we had a case like
3161 * %p which means the current locale doesn't use a.m. or
3162 * p.m., and that is valid */
3165 /* Here, still didn't work. If we get well beyond a
3166 * reasonable size, bail out to prevent an infinite
3169 if (PL_langinfo_bufsize > 100 * format_size) {
3170 *PL_langinfo_buf = '\0';
3173 /* Double the buffer size to retry; Add 1 in case
3174 * original was 0, so we aren't stuck at 0. */
3175 PL_langinfo_bufsize *= 2;
3176 PL_langinfo_bufsize++;
3177 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3185 /* Here, we got a result.
3187 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3188 * alternate format for wday 0. If the value is the same as
3189 * the normal 0, there isn't an alternate, so clear the buffer.
3191 if ( item == ALT_DIGITS
3192 && strEQ(PL_langinfo_buf, "0"))
3194 *PL_langinfo_buf = '\0';
3197 /* ALT_DIGITS is problematic. Experiments on it showed that
3198 * strftime() did not always work properly when going from
3199 * alt-9 to alt-10. Only a few locales have this item defined,
3200 * and in all of them on Linux that khw was able to find,
3201 * nl_langinfo() merely returned the alt-0 character, possibly
3202 * doubled. Most Unicode digits are in blocks of 10
3203 * consecutive code points, so that is sufficient information
3204 * for those scripts, as we can infer alt-1, alt-2, .... But
3205 * for a Japanese locale, a CJK ideographic 0 is returned, and
3206 * the CJK digits are not in code point order, so you can't
3207 * really infer anything. The localedef for this locale did
3208 * specify the succeeding digits, so that strftime() works
3209 * properly on them, without needing to infer anything. But
3210 * the nl_langinfo() return did not give sufficient information
3211 * for the caller to understand what's going on. So until
3212 * there is evidence that it should work differently, this
3213 * returns the alt-0 string for ALT_DIGITS.
3215 * wday was chosen because its range is all a single digit.
3216 * Things like tm_sec have two digits as the minimum: '00' */
3220 retval = PL_langinfo_buf;
3222 /* If to return the format, not the value, overwrite the buffer
3223 * with it. But some strftime()s will keep the original format
3224 * if illegal, so change those to "" */
3225 if (return_format) {
3226 if (strEQ(PL_langinfo_buf, format)) {
3227 *PL_langinfo_buf = '\0';
3230 retval = save_to_buffer(format, &PL_langinfo_buf,
3231 &PL_langinfo_bufsize, 0);
3249 * Initialize locale awareness.
3252 Perl_init_i18nl10n(pTHX_ int printwarn)
3256 * 0 if not to output warning when setup locale is bad
3257 * 1 if to output warning based on value of PERL_BADLANG
3258 * >1 if to output regardless of PERL_BADLANG
3261 * 1 = set ok or not applicable,
3262 * 0 = fallback to a locale of lower priority
3263 * -1 = fallback to all locales failed, not even to the C locale
3265 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3266 * set, debugging information is output.
3268 * This looks more complicated than it is, mainly due to the #ifdefs.
3270 * We try to set LC_ALL to the value determined by the environment. If
3271 * there is no LC_ALL on this platform, we try the individual categories we
3272 * know about. If this works, we are done.
3274 * But if it doesn't work, we have to do something else. We search the
3275 * environment variables ourselves instead of relying on the system to do
3276 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3277 * think there is one), and the ultimate fallback "C". This is all done in
3278 * the same loop as above to avoid duplicating code, but it makes things
3279 * more complex. The 'trial_locales' array is initialized with just one
3280 * element; it causes the behavior described in the paragraph above this to
3281 * happen. If that fails, we add elements to 'trial_locales', and do extra
3282 * loop iterations to cause the behavior described in this paragraph.
3284 * On Ultrix, the locale MUST come from the environment, so there is
3285 * preliminary code to set it. I (khw) am not sure that it is necessary,
3286 * and that this couldn't be folded into the loop, but barring any real
3287 * platforms to test on, it's staying as-is
3289 * A slight complication is that in embedded Perls, the locale may already
3290 * be set-up, and we don't want to get it from the normal environment
3291 * variables. This is handled by having a special environment variable
3292 * indicate we're in this situation. We simply set setlocale's 2nd
3293 * parameter to be a NULL instead of "". That indicates to setlocale that
3294 * it is not to change anything, but to return the current value,
3295 * effectively initializing perl's db to what the locale already is.
3297 * We play the same trick with NULL if a LC_ALL succeeds. We call
3298 * setlocale() on the individual categores with NULL to get their existing
3299 * values for our db, instead of trying to change them.
3308 PERL_UNUSED_ARG(printwarn);
3310 #else /* USE_LOCALE */
3313 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3317 /* NULL uses the existing already set up locale */
3318 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3321 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3322 unsigned int trial_locales_count;
3323 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3324 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3325 bool setlocale_failure = FALSE;
3328 /* A later getenv() could zap this, so only use here */
3329 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3331 const bool locwarn = (printwarn > 1
3333 && ( ! bad_lang_use_once
3335 /* disallow with "" or "0" */
3337 && strNE("0", bad_lang_use_once)))));
3339 /* setlocale() return vals; not copied so must be looked at immediately */
3340 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3342 /* current locale for given category; should have been copied so aren't
3344 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3348 /* In some systems you can find out the system default locale
3349 * and use that as the fallback locale. */
3350 # define SYSTEM_DEFAULT_LOCALE
3352 # ifdef SYSTEM_DEFAULT_LOCALE
3354 const char *system_default_locale = NULL;
3359 # define DEBUG_LOCALE_INIT(a,b,c)
3362 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3364 # define DEBUG_LOCALE_INIT(category, locale, result) \
3366 if (debug_initialization) { \
3367 PerlIO_printf(Perl_debug_log, \
3369 __FILE__, __LINE__, \
3370 setlocale_debug_string(category, \
3376 /* Make sure the parallel arrays are properly set up */
3377 # ifdef USE_LOCALE_NUMERIC
3378 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3379 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3380 # ifdef USE_POSIX_2008_LOCALE
3381 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3384 # ifdef USE_LOCALE_CTYPE
3385 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3386 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3387 # ifdef USE_POSIX_2008_LOCALE
3388 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3391 # ifdef USE_LOCALE_COLLATE
3392 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3393 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3394 # ifdef USE_POSIX_2008_LOCALE
3395 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3398 # ifdef USE_LOCALE_TIME
3399 assert(categories[LC_TIME_INDEX] == LC_TIME);
3400 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3401 # ifdef USE_POSIX_2008_LOCALE
3402 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3405 # ifdef USE_LOCALE_MESSAGES
3406 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3407 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3408 # ifdef USE_POSIX_2008_LOCALE
3409 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3412 # ifdef USE_LOCALE_MONETARY
3413 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3414 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3415 # ifdef USE_POSIX_2008_LOCALE
3416 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3419 # ifdef USE_LOCALE_ADDRESS
3420 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3421 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3422 # ifdef USE_POSIX_2008_LOCALE
3423 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3426 # ifdef USE_LOCALE_IDENTIFICATION
3427 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3428 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3429 # ifdef USE_POSIX_2008_LOCALE
3430 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3433 # ifdef USE_LOCALE_MEASUREMENT
3434 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3435 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3436 # ifdef USE_POSIX_2008_LOCALE
3437 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3440 # ifdef USE_LOCALE_PAPER
3441 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3442 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3443 # ifdef USE_POSIX_2008_LOCALE
3444 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3447 # ifdef USE_LOCALE_TELEPHONE
3448 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3449 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3450 # ifdef USE_POSIX_2008_LOCALE
3451 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3455 assert(categories[LC_ALL_INDEX] == LC_ALL);
3456 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3457 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3458 # ifdef USE_POSIX_2008_LOCALE
3459 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3462 # endif /* DEBUGGING */
3464 /* Initialize the cache of the program's UTF-8ness for the always known
3465 * locales C and POSIX */
3466 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3467 sizeof(PL_locale_utf8ness));
3469 # ifdef USE_THREAD_SAFE_LOCALE
3472 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3476 # ifdef USE_POSIX_2008_LOCALE
3478 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3479 if (! PL_C_locale_obj) {
3480 Perl_croak_nocontext(
3481 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3483 if (DEBUG_Lv_TEST || debug_initialization) {
3484 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3489 # ifdef USE_LOCALE_NUMERIC
3491 PL_numeric_radix_sv = newSVpvs(".");
3495 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3497 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3498 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3501 # ifdef LOCALE_ENVIRON_REQUIRED
3504 * Ultrix setlocale(..., "") fails if there are no environment
3505 * variables from which to get a locale name.
3509 # error Ultrix without LC_ALL not implemented
3515 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3516 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3517 if (sl_result[LC_ALL_INDEX])
3520 setlocale_failure = TRUE;
3522 if (! setlocale_failure) {
3523 const char * locale_param;
3524 for (i = 0; i < LC_ALL_INDEX; i++) {
3525 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3528 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3529 if (! sl_result[i]) {
3530 setlocale_failure = TRUE;
3532 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3537 # endif /* LC_ALL */
3538 # endif /* LOCALE_ENVIRON_REQUIRED */
3540 /* We try each locale in the list until we get one that works, or exhaust
3541 * the list. Normally the loop is executed just once. But if setting the
3542 * locale fails, inside the loop we add fallback trials to the array and so
3543 * will execute the loop multiple times */
3544 trial_locales[0] = setlocale_init;
3545 trial_locales_count = 1;
3547 for (i= 0; i < trial_locales_count; i++) {
3548 const char * trial_locale = trial_locales[i];
3552 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3553 * when i==0, but I (khw) don't think that behavior makes much
3555 setlocale_failure = FALSE;
3557 # ifdef SYSTEM_DEFAULT_LOCALE
3558 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3560 /* On Windows machines, an entry of "" after the 0th means to use
3561 * the system default locale, which we now proceed to get. */
3562 if (strEQ(trial_locale, "")) {
3565 /* Note that this may change the locale, but we are going to do
3566 * that anyway just below */
3567 system_default_locale = do_setlocale_c(LC_ALL, "");
3568 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3570 /* Skip if invalid or if it's already on the list of locales to
3572 if (! system_default_locale) {
3573 goto next_iteration;
3575 for (j = 0; j < trial_locales_count; j++) {
3576 if (strEQ(system_default_locale, trial_locales[j])) {
3577 goto next_iteration;
3581 trial_locale = system_default_locale;
3584 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3586 # endif /* SYSTEM_DEFAULT_LOCALE */
3592 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3593 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3594 if (! sl_result[LC_ALL_INDEX]) {
3595 setlocale_failure = TRUE;
3598 /* Since LC_ALL succeeded, it should have changed all the other
3599 * categories it can to its value; so we massage things so that the
3600 * setlocales below just return their category's current values.
3601 * This adequately handles the case in NetBSD where LC_COLLATE may
3602 * not be defined for a locale, and setting it individually will
3603 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3604 * the POSIX locale. */
3605 trial_locale = NULL;
3608 # endif /* LC_ALL */
3610 if (! setlocale_failure) {
3612 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3614 = savepv(do_setlocale_r(categories[j], trial_locale));
3615 if (! curlocales[j]) {
3616 setlocale_failure = TRUE;
3618 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3621 if (! setlocale_failure) { /* All succeeded */
3622 break; /* Exit trial_locales loop */
3626 /* Here, something failed; will need to try a fallback. */
3632 if (locwarn) { /* Output failure info only on the first one */
3636 PerlIO_printf(Perl_error_log,
3637 "perl: warning: Setting locale failed.\n");
3639 # else /* !LC_ALL */
3641 PerlIO_printf(Perl_error_log,
3642 "perl: warning: Setting locale failed for the categories:\n\t");
3644 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3645 if (! curlocales[j]) {
3646 PerlIO_printf(Perl_error_log, category_names[j]);
3649 Safefree(curlocales[j]);
3653 # endif /* LC_ALL */
3655 PerlIO_printf(Perl_error_log,
3656 "perl: warning: Please check that your locale settings:\n");
3660 PerlIO_printf(Perl_error_log,
3661 "\tLANGUAGE = %c%s%c,\n",
3662 language ? '"' : '(',
3663 language ? language : "unset",
3664 language ? '"' : ')');
3667 PerlIO_printf(Perl_error_log,
3668 "\tLC_ALL = %c%s%c,\n",
3670 lc_all ? lc_all : "unset",
3671 lc_all ? '"' : ')');
3673 # if defined(USE_ENVIRON_ARRAY)
3678 /* Look through the environment for any variables of the
3679 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3680 * already handled above. These are assumed to be locale
3681 * settings. Output them and their values. */
3682 for (e = environ; *e; e++) {
3683 const STRLEN prefix_len = sizeof("LC_") - 1;
3686 if ( strBEGINs(*e, "LC_")
3687 && ! strBEGINs(*e, "LC_ALL=")
3688 && (uppers_len = strspn(*e + prefix_len,
3689 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3690 && ((*e)[prefix_len + uppers_len] == '='))
3692 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3693 (int) (prefix_len + uppers_len), *e,
3694 *e + prefix_len + uppers_len + 1);
3701 PerlIO_printf(Perl_error_log,
3702 "\t(possibly more locale environment variables)\n");
3706 PerlIO_printf(Perl_error_log,
3707 "\tLANG = %c%s%c\n",
3709 lang ? lang : "unset",
3712 PerlIO_printf(Perl_error_log,
3713 " are supported and installed on your system.\n");
3716 /* Calculate what fallback locales to try. We have avoided this
3717 * until we have to, because failure is quite unlikely. This will
3718 * usually change the upper bound of the loop we are in.
3720 * Since the system's default way of setting the locale has not
3721 * found one that works, We use Perl's defined ordering: LC_ALL,
3722 * LANG, and the C locale. We don't try the same locale twice, so
3723 * don't add to the list if already there. (On POSIX systems, the
3724 * LC_ALL element will likely be a repeat of the 0th element "",
3725 * but there's no harm done by doing it explicitly.
3727 * Note that this tries the LC_ALL environment variable even on
3728 * systems which have no LC_ALL locale setting. This may or may
3729 * not have been originally intentional, but there's no real need
3730 * to change the behavior. */
3732 for (j = 0; j < trial_locales_count; j++) {
3733 if (strEQ(lc_all, trial_locales[j])) {
3737 trial_locales[trial_locales_count++] = lc_all;
3742 for (j = 0; j < trial_locales_count; j++) {
3743 if (strEQ(lang, trial_locales[j])) {
3747 trial_locales[trial_locales_count++] = lang;
3751 # if defined(WIN32) && defined(LC_ALL)
3753 /* For Windows, we also try the system default locale before "C".
3754 * (If there exists a Windows without LC_ALL we skip this because
3755 * it gets too complicated. For those, the "C" is the next
3756 * fallback possibility). The "" is the same as the 0th element of
3757 * the array, but the code at the loop above knows to treat it
3758 * differently when not the 0th */
3759 trial_locales[trial_locales_count++] = "";
3763 for (j = 0; j < trial_locales_count; j++) {
3764 if (strEQ("C", trial_locales[j])) {
3768 trial_locales[trial_locales_count++] = "C";
3771 } /* end of first time through the loop */
3779 } /* end of looping through the trial locales */
3781 if (ok < 1) { /* If we tried to fallback */
3783 if (! setlocale_failure) { /* fallback succeeded */
3784 msg = "Falling back to";
3786 else { /* fallback failed */
3789 /* We dropped off the end of the loop, so have to decrement i to
3790 * get back to the value the last time through */
3794 msg = "Failed to fall back to";
3796 /* To continue, we should use whatever values we've got */
3798 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3799 Safefree(curlocales[j]);
3800 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3801 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3806 const char * description;
3807 const char * name = "";
3808 if (strEQ(trial_locales[i], "C")) {
3809 description = "the standard locale";
3813 # ifdef SYSTEM_DEFAULT_LOCALE
3815 else if (strEQ(trial_locales[i], "")) {
3816 description = "the system default locale";
3817 if (system_default_locale) {
3818 name = system_default_locale;
3822 # endif /* SYSTEM_DEFAULT_LOCALE */
3825 description = "a fallback locale";
3826 name = trial_locales[i];
3828 if (name && strNE(name, "")) {
3829 PerlIO_printf(Perl_error_log,
3830 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3833 PerlIO_printf(Perl_error_log,
3834 "perl: warning: %s %s.\n", msg, description);
3837 } /* End of tried to fallback */
3839 /* Done with finding the locales; update our records */
3841 # ifdef USE_LOCALE_CTYPE
3843 new_ctype(curlocales[LC_CTYPE_INDEX]);
3846 # ifdef USE_LOCALE_COLLATE
3848 new_collate(curlocales[LC_COLLATE_INDEX]);
3851 # ifdef USE_LOCALE_NUMERIC
3853 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3857 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3859 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3861 /* This caches whether each category's locale is UTF-8 or not. This
3862 * may involve changing the locale. It is ok to do this at
3863 * initialization time before any threads have started, but not later
3864 * unless thread-safe operations are used.
3865 * Caching means that if the program heeds our dictate not to change
3866 * locales in threaded applications, this data will remain valid, and
3867 * it may get queried without having to change locales. If the
3868 * environment is such that all categories have the same locale, this
3869 * isn't needed, as the code will not change the locale; but this
3870 * handles the uncommon case where the environment has disparate
3871 * locales for the categories */
3872 (void) _is_cur_LC_category_utf8(categories[i]);
3876 Safefree(curlocales[i]);
3879 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3881 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3882 * locale is UTF-8. The call to new_ctype() just above has already
3883 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3884 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3885 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3886 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3887 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3889 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3890 This is an alternative to using the -C command line switch
3891 (the -C if present will override this). */
3893 const char *p = PerlEnv_getenv("PERL_UNICODE");
3894 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3895 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3909 #endif /* USE_LOCALE */
3912 /* So won't continue to output stuff */
3913 DEBUG_INITIALIZATION_set(FALSE);
3920 #ifdef USE_LOCALE_COLLATE
3923 Perl__mem_collxfrm(pTHX_ const char *input_string,
3924 STRLEN len, /* Length of 'input_string' */
3925 STRLEN *xlen, /* Set to length of returned string
3926 (not including the collation index
3928 bool utf8 /* Is the input in UTF-8? */
3932 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3933 * differences. First, it handles embedded NULs. Second, it allocates a bit
3934 * more memory than needed for the transformed data itself. The real
3935 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3936 * the length of that, and doesn't include the collation index size.
3937 * Please see sv_collxfrm() to see how this is used. */
3939 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3941 char * s = (char *) input_string;
3942 STRLEN s_strlen = strlen(input_string);
3944 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3945 STRLEN length_in_chars;
3946 bool first_time = TRUE; /* Cleared after first loop iteration */
3948 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3950 /* Must be NUL-terminated */
3951 assert(*(input_string + len) == '\0');
3953 /* If this locale has defective collation, skip */
3954 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3955 DEBUG_L(PerlIO_printf(Perl_debug_log,
3956 "_mem_collxfrm: locale's collation is defective\n"));
3960 /* Replace any embedded NULs with the control that sorts before any others.
3961 * This will give as good as possible results on strings that don't
3962 * otherwise contain that character, but otherwise there may be
3963 * less-than-perfect results with that character and NUL. This is
3964 * unavoidable unless we replace strxfrm with our own implementation. */
3965 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3969 STRLEN sans_nuls_len;
3970 int try_non_controls;
3971 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3972 making sure 2nd byte is NUL.
3974 STRLEN this_replacement_len;
3976 /* If we don't know what non-NUL control character sorts lowest for
3977 * this locale, find it */
3978 if (PL_strxfrm_NUL_replacement == '\0') {
3980 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3981 includes the collation index
3984 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3986 /* Unlikely, but it may be that no control will work to replace
3987 * NUL, in which case we instead look for any character. Controls
3988 * are preferred because collation order is, in general, context
3989 * sensitive, with adjoining characters affecting the order, and
3990 * controls are less likely to have such interactions, allowing the
3991 * NUL-replacement to stand on its own. (Another way to look at it
3992 * is to imagine what would happen if the NUL were replaced by a
3993 * combining character; it wouldn't work out all that well.) */
3994 for (try_non_controls = 0;
3995 try_non_controls < 2;
3998 /* Look through all legal code points (NUL isn't) */
3999 for (j = 1; j < 256; j++) {
4000 char * x; /* j's xfrm plus collation index */
4001 STRLEN x_len; /* length of 'x' */
4002 STRLEN trial_len = 1;
4003 char cur_source[] = { '\0', '\0' };
4005 /* Skip non-controls the first time through the loop. The
4006 * controls in a UTF-8 locale are the L1 ones */
4007 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
4014 /* Create a 1-char string of the current code point */
4015 cur_source[0] = (char) j;
4017 /* Then transform it */
4018 x = _mem_collxfrm(cur_source, trial_len, &x_len,
4019 0 /* The string is not in UTF-8 */);
4021 /* Ignore any character that didn't successfully transform.
4027 /* If this character's transformation is lower than
4028 * the current lowest, this one becomes the lowest */
4029 if ( cur_min_x == NULL
4030 || strLT(x + COLLXFRM_HDR_LEN,
4031 cur_min_x + COLLXFRM_HDR_LEN))
4033 PL_strxfrm_NUL_replacement = j;
4034 Safefree(cur_min_x);
4040 } /* end of loop through all 255 characters */
4042 /* Stop looking if found */
4047 /* Unlikely, but possible, if there aren't any controls that
4048 * work in the locale, repeat the loop, looking for any
4049 * character that works */
4050 DEBUG_L(PerlIO_printf(Perl_debug_log,
4051 "_mem_collxfrm: No control worked. Trying non-controls\n"));
4052 } /* End of loop to try first the controls, then any char */
4055 DEBUG_L(PerlIO_printf(Perl_debug_log,
4056 "_mem_collxfrm: Couldn't find any character to replace"
4057 " embedded NULs in locale %s with", PL_collation_name));
4061 DEBUG_L(PerlIO_printf(Perl_debug_log,
4062 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
4063 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
4065 Safefree(cur_min_x);
4066 } /* End of determining the character that is to replace NULs */
4068 /* If the replacement is variant under UTF-8, it must match the
4069 * UTF8-ness of the original */
4070 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
4071 this_replacement_char[0] =
4072 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
4073 this_replacement_char[1] =
4074 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
4075 this_replacement_len = 2;
4078 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
4079 /* this_replacement_char[1] = '\0' was done at initialization */
4080 this_replacement_len = 1;
4083 /* The worst case length for the replaced string would be if every
4084 * character in it is NUL. Multiply that by the length of each
4085 * replacement, and allow for a trailing NUL */
4086 sans_nuls_len = (len * this_replacement_len) + 1;
4087 Newx(sans_nuls, sans_nuls_len, char);
4090 /* Replace each NUL with the lowest collating control. Loop until have
4091 * exhausted all the NULs */
4092 while (s + s_strlen < e) {
4093 my_strlcat(sans_nuls, s, sans_nuls_len);
4095 /* Do the actual replacement */
4096 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
4098 /* Move past the input NUL */
4100 s_strlen = strlen(s);
4103 /* And add anything that trails the final NUL */
4104 my_strlcat(sans_nuls, s, sans_nuls_len);
4106 /* Switch so below we transform this modified string */
4109 } /* End of replacing NULs */
4111 /* Make sure the UTF8ness of the string and locale match */
4112 if (utf8 != PL_in_utf8_COLLATE_locale) {
4113 /* XXX convert above Unicode to 10FFFF? */
4114 const char * const t = s; /* Temporary so we can later find where the
4117 /* Here they don't match. Change the string's to be what the locale is
4120 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
4121 s = (char *) bytes_to_utf8((const U8 *) s, &len);
4124 else { /* locale is not UTF-8; but input is; downgrade the input */
4126 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
4128 /* If the downgrade was successful we are done, but if the input
4129 * contains things that require UTF-8 to represent, have to do
4130 * damage control ... */
4131 if (UNLIKELY(utf8)) {
4133 /* What we do is construct a non-UTF-8 string with
4134 * 1) the characters representable by a single byte converted
4135 * to be so (if necessary);
4136 * 2) and the rest converted to collate the same as the
4137 * highest collating representable character. That makes
4138 * them collate at the end. This is similar to how we
4139 * handle embedded NULs, but we use the highest collating
4140 * code point instead of the smallest. Like the NUL case,
4141 * this isn't perfect, but is the best we can reasonably
4142 * do. Every above-255 code point will sort the same as
4143 * the highest-sorting 0-255 code point. If that code
4144 * point can combine in a sequence with some other code
4145 * points for weight calculations, us changing something to
4146 * be it can adversely affect the results. But in most
4147 * cases, it should work reasonably. And note that this is
4148 * really an illegal situation: using code points above 255
4149 * on a locale where only 0-255 are valid. If two strings
4150 * sort entirely equal, then the sort order for the
4151 * above-255 code points will be in code point order. */
4155 /* If we haven't calculated the code point with the maximum
4156 * collating order for this locale, do so now */
4157 if (! PL_strxfrm_max_cp) {
4160 /* The current transformed string that collates the
4161 * highest (except it also includes the prefixed collation
4163 char * cur_max_x = NULL;
4165 /* Look through all legal code points (NUL isn't) */
4166 for (j = 1; j < 256; j++) {
4169 char cur_source[] = { '\0', '\0' };
4171 /* Create a 1-char string of the current code point */
4172 cur_source[0] = (char) j;
4174 /* Then transform it */
4175 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4177 /* If something went wrong (which it shouldn't), just
4178 * ignore this code point */
4183 /* If this character's transformation is higher than
4184 * the current highest, this one becomes the highest */
4185 if ( cur_max_x == NULL
4186 || strGT(x + COLLXFRM_HDR_LEN,
4187 cur_max_x + COLLXFRM_HDR_LEN))
4189 PL_strxfrm_max_cp = j;
4190 Safefree(cur_max_x);
4199 DEBUG_L(PerlIO_printf(Perl_debug_log,
4200 "_mem_collxfrm: Couldn't find any character to"
4201 " replace above-Latin1 chars in locale %s with",
4202 PL_collation_name));
4206 DEBUG_L(PerlIO_printf(Perl_debug_log,
4207 "_mem_collxfrm: highest 1-byte collating character"
4208 " in locale %s is 0x%02X\n",
4210 PL_strxfrm_max_cp));
4212 Safefree(cur_max_x);
4215 /* Here we know which legal code point collates the highest.
4216 * We are ready to construct the non-UTF-8 string. The length
4217 * will be at least 1 byte smaller than the input string
4218 * (because we changed at least one 2-byte character into a
4219 * single byte), but that is eaten up by the trailing NUL */
4225 char * e = (char *) t + len;
4227 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4229 if (UTF8_IS_INVARIANT(cur_char)) {
4232 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4233 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4235 else { /* Replace illegal cp with highest collating
4237 s[d++] = PL_strxfrm_max_cp;
4241 Renew(s, d, char); /* Free up unused space */
4246 /* Here, we have constructed a modified version of the input. It could
4247 * be that we already had a modified copy before we did this version.
4248 * If so, that copy is no longer needed */
4249 if (t != input_string) {
4254 length_in_chars = (utf8)
4255 ? utf8_length((U8 *) s, (U8 *) s + len)
4258 /* The first element in the output is the collation id, used by
4259 * sv_collxfrm(); then comes the space for the transformed string. The
4260 * equation should give us a good estimate as to how much is needed */
4261 xAlloc = COLLXFRM_HDR_LEN
4263 + (PL_collxfrm_mult * length_in_chars);
4264 Newx(xbuf, xAlloc, char);
4265 if (UNLIKELY(! xbuf)) {
4266 DEBUG_L(PerlIO_printf(Perl_debug_log,
4267 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4271 /* Store the collation id */
4272 *(U32*)xbuf = PL_collation_ix;
4274 /* Then the transformation of the input. We loop until successful, or we
4278 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4280 /* If the transformed string occupies less space than we told strxfrm()
4281 * was available, it means it successfully transformed the whole
4283 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4285 /* Some systems include a trailing NUL in the returned length.
4286 * Ignore it, using a loop in case multiple trailing NULs are
4289 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4294 /* If the first try didn't get it, it means our prediction was low.
4295 * Modify the coefficients so that we predict a larger value in any
4296 * future transformations */
4298 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4299 STRLEN computed_guess = PL_collxfrm_base
4300 + (PL_collxfrm_mult * length_in_chars);
4302 /* On zero-length input, just keep current slope instead of
4304 const STRLEN new_m = (length_in_chars != 0)
4305 ? needed / length_in_chars
4308 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4309 "%s: %d: initial size of %zu bytes for a length "
4310 "%zu string was insufficient, %zu needed\n",
4312 computed_guess, length_in_chars, needed));
4314 /* If slope increased, use it, but discard this result for
4315 * length 1 strings, as we can't be sure that it's a real slope
4317 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4321 STRLEN old_m = PL_collxfrm_mult;
4322 STRLEN old_b = PL_collxfrm_base;
4326 PL_collxfrm_mult = new_m;
4327 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4328 computed_guess = PL_collxfrm_base
4329 + (PL_collxfrm_mult * length_in_chars);
4330 if (computed_guess < needed) {
4331 PL_collxfrm_base += needed - computed_guess;
4334 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4335 "%s: %d: slope is now %zu; was %zu, base "
4336 "is now %zu; was %zu\n",
4338 PL_collxfrm_mult, old_m,
4339 PL_collxfrm_base, old_b));
4341 else { /* Slope didn't change, but 'b' did */
4342 const STRLEN new_b = needed
4345 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4346 "%s: %d: base is now %zu; was %zu\n",
4348 new_b, PL_collxfrm_base));
4349 PL_collxfrm_base = new_b;
4356 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4357 DEBUG_L(PerlIO_printf(Perl_debug_log,
4358 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4359 *xlen, PERL_INT_MAX));
4363 /* A well-behaved strxfrm() returns exactly how much space it needs
4364 * (usually not including the trailing NUL) when it fails due to not
4365 * enough space being provided. Assume that this is the case unless
4366 * it's been proven otherwise */
4367 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4368 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4370 else { /* Here, either:
4371 * 1) The strxfrm() has previously shown bad behavior; or
4372 * 2) It isn't the first time through the loop, which means
4373 * that the strxfrm() is now showing bad behavior, because
4374 * we gave it what it said was needed in the previous
4375 * iteration, and it came back saying it needed still more.
4376 * (Many versions of cygwin fit this. When the buffer size
4377 * isn't sufficient, they return the input size instead of
4378 * how much is needed.)
4379 * Increase the buffer size by a fixed percentage and try again.
4381 xAlloc += (xAlloc / 4) + 1;
4382 PL_strxfrm_is_behaved = FALSE;
4386 if (DEBUG_Lv_TEST || debug_initialization) {
4387 PerlIO_printf(Perl_debug_log,
4388 "_mem_collxfrm required more space than previously calculated"
4389 " for locale %s, trying again with new guess=%d+%zu\n",
4390 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4391 xAlloc - COLLXFRM_HDR_LEN);
4398 Renew(xbuf, xAlloc, char);
4399 if (UNLIKELY(! xbuf)) {
4400 DEBUG_L(PerlIO_printf(Perl_debug_log,
4401 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4411 if (DEBUG_Lv_TEST || debug_initialization) {
4413 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4414 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4415 PerlIO_printf(Perl_debug_log, "%s\n",
4416 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4422 /* Free up unneeded space; retain ehough for trailing NUL */
4423 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4425 if (s != input_string) {
4435 if (DEBUG_Lv_TEST || debug_initialization) {
4436 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4442 if (s != input_string) {
4453 S_print_collxfrm_input_and_return(pTHX_
4454 const char * const s,
4455 const char * const e,
4456 const STRLEN * const xlen,
4460 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4462 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4463 (UV)PL_collation_ix);
4465 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4468 PerlIO_printf(Perl_debug_log, "NULL");
4470 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4472 print_bytes_for_locale(s, e, is_utf8);
4474 PerlIO_printf(Perl_debug_log, "'\n");
4477 # endif /* DEBUGGING */
4478 #endif /* USE_LOCALE_COLLATE */
4483 S_print_bytes_for_locale(pTHX_
4484 const char * const s,
4485 const char * const e,
4489 bool prev_was_printable = TRUE;
4490 bool first_time = TRUE;
4492 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4496 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4499 if (! prev_was_printable) {
4500 PerlIO_printf(Perl_debug_log, " ");
4502 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4503 prev_was_printable = TRUE;
4507 PerlIO_printf(Perl_debug_log, " ");
4509 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4510 prev_was_printable = FALSE;
4512 t += (is_utf8) ? UTF8SKIP(t) : 1;
4517 # endif /* #ifdef DEBUGGING */
4520 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4522 /* Changes the locale for LC_'switch_category" to that of
4523 * LC_'template_category', if they aren't already the same. If not NULL,
4524 * 'template_locale' is the locale that 'template_category' is in.
4526 * Returns a copy of the name of the original locale for 'switch_category'
4527 * so can be switched back to with the companion function
4528 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4530 char * restore_to_locale = NULL;
4532 if (switch_category == template_category) { /* No changes needed */
4536 /* Find the original locale of the category we may need to change, so that
4537 * it can be restored to later */
4538 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4540 if (! restore_to_locale) {
4542 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4543 __FILE__, __LINE__, category_name(switch_category), errno);
4546 /* If the locale of the template category wasn't passed in, find it now */
4547 if (template_locale == NULL) {
4548 template_locale = do_setlocale_r(template_category, NULL);
4549 if (! template_locale) {
4551 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4552 __FILE__, __LINE__, category_name(template_category), errno);
4556 /* It the locales are the same, there's nothing to do */
4557 if (strEQ(restore_to_locale, template_locale)) {
4558 Safefree(restore_to_locale);
4560 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4561 category_name(switch_category), restore_to_locale));
4566 /* Finally, change the locale to the template one */
4567 if (! do_setlocale_r(switch_category, template_locale)) {
4569 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4570 __FILE__, __LINE__, category_name(switch_category),
4571 template_locale, errno);
4574 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4575 category_name(switch_category), template_locale));
4577 return restore_to_locale;
4581 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4583 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4584 * copy that will be freed by this function), or do nothing if the latter
4585 * parameter is NULL */
4587 if (original_locale == NULL) {
4591 if (! do_setlocale_r(category, original_locale)) {
4593 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4595 category_name(category), original_locale, errno);
4598 Safefree(original_locale);
4601 /* is_cur_LC_category_utf8 uses a small char buffer to avoid malloc/free */
4602 #define CUR_LC_BUFFER_SIZE 64
4605 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4607 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4608 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4609 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4610 * could give the wrong result. The result will very likely be correct for
4611 * languages that have commonly used non-ASCII characters, but for notably
4612 * English, it comes down to if the locale's name ends in something like
4613 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4615 * If the platform is early C89, not containing mbtowc(), or we are
4616 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4617 * These work very well for non-Latin locales or those whose currency
4618 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4619 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4620 * on the name containing UTF-8 or not. */
4622 /* Name of current locale corresponding to the input category */
4623 const char *save_input_locale = NULL;
4625 bool is_utf8 = FALSE; /* The return value */
4627 /* The variables below are for the cache of previous lookups using this
4628 * function. The cache is a C string, described at the definition for
4629 * 'C_and_POSIX_utf8ness'.
4631 * The first part of the cache is fixed, for the C and POSIX locales. The
4632 * varying part starts just after them. */
4633 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4635 Size_t utf8ness_cache_size; /* Size of the varying portion */
4636 Size_t input_name_len; /* Length in bytes of save_input_locale */
4637 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4638 the name in the cache */
4639 char * delimited; /* The name plus the delimiters used to store
4641 char buffer[CUR_LC_BUFFER_SIZE]; /* small buffer */
4642 char * name_pos; /* position of 'delimited' in the cache, or 0
4648 assert(category != LC_ALL);
4652 /* Get the desired category's locale */
4653 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4654 if (! save_input_locale) {
4656 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4657 __FILE__, __LINE__, category_name(category), errno);
4660 DEBUG_L(PerlIO_printf(Perl_debug_log,
4661 "Current locale for %s is %s\n",
4662 category_name(category), save_input_locale));
4664 input_name_len = strlen(save_input_locale);
4666 /* In our cache, each name is accompanied by two delimiters and a single
4668 input_name_len_with_overhead = input_name_len + 3;
4670 if ( input_name_len_with_overhead <= CUR_LC_BUFFER_SIZE ) {
4671 /* we can use the buffer, avoid a malloc */
4673 } else { /* need a malloc */
4674 /* Allocate and populate space for a copy of the name surrounded by the
4676 Newx(delimited, input_name_len_with_overhead, char);
4679 delimited[0] = UTF8NESS_SEP[0];
4680 Copy(save_input_locale, delimited + 1, input_name_len, char);
4681 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4682 delimited[input_name_len+2] = '\0';
4684 /* And see if that is in the cache */
4685 name_pos = instr(PL_locale_utf8ness, delimited);
4687 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4691 if (DEBUG_Lv_TEST || debug_initialization) {
4692 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4693 save_input_locale, is_utf8);
4698 /* And, if not already in that position, move it to the beginning of
4699 * the non-constant portion of the list, since it is the most recently
4700 * used. (We don't have to worry about overflow, since just moving
4701 * existing names around) */
4702 if (name_pos > utf8ness_cache) {
4703 Move(utf8ness_cache,
4704 utf8ness_cache + input_name_len_with_overhead,
4705 name_pos - utf8ness_cache, char);
4708 input_name_len_with_overhead - 1, char);
4709 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4712 /* free only when not using the buffer */
4713 if ( delimited != buffer ) Safefree(delimited);
4714 Safefree(save_input_locale);
4718 /* Here we don't have stored the utf8ness for the input locale. We have to
4721 # if defined(USE_LOCALE_CTYPE) \
4722 && ( defined(HAS_NL_LANGINFO) \
4723 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4726 const char *original_ctype_locale
4727 = switch_category_locale_to_template(LC_CTYPE,
4731 /* Here the current LC_CTYPE is set to the locale of the category whose
4732 * information is desired. This means that nl_langinfo() and mbtowc()
4733 * should give the correct results */
4735 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4736 calling the functions if we have this */
4738 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4739 * Unicode code point. */
4741 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4742 __FILE__, __LINE__, (int) MB_CUR_MAX));
4743 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
4745 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4746 goto finish_and_return;
4750 # if defined(HAS_NL_LANGINFO)
4752 { /* The task is easiest if the platform has this POSIX 2001 function.
4753 Except on some platforms it can wrongly return "", so have to have
4754 a fallback. And it can return that it's UTF-8, even if there are
4755 variances from that. For example, Turkish locales may use the
4756 alternate dotted I rules, and sometimes it appears to be a
4757 defective locale definition. XXX We should probably check for
4758 these in the Latin1 range and warn (but on glibc, requires
4759 iswalnum() etc. due to their not handling 80-FF correctly */
4760 const char *codeset = my_nl_langinfo(CODESET, FALSE);
4761 /* FALSE => already in dest locale */
4763 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4764 "\tnllanginfo returned CODESET '%s'\n", codeset));
4766 if (codeset && strNE(codeset, "")) {
4768 /* If the implementation of foldEQ() somehow were
4769 * to change to not go byte-by-byte, this could
4770 * read past end of string, as only one length is
4771 * checked. But currently, a premature NUL will
4772 * compare false, and it will stop there */
4773 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
4774 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
4776 DEBUG_L(PerlIO_printf(Perl_debug_log,
4777 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
4779 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4780 goto finish_and_return;
4785 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
4786 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
4787 * late adder to C89, so very likely to have it. However, testing has
4788 * shown that, like nl_langinfo() above, there are locales that are not
4789 * strictly UTF-8 that this will return that they are */
4796 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4802 /* mbrtowc() and mbtowc() convert a byte string to a wide
4803 * character. Feed a byte string to one of them and check that the
4804 * result is the expected Unicode code point */
4806 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4807 /* Prefer this function if available, as it's reentrant */
4809 memset(&ps, 0, sizeof(ps));;
4810 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
4813 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
4819 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
4821 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
4828 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4829 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
4830 len, (unsigned int) wc, GET_ERRNO));
4832 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
4833 && wc == (wchar_t) UNICODE_REPLACEMENT);
4838 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4839 goto finish_and_return;
4844 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
4845 * try looking at the currency symbol to see if it disambiguates
4846 * things. Often that will be in the native script, and if the symbol
4847 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
4848 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
4849 * string being valid UTF-8 are quite small */
4851 # ifdef USE_LOCALE_MONETARY
4853 /* If have LC_MONETARY, we can look at the currency symbol. Often that
4854 * will be in the native script. We do this one first because there is
4855 * just one string to examine, so potentially avoids work */
4858 const char *original_monetary_locale
4859 = switch_category_locale_to_template(LC_MONETARY,
4862 bool only_ascii = FALSE;
4863 const U8 * currency_string
4864 = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE);
4865 /* 2nd param not relevant for this item */
4866 const U8 * first_variant;
4868 assert( *currency_string == '-'
4869 || *currency_string == '+'
4870 || *currency_string == '.');
4874 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
4876 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));
4880 is_utf8 = is_strict_utf8_string(first_variant, 0);
4883 restore_switched_locale(LC_MONETARY, original_monetary_locale);
4887 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
4888 * otherwise assume the locale is UTF-8 if and only if the symbol
4889 * is non-ascii UTF-8. */
4890 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
4891 save_input_locale, is_utf8));
4892 goto finish_and_return;
4896 # endif /* USE_LOCALE_MONETARY */
4897 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
4899 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
4900 * the names of the months and weekdays, timezone, and am/pm indicator */
4902 const char *original_time_locale
4903 = switch_category_locale_to_template(LC_TIME,
4907 bool is_dst = FALSE;
4911 char * formatted_time;
4913 /* Here the current LC_TIME is set to the locale of the category
4914 * whose information is desired. Look at all the days of the week and
4915 * month names, and the timezone and am/pm indicator for UTF-8 variant
4916 * characters. The first such a one found will tell us if the locale
4917 * is UTF-8 or not */
4919 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
4920 formatted_time = my_strftime("%A %B %Z %p",
4921 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
4922 if ( ! formatted_time
4923 || is_utf8_invariant_string((U8 *) formatted_time, 0))
4926 /* Here, we didn't find a non-ASCII. Try the next time through
4927 * with the complemented dst and am/pm, and try with the next
4928 * weekday. After we have gotten all weekdays, try the next
4931 hour = (hour + 12) % 24;
4939 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
4940 * false otherwise. But first, restore LC_TIME to its original
4941 * locale if we changed it */
4942 restore_switched_locale(LC_TIME, original_time_locale);
4944 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
4946 is_utf8_string((U8 *) formatted_time, 0)));
4947 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
4948 goto finish_and_return;
4951 /* Falling off the end of the loop indicates all the names were just
4952 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
4953 * to its original locale */
4954 restore_switched_locale(LC_TIME, original_time_locale);
4955 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));
4960 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
4962 /* This code is ifdefd out because it was found to not be necessary in testing
4963 * on our dromedary test machine, which has over 700 locales. There, this
4964 * added no value to looking at the currency symbol and the time strings. I
4965 * left it in so as to avoid rewriting it if real-world experience indicates
4966 * that dromedary is an outlier. Essentially, instead of returning abpve if we
4967 * haven't found illegal utf8, we continue on and examine all the strerror()
4968 * messages on the platform for utf8ness. If all are ASCII, we still don't
4969 * know the answer; but otherwise we have a pretty good indication of the
4970 * utf8ness. The reason this doesn't help much is that the messages may not
4971 * have been translated into the locale. The currency symbol and time strings
4972 * are much more likely to have been translated. */
4975 bool non_ascii = FALSE;
4976 const char *original_messages_locale
4977 = switch_category_locale_to_template(LC_MESSAGES,
4980 const char * errmsg = NULL;
4982 /* Here the current LC_MESSAGES is set to the locale of the category
4983 * whose information is desired. Look through all the messages. We
4984 * can't use Strerror() here because it may expand to code that
4985 * segfaults in miniperl */
4987 for (e = 0; e <= sys_nerr; e++) {
4989 errmsg = sys_errlist[e];
4990 if (errno || !errmsg) {
4993 errmsg = savepv(errmsg);
4994 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
4996 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
5002 restore_switched_locale(LC_MESSAGES, original_messages_locale);
5006 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
5007 * any non-ascii means it is one; otherwise we assume it isn't */
5008 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
5011 goto finish_and_return;
5014 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));
5018 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
5021 /* As a last resort, look at the locale name to see if it matches
5022 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
5023 * return of setlocale(), is actually defined to be opaque, so we can't
5024 * really rely on the absence of various substrings in the name to indicate
5025 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
5026 * be a UTF-8 locale. Similarly for the other common names */
5029 const Size_t final_pos = strlen(save_input_locale) - 1;
5031 if (final_pos >= 3) {
5032 const char *name = save_input_locale;
5034 /* Find next 'U' or 'u' and look from there */
5035 while ((name += strcspn(name, "Uu") + 1)
5036 <= save_input_locale + final_pos - 2)
5038 if ( isALPHA_FOLD_NE(*name, 't')
5039 || isALPHA_FOLD_NE(*(name + 1), 'f'))
5044 if (*(name) == '-') {
5045 if ((name > save_input_locale + final_pos - 1)) {
5050 if (*(name) == '8') {
5051 DEBUG_L(PerlIO_printf(Perl_debug_log,
5052 "Locale %s ends with UTF-8 in name\n",
5053 save_input_locale));
5055 goto finish_and_return;
5058 DEBUG_L(PerlIO_printf(Perl_debug_log,
5059 "Locale %s doesn't end with UTF-8 in name\n",
5060 save_input_locale));
5065 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
5066 if (memENDs(save_input_locale, final_pos, "65001")) {
5067 DEBUG_L(PerlIO_printf(Perl_debug_log,
5068 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
5069 save_input_locale));
5071 goto finish_and_return;
5078 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
5079 * since we are about to return FALSE anyway, there is no point in doing
5080 * this extra work */
5083 if (instr(save_input_locale, "8859")) {
5084 DEBUG_L(PerlIO_printf(Perl_debug_log,
5085 "Locale %s has 8859 in name, not UTF-8 locale\n",
5086 save_input_locale));
5088 goto finish_and_return;
5092 DEBUG_L(PerlIO_printf(Perl_debug_log,
5093 "Assuming locale %s is not a UTF-8 locale\n",
5094 save_input_locale));
5097 # endif /* the code that is compiled when no modern LC_CTYPE */
5101 /* Cache this result so we don't have to go through all this next time. */
5102 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
5103 - (utf8ness_cache - PL_locale_utf8ness);
5105 /* But we can't save it if it is too large for the total space available */
5106 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
5107 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
5109 /* Here it can fit, but we may need to clear out the oldest cached
5110 * result(s) to do so. Check */
5111 if (utf8ness_cache_len + input_name_len_with_overhead
5112 >= utf8ness_cache_size)
5114 /* Here we have to clear something out to make room for this.
5115 * Start looking at the rightmost place where it could fit and find
5116 * the beginning of the entry that extends past that. */
5117 char * cutoff = (char *) my_memrchr(utf8ness_cache,
5120 - input_name_len_with_overhead);
5123 assert(cutoff >= utf8ness_cache);
5125 /* This and all subsequent entries must be removed */
5127 utf8ness_cache_len = strlen(utf8ness_cache);
5130 /* Make space for the new entry */
5131 Move(utf8ness_cache,
5132 utf8ness_cache + input_name_len_with_overhead,
5133 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
5136 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
5137 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
5139 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1] & ~1) != '0') {
5141 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu,"
5142 " inserted_name=%s, its_len=%zu\n",
5144 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
5145 delimited, input_name_len_with_overhead);
5151 if (DEBUG_Lv_TEST) {
5152 const char * s = PL_locale_utf8ness;
5154 /* Audit the structure */
5155 while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) {
5158 if (*s != UTF8NESS_SEP[0]) {
5160 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5161 " separator %.*s<-- HERE %s\n",
5163 (int) (s - PL_locale_utf8ness), PL_locale_utf8ness,
5167 e = strchr(s, UTF8NESS_PREFIX[0]);
5169 e = PL_locale_utf8ness + strlen(PL_locale_utf8ness);
5171 "panic: %s: %d: Corrupt utf8ness_cache: missing"
5172 " separator %.*s<-- HERE %s\n",
5174 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5178 if (*e != '0' && *e != '1') {
5180 "panic: %s: %d: Corrupt utf8ness_cache: utf8ness"
5181 " must be [01] %.*s<-- HERE %s\n",
5183 (int) (e + 1 - PL_locale_utf8ness),
5184 PL_locale_utf8ness, e + 1);
5186 if (ninstr(PL_locale_utf8ness, s, s-1, e)) {
5188 "panic: %s: %d: Corrupt utf8ness_cache: entry"
5189 " has duplicate %.*s<-- HERE %s\n",
5191 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
5198 if (DEBUG_Lv_TEST || debug_initialization) {
5200 PerlIO_printf(Perl_debug_log,
5201 "PL_locale_utf8ness is now %s; returning %d\n",
5202 PL_locale_utf8ness, is_utf8);
5207 /* free only when not using the buffer */
5208 if ( delimited != buffer ) Safefree(delimited);
5209 Safefree(save_input_locale);
5216 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
5219 /* Internal function which returns if we are in the scope of a pragma that
5220 * enables the locale category 'category'. 'compiling' should indicate if
5221 * this is during the compilation phase (TRUE) or not (FALSE). */
5223 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
5225 SV *these_categories = cop_hints_fetch_pvs(cop, "locale", 0);
5226 if (! these_categories || these_categories == &PL_sv_placeholder) {
5230 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
5231 * a valid unsigned */
5232 assert(category >= -1);
5233 return cBOOL(SvUV(these_categories) & (1U << (category + 1)));
5237 Perl_my_strerror(pTHX_ const int errnum)
5239 /* Returns a mortalized copy of the text of the error message associated
5240 * with 'errnum'. It uses the current locale's text unless the platform
5241 * doesn't have the LC_MESSAGES category or we are not being called from
5242 * within the scope of 'use locale'. In the former case, it uses whatever
5243 * strerror returns; in the latter case it uses the text from the C locale.
5245 * The function just calls strerror(), but temporarily switches, if needed,
5246 * to the C locale */
5251 #ifndef USE_LOCALE_MESSAGES
5253 /* If platform doesn't have messages category, we don't do any switching to
5254 * the C locale; we just use whatever strerror() returns */
5256 errstr = savepv(Strerror(errnum));
5258 #else /* Has locale messages */
5260 const bool within_locale_scope = IN_LC(LC_MESSAGES);
5262 # ifndef USE_ITHREADS
5264 /* This function is trivial without threads. */
5265 if (within_locale_scope) {
5266 errstr = savepv(strerror(errnum));
5269 const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL));
5271 do_setlocale_c(LC_MESSAGES, "C");
5272 errstr = savepv(strerror(errnum));
5273 do_setlocale_c(LC_MESSAGES, save_locale);
5274 Safefree(save_locale);
5277 # elif defined(USE_POSIX_2008_LOCALE) \
5278 && defined(HAS_STRERROR_L) \
5279 && defined(HAS_DUPLOCALE)
5281 /* This function is also trivial if we don't have to worry about thread
5282 * safety and have strerror_l(), as it handles the switch of locales so we
5283 * don't have to deal with that. We don't have to worry about thread
5284 * safety if strerror_r() is also available. Both it and strerror_l() are
5285 * thread-safe. Plain strerror() isn't thread safe. But on threaded
5286 * builds when strerror_r() is available, the apparent call to strerror()
5287 * below is actually a macro that behind-the-scenes calls strerror_r(). */
5289 # ifdef HAS_STRERROR_R
5291 if (within_locale_scope) {
5292 errstr = savepv(strerror(errnum));
5295 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
5300 /* Here we have strerror_l(), but not strerror_r() and we are on a
5301 * threaded-build. We use strerror_l() for everything, constructing a
5302 * locale to pass to it if necessary */
5304 bool do_free = FALSE;
5305 locale_t locale_to_use;
5307 if (within_locale_scope) {
5308 locale_to_use = uselocale((locale_t) 0);
5309 if (locale_to_use == LC_GLOBAL_LOCALE) {
5310 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
5314 else { /* Use C locale if not within 'use locale' scope */
5315 locale_to_use = PL_C_locale_obj;
5318 errstr = savepv(strerror_l(errnum, locale_to_use));
5321 freelocale(locale_to_use);
5325 # else /* Doesn't have strerror_l() */
5327 const char * save_locale = NULL;
5328 bool locale_is_C = FALSE;
5330 /* We have a critical section to prevent another thread from executing this
5331 * same code at the same time. (On thread-safe perls, the LOCK is a
5332 * no-op.) Since this is the only place in core that changes LC_MESSAGES
5333 * (unless the user has called setlocale(), this works to prevent races. */
5336 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5337 "my_strerror called with errnum %d\n", errnum));
5338 if (! within_locale_scope) {
5339 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
5340 if (! save_locale) {
5342 "panic: %s: %d: Could not find current LC_MESSAGES locale,"
5343 " errno=%d\n", __FILE__, __LINE__, errno);
5346 locale_is_C = isNAME_C_OR_POSIX(save_locale);
5348 /* Switch to the C locale if not already in it */
5349 if (! locale_is_C) {
5351 /* The setlocale() just below likely will zap 'save_locale', so
5353 save_locale = savepv(save_locale);
5354 do_setlocale_c(LC_MESSAGES, "C");
5357 } /* end of ! within_locale_scope */
5359 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
5360 __FILE__, __LINE__));
5363 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5364 "Any locale change has been done; about to call Strerror\n"));
5365 errstr = savepv(Strerror(errnum));
5367 if (! within_locale_scope) {
5368 if (save_locale && ! locale_is_C) {
5369 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
5371 "panic: %s: %d: setlocale restore failed, errno=%d\n",
5372 __FILE__, __LINE__, errno);
5374 Safefree(save_locale);
5380 # endif /* End of doesn't have strerror_l */
5383 if (DEBUG_Lv_TEST) {
5384 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
5385 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
5386 PerlIO_printf(Perl_debug_log, "'\n");
5390 #endif /* End of does have locale messages */
5398 =for apidoc switch_to_global_locale
5400 On systems without locale support, or on typical single-threaded builds, or on
5401 platforms that do not support per-thread locale operations, this function does
5402 nothing. On such systems that do have locale support, only a locale global to
5403 the whole program is available.
5405 On multi-threaded builds on systems that do have per-thread locale operations,
5406 this function converts the thread it is running in to use the global locale.
5407 This is for code that has not yet or cannot be updated to handle multi-threaded
5408 locale operation. As long as only a single thread is so-converted, everything
5409 works fine, as all the other threads continue to ignore the global one, so only
5410 this thread looks at it.
5412 However, on Windows systems this isn't quite true prior to Visual Studio 15,
5413 at which point Microsoft fixed a bug. A race can occur if you use the
5414 following operations on earlier Windows platforms:
5418 =item L<POSIX::localeconv|POSIX/localeconv>
5420 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5422 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5426 The first item is not fixable (except by upgrading to a later Visual Studio
5427 release), but it would be possible to work around the latter two items by using
5428 the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches
5431 Without this function call, threads that use the L<C<setlocale(3)>> system
5432 function will not work properly, as all the locale-sensitive functions will
5433 look at the per-thread locale, and C<setlocale> will have no effect on this
5436 Perl code should convert to either call
5437 L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system
5438 C<setlocale>) or use the methods given in L<perlcall> to call
5439 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly
5440 handle all cases of single- vs multi-thread, POSIX 2008-supported or not.
5442 Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can
5443 continue to work if this function is called before transferring control to the
5446 Upon return from the code that needs to use the global locale,
5447 L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe
5448 multi-thread operation.
5454 Perl_switch_to_global_locale()
5457 #ifdef USE_THREAD_SAFE_LOCALE
5460 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
5463 # ifdef HAS_QUERYLOCALE
5465 setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0)));
5472 for (i = 0; i < LC_ALL_INDEX; i++) {
5473 setlocale(categories[i], do_setlocale_r(categories[i], NULL));
5479 uselocale(LC_GLOBAL_LOCALE);
5488 =for apidoc sync_locale
5490 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or
5491 change the locale (though changing the locale is antisocial and dangerous on
5492 multi-threaded systems that don't have multi-thread safe locale operations.
5493 (See L<perllocale/Multi-threaded operation>). Using the system
5494 L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries
5495 called from XS, such as C<Gtk> do so, and this can't be changed. When the
5496 locale is changed by XS code that didn't use
5497 L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the
5498 locale has changed. Use this function to do so, before returning to Perl.
5500 The return value is a boolean: TRUE if the global locale at the time of call
5501 was in effect; and FALSE if a per-thread locale was in effect. This can be
5502 used by the caller that needs to restore things as-they-were to decide whether
5504 L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>.
5519 const char * newlocale;
5522 # ifdef USE_POSIX_2008_LOCALE
5524 bool was_in_global_locale = FALSE;
5525 locale_t cur_obj = uselocale((locale_t) 0);
5527 /* On Windows, unless the foreign code has turned off the thread-safe
5528 * locale setting, any plain setlocale() will have affected what we see, so
5529 * no need to worry. Otherwise, If the foreign code has done a plain
5530 * setlocale(), it will only affect the global locale on POSIX systems, but
5531 * will affect the */
5532 if (cur_obj == LC_GLOBAL_LOCALE) {
5534 # ifdef HAS_QUERY_LOCALE
5536 do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL));
5542 /* We can't trust that we can read the LC_ALL format on the
5543 * platform, so do them individually */
5544 for (i = 0; i < LC_ALL_INDEX; i++) {
5545 do_setlocale_r(categories[i], setlocale(categories[i], NULL));
5550 was_in_global_locale = TRUE;
5555 bool was_in_global_locale = TRUE;
5558 # ifdef USE_LOCALE_CTYPE
5560 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
5561 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5562 "%s:%d: %s\n", __FILE__, __LINE__,
5563 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
5564 new_ctype(newlocale);
5565 Safefree(newlocale);
5567 # endif /* USE_LOCALE_CTYPE */
5568 # ifdef USE_LOCALE_COLLATE
5570 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
5571 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5572 "%s:%d: %s\n", __FILE__, __LINE__,
5573 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
5574 new_collate(newlocale);
5575 Safefree(newlocale);
5578 # ifdef USE_LOCALE_NUMERIC
5580 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
5581 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5582 "%s:%d: %s\n", __FILE__, __LINE__,
5583 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
5584 new_numeric(newlocale);
5585 Safefree(newlocale);
5587 # endif /* USE_LOCALE_NUMERIC */
5589 return was_in_global_locale;
5595 #if defined(DEBUGGING) && defined(USE_LOCALE)
5598 S_setlocale_debug_string(const int category, /* category number,
5600 const char* const locale, /* locale name */
5602 /* return value from setlocale() when attempting to
5603 * set 'category' to 'locale' */
5604 const char* const retval)
5606 /* Returns a pointer to a NUL-terminated string in static storage with
5607 * added text about the info passed in. This is not thread safe and will
5608 * be overwritten by the next call, so this should be used just to
5609 * formulate a string to immediately print or savepv() on. */
5611 /* initialise to a non-null value to keep it out of BSS and so keep
5612 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
5613 static char ret[256] = "If you can read this, thank your buggy C"
5614 " library strlcpy(), and change your hints file"
5617 my_strlcpy(ret, "setlocale(", sizeof(ret));
5618 my_strlcat(ret, category_name(category), sizeof(ret));
5619 my_strlcat(ret, ", ", sizeof(ret));
5622 my_strlcat(ret, "\"", sizeof(ret));
5623 my_strlcat(ret, locale, sizeof(ret));
5624 my_strlcat(ret, "\"", sizeof(ret));
5627 my_strlcat(ret, "NULL", sizeof(ret));
5630 my_strlcat(ret, ") returned ", sizeof(ret));
5633 my_strlcat(ret, "\"", sizeof(ret));
5634 my_strlcat(ret, retval, sizeof(ret));
5635 my_strlcat(ret, "\"", sizeof(ret));
5638 my_strlcat(ret, "NULL", sizeof(ret));
5641 assert(strlen(ret) < sizeof(ret));
5649 Perl_thread_locale_init()
5651 /* Called from a thread on startup*/
5653 #ifdef USE_THREAD_SAFE_LOCALE
5657 /* C starts the new thread in the global C locale. If we are thread-safe,
5658 * we want to not be in the global locale */
5660 DEBUG_L(PerlIO_printf(Perl_debug_log,
5661 "%s:%d: new thread, initial locale is %s; calling setlocale\n",
5662 __FILE__, __LINE__, setlocale(LC_ALL, NULL)));
5666 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
5670 Perl_setlocale(LC_ALL, "C");
5678 Perl_thread_locale_term()
5680 /* Called from a thread as it gets ready to terminate */
5682 #ifdef USE_THREAD_SAFE_LOCALE
5684 /* C starts the new thread in the global C locale. If we are thread-safe,
5685 * we want to not be in the global locale */
5691 locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE);
5692 if (cur_obj != LC_GLOBAL_LOCALE && cur_obj != PL_C_locale_obj) {
5693 freelocale(cur_obj);
5703 * ex: set ts=8 sts=4 sw=4 et: