3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * A Elbereth Gilthoniel,
13 * silivren penna míriel
14 * o menel aglar elenath!
15 * Na-chaered palan-díriel
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
47 #define PERL_IN_LOCALE_C
48 #include "perl_langinfo.h"
57 /* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
60 #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61 # define debug_initialization 0
62 # define DEBUG_INITIALIZATION_set(v)
64 static bool debug_initialization = FALSE;
65 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
69 /* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
72 #define GET_ERRNO saved_errno
74 /* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
77 #define STRLENs(s) (sizeof("" s "") - 1)
79 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89 #define isNAME_C_OR_POSIX(name) \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
96 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
99 #define UTF8NESS_SEP "\v"
100 #define UTF8NESS_PREFIX "\f"
102 /* So, the string looks like:
104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
109 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
112 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
115 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
120 S_stdize_locale(pTHX_ char *locs)
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
136 const char * const s = strchr(locs, '=');
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
142 const char * const t = strchr(s, '.');
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
161 /* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
165 const int categories[] = {
167 # ifdef USE_LOCALE_NUMERIC
170 # ifdef USE_LOCALE_CTYPE
173 # ifdef USE_LOCALE_COLLATE
176 # ifdef USE_LOCALE_TIME
179 # ifdef USE_LOCALE_MESSAGES
182 # ifdef USE_LOCALE_MONETARY
185 # ifdef USE_LOCALE_ADDRESS
188 # ifdef USE_LOCALE_IDENTIFICATION
191 # ifdef USE_LOCALE_MEASUREMENT
194 # ifdef USE_LOCALE_PAPER
197 # ifdef USE_LOCALE_TELEPHONE
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
208 /* The top-most real element is LC_ALL */
210 const char * category_names[] = {
212 # ifdef USE_LOCALE_NUMERIC
215 # ifdef USE_LOCALE_CTYPE
218 # ifdef USE_LOCALE_COLLATE
221 # ifdef USE_LOCALE_TIME
224 # ifdef USE_LOCALE_MESSAGES
227 # ifdef USE_LOCALE_MONETARY
230 # ifdef USE_LOCALE_ADDRESS
233 # ifdef USE_LOCALE_IDENTIFICATION
236 # ifdef USE_LOCALE_MEASUREMENT
239 # ifdef USE_LOCALE_PAPER
242 # ifdef USE_LOCALE_TELEPHONE
248 NULL /* Placeholder */
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
265 /* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
274 S_category_name(const int category)
280 if (category == LC_ALL) {
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
293 const char suffix[] = " (unknown)";
295 Size_t length = sizeof(suffix) + 1;
304 /* Calculate the number of digits */
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
317 /* Now create LC_foo_INDEX #defines for just those categories on this system */
318 # ifdef USE_LOCALE_NUMERIC
319 # define LC_NUMERIC_INDEX 0
320 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
322 # define _DUMMY_NUMERIC -1
324 # ifdef USE_LOCALE_CTYPE
325 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326 # define _DUMMY_CTYPE LC_CTYPE_INDEX
328 # define _DUMMY_CTYPE _DUMMY_NUMERIC
330 # ifdef USE_LOCALE_COLLATE
331 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332 # define _DUMMY_COLLATE LC_COLLATE_INDEX
334 # define _DUMMY_COLLATE _DUMMY_COLLATE
336 # ifdef USE_LOCALE_TIME
337 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
338 # define _DUMMY_TIME LC_TIME_INDEX
340 # define _DUMMY_TIME _DUMMY_COLLATE
342 # ifdef USE_LOCALE_MESSAGES
343 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
346 # define _DUMMY_MESSAGES _DUMMY_TIME
348 # ifdef USE_LOCALE_MONETARY
349 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350 # define _DUMMY_MONETARY LC_MONETARY_INDEX
352 # define _DUMMY_MONETARY _DUMMY_MESSAGES
354 # ifdef USE_LOCALE_ADDRESS
355 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
358 # define _DUMMY_ADDRESS _DUMMY_MONETARY
360 # ifdef USE_LOCALE_IDENTIFICATION
361 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
364 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
366 # ifdef USE_LOCALE_MEASUREMENT
367 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
370 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
372 # ifdef USE_LOCALE_PAPER
373 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374 # define _DUMMY_PAPER LC_PAPER_INDEX
376 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
378 # ifdef USE_LOCALE_TELEPHONE
379 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
382 # define _DUMMY_TELEPHONE _DUMMY_PAPER
385 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
387 #endif /* ifdef USE_LOCALE */
389 /* Windows requres a customized base-level setlocale() */
391 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
393 # define my_setlocale(cat, locale) setlocale(cat, locale)
396 #ifndef USE_POSIX_2008_LOCALE
398 /* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
403 #else /* Below uses POSIX 2008 */
405 /* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411 # define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
415 /* A third array, parallel to the ones above to map from category to its
417 const int category_masks[] = {
418 # ifdef USE_LOCALE_NUMERIC
421 # ifdef USE_LOCALE_CTYPE
424 # ifdef USE_LOCALE_COLLATE
427 # ifdef USE_LOCALE_TIME
430 # ifdef USE_LOCALE_MESSAGES
433 # ifdef USE_LOCALE_MONETARY
436 # ifdef USE_LOCALE_ADDRESS
439 # ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
442 # ifdef USE_LOCALE_MEASUREMENT
445 # ifdef USE_LOCALE_PAPER
448 # ifdef USE_LOCALE_TELEPHONE
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
460 S_emulate_setlocale(const int category,
463 const bool is_index_valid
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
554 mask = category_masks[index];
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
580 # ifdef HAS_QUERYLOCALE
582 return (char *) querylocale(mask, cur_obj);
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
589 # if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
608 # ifdef USE_C_BACKTRACE
610 dump_c_backtrace(Perl_debug_log, 20, 1);
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
626 /* Without querylocale(), we have to use our record-keeping we've
629 if (category != LC_ALL) {
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
639 return PL_curlocales[index];
641 else { /* For LC_ALL */
643 Size_t names_len = 0;
645 bool are_all_categories_the_same_locale = TRUE;
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
658 return PL_curlocales[LC_ALL_INDEX];
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
666 for (i = 0; i < LC_ALL_INDEX; i++) {
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
676 names_len += strlen(category_names[i])
678 + strlen(PL_curlocales[i])
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
728 SETERRNO(EINVAL, LIB_INVARG);
738 /* Here, we are switching locales. */
740 # ifndef HAS_QUERYLOCALE
742 if (strEQ(locale, "")) {
744 /* For non-querylocale() systems, we do the setting of "" ourselves to
745 * be sure that we really know what's going on. We follow the Linux
746 * documented behavior (but if that differs from the actual behavior,
747 * this won't work exactly as the OS implements). We go out and
748 * examine the environment based on our understanding of how the system
749 * works, and use that to figure things out */
751 const char * const lc_all = PerlEnv_getenv("LC_ALL");
753 /* Use any "LC_ALL" environment variable, as it overrides everything
755 if (lc_all && strNE(lc_all, "")) {
760 /* Otherwise, we need to dig deeper. Unless overridden, the
761 * default is the LANG environment variable; if it doesn't exist,
764 const char * default_name;
766 /* To minimize other threads messing with the environment, we copy
767 * the variable, making it a temporary. But this doesn't work upon
768 * program initialization before any scopes are created, and at
769 * this time, there's nothing else going on that would interfere.
770 * So skip the copy in that case */
771 if (PL_scopestack_ix == 0) {
772 default_name = PerlEnv_getenv("LANG");
775 default_name = savepv(PerlEnv_getenv("LANG"));
778 if (! default_name || strEQ(default_name, "")) {
781 else if (PL_scopestack_ix != 0) {
782 SAVEFREEPV(default_name);
785 if (category != LC_ALL) {
786 const char * const name = PerlEnv_getenv(category_names[index]);
788 /* Here we are setting a single category. Assume will have the
790 locale = default_name;
792 /* But then look for an overriding environment variable */
793 if (name && strNE(name, "")) {
798 bool did_override = FALSE;
801 /* Here, we are getting LC_ALL. Any categories that don't have
802 * a corresponding environment variable set should be set to
803 * LANG, or to "C" if there is no LANG. If no individual
804 * categories differ from this, we can just set LC_ALL. This
805 * is buggy on systems that have extra categories that we don't
806 * know about. If there is an environment variable that sets
807 * that category, we won't know to look for it, and so our use
808 * of LANG or "C" improperly overrides it. On the other hand,
809 * if we don't do what is done here, and there is no
810 * environment variable, the category's locale should be set to
811 * LANG or "C". So there is no good solution. khw thinks the
812 * best is to look at systems to see what categories they have,
813 * and include them, and then to assume that we know the
816 for (i = 0; i < LC_ALL_INDEX; i++) {
817 const char * const env_override
818 = savepv(PerlEnv_getenv(category_names[i]));
819 const char * this_locale = ( env_override
820 && strNE(env_override, ""))
823 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
825 Safefree(env_override);
829 if (strNE(this_locale, default_name)) {
833 Safefree(env_override);
836 /* If all the categories are the same, we can set LC_ALL to
838 if (! did_override) {
839 locale = default_name;
843 /* Here, LC_ALL is no longer valid, as some individual
844 * categories don't match it. We call ourselves
845 * recursively, as that will execute the code that
846 * generates the proper locale string for this situation.
847 * We don't do the remainder of this function, as that is
848 * to update our records, and we've just done that for the
849 * individual categories in the loop above, and doing so
850 * would cause LC_ALL to be done as well */
851 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
856 else if (strchr(locale, ';')) {
858 /* LC_ALL may actually incude a conglomeration of various categories.
859 * Without querylocale, this code uses the glibc (as of this writing)
860 * syntax for representing that, but that is not a stable API, and
861 * other platforms do it differently, so we have to handle all cases
865 const char * s = locale;
866 const char * e = locale + strlen(locale);
868 const char * category_end;
869 const char * name_start;
870 const char * name_end;
872 /* If the string that gives what to set doesn't include all categories,
873 * the omitted ones get set to "C". To get this behavior, first set
874 * all the individual categories to "C", and override the furnished
876 for (i = 0; i < LC_ALL_INDEX; i++) {
877 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
884 /* Parse through the category */
885 while (isWORDCHAR(*p)) {
892 "panic: %s: %d: Unexpected character in locale name '%02X",
893 __FILE__, __LINE__, *(p-1));
896 /* Parse through the locale name */
898 while (p < e && *p != ';') {
901 "panic: %s: %d: Unexpected character in locale name '%02X",
902 __FILE__, __LINE__, *(p-1));
908 /* Space past the semi-colon */
913 /* Find the index of the category name in our lists */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
915 char * individ_locale;
917 /* Keep going if this isn't the index. The strnNE() avoids a
918 * Perl_form(), but would fail if ever a category name could be
919 * a substring of another one, like if there were a
921 if strnNE(s, category_names[i], category_end - s) {
925 /* If this index is for the single category we're changing, we
926 * have found the locale to set it to. */
927 if (category == categories[i]) {
928 locale = Perl_form(aTHX_ "%.*s",
929 (int) (name_end - name_start),
934 assert(category == LC_ALL);
935 individ_locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start), name_start);
937 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
946 /* Here we have set all the individual categories by recursive calls.
947 * These collectively should have fixed up LC_ALL, so can just query
948 * what that now is */
949 assert(category == LC_ALL);
951 return do_setlocale_c(LC_ALL, NULL);
956 /* Here at the end of having to deal with the absence of querylocale().
957 * Some cases have already been fully handled by recursive calls to this
958 * function. But at this point, we haven't dealt with those, but are now
959 * prepared to, knowing what the locale name to set this category to is.
960 * This would have come for free if this system had had querylocale() */
962 # endif /* end of ! querylocale */
964 assert(PL_C_locale_obj);
966 /* Switching locales generally entails freeing the current one's space (at
967 * the C library's discretion). We need to stop using that locale before
968 * the switch. So switch to a known locale object that we don't otherwise
969 * mess with. This returns the locale object in effect at the time of the
971 old_obj = uselocale(PL_C_locale_obj);
975 if (DEBUG_Lv_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
985 if (DEBUG_L_TEST || debug_initialization) {
987 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
998 if (DEBUG_Lv_TEST || debug_initialization) {
999 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1004 /* If we weren't in a thread safe locale, set so that newlocale() below
1005 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1006 The latter is defensive coding, so that, even if there is some bug, we
1007 will never end up trying to modify either of these, as if passed to
1008 newlocale(), they can be. */
1009 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1010 old_obj = (locale_t) 0;
1013 /* Ready to create a new locale by modification of the exising one */
1014 new_obj = newlocale(mask, locale, old_obj);
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1027 if (! uselocale(old_obj)) {
1031 if (DEBUG_L_TEST || debug_initialization) {
1032 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1044 if (DEBUG_Lv_TEST || debug_initialization) {
1045 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p; should have freed %p\n", __FILE__, __LINE__, new_obj, old_obj);
1050 /* And switch into it */
1051 if (! uselocale(new_obj)) {
1056 if (DEBUG_L_TEST || debug_initialization) {
1057 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1062 if (! uselocale(old_obj)) {
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1073 freelocale(new_obj);
1080 if (DEBUG_Lv_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1086 /* We are done, except for updating our records (if the system doesn't keep
1087 * them) and in the case of locale "", we don't actually know what the
1088 * locale that got switched to is, as it came from the environment. So
1089 * have to find it */
1091 # ifdef HAS_QUERYLOCALE
1093 if (strEQ(locale, "")) {
1094 locale = querylocale(mask, new_obj);
1099 /* Here, 'locale' is the return value */
1101 /* Without querylocale(), we have to update our records */
1103 if (category == LC_ALL) {
1106 /* For LC_ALL, we change all individual categories to correspond */
1107 /* PL_curlocales is a parallel array, so has same
1108 * length as 'categories' */
1109 for (i = 0; i <= LC_ALL_INDEX; i++) {
1110 Safefree(PL_curlocales[i]);
1111 PL_curlocales[i] = savepv(locale);
1116 /* For a single category, if it's not the same as the one in LC_ALL, we
1119 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1120 Safefree(PL_curlocales[LC_ALL_INDEX]);
1121 PL_curlocales[LC_ALL_INDEX] = NULL;
1124 /* Then update the category's record */
1125 Safefree(PL_curlocales[index]);
1126 PL_curlocales[index] = savepv(locale);
1134 #endif /* USE_POSIX_2008_LOCALE */
1136 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1137 didn't natively support them */
1139 /* The way this would work is that we would keep a per-thread list of the
1140 * correct locale for that thread. Any operation that was locale-sensitive
1141 * would have to be changed so that it would look like this:
1144 * setlocale to the correct locale for this operation
1148 * This leaves the global locale in the most recently used operation's, but it
1149 * was locked long enough to get the result. If that result is static, it
1150 * needs to be copied before the unlock.
1152 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1153 * the setup, but are no-ops when not needed, and similarly,
1154 * END_LOCALE_DEPENDENT_OP for the tear-down
1156 * But every call to a locale-sensitive function would have to be changed, and
1157 * if a module didn't cooperate by using the mutex, things would break.
1159 * This code was abandoned before being completed or tested, and is left as-is
1162 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1163 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1166 S_locking_setlocale(pTHX_
1168 const char * locale,
1170 const bool is_index_valid
1173 /* This function kind of performs a setlocale() on just the current thread;
1174 * thus it is kind of thread-safe. It does this by keeping a thread-level
1175 * array of the current locales for each category. Every time a locale is
1176 * switched to, it does the switch globally, but updates the thread's
1177 * array. A query as to what the current locale is just returns the
1178 * appropriate element from the array, and doesn't actually call the system
1179 * setlocale(). The saving into the array is done in an uninterruptible
1180 * section of code, so is unaffected by whatever any other threads might be
1183 * All locale-sensitive operations must work by first starting a critical
1184 * section, then switching to the thread's locale as kept by this function,
1185 * and then doing the operation, then ending the critical section. Thus,
1186 * each gets done in the appropriate locale. simulating thread-safety.
1188 * This function takes the same parameters, 'category' and 'locale', that
1189 * the regular setlocale() function does, but it also takes two additional
1190 * ones. This is because as described earlier. If we know on input the
1191 * index corresponding to the category into the array where we store the
1192 * current locales, we don't have to calculate it. If the caller knows at
1193 * compile time what the index is, it it can pass it, setting
1194 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1198 /* If the input index might be incorrect, calculate the correct one */
1199 if (! is_index_valid) {
1202 if (DEBUG_Lv_TEST || debug_initialization) {
1203 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1206 for (i = 0; i <= LC_ALL_INDEX; i++) {
1207 if (category == categories[i]) {
1213 /* Here, we don't know about this category, so can't handle it.
1214 * XXX best we can do is to unsafely set this
1217 return my_setlocale(category, locale);
1221 if (DEBUG_Lv_TEST || debug_initialization) {
1222 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1226 /* For a query, just return what's in our records */
1227 if (new_locale == NULL) {
1228 return curlocales[index];
1232 /* Otherwise, we need to do the switch, and save the result, all in a
1233 * critical section */
1235 Safefree(curlocales[[index]]);
1237 /* It might be that this is called from an already-locked section of code.
1238 * We would have to detect and skip the LOCK/UNLOCK if so */
1241 curlocales[index] = savepv(my_setlocale(category, new_locale));
1243 if (strEQ(new_locale, "")) {
1247 /* The locale values come from the environment, and may not all be the
1248 * same, so for LC_ALL, we have to update all the others, while the
1249 * mutex is still locked */
1251 if (category == LC_ALL) {
1253 for (i = 0; i < LC_ALL_INDEX) {
1254 curlocales[i] = my_setlocale(categories[i], NULL);
1263 return curlocales[index];
1269 S_set_numeric_radix(pTHX_ const bool use_locale)
1271 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1272 * TRUE, use the radix character derived from the current locale */
1274 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1275 || defined(HAS_NL_LANGINFO))
1277 const char * radix = (use_locale)
1278 ? my_nl_langinfo(RADIXCHAR, FALSE)
1279 /* FALSE => already in dest locale */
1282 sv_setpv(PL_numeric_radix_sv, radix);
1284 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1285 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1286 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1287 SvCUR(PL_numeric_radix_sv))
1288 && _is_cur_LC_category_utf8(LC_NUMERIC))
1290 SvUTF8_on(PL_numeric_radix_sv);
1295 if (DEBUG_L_TEST || debug_initialization) {
1296 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1297 SvPVX(PL_numeric_radix_sv),
1298 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1302 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1307 S_new_numeric(pTHX_ const char *newnum)
1310 #ifndef USE_LOCALE_NUMERIC
1312 PERL_UNUSED_ARG(newnum);
1316 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1317 * core Perl this and that 'newnum' is the name of the new locale.
1318 * It installs this locale as the current underlying default.
1320 * The default locale and the C locale can be toggled between by use of the
1321 * set_numeric_underlying() and set_numeric_standard() functions, which
1322 * should probably not be called directly, but only via macros like
1323 * SET_NUMERIC_STANDARD() in perl.h.
1325 * The toggling is necessary mainly so that a non-dot radix decimal point
1326 * character can be output, while allowing internal calculations to use a
1329 * This sets several interpreter-level variables:
1330 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1331 * PL_numeric_underlying A boolean indicating if the toggled state is such
1332 * that the current locale is the program's underlying
1334 * PL_numeric_standard An int indicating if the toggled state is such
1335 * that the current locale is the C locale or
1336 * indistinguishable from the C locale. If non-zero, it
1337 * is in C; if > 1, it means it may not be toggled away
1339 * PL_numeric_underlying_is_standard A bool kept by this function
1340 * indicating that the underlying locale and the standard
1341 * C locale are indistinguishable for the purposes of
1342 * LC_NUMERIC. This happens when both of the above two
1343 * variables are true at the same time. (Toggling is a
1344 * no-op under these circumstances.) This variable is
1345 * used to avoid having to recalculate.
1351 Safefree(PL_numeric_name);
1352 PL_numeric_name = NULL;
1353 PL_numeric_standard = TRUE;
1354 PL_numeric_underlying = TRUE;
1355 PL_numeric_underlying_is_standard = TRUE;
1359 save_newnum = stdize_locale(savepv(newnum));
1360 PL_numeric_underlying = TRUE;
1361 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1363 #ifndef TS_W32_BROKEN_LOCALECONV
1365 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1366 * them. But on broken Windows systems calling my_nl_langinfo() for
1367 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1368 * and just always change the locale if not C nor POSIX on those systems */
1369 if (! PL_numeric_standard) {
1370 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1371 FALSE /* Don't toggle locale */ ))
1372 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1377 /* Save the new name if it isn't the same as the previous one, if any */
1378 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1379 Safefree(PL_numeric_name);
1380 PL_numeric_name = save_newnum;
1383 Safefree(save_newnum);
1386 PL_numeric_underlying_is_standard = PL_numeric_standard;
1388 # ifdef HAS_POSIX_2008_LOCALE
1390 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1392 PL_underlying_numeric_obj);
1396 if (DEBUG_L_TEST || debug_initialization) {
1397 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1400 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1401 * have to worry about the radix being a non-dot. (Core operations that
1402 * need the underlying locale change to it temporarily). */
1403 if (PL_numeric_standard) {
1404 set_numeric_radix(0);
1407 set_numeric_standard();
1410 #endif /* USE_LOCALE_NUMERIC */
1415 Perl_set_numeric_standard(pTHX)
1418 #ifdef USE_LOCALE_NUMERIC
1420 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1421 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1422 * macro avoids calling this routine if toggling isn't necessary according
1423 * to our records (which could be wrong if some XS code has changed the
1424 * locale behind our back) */
1426 do_setlocale_c(LC_NUMERIC, "C");
1427 PL_numeric_standard = TRUE;
1428 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1429 set_numeric_radix(0);
1433 if (DEBUG_L_TEST || debug_initialization) {
1434 PerlIO_printf(Perl_debug_log,
1435 "LC_NUMERIC locale now is standard C\n");
1439 #endif /* USE_LOCALE_NUMERIC */
1444 Perl_set_numeric_underlying(pTHX)
1447 #ifdef USE_LOCALE_NUMERIC
1449 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1450 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1451 * instead of calling this directly. The macro avoids calling this routine
1452 * if toggling isn't necessary according to our records (which could be
1453 * wrong if some XS code has changed the locale behind our back) */
1455 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1456 PL_numeric_standard = PL_numeric_underlying_is_standard;
1457 PL_numeric_underlying = TRUE;
1458 set_numeric_radix(! PL_numeric_standard);
1462 if (DEBUG_L_TEST || debug_initialization) {
1463 PerlIO_printf(Perl_debug_log,
1464 "LC_NUMERIC locale now is %s\n",
1469 #endif /* USE_LOCALE_NUMERIC */
1474 * Set up for a new ctype locale.
1477 S_new_ctype(pTHX_ const char *newctype)
1480 #ifndef USE_LOCALE_CTYPE
1482 PERL_ARGS_ASSERT_NEW_CTYPE;
1483 PERL_UNUSED_ARG(newctype);
1484 PERL_UNUSED_CONTEXT;
1488 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1489 * core Perl this and that 'newctype' is the name of the new locale.
1491 * This function sets up the folding arrays for all 256 bytes, assuming
1492 * that tofold() is tolc() since fold case is not a concept in POSIX,
1494 * Any code changing the locale (outside this file) should use
1495 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1496 * this function should be called directly only from this file and from
1497 * POSIX::setlocale() */
1502 /* Don't check for problems if we are suppressing the warnings */
1503 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1505 PERL_ARGS_ASSERT_NEW_CTYPE;
1507 /* We will replace any bad locale warning with 1) nothing if the new one is
1508 * ok; or 2) a new warning for the bad new locale */
1509 if (PL_warn_locale) {
1510 SvREFCNT_dec_NN(PL_warn_locale);
1511 PL_warn_locale = NULL;
1514 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1516 /* A UTF-8 locale gets standard rules. But note that code still has to
1517 * handle this specially because of the three problematic code points */
1518 if (PL_in_utf8_CTYPE_locale) {
1519 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1522 /* We don't populate the other lists if a UTF-8 locale, but do check that
1523 * everything works as expected, unless checking turned off */
1524 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1525 /* Assume enough space for every character being bad. 4 spaces each
1526 * for the 94 printable characters that are output like "'x' "; and 5
1527 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1529 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1530 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1532 unsigned int bad_count = 0; /* Count of bad characters */
1534 for (i = 0; i < 256; i++) {
1535 if (! PL_in_utf8_CTYPE_locale) {
1537 PL_fold_locale[i] = (U8) tolower(i);
1538 else if (islower(i))
1539 PL_fold_locale[i] = (U8) toupper(i);
1541 PL_fold_locale[i] = (U8) i;
1544 /* If checking for locale problems, see if the native ASCII-range
1545 * printables plus \n and \t are in their expected categories in
1546 * the new locale. If not, this could mean big trouble, upending
1547 * Perl's and most programs' assumptions, like having a
1548 * metacharacter with special meaning become a \w. Fortunately,
1549 * it's very rare to find locales that aren't supersets of ASCII
1550 * nowadays. It isn't a problem for most controls to be changed
1551 * into something else; we check only \n and \t, though perhaps \r
1552 * could be an issue as well. */
1553 if ( check_for_problems
1554 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1556 bool is_bad = FALSE;
1557 char name[4] = { '\0' };
1559 /* Convert the name into a string */
1564 else if (i == '\n') {
1565 my_strlcpy(name, "\\n", sizeof(name));
1567 else if (i == '\t') {
1568 my_strlcpy(name, "\\t", sizeof(name));
1572 my_strlcpy(name, "' '", sizeof(name));
1575 /* Check each possibe class */
1576 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1578 DEBUG_L(PerlIO_printf(Perl_debug_log,
1579 "isalnum('%s') unexpectedly is %d\n",
1580 name, cBOOL(isalnum(i))));
1582 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isalpha('%s') unexpectedly is %d\n",
1586 name, cBOOL(isalpha(i))));
1588 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isdigit('%s') unexpectedly is %d\n",
1592 name, cBOOL(isdigit(i))));
1594 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isgraph('%s') unexpectedly is %d\n",
1598 name, cBOOL(isgraph(i))));
1600 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "islower('%s') unexpectedly is %d\n",
1604 name, cBOOL(islower(i))));
1606 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "isprint('%s') unexpectedly is %d\n",
1610 name, cBOOL(isprint(i))));
1612 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "ispunct('%s') unexpectedly is %d\n",
1616 name, cBOOL(ispunct(i))));
1618 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "isspace('%s') unexpectedly is %d\n",
1622 name, cBOOL(isspace(i))));
1624 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1626 DEBUG_L(PerlIO_printf(Perl_debug_log,
1627 "isupper('%s') unexpectedly is %d\n",
1628 name, cBOOL(isupper(i))));
1630 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1632 DEBUG_L(PerlIO_printf(Perl_debug_log,
1633 "isxdigit('%s') unexpectedly is %d\n",
1634 name, cBOOL(isxdigit(i))));
1636 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1638 DEBUG_L(PerlIO_printf(Perl_debug_log,
1639 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1640 name, tolower(i), (int) toLOWER_A(i)));
1642 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1644 DEBUG_L(PerlIO_printf(Perl_debug_log,
1645 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1646 name, toupper(i), (int) toUPPER_A(i)));
1648 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1650 DEBUG_L(PerlIO_printf(Perl_debug_log,
1651 "'\\n' (=%02X) is not a control\n", (int) i));
1654 /* Add to the list; Separate multiple entries with a blank */
1657 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1659 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1667 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1668 * this locale requires more than one byte, there are going to be
1670 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1671 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1672 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1674 if ( check_for_problems && MB_CUR_MAX > 1
1675 && ! PL_in_utf8_CTYPE_locale
1677 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1678 * locale. Just assume that the implementation for them (plus
1679 * for POSIX) is correct and the > 1 value is spurious. (Since
1680 * these are specially handled to never be considered UTF-8
1681 * locales, as long as this is the only problem, everything
1682 * should work fine */
1683 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1685 multi_byte_locale = TRUE;
1690 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1691 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1692 PL_warn_locale = Perl_newSVpvf(aTHX_
1693 "Locale '%s' contains (at least) the following characters"
1694 " which have\nunexpected meanings: %s\nThe Perl program"
1695 " will use the expected meanings",
1696 newctype, bad_chars_list);
1699 PL_warn_locale = Perl_newSVpvf(aTHX_
1700 "Locale '%s' may not work well.%s%s%s\n",
1703 ? " Some characters in it are not recognized by"
1707 ? "\nThe following characters (and maybe others)"
1708 " may not have the same meaning as the Perl"
1709 " program expects:\n"
1717 # ifdef HAS_NL_LANGINFO
1719 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1720 /* parameter FALSE is a don't care here */
1721 my_nl_langinfo(CODESET, FALSE));
1725 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1727 /* If we are actually in the scope of the locale or are debugging,
1728 * output the message now. If not in that scope, we save the
1729 * message to be output at the first operation using this locale,
1730 * if that actually happens. Most programs don't use locales, so
1731 * they are immune to bad ones. */
1732 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1734 /* The '0' below suppresses a bogus gcc compiler warning */
1735 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1737 if (IN_LC(LC_CTYPE)) {
1738 SvREFCNT_dec_NN(PL_warn_locale);
1739 PL_warn_locale = NULL;
1745 #endif /* USE_LOCALE_CTYPE */
1750 Perl__warn_problematic_locale()
1753 #ifdef USE_LOCALE_CTYPE
1757 /* Internal-to-core function that outputs the message in PL_warn_locale,
1758 * and then NULLS it. Should be called only through the macro
1759 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1761 if (PL_warn_locale) {
1762 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1763 SvPVX(PL_warn_locale),
1764 0 /* dummy to avoid compiler warning */ );
1765 SvREFCNT_dec_NN(PL_warn_locale);
1766 PL_warn_locale = NULL;
1774 S_new_collate(pTHX_ const char *newcoll)
1777 #ifndef USE_LOCALE_COLLATE
1779 PERL_UNUSED_ARG(newcoll);
1780 PERL_UNUSED_CONTEXT;
1784 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1785 * core Perl this and that 'newcoll' is the name of the new locale.
1787 * The design of locale collation is that every locale change is given an
1788 * index 'PL_collation_ix'. The first time a string particpates in an
1789 * operation that requires collation while locale collation is active, it
1790 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1791 * magic includes the collation index, and the transformation of the string
1792 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1793 * instead of the string itself. If a string changes, the magic is
1794 * cleared. The next time the locale changes, the index is incremented,
1795 * and so we know during a comparison that the transformation is not
1796 * necessarily still valid, and so is recomputed. Note that if the locale
1797 * changes enough times, the index could wrap (a U32), and it is possible
1798 * that a transformation would improperly be considered valid, leading to
1799 * an unlikely bug */
1802 if (PL_collation_name) {
1804 Safefree(PL_collation_name);
1805 PL_collation_name = NULL;
1807 PL_collation_standard = TRUE;
1808 is_standard_collation:
1809 PL_collxfrm_base = 0;
1810 PL_collxfrm_mult = 2;
1811 PL_in_utf8_COLLATE_locale = FALSE;
1812 PL_strxfrm_NUL_replacement = '\0';
1813 PL_strxfrm_max_cp = 0;
1817 /* If this is not the same locale as currently, set the new one up */
1818 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1820 Safefree(PL_collation_name);
1821 PL_collation_name = stdize_locale(savepv(newcoll));
1822 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1823 if (PL_collation_standard) {
1824 goto is_standard_collation;
1827 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1828 PL_strxfrm_NUL_replacement = '\0';
1829 PL_strxfrm_max_cp = 0;
1831 /* A locale collation definition includes primary, secondary, tertiary,
1832 * etc. weights for each character. To sort, the primary weights are
1833 * used, and only if they compare equal, then the secondary weights are
1834 * used, and only if they compare equal, then the tertiary, etc.
1836 * strxfrm() works by taking the input string, say ABC, and creating an
1837 * output transformed string consisting of first the primary weights,
1838 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1839 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1840 * may not have weights at every level. In our example, let's say B
1841 * doesn't have a tertiary weight, and A doesn't have a secondary
1842 * weight. The constructed string is then going to be
1843 * A¹B¹C¹ B²C² A³C³ ....
1844 * This has the desired effect that strcmp() will look at the secondary
1845 * or tertiary weights only if the strings compare equal at all higher
1846 * priority weights. The spaces shown here, like in
1848 * are not just for readability. In the general case, these must
1849 * actually be bytes, which we will call here 'separator weights'; and
1850 * they must be smaller than any other weight value, but since these
1851 * are C strings, only the terminating one can be a NUL (some
1852 * implementations may include a non-NUL separator weight just before
1853 * the NUL). Implementations tend to reserve 01 for the separator
1854 * weights. They are needed so that a shorter string's secondary
1855 * weights won't be misconstrued as primary weights of a longer string,
1856 * etc. By making them smaller than any other weight, the shorter
1857 * string will sort first. (Actually, if all secondary weights are
1858 * smaller than all primary ones, there is no need for a separator
1859 * weight between those two levels, etc.)
1861 * The length of the transformed string is roughly a linear function of
1862 * the input string. It's not exactly linear because some characters
1863 * don't have weights at all levels. When we call strxfrm() we have to
1864 * allocate some memory to hold the transformed string. The
1865 * calculations below try to find coefficients 'm' and 'b' for this
1866 * locale so that m*x + b equals how much space we need, given the size
1867 * of the input string in 'x'. If we calculate too small, we increase
1868 * the size as needed, and call strxfrm() again, but it is better to
1869 * get it right the first time to avoid wasted expensive string
1870 * transformations. */
1873 /* We use the string below to find how long the tranformation of it
1874 * is. Almost all locales are supersets of ASCII, or at least the
1875 * ASCII letters. We use all of them, half upper half lower,
1876 * because if we used fewer, we might hit just the ones that are
1877 * outliers in a particular locale. Most of the strings being
1878 * collated will contain a preponderance of letters, and even if
1879 * they are above-ASCII, they are likely to have the same number of
1880 * weight levels as the ASCII ones. It turns out that digits tend
1881 * to have fewer levels, and some punctuation has more, but those
1882 * are relatively sparse in text, and khw believes this gives a
1883 * reasonable result, but it could be changed if experience so
1885 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1886 char * x_longer; /* Transformed 'longer' */
1887 Size_t x_len_longer; /* Length of 'x_longer' */
1889 char * x_shorter; /* We also transform a substring of 'longer' */
1890 Size_t x_len_shorter;
1892 /* _mem_collxfrm() is used get the transformation (though here we
1893 * are interested only in its length). It is used because it has
1894 * the intelligence to handle all cases, but to work, it needs some
1895 * values of 'm' and 'b' to get it started. For the purposes of
1896 * this calculation we use a very conservative estimate of 'm' and
1897 * 'b'. This assumes a weight can be multiple bytes, enough to
1898 * hold any UV on the platform, and there are 5 levels, 4 weight
1899 * bytes, and a trailing NUL. */
1900 PL_collxfrm_base = 5;
1901 PL_collxfrm_mult = 5 * sizeof(UV);
1903 /* Find out how long the transformation really is */
1904 x_longer = _mem_collxfrm(longer,
1908 /* We avoid converting to UTF-8 in the
1909 * called function by telling it the
1910 * string is in UTF-8 if the locale is a
1911 * UTF-8 one. Since the string passed
1912 * here is invariant under UTF-8, we can
1913 * claim it's UTF-8 even though it isn't.
1915 PL_in_utf8_COLLATE_locale);
1918 /* Find out how long the transformation of a substring of 'longer'
1919 * is. Together the lengths of these transformations are
1920 * sufficient to calculate 'm' and 'b'. The substring is all of
1921 * 'longer' except the first character. This minimizes the chances
1922 * of being swayed by outliers */
1923 x_shorter = _mem_collxfrm(longer + 1,
1926 PL_in_utf8_COLLATE_locale);
1927 Safefree(x_shorter);
1929 /* If the results are nonsensical for this simple test, the whole
1930 * locale definition is suspect. Mark it so that locale collation
1931 * is not active at all for it. XXX Should we warn? */
1932 if ( x_len_shorter == 0
1933 || x_len_longer == 0
1934 || x_len_shorter >= x_len_longer)
1936 PL_collxfrm_mult = 0;
1937 PL_collxfrm_base = 0;
1940 SSize_t base; /* Temporary */
1942 /* We have both: m * strlen(longer) + b = x_len_longer
1943 * m * strlen(shorter) + b = x_len_shorter;
1944 * subtracting yields:
1945 * m * (strlen(longer) - strlen(shorter))
1946 * = x_len_longer - x_len_shorter
1947 * But we have set things up so that 'shorter' is 1 byte smaller
1948 * than 'longer'. Hence:
1949 * m = x_len_longer - x_len_shorter
1951 * But if something went wrong, make sure the multiplier is at
1954 if (x_len_longer > x_len_shorter) {
1955 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1958 PL_collxfrm_mult = 1;
1963 * but in case something has gone wrong, make sure it is
1965 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1970 /* Add 1 for the trailing NUL */
1971 PL_collxfrm_base = base + 1;
1976 if (DEBUG_L_TEST || debug_initialization) {
1977 PerlIO_printf(Perl_debug_log,
1978 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1980 " collate multipler=%zu, collate base=%zu\n",
1982 PL_in_utf8_COLLATE_locale,
1983 x_len_shorter, x_len_longer,
1984 PL_collxfrm_mult, PL_collxfrm_base);
1991 #endif /* USE_LOCALE_COLLATE */
1998 S_win32_setlocale(pTHX_ int category, const char* locale)
2000 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2001 * difference between the two unless the input locale is "", which normally
2002 * means on Windows to get the machine default, which is set via the
2003 * computer's "Regional and Language Options" (or its current equivalent).
2004 * In POSIX, it instead means to find the locale from the user's
2005 * environment. This routine changes the Windows behavior to first look in
2006 * the environment, and, if anything is found, use that instead of going to
2007 * the machine default. If there is no environment override, the machine
2008 * default is used, by calling the real setlocale() with "".
2010 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2011 * use the particular category's variable if set; otherwise to use the LANG
2014 bool override_LC_ALL = FALSE;
2018 if (locale && strEQ(locale, "")) {
2022 locale = PerlEnv_getenv("LC_ALL");
2024 if (category == LC_ALL) {
2025 override_LC_ALL = TRUE;
2031 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2032 if (category == categories[i]) {
2033 locale = PerlEnv_getenv(category_names[i]);
2038 locale = PerlEnv_getenv("LANG");
2054 result = setlocale(category, locale);
2055 DEBUG_L(STMT_START {
2057 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2058 setlocale_debug_string(category, locale, result));
2062 if (! override_LC_ALL) {
2066 /* Here the input category was LC_ALL, and we have set it to what is in the
2067 * LANG variable or the system default if there is no LANG. But these have
2068 * lower priority than the other LC_foo variables, so override it for each
2069 * one that is set. (If they are set to "", it means to use the same thing
2070 * we just set LC_ALL to, so can skip) */
2072 for (i = 0; i < LC_ALL_INDEX; i++) {
2073 result = PerlEnv_getenv(category_names[i]);
2074 if (result && strNE(result, "")) {
2075 setlocale(categories[i], result);
2076 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2078 setlocale_debug_string(categories[i], result, "not captured")));
2082 result = setlocale(LC_ALL, NULL);
2083 DEBUG_L(STMT_START {
2085 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2087 setlocale_debug_string(LC_ALL, NULL, result));
2098 =head1 Locale-related functions and macros
2100 =for apidoc Perl_setlocale
2102 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2103 taking the same parameters, and returning the same information, except that it
2104 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2105 instead return C<C> if the underlying locale has a non-dot decimal point
2106 character, or a non-empty thousands separator for displaying floating point
2107 numbers. This is because perl keeps that locale category such that it has a
2108 dot and empty separator, changing the locale briefly during the operations
2109 where the underlying one is required. C<Perl_setlocale> knows about this, and
2110 compensates; regular C<setlocale> doesn't.
2112 Another reason it isn't completely a drop-in replacement is that it is
2113 declared to return S<C<const char *>>, whereas the system setlocale omits the
2114 C<const> (presumably because its API was specified long ago, and can't be
2115 updated; it is illegal to change the information C<setlocale> returns; doing
2116 so leads to segfaults.)
2118 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2119 C<setlocale> can be completely ineffective on some platforms under some
2122 C<Perl_setlocale> should not be used to change the locale except on systems
2123 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2124 the system C<setlocale()> is ineffective, returning the wrong information, and
2125 failing to actually change the locale. C<Perl_setlocale>, however works
2126 properly in all circumstances.
2128 The return points to a per-thread static buffer, which is overwritten the next
2129 time C<Perl_setlocale> is called from the same thread.
2136 Perl_setlocale(const int category, const char * locale)
2138 /* This wraps POSIX::setlocale() */
2140 const char * retval;
2141 const char * newlocale;
2143 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2146 #ifdef USE_LOCALE_NUMERIC
2148 /* A NULL locale means only query what the current one is. We have the
2149 * LC_NUMERIC name saved, because we are normally switched into the C
2150 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2151 * the correct results. All other categories don't require special
2153 if (locale == NULL) {
2154 if (category == LC_NUMERIC) {
2156 /* We don't have to copy this return value, as it is a per-thread
2157 * variable, and won't change until a future setlocale */
2158 return PL_numeric_name;
2163 else if (category == LC_ALL) {
2164 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2173 retval = save_to_buffer(do_setlocale_r(category, locale),
2174 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2177 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2179 if (locale == NULL && category == LC_ALL) {
2180 RESTORE_LC_NUMERIC();
2185 DEBUG_L(PerlIO_printf(Perl_debug_log,
2186 "%s:%d: %s\n", __FILE__, __LINE__,
2187 setlocale_debug_string(category, locale, retval)));
2195 /* If locale == NULL, we are just querying the state */
2196 if (locale == NULL) {
2200 /* Now that have switched locales, we have to update our records to
2205 #ifdef USE_LOCALE_CTYPE
2212 #ifdef USE_LOCALE_COLLATE
2215 new_collate(retval);
2219 #ifdef USE_LOCALE_NUMERIC
2222 new_numeric(retval);
2230 /* LC_ALL updates all the things we care about. The values may not
2231 * be the same as 'retval', as the locale "" may have set things
2234 # ifdef USE_LOCALE_CTYPE
2236 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2237 new_ctype(newlocale);
2238 Safefree(newlocale);
2240 # endif /* USE_LOCALE_CTYPE */
2241 # ifdef USE_LOCALE_COLLATE
2243 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2244 new_collate(newlocale);
2245 Safefree(newlocale);
2248 # ifdef USE_LOCALE_NUMERIC
2250 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2251 new_numeric(newlocale);
2252 Safefree(newlocale);
2254 # endif /* USE_LOCALE_NUMERIC */
2265 PERL_STATIC_INLINE const char *
2266 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2268 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2269 * growing it if necessary */
2273 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2279 string_size = strlen(string) + offset + 1;
2281 if (*buf_size == 0) {
2282 Newx(*buf, string_size, char);
2283 *buf_size = string_size;
2285 else if (string_size > *buf_size) {
2286 Renew(*buf, string_size, char);
2287 *buf_size = string_size;
2290 Copy(string, *buf + offset, string_size - offset, char);
2296 =for apidoc Perl_langinfo
2298 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2299 taking the same C<item> parameter values, and returning the same information.
2300 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2301 of Perl's locale handling from your code, and can be used on systems that lack
2302 a native C<nl_langinfo>.
2310 The reason it isn't quite a drop-in replacement is actually an advantage. The
2311 only difference is that it returns S<C<const char *>>, whereas plain
2312 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2313 forbidden to write into the buffer. By declaring this C<const>, the compiler
2314 enforces this restriction, so if it is violated, you know at compilation time,
2315 rather than getting segfaults at runtime.
2319 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2320 without you having to write extra code. The reason for the extra code would be
2321 because these are from the C<LC_NUMERIC> locale category, which is normally
2322 kept set by Perl so that the radix is a dot, and the separator is the empty
2323 string, no matter what the underlying locale is supposed to be, and so to get
2324 the expected results, you have to temporarily toggle into the underlying
2325 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2326 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2327 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2328 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2329 (decimal point) character to be a dot.)
2333 The system function it replaces can have its static return buffer trashed,
2334 not only by a subesequent call to that function, but by a C<freelocale>,
2335 C<setlocale>, or other locale change. The returned buffer of this function is
2336 not changed until the next call to it, so the buffer is never in a trashed
2341 Its return buffer is per-thread, so it also is never overwritten by a call to
2342 this function from another thread; unlike the function it replaces.
2346 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2347 as Windows, hence makes your code more portable. Of the fifty-some possible
2348 items specified by the POSIX 2008 standard,
2349 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2350 only one is completely unimplemented, though on non-Windows platforms, another
2351 significant one is also not implemented). It uses various techniques to
2352 recover the other items, including calling C<L<localeconv(3)>>, and
2353 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2354 available. Later C<strftime()> versions have additional capabilities; C<""> is
2355 returned for those not available on your system.
2357 It is important to note that when called with an item that is recovered by
2358 using C<localeconv>, the buffer from any previous explicit call to
2359 C<localeconv> will be overwritten. This means you must save that buffer's
2360 contents if you need to access them after a call to this function. (But note
2361 that you might not want to be using C<localeconv()> directly anyway, because of
2362 issues like the ones listed in the second item of this list (above) for
2363 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2364 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2367 The details for those items which may deviate from what this emulation returns
2368 and what a native C<nl_langinfo()> would return are specified in
2373 When using C<Perl_langinfo> on systems that don't have a native
2374 C<nl_langinfo()>, you must
2376 #include "perl_langinfo.h"
2378 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2379 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2380 C<langinfo.h> would try to import into the namespace for code that doesn't need
2383 The original impetus for C<Perl_langinfo()> was so that code that needs to
2384 find out the current currency symbol, floating point radix character, or digit
2385 grouping separator can use, on all systems, the simpler and more
2386 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2387 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2388 is better to use the methods given in L<perlcall> to call
2389 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2396 #ifdef HAS_NL_LANGINFO
2397 Perl_langinfo(const nl_item item)
2399 Perl_langinfo(const int item)
2402 return my_nl_langinfo(item, TRUE);
2406 #ifdef HAS_NL_LANGINFO
2407 S_my_nl_langinfo(const nl_item item, bool toggle)
2409 S_my_nl_langinfo(const int item, bool toggle)
2413 const char * retval;
2415 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2416 * two items, and only if not already there */
2417 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2418 || PL_numeric_underlying))
2423 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2424 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2425 || ! defined(HAS_POSIX_2008_LOCALE) \
2426 || ! defined(DUPLOCALE)
2428 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2429 * for those items dependent on it. This must be copied to a buffer before
2430 * switching back, as some systems destroy the buffer when setlocale() is
2434 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2437 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2440 LOCALE_LOCK; /* Prevent interference from another thread executing
2441 this code section (the only call to nl_langinfo in
2445 /* Copy to a per-thread buffer, which is also one that won't be
2446 * destroyed by a subsequent setlocale(), such as the
2447 * RESTORE_LC_NUMERIC may do just below. */
2448 retval = save_to_buffer(nl_langinfo(item),
2449 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2454 RESTORE_LC_NUMERIC();
2458 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2461 bool do_free = FALSE;
2462 locale_t cur = uselocale((locale_t) 0);
2464 if (cur == LC_GLOBAL_LOCALE) {
2465 cur = duplocale(LC_GLOBAL_LOCALE);
2470 if (PL_underlying_numeric_obj) {
2471 cur = PL_underlying_numeric_obj;
2474 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2479 /* We have to save it to a buffer, because the freelocale() just below
2480 * can invalidate the internal one */
2481 retval = save_to_buffer(nl_langinfo_l(item, cur),
2482 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2491 if (strEQ(retval, "")) {
2492 if (item == YESSTR) {
2495 if (item == NOSTR) {
2502 #else /* Below, emulate nl_langinfo as best we can */
2506 # ifdef HAS_LOCALECONV
2508 const struct lconv* lc;
2510 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2512 # ifdef TS_W32_BROKEN_LOCALECONV
2514 const char * save_global;
2515 const char * save_thread;
2523 # ifdef HAS_STRFTIME
2526 bool return_format = FALSE; /* Return the %format, not the value */
2527 const char * format;
2531 /* We copy the results to a per-thread buffer, even if not
2532 * multi-threaded. This is in part to simplify this code, and partly
2533 * because we need a buffer anyway for strftime(), and partly because a
2534 * call of localeconv() could otherwise wipe out the buffer, and the
2535 * programmer would not be expecting this, as this is a nl_langinfo()
2536 * substitute after all, so s/he might be thinking their localeconv()
2537 * is safe until another localeconv() call. */
2542 /* This is unimplemented */
2543 case ERA: /* For use with strftime() %E modifier */
2548 /* We use only an English set, since we don't know any more */
2549 case YESEXPR: return "^[+1yY]";
2550 case YESSTR: return "yes";
2551 case NOEXPR: return "^[-0nN]";
2552 case NOSTR: return "no";
2558 /* On non-windows, this is unimplemented, in part because of
2559 * inconsistencies between vendors. The Darwin native
2560 * nl_langinfo() implementation simply looks at everything past
2561 * any dot in the name, but that doesn't work for other
2562 * vendors. Many Linux locales that don't have UTF-8 in their
2563 * names really are UTF-8, for example; z/OS locales that do
2564 * have UTF-8 in their names, aren't really UTF-8 */
2569 { /* But on Windows, the name does seem to be consistent, so
2574 const char * name = my_setlocale(LC_CTYPE, NULL);
2576 if (isNAME_C_OR_POSIX(name)) {
2577 return "ANSI_X3.4-1968";
2580 /* Find the dot in the locale name */
2581 first = (const char *) strchr(name, '.');
2587 /* Look at everything past the dot */
2592 if (! isDIGIT(*p)) {
2599 /* Here everything past the dot is a digit. Treat it as a
2601 retval = save_to_buffer("CP", &PL_langinfo_buf,
2602 &PL_langinfo_bufsize, 0);
2603 offset = STRLENs("CP");
2607 retval = save_to_buffer(first, &PL_langinfo_buf,
2608 &PL_langinfo_bufsize, offset);
2614 # ifdef HAS_LOCALECONV
2618 /* We don't bother with localeconv_l() because any system that
2619 * has it is likely to also have nl_langinfo() */
2621 LOCALE_LOCK_V; /* Prevent interference with other threads
2622 using localeconv() */
2624 # ifdef TS_W32_BROKEN_LOCALECONV
2626 /* This is a workaround for a Windows bug prior to VS 15.
2627 * What we do here is, while locked, switch to the global
2628 * locale so localeconv() works; then switch back just before
2629 * the unlock. This can screw things up if some thread is
2630 * already using the global locale while assuming no other is.
2631 * A different workaround would be to call GetCurrencyFormat on
2632 * a known value, and parse it; patches welcome
2634 * We have to use LC_ALL instead of LC_MONETARY because of
2635 * another bug in Windows */
2637 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2638 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2639 save_global= savepv(my_setlocale(LC_ALL, NULL));
2640 my_setlocale(LC_ALL, save_thread);
2646 || ! lc->currency_symbol
2647 || strEQ("", lc->currency_symbol))
2653 /* Leave the first spot empty to be filled in below */
2654 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2655 &PL_langinfo_bufsize, 1);
2656 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2657 { /* khw couldn't figure out how the localedef specifications
2658 would show that the $ should replace the radix; this is
2659 just a guess as to how it might work.*/
2660 PL_langinfo_buf[0] = '.';
2662 else if (lc->p_cs_precedes) {
2663 PL_langinfo_buf[0] = '-';
2666 PL_langinfo_buf[0] = '+';
2669 # ifdef TS_W32_BROKEN_LOCALECONV
2671 my_setlocale(LC_ALL, save_global);
2672 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2673 my_setlocale(LC_ALL, save_thread);
2674 Safefree(save_global);
2675 Safefree(save_thread);
2682 # ifdef TS_W32_BROKEN_LOCALECONV
2686 /* For this, we output a known simple floating point number to
2687 * a buffer, and parse it, looking for the radix */
2690 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2693 if (PL_langinfo_bufsize < 10) {
2694 PL_langinfo_bufsize = 10;
2695 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2698 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2700 if (needed_size >= (int) PL_langinfo_bufsize) {
2701 PL_langinfo_bufsize = needed_size + 1;
2702 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2703 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2705 assert(needed_size < (int) PL_langinfo_bufsize);
2708 ptr = PL_langinfo_buf;
2709 e = PL_langinfo_buf + PL_langinfo_bufsize;
2712 while (ptr < e && *ptr != '1') {
2719 while (ptr < e && *ptr != '5') {
2723 /* Everything in between is the radix string */
2725 PL_langinfo_buf[0] = '?';
2726 PL_langinfo_buf[1] = '\0';
2730 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2734 RESTORE_LC_NUMERIC();
2737 retval = PL_langinfo_buf;
2742 case RADIXCHAR: /* No special handling needed */
2749 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2752 LOCALE_LOCK_V; /* Prevent interference with other threads
2753 using localeconv() */
2755 # ifdef TS_W32_BROKEN_LOCALECONV
2757 /* This should only be for the thousands separator. A
2758 * different work around would be to use GetNumberFormat on a
2759 * known value and parse the result to find the separator */
2760 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2761 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2762 save_global = savepv(my_setlocale(LC_ALL, NULL));
2763 my_setlocale(LC_ALL, save_thread);
2765 /* This is the start of code that for broken Windows replaces
2766 * the above and below code, and instead calls
2767 * GetNumberFormat() and then would parse that to find the
2768 * thousands separator. It needs to handle UTF-16 vs -8
2771 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2772 DEBUG_L(PerlIO_printf(Perl_debug_log,
2773 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2774 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2784 temp = (item == RADIXCHAR)
2786 : lc->thousands_sep;
2792 retval = save_to_buffer(temp, &PL_langinfo_buf,
2793 &PL_langinfo_bufsize, 0);
2795 # ifdef TS_W32_BROKEN_LOCALECONV
2797 my_setlocale(LC_ALL, save_global);
2798 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2799 my_setlocale(LC_ALL, save_thread);
2800 Safefree(save_global);
2801 Safefree(save_thread);
2808 RESTORE_LC_NUMERIC();
2814 # ifdef HAS_STRFTIME
2816 /* These are defined by C89, so we assume that strftime supports
2817 * them, and so are returned unconditionally; they may not be what
2818 * the locale actually says, but should give good enough results
2819 * for someone using them as formats (as opposed to trying to parse
2820 * them to figure out what the locale says). The other format
2821 * items are actually tested to verify they work on the platform */
2822 case D_FMT: return "%x";
2823 case T_FMT: return "%X";
2824 case D_T_FMT: return "%c";
2826 /* These formats are only available in later strfmtime's */
2827 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2829 /* The rest can be gotten from most versions of strftime(). */
2830 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2831 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2833 case AM_STR: case PM_STR:
2834 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2835 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2836 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2837 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2838 case DAY_5: case DAY_6: case DAY_7:
2839 case MON_1: case MON_2: case MON_3: case MON_4:
2840 case MON_5: case MON_6: case MON_7: case MON_8:
2841 case MON_9: case MON_10: case MON_11: case MON_12:
2845 init_tm(&tm); /* Precaution against core dumps */
2849 tm.tm_year = 2017 - 1900;
2856 "panic: %s: %d: switch case: %d problem",
2857 __FILE__, __LINE__, item);
2858 NOT_REACHED; /* NOTREACHED */
2860 case PM_STR: tm.tm_hour = 18;
2865 case ABDAY_7: tm.tm_wday++;
2866 case ABDAY_6: tm.tm_wday++;
2867 case ABDAY_5: tm.tm_wday++;
2868 case ABDAY_4: tm.tm_wday++;
2869 case ABDAY_3: tm.tm_wday++;
2870 case ABDAY_2: tm.tm_wday++;
2875 case DAY_7: tm.tm_wday++;
2876 case DAY_6: tm.tm_wday++;
2877 case DAY_5: tm.tm_wday++;
2878 case DAY_4: tm.tm_wday++;
2879 case DAY_3: tm.tm_wday++;
2880 case DAY_2: tm.tm_wday++;
2885 case ABMON_12: tm.tm_mon++;
2886 case ABMON_11: tm.tm_mon++;
2887 case ABMON_10: tm.tm_mon++;
2888 case ABMON_9: tm.tm_mon++;
2889 case ABMON_8: tm.tm_mon++;
2890 case ABMON_7: tm.tm_mon++;
2891 case ABMON_6: tm.tm_mon++;
2892 case ABMON_5: tm.tm_mon++;
2893 case ABMON_4: tm.tm_mon++;
2894 case ABMON_3: tm.tm_mon++;
2895 case ABMON_2: tm.tm_mon++;
2900 case MON_12: tm.tm_mon++;
2901 case MON_11: tm.tm_mon++;
2902 case MON_10: tm.tm_mon++;
2903 case MON_9: tm.tm_mon++;
2904 case MON_8: tm.tm_mon++;
2905 case MON_7: tm.tm_mon++;
2906 case MON_6: tm.tm_mon++;
2907 case MON_5: tm.tm_mon++;
2908 case MON_4: tm.tm_mon++;
2909 case MON_3: tm.tm_mon++;
2910 case MON_2: tm.tm_mon++;
2917 return_format = TRUE;
2922 return_format = TRUE;
2927 return_format = TRUE;
2932 return_format = TRUE;
2937 format = "%Ow"; /* Find the alternate digit for 0 */
2941 /* We can't use my_strftime() because it doesn't look at
2943 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2946 /* A zero return means one of:
2947 * a) there wasn't enough space in PL_langinfo_buf
2948 * b) the format, like a plain %p, returns empty
2949 * c) it was an illegal format, though some
2950 * implementations of strftime will just return the
2951 * illegal format as a plain character sequence.
2953 * To quickly test for case 'b)', try again but precede
2954 * the format with a plain character. If that result is
2955 * still empty, the problem is either 'a)' or 'c)' */
2957 Size_t format_size = strlen(format) + 1;
2958 Size_t mod_size = format_size + 1;
2962 Newx(mod_format, mod_size, char);
2963 Newx(temp_result, PL_langinfo_bufsize, char);
2965 my_strlcpy(mod_format + 1, format, mod_size);
2966 len = strftime(temp_result,
2967 PL_langinfo_bufsize,
2969 Safefree(mod_format);
2970 Safefree(temp_result);
2972 /* If 'len' is non-zero, it means that we had a case like
2973 * %p which means the current locale doesn't use a.m. or
2974 * p.m., and that is valid */
2977 /* Here, still didn't work. If we get well beyond a
2978 * reasonable size, bail out to prevent an infinite
2981 if (PL_langinfo_bufsize > 100 * format_size) {
2982 *PL_langinfo_buf = '\0';
2985 /* Double the buffer size to retry; Add 1 in case
2986 * original was 0, so we aren't stuck at 0. */
2987 PL_langinfo_bufsize *= 2;
2988 PL_langinfo_bufsize++;
2989 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2997 /* Here, we got a result.
2999 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3000 * alternate format for wday 0. If the value is the same as
3001 * the normal 0, there isn't an alternate, so clear the buffer.
3003 if ( item == ALT_DIGITS
3004 && strEQ(PL_langinfo_buf, "0"))
3006 *PL_langinfo_buf = '\0';
3009 /* ALT_DIGITS is problematic. Experiments on it showed that
3010 * strftime() did not always work properly when going from
3011 * alt-9 to alt-10. Only a few locales have this item defined,
3012 * and in all of them on Linux that khw was able to find,
3013 * nl_langinfo() merely returned the alt-0 character, possibly
3014 * doubled. Most Unicode digits are in blocks of 10
3015 * consecutive code points, so that is sufficient information
3016 * for those scripts, as we can infer alt-1, alt-2, .... But
3017 * for a Japanese locale, a CJK ideographic 0 is returned, and
3018 * the CJK digits are not in code point order, so you can't
3019 * really infer anything. The localedef for this locale did
3020 * specify the succeeding digits, so that strftime() works
3021 * properly on them, without needing to infer anything. But
3022 * the nl_langinfo() return did not give sufficient information
3023 * for the caller to understand what's going on. So until
3024 * there is evidence that it should work differently, this
3025 * returns the alt-0 string for ALT_DIGITS.
3027 * wday was chosen because its range is all a single digit.
3028 * Things like tm_sec have two digits as the minimum: '00' */
3032 retval = PL_langinfo_buf;
3034 /* If to return the format, not the value, overwrite the buffer
3035 * with it. But some strftime()s will keep the original format
3036 * if illegal, so change those to "" */
3037 if (return_format) {
3038 if (strEQ(PL_langinfo_buf, format)) {
3039 *PL_langinfo_buf = '\0';
3042 retval = save_to_buffer(format, &PL_langinfo_buf,
3043 &PL_langinfo_bufsize, 0);
3061 * Initialize locale awareness.
3064 Perl_init_i18nl10n(pTHX_ int printwarn)
3068 * 0 if not to output warning when setup locale is bad
3069 * 1 if to output warning based on value of PERL_BADLANG
3070 * >1 if to output regardless of PERL_BADLANG
3073 * 1 = set ok or not applicable,
3074 * 0 = fallback to a locale of lower priority
3075 * -1 = fallback to all locales failed, not even to the C locale
3077 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3078 * set, debugging information is output.
3080 * This looks more complicated than it is, mainly due to the #ifdefs.
3082 * We try to set LC_ALL to the value determined by the environment. If
3083 * there is no LC_ALL on this platform, we try the individual categories we
3084 * know about. If this works, we are done.
3086 * But if it doesn't work, we have to do something else. We search the
3087 * environment variables ourselves instead of relying on the system to do
3088 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3089 * think there is one), and the ultimate fallback "C". This is all done in
3090 * the same loop as above to avoid duplicating code, but it makes things
3091 * more complex. The 'trial_locales' array is initialized with just one
3092 * element; it causes the behavior described in the paragraph above this to
3093 * happen. If that fails, we add elements to 'trial_locales', and do extra
3094 * loop iterations to cause the behavior described in this paragraph.
3096 * On Ultrix, the locale MUST come from the environment, so there is
3097 * preliminary code to set it. I (khw) am not sure that it is necessary,
3098 * and that this couldn't be folded into the loop, but barring any real
3099 * platforms to test on, it's staying as-is
3101 * A slight complication is that in embedded Perls, the locale may already
3102 * be set-up, and we don't want to get it from the normal environment
3103 * variables. This is handled by having a special environment variable
3104 * indicate we're in this situation. We simply set setlocale's 2nd
3105 * parameter to be a NULL instead of "". That indicates to setlocale that
3106 * it is not to change anything, but to return the current value,
3107 * effectively initializing perl's db to what the locale already is.
3109 * We play the same trick with NULL if a LC_ALL succeeds. We call
3110 * setlocale() on the individual categores with NULL to get their existing
3111 * values for our db, instead of trying to change them.
3118 PERL_UNUSED_ARG(printwarn);
3120 #else /* USE_LOCALE */
3123 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3127 /* NULL uses the existing already set up locale */
3128 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3131 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3132 unsigned int trial_locales_count;
3133 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3134 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3135 bool setlocale_failure = FALSE;
3138 /* A later getenv() could zap this, so only use here */
3139 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3141 const bool locwarn = (printwarn > 1
3143 && ( ! bad_lang_use_once
3145 /* disallow with "" or "0" */
3147 && strNE("0", bad_lang_use_once)))));
3149 /* setlocale() return vals; not copied so must be looked at immediately */
3150 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3152 /* current locale for given category; should have been copied so aren't
3154 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3158 /* In some systems you can find out the system default locale
3159 * and use that as the fallback locale. */
3160 # define SYSTEM_DEFAULT_LOCALE
3162 # ifdef SYSTEM_DEFAULT_LOCALE
3164 const char *system_default_locale = NULL;
3169 # define DEBUG_LOCALE_INIT(a,b,c)
3172 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3174 # define DEBUG_LOCALE_INIT(category, locale, result) \
3176 if (debug_initialization) { \
3177 PerlIO_printf(Perl_debug_log, \
3179 __FILE__, __LINE__, \
3180 setlocale_debug_string(category, \
3186 /* Make sure the parallel arrays are properly set up */
3187 # ifdef USE_LOCALE_NUMERIC
3188 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3189 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3190 # ifdef USE_POSIX_2008_LOCALE
3191 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3194 # ifdef USE_LOCALE_CTYPE
3195 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3196 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3197 # ifdef USE_POSIX_2008_LOCALE
3198 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3201 # ifdef USE_LOCALE_COLLATE
3202 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3203 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3204 # ifdef USE_POSIX_2008_LOCALE
3205 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3208 # ifdef USE_LOCALE_TIME
3209 assert(categories[LC_TIME_INDEX] == LC_TIME);
3210 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3211 # ifdef USE_POSIX_2008_LOCALE
3212 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3215 # ifdef USE_LOCALE_MESSAGES
3216 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3217 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3218 # ifdef USE_POSIX_2008_LOCALE
3219 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3222 # ifdef USE_LOCALE_MONETARY
3223 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3224 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3225 # ifdef USE_POSIX_2008_LOCALE
3226 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3229 # ifdef USE_LOCALE_ADDRESS
3230 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3231 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3232 # ifdef USE_POSIX_2008_LOCALE
3233 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3236 # ifdef USE_LOCALE_IDENTIFICATION
3237 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3238 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3239 # ifdef USE_POSIX_2008_LOCALE
3240 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3243 # ifdef USE_LOCALE_MEASUREMENT
3244 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3245 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3246 # ifdef USE_POSIX_2008_LOCALE
3247 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3250 # ifdef USE_LOCALE_PAPER
3251 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3252 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3253 # ifdef USE_POSIX_2008_LOCALE
3254 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3257 # ifdef USE_LOCALE_TELEPHONE
3258 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3259 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3260 # ifdef USE_POSIX_2008_LOCALE
3261 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3265 assert(categories[LC_ALL_INDEX] == LC_ALL);
3266 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3267 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3268 # ifdef USE_POSIX_2008_LOCALE
3269 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3272 # endif /* DEBUGGING */
3274 /* Initialize the cache of the program's UTF-8ness for the always known
3275 * locales C and POSIX */
3276 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3277 sizeof(PL_locale_utf8ness));
3279 # ifdef USE_THREAD_SAFE_LOCALE
3282 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3286 # ifdef USE_POSIX_2008_LOCALE
3288 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3289 if (! PL_C_locale_obj) {
3290 Perl_croak_nocontext(
3291 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3293 if (DEBUG_Lv_TEST || debug_initialization) {
3294 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3299 PL_numeric_radix_sv = newSVpvs(".");
3301 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3303 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3304 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3307 # ifdef LOCALE_ENVIRON_REQUIRED
3310 * Ultrix setlocale(..., "") fails if there are no environment
3311 * variables from which to get a locale name.
3315 # error Ultrix without LC_ALL not implemented
3321 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3322 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3323 if (sl_result[LC_ALL_INDEX])
3326 setlocale_failure = TRUE;
3328 if (! setlocale_failure) {
3329 const char * locale_param;
3330 for (i = 0; i < LC_ALL_INDEX; i++) {
3331 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3334 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3335 if (! sl_result[i]) {
3336 setlocale_failure = TRUE;
3338 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3343 # endif /* LC_ALL */
3344 # endif /* LOCALE_ENVIRON_REQUIRED */
3346 /* We try each locale in the list until we get one that works, or exhaust
3347 * the list. Normally the loop is executed just once. But if setting the
3348 * locale fails, inside the loop we add fallback trials to the array and so
3349 * will execute the loop multiple times */
3350 trial_locales[0] = setlocale_init;
3351 trial_locales_count = 1;
3353 for (i= 0; i < trial_locales_count; i++) {
3354 const char * trial_locale = trial_locales[i];
3358 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3359 * when i==0, but I (khw) don't think that behavior makes much
3361 setlocale_failure = FALSE;
3363 # ifdef SYSTEM_DEFAULT_LOCALE
3364 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3366 /* On Windows machines, an entry of "" after the 0th means to use
3367 * the system default locale, which we now proceed to get. */
3368 if (strEQ(trial_locale, "")) {
3371 /* Note that this may change the locale, but we are going to do
3372 * that anyway just below */
3373 system_default_locale = do_setlocale_c(LC_ALL, "");
3374 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3376 /* Skip if invalid or if it's already on the list of locales to
3378 if (! system_default_locale) {
3379 goto next_iteration;
3381 for (j = 0; j < trial_locales_count; j++) {
3382 if (strEQ(system_default_locale, trial_locales[j])) {
3383 goto next_iteration;
3387 trial_locale = system_default_locale;
3390 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3392 # endif /* SYSTEM_DEFAULT_LOCALE */
3398 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3399 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3400 if (! sl_result[LC_ALL_INDEX]) {
3401 setlocale_failure = TRUE;
3404 /* Since LC_ALL succeeded, it should have changed all the other
3405 * categories it can to its value; so we massage things so that the
3406 * setlocales below just return their category's current values.
3407 * This adequately handles the case in NetBSD where LC_COLLATE may
3408 * not be defined for a locale, and setting it individually will
3409 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3410 * the POSIX locale. */
3411 trial_locale = NULL;
3414 # endif /* LC_ALL */
3416 if (! setlocale_failure) {
3418 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3420 = savepv(do_setlocale_r(categories[j], trial_locale));
3421 if (! curlocales[j]) {
3422 setlocale_failure = TRUE;
3424 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3427 if (! setlocale_failure) { /* All succeeded */
3428 break; /* Exit trial_locales loop */
3432 /* Here, something failed; will need to try a fallback. */
3438 if (locwarn) { /* Output failure info only on the first one */
3442 PerlIO_printf(Perl_error_log,
3443 "perl: warning: Setting locale failed.\n");
3445 # else /* !LC_ALL */
3447 PerlIO_printf(Perl_error_log,
3448 "perl: warning: Setting locale failed for the categories:\n\t");
3450 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3451 if (! curlocales[j]) {
3452 PerlIO_printf(Perl_error_log, category_names[j]);
3455 Safefree(curlocales[j]);
3459 # endif /* LC_ALL */
3461 PerlIO_printf(Perl_error_log,
3462 "perl: warning: Please check that your locale settings:\n");
3466 PerlIO_printf(Perl_error_log,
3467 "\tLANGUAGE = %c%s%c,\n",
3468 language ? '"' : '(',
3469 language ? language : "unset",
3470 language ? '"' : ')');
3473 PerlIO_printf(Perl_error_log,
3474 "\tLC_ALL = %c%s%c,\n",
3476 lc_all ? lc_all : "unset",
3477 lc_all ? '"' : ')');
3479 # if defined(USE_ENVIRON_ARRAY)
3484 /* Look through the environment for any variables of the
3485 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3486 * already handled above. These are assumed to be locale
3487 * settings. Output them and their values. */
3488 for (e = environ; *e; e++) {
3489 const STRLEN prefix_len = sizeof("LC_") - 1;
3492 if ( strBEGINs(*e, "LC_")
3493 && ! strBEGINs(*e, "LC_ALL=")
3494 && (uppers_len = strspn(*e + prefix_len,
3495 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3496 && ((*e)[prefix_len + uppers_len] == '='))
3498 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3499 (int) (prefix_len + uppers_len), *e,
3500 *e + prefix_len + uppers_len + 1);
3507 PerlIO_printf(Perl_error_log,
3508 "\t(possibly more locale environment variables)\n");
3512 PerlIO_printf(Perl_error_log,
3513 "\tLANG = %c%s%c\n",
3515 lang ? lang : "unset",
3518 PerlIO_printf(Perl_error_log,
3519 " are supported and installed on your system.\n");
3522 /* Calculate what fallback locales to try. We have avoided this
3523 * until we have to, because failure is quite unlikely. This will
3524 * usually change the upper bound of the loop we are in.
3526 * Since the system's default way of setting the locale has not
3527 * found one that works, We use Perl's defined ordering: LC_ALL,
3528 * LANG, and the C locale. We don't try the same locale twice, so
3529 * don't add to the list if already there. (On POSIX systems, the
3530 * LC_ALL element will likely be a repeat of the 0th element "",
3531 * but there's no harm done by doing it explicitly.
3533 * Note that this tries the LC_ALL environment variable even on
3534 * systems which have no LC_ALL locale setting. This may or may
3535 * not have been originally intentional, but there's no real need
3536 * to change the behavior. */
3538 for (j = 0; j < trial_locales_count; j++) {
3539 if (strEQ(lc_all, trial_locales[j])) {
3543 trial_locales[trial_locales_count++] = lc_all;
3548 for (j = 0; j < trial_locales_count; j++) {
3549 if (strEQ(lang, trial_locales[j])) {
3553 trial_locales[trial_locales_count++] = lang;
3557 # if defined(WIN32) && defined(LC_ALL)
3559 /* For Windows, we also try the system default locale before "C".
3560 * (If there exists a Windows without LC_ALL we skip this because
3561 * it gets too complicated. For those, the "C" is the next
3562 * fallback possibility). The "" is the same as the 0th element of
3563 * the array, but the code at the loop above knows to treat it
3564 * differently when not the 0th */
3565 trial_locales[trial_locales_count++] = "";
3569 for (j = 0; j < trial_locales_count; j++) {
3570 if (strEQ("C", trial_locales[j])) {
3574 trial_locales[trial_locales_count++] = "C";
3577 } /* end of first time through the loop */
3585 } /* end of looping through the trial locales */
3587 if (ok < 1) { /* If we tried to fallback */
3589 if (! setlocale_failure) { /* fallback succeeded */
3590 msg = "Falling back to";
3592 else { /* fallback failed */
3595 /* We dropped off the end of the loop, so have to decrement i to
3596 * get back to the value the last time through */
3600 msg = "Failed to fall back to";
3602 /* To continue, we should use whatever values we've got */
3604 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3605 Safefree(curlocales[j]);
3606 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3607 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3612 const char * description;
3613 const char * name = "";
3614 if (strEQ(trial_locales[i], "C")) {
3615 description = "the standard locale";
3619 # ifdef SYSTEM_DEFAULT_LOCALE
3621 else if (strEQ(trial_locales[i], "")) {
3622 description = "the system default locale";
3623 if (system_default_locale) {
3624 name = system_default_locale;
3628 # endif /* SYSTEM_DEFAULT_LOCALE */
3631 description = "a fallback locale";
3632 name = trial_locales[i];
3634 if (name && strNE(name, "")) {
3635 PerlIO_printf(Perl_error_log,
3636 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3639 PerlIO_printf(Perl_error_log,
3640 "perl: warning: %s %s.\n", msg, description);
3643 } /* End of tried to fallback */
3645 /* Done with finding the locales; update our records */
3647 # ifdef USE_LOCALE_CTYPE
3649 new_ctype(curlocales[LC_CTYPE_INDEX]);
3652 # ifdef USE_LOCALE_COLLATE
3654 new_collate(curlocales[LC_COLLATE_INDEX]);
3657 # ifdef USE_LOCALE_NUMERIC
3659 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3663 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3665 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3667 /* This caches whether each category's locale is UTF-8 or not. This
3668 * may involve changing the locale. It is ok to do this at
3669 * initialization time before any threads have started, but not later
3670 * unless thread-safe operations are used.
3671 * Caching means that if the program heeds our dictate not to change
3672 * locales in threaded applications, this data will remain valid, and
3673 * it may get queried without having to change locales. If the
3674 * environment is such that all categories have the same locale, this
3675 * isn't needed, as the code will not change the locale; but this
3676 * handles the uncommon case where the environment has disparate
3677 * locales for the categories */
3678 (void) _is_cur_LC_category_utf8(categories[i]);
3682 Safefree(curlocales[i]);
3685 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3687 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3688 * locale is UTF-8. The call to new_ctype() just above has already
3689 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3690 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3691 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3692 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3693 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3695 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3696 This is an alternative to using the -C command line switch
3697 (the -C if present will override this). */
3699 const char *p = PerlEnv_getenv("PERL_UNICODE");
3700 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3701 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3715 #endif /* USE_LOCALE */
3718 /* So won't continue to output stuff */
3719 DEBUG_INITIALIZATION_set(FALSE);
3726 #ifdef USE_LOCALE_COLLATE
3729 Perl__mem_collxfrm(pTHX_ const char *input_string,
3730 STRLEN len, /* Length of 'input_string' */
3731 STRLEN *xlen, /* Set to length of returned string
3732 (not including the collation index
3734 bool utf8 /* Is the input in UTF-8? */
3738 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3739 * differences. First, it handles embedded NULs. Second, it allocates a bit
3740 * more memory than needed for the transformed data itself. The real
3741 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3742 * the length of that, and doesn't include the collation index size.
3743 * Please see sv_collxfrm() to see how this is used. */
3745 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3747 char * s = (char *) input_string;
3748 STRLEN s_strlen = strlen(input_string);
3750 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3751 STRLEN length_in_chars;
3752 bool first_time = TRUE; /* Cleared after first loop iteration */
3754 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3756 /* Must be NUL-terminated */
3757 assert(*(input_string + len) == '\0');
3759 /* If this locale has defective collation, skip */
3760 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3761 DEBUG_L(PerlIO_printf(Perl_debug_log,
3762 "_mem_collxfrm: locale's collation is defective\n"));
3766 /* Replace any embedded NULs with the control that sorts before any others.
3767 * This will give as good as possible results on strings that don't
3768 * otherwise contain that character, but otherwise there may be
3769 * less-than-perfect results with that character and NUL. This is
3770 * unavoidable unless we replace strxfrm with our own implementation. */
3771 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3775 STRLEN sans_nuls_len;
3776 int try_non_controls;
3777 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3778 making sure 2nd byte is NUL.
3780 STRLEN this_replacement_len;
3782 /* If we don't know what non-NUL control character sorts lowest for
3783 * this locale, find it */
3784 if (PL_strxfrm_NUL_replacement == '\0') {
3786 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3787 includes the collation index
3790 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3792 /* Unlikely, but it may be that no control will work to replace
3793 * NUL, in which case we instead look for any character. Controls
3794 * are preferred because collation order is, in general, context
3795 * sensitive, with adjoining characters affecting the order, and
3796 * controls are less likely to have such interactions, allowing the
3797 * NUL-replacement to stand on its own. (Another way to look at it
3798 * is to imagine what would happen if the NUL were replaced by a
3799 * combining character; it wouldn't work out all that well.) */
3800 for (try_non_controls = 0;
3801 try_non_controls < 2;
3804 /* Look through all legal code points (NUL isn't) */
3805 for (j = 1; j < 256; j++) {
3806 char * x; /* j's xfrm plus collation index */
3807 STRLEN x_len; /* length of 'x' */
3808 STRLEN trial_len = 1;
3809 char cur_source[] = { '\0', '\0' };
3811 /* Skip non-controls the first time through the loop. The
3812 * controls in a UTF-8 locale are the L1 ones */
3813 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3820 /* Create a 1-char string of the current code point */
3821 cur_source[0] = (char) j;
3823 /* Then transform it */
3824 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3825 0 /* The string is not in UTF-8 */);
3827 /* Ignore any character that didn't successfully transform.
3833 /* If this character's transformation is lower than
3834 * the current lowest, this one becomes the lowest */
3835 if ( cur_min_x == NULL
3836 || strLT(x + COLLXFRM_HDR_LEN,
3837 cur_min_x + COLLXFRM_HDR_LEN))
3839 PL_strxfrm_NUL_replacement = j;
3845 } /* end of loop through all 255 characters */
3847 /* Stop looking if found */
3852 /* Unlikely, but possible, if there aren't any controls that
3853 * work in the locale, repeat the loop, looking for any
3854 * character that works */
3855 DEBUG_L(PerlIO_printf(Perl_debug_log,
3856 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3857 } /* End of loop to try first the controls, then any char */
3860 DEBUG_L(PerlIO_printf(Perl_debug_log,
3861 "_mem_collxfrm: Couldn't find any character to replace"
3862 " embedded NULs in locale %s with", PL_collation_name));
3866 DEBUG_L(PerlIO_printf(Perl_debug_log,
3867 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3868 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3870 Safefree(cur_min_x);
3871 } /* End of determining the character that is to replace NULs */
3873 /* If the replacement is variant under UTF-8, it must match the
3874 * UTF8-ness of the original */
3875 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3876 this_replacement_char[0] =
3877 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3878 this_replacement_char[1] =
3879 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3880 this_replacement_len = 2;
3883 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3884 /* this_replacement_char[1] = '\0' was done at initialization */
3885 this_replacement_len = 1;
3888 /* The worst case length for the replaced string would be if every
3889 * character in it is NUL. Multiply that by the length of each
3890 * replacement, and allow for a trailing NUL */
3891 sans_nuls_len = (len * this_replacement_len) + 1;
3892 Newx(sans_nuls, sans_nuls_len, char);
3895 /* Replace each NUL with the lowest collating control. Loop until have
3896 * exhausted all the NULs */
3897 while (s + s_strlen < e) {
3898 my_strlcat(sans_nuls, s, sans_nuls_len);
3900 /* Do the actual replacement */
3901 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3903 /* Move past the input NUL */
3905 s_strlen = strlen(s);
3908 /* And add anything that trails the final NUL */
3909 my_strlcat(sans_nuls, s, sans_nuls_len);
3911 /* Switch so below we transform this modified string */
3914 } /* End of replacing NULs */
3916 /* Make sure the UTF8ness of the string and locale match */
3917 if (utf8 != PL_in_utf8_COLLATE_locale) {
3918 /* XXX convert above Unicode to 10FFFF? */
3919 const char * const t = s; /* Temporary so we can later find where the
3922 /* Here they don't match. Change the string's to be what the locale is
3925 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3926 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3929 else { /* locale is not UTF-8; but input is; downgrade the input */
3931 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3933 /* If the downgrade was successful we are done, but if the input
3934 * contains things that require UTF-8 to represent, have to do
3935 * damage control ... */
3936 if (UNLIKELY(utf8)) {
3938 /* What we do is construct a non-UTF-8 string with
3939 * 1) the characters representable by a single byte converted
3940 * to be so (if necessary);
3941 * 2) and the rest converted to collate the same as the
3942 * highest collating representable character. That makes
3943 * them collate at the end. This is similar to how we
3944 * handle embedded NULs, but we use the highest collating
3945 * code point instead of the smallest. Like the NUL case,
3946 * this isn't perfect, but is the best we can reasonably
3947 * do. Every above-255 code point will sort the same as
3948 * the highest-sorting 0-255 code point. If that code
3949 * point can combine in a sequence with some other code
3950 * points for weight calculations, us changing something to
3951 * be it can adversely affect the results. But in most
3952 * cases, it should work reasonably. And note that this is
3953 * really an illegal situation: using code points above 255
3954 * on a locale where only 0-255 are valid. If two strings
3955 * sort entirely equal, then the sort order for the
3956 * above-255 code points will be in code point order. */
3960 /* If we haven't calculated the code point with the maximum
3961 * collating order for this locale, do so now */
3962 if (! PL_strxfrm_max_cp) {
3965 /* The current transformed string that collates the
3966 * highest (except it also includes the prefixed collation
3968 char * cur_max_x = NULL;
3970 /* Look through all legal code points (NUL isn't) */
3971 for (j = 1; j < 256; j++) {
3974 char cur_source[] = { '\0', '\0' };
3976 /* Create a 1-char string of the current code point */
3977 cur_source[0] = (char) j;
3979 /* Then transform it */
3980 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3982 /* If something went wrong (which it shouldn't), just
3983 * ignore this code point */
3988 /* If this character's transformation is higher than
3989 * the current highest, this one becomes the highest */
3990 if ( cur_max_x == NULL
3991 || strGT(x + COLLXFRM_HDR_LEN,
3992 cur_max_x + COLLXFRM_HDR_LEN))
3994 PL_strxfrm_max_cp = j;
4003 DEBUG_L(PerlIO_printf(Perl_debug_log,
4004 "_mem_collxfrm: Couldn't find any character to"
4005 " replace above-Latin1 chars in locale %s with",
4006 PL_collation_name));
4010 DEBUG_L(PerlIO_printf(Perl_debug_log,
4011 "_mem_collxfrm: highest 1-byte collating character"
4012 " in locale %s is 0x%02X\n",
4014 PL_strxfrm_max_cp));
4016 Safefree(cur_max_x);
4019 /* Here we know which legal code point collates the highest.
4020 * We are ready to construct the non-UTF-8 string. The length
4021 * will be at least 1 byte smaller than the input string
4022 * (because we changed at least one 2-byte character into a
4023 * single byte), but that is eaten up by the trailing NUL */
4029 char * e = (char *) t + len;
4031 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4033 if (UTF8_IS_INVARIANT(cur_char)) {
4036 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4037 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4039 else { /* Replace illegal cp with highest collating
4041 s[d++] = PL_strxfrm_max_cp;
4045 Renew(s, d, char); /* Free up unused space */
4050 /* Here, we have constructed a modified version of the input. It could
4051 * be that we already had a modified copy before we did this version.
4052 * If so, that copy is no longer needed */
4053 if (t != input_string) {
4058 length_in_chars = (utf8)
4059 ? utf8_length((U8 *) s, (U8 *) s + len)
4062 /* The first element in the output is the collation id, used by
4063 * sv_collxfrm(); then comes the space for the transformed string. The
4064 * equation should give us a good estimate as to how much is needed */
4065 xAlloc = COLLXFRM_HDR_LEN
4067 + (PL_collxfrm_mult * length_in_chars);
4068 Newx(xbuf, xAlloc, char);
4069 if (UNLIKELY(! xbuf)) {
4070 DEBUG_L(PerlIO_printf(Perl_debug_log,
4071 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4075 /* Store the collation id */
4076 *(U32*)xbuf = PL_collation_ix;
4078 /* Then the transformation of the input. We loop until successful, or we
4082 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4084 /* If the transformed string occupies less space than we told strxfrm()
4085 * was available, it means it successfully transformed the whole
4087 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4089 /* Some systems include a trailing NUL in the returned length.
4090 * Ignore it, using a loop in case multiple trailing NULs are
4093 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4098 /* If the first try didn't get it, it means our prediction was low.
4099 * Modify the coefficients so that we predict a larger value in any
4100 * future transformations */
4102 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4103 STRLEN computed_guess = PL_collxfrm_base
4104 + (PL_collxfrm_mult * length_in_chars);
4106 /* On zero-length input, just keep current slope instead of
4108 const STRLEN new_m = (length_in_chars != 0)
4109 ? needed / length_in_chars
4112 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4113 "%s: %d: initial size of %zu bytes for a length "
4114 "%zu string was insufficient, %zu needed\n",
4116 computed_guess, length_in_chars, needed));
4118 /* If slope increased, use it, but discard this result for
4119 * length 1 strings, as we can't be sure that it's a real slope
4121 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4125 STRLEN old_m = PL_collxfrm_mult;
4126 STRLEN old_b = PL_collxfrm_base;
4130 PL_collxfrm_mult = new_m;
4131 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4132 computed_guess = PL_collxfrm_base
4133 + (PL_collxfrm_mult * length_in_chars);
4134 if (computed_guess < needed) {
4135 PL_collxfrm_base += needed - computed_guess;
4138 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4139 "%s: %d: slope is now %zu; was %zu, base "
4140 "is now %zu; was %zu\n",
4142 PL_collxfrm_mult, old_m,
4143 PL_collxfrm_base, old_b));
4145 else { /* Slope didn't change, but 'b' did */
4146 const STRLEN new_b = needed
4149 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4150 "%s: %d: base is now %zu; was %zu\n",
4152 new_b, PL_collxfrm_base));
4153 PL_collxfrm_base = new_b;
4160 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4161 DEBUG_L(PerlIO_printf(Perl_debug_log,
4162 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4163 *xlen, PERL_INT_MAX));
4167 /* A well-behaved strxfrm() returns exactly how much space it needs
4168 * (usually not including the trailing NUL) when it fails due to not
4169 * enough space being provided. Assume that this is the case unless
4170 * it's been proven otherwise */
4171 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4172 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4174 else { /* Here, either:
4175 * 1) The strxfrm() has previously shown bad behavior; or
4176 * 2) It isn't the first time through the loop, which means
4177 * that the strxfrm() is now showing bad behavior, because
4178 * we gave it what it said was needed in the previous
4179 * iteration, and it came back saying it needed still more.
4180 * (Many versions of cygwin fit this. When the buffer size
4181 * isn't sufficient, they return the input size instead of
4182 * how much is needed.)
4183 * Increase the buffer size by a fixed percentage and try again.
4185 xAlloc += (xAlloc / 4) + 1;
4186 PL_strxfrm_is_behaved = FALSE;
4190 if (DEBUG_Lv_TEST || debug_initialization) {
4191 PerlIO_printf(Perl_debug_log,
4192 "_mem_collxfrm required more space than previously calculated"
4193 " for locale %s, trying again with new guess=%d+%zu\n",
4194 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4195 xAlloc - COLLXFRM_HDR_LEN);
4202 Renew(xbuf, xAlloc, char);
4203 if (UNLIKELY(! xbuf)) {
4204 DEBUG_L(PerlIO_printf(Perl_debug_log,
4205 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4215 if (DEBUG_Lv_TEST || debug_initialization) {
4217 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4218 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4219 PerlIO_printf(Perl_debug_log, "%s\n",
4220 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4226 /* Free up unneeded space; retain ehough for trailing NUL */
4227 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4229 if (s != input_string) {
4237 if (s != input_string) {
4244 if (DEBUG_Lv_TEST || debug_initialization) {
4245 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4256 S_print_collxfrm_input_and_return(pTHX_
4257 const char * const s,
4258 const char * const e,
4259 const STRLEN * const xlen,
4263 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4265 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4266 (UV)PL_collation_ix);
4268 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4271 PerlIO_printf(Perl_debug_log, "NULL");
4273 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4275 print_bytes_for_locale(s, e, is_utf8);
4277 PerlIO_printf(Perl_debug_log, "'\n");
4281 S_print_bytes_for_locale(pTHX_
4282 const char * const s,
4283 const char * const e,
4287 bool prev_was_printable = TRUE;
4288 bool first_time = TRUE;
4290 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4294 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4297 if (! prev_was_printable) {
4298 PerlIO_printf(Perl_debug_log, " ");
4300 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4301 prev_was_printable = TRUE;
4305 PerlIO_printf(Perl_debug_log, " ");
4307 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4308 prev_was_printable = FALSE;
4310 t += (is_utf8) ? UTF8SKIP(t) : 1;
4315 # endif /* #ifdef DEBUGGING */
4316 #endif /* USE_LOCALE_COLLATE */
4321 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4323 /* Changes the locale for LC_'switch_category" to that of
4324 * LC_'template_category', if they aren't already the same. If not NULL,
4325 * 'template_locale' is the locale that 'template_category' is in.
4327 * Returns a copy of the name of the original locale for 'switch_category'
4328 * so can be switched back to with the companion function
4329 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4331 char * restore_to_locale = NULL;
4333 if (switch_category == template_category) { /* No changes needed */
4337 /* Find the original locale of the category we may need to change, so that
4338 * it can be restored to later */
4339 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4341 if (! restore_to_locale) {
4343 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4344 __FILE__, __LINE__, category_name(switch_category), errno);
4347 /* If the locale of the template category wasn't passed in, find it now */
4348 if (template_locale == NULL) {
4349 template_locale = do_setlocale_r(template_category, NULL);
4350 if (! template_locale) {
4352 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4353 __FILE__, __LINE__, category_name(template_category), errno);
4357 /* It the locales are the same, there's nothing to do */
4358 if (strEQ(restore_to_locale, template_locale)) {
4359 Safefree(restore_to_locale);
4361 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4362 category_name(switch_category), restore_to_locale));
4367 /* Finally, change the locale to the template one */
4368 if (! do_setlocale_r(switch_category, template_locale)) {
4370 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4371 __FILE__, __LINE__, category_name(switch_category),
4372 template_locale, errno);
4375 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4376 category_name(switch_category), template_locale));
4378 return restore_to_locale;
4382 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4384 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4385 * copy that will be freed by this function), or do nothing if the latter
4386 * parameter is NULL */
4388 if (original_locale == NULL) {
4392 if (! do_setlocale_r(category, original_locale)) {
4394 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4396 category_name(category), original_locale, errno);
4399 Safefree(original_locale);
4403 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4405 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4406 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4407 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4408 * could give the wrong result. The result will very likely be correct for
4409 * languages that have commonly used non-ASCII characters, but for notably
4410 * English, it comes down to if the locale's name ends in something like
4411 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4413 * If the platform is early C89, not containing mbtowc(), or we are
4414 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4415 * These work very well for non-Latin locales or those whose currency
4416 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4417 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4418 * on the name containing UTF-8 or not. */
4420 /* Name of current locale corresponding to the input category */
4421 const char *save_input_locale = NULL;
4423 bool is_utf8 = FALSE; /* The return value */
4425 /* The variables below are for the cache of previous lookups using this
4426 * function. The cache is a C string, described at the definition for
4427 * 'C_and_POSIX_utf8ness'.
4429 * The first part of the cache is fixed, for the C and POSIX locales. The
4430 * varying part starts just after them. */
4431 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4433 Size_t utf8ness_cache_size; /* Size of the varying portion */
4434 Size_t input_name_len; /* Length in bytes of save_input_locale */
4435 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4436 the name in the cache */
4437 char * delimited; /* The name plus the delimiters used to store
4439 char * name_pos; /* position of 'delimited' in the cache, or 0
4445 assert(category != LC_ALL);
4449 /* Get the desired category's locale */
4450 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4451 if (! save_input_locale) {
4453 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4454 __FILE__, __LINE__, category_name(category), errno);
4457 DEBUG_L(PerlIO_printf(Perl_debug_log,
4458 "Current locale for %s is %s\n",
4459 category_name(category), save_input_locale));
4461 input_name_len = strlen(save_input_locale);
4463 /* In our cache, each name is accompanied by two delimiters and a single
4465 input_name_len_with_overhead = input_name_len + 3;
4467 /* Allocate and populate space for a copy of the name surrounded by the
4469 Newx(delimited, input_name_len_with_overhead, char);
4470 delimited[0] = UTF8NESS_SEP[0];
4471 Copy(save_input_locale, delimited + 1, input_name_len, char);
4472 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4473 delimited[input_name_len+2] = '\0';
4475 /* And see if that is in the cache */
4476 name_pos = instr(PL_locale_utf8ness, delimited);
4478 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4482 if (DEBUG_Lv_TEST || debug_initialization) {
4483 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4484 save_input_locale, is_utf8);
4489 /* And, if not already in that position, move it to the beginning of
4490 * the non-constant portion of the list, since it is the most recently
4491 * used. (We don't have to worry about overflow, since just moving
4492 * existing names around) */
4493 if (name_pos > utf8ness_cache) {
4494 Move(utf8ness_cache,
4495 utf8ness_cache + input_name_len_with_overhead,
4496 name_pos - utf8ness_cache, char);
4499 input_name_len_with_overhead - 1, char);
4500 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4503 Safefree(delimited);
4504 Safefree(save_input_locale);
4508 /* Here we don't have stored the utf8ness for the input locale. We have to
4511 # if defined(USE_LOCALE_CTYPE) \
4512 && ( defined(HAS_NL_LANGINFO) \
4513 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4516 const char *original_ctype_locale
4517 = switch_category_locale_to_template(LC_CTYPE,
4521 /* Here the current LC_CTYPE is set to the locale of the category whose
4522 * information is desired. This means that nl_langinfo() and mbtowc()
4523 * should give the correct results */
4525 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4526 calling the functions if we have this */
4528 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4529 * Unicode code point. */
4531 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4532 __FILE__, __LINE__, (int) MB_CUR_MAX));
4533 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
4535 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4536 goto finish_and_return;
4540 # if defined(HAS_NL_LANGINFO)
4542 { /* The task is easiest if the platform has this POSIX 2001 function.
4543 Except on some platforms it can wrongly return "", so have to have
4544 a fallback. And it can return that it's UTF-8, even if there are
4545 variances from that. For example, Turkish locales may use the
4546 alternate dotted I rules, and sometimes it appears to be a
4547 defective locale definition. XXX We should probably check for
4548 these in the Latin1 range and warn (but on glibc, requires
4549 iswalnum() etc. due to their not handling 80-FF correctly */
4550 const char *codeset = my_nl_langinfo(CODESET, FALSE);
4551 /* FALSE => already in dest locale */
4553 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4554 "\tnllanginfo returned CODESET '%s'\n", codeset));
4556 if (codeset && strNE(codeset, "")) {
4558 /* If the implementation of foldEQ() somehow were
4559 * to change to not go byte-by-byte, this could
4560 * read past end of string, as only one length is
4561 * checked. But currently, a premature NUL will
4562 * compare false, and it will stop there */
4563 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
4564 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
4566 DEBUG_L(PerlIO_printf(Perl_debug_log,
4567 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
4569 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4570 goto finish_and_return;
4575 # if defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)
4576 /* We can see if this is a UTF-8-like locale if have mbtowc(). It was a
4577 * late adder to C89, so very likely to have it. However, testing has
4578 * shown that, like nl_langinfo() above, there are locales that are not
4579 * strictly UTF-8 that this will return that they are */
4586 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4592 /* mbrtowc() and mbtowc() convert a byte string to a wide
4593 * character. Feed a byte string to one of them and check that the
4594 * result is the expected Unicode code point */
4596 # if defined(HAS_MBRTOWC) && defined(USE_ITHREADS)
4597 /* Prefer this function if available, as it's reentrant */
4599 memset(&ps, 0, sizeof(ps));;
4600 PERL_UNUSED_RESULT(mbrtowc(&wc, NULL, 0, &ps)); /* Reset any shift
4603 len = mbrtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8), &ps);
4609 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
4611 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
4618 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4619 "\treturn from mbtowc; len=%d; code_point=%x; errno=%d\n",
4620 len, (unsigned int) wc, GET_ERRNO));
4622 is_utf8 = cBOOL( len == STRLENs(REPLACEMENT_CHARACTER_UTF8)
4623 && wc == (wchar_t) UNICODE_REPLACEMENT);
4626 restore_switched_locale(LC_CTYPE, original_ctype_locale);
4627 goto finish_and_return;
4633 /* Here, we must have a C89 compiler that doesn't have mbtowc(). Next
4634 * try looking at the currency symbol to see if it disambiguates
4635 * things. Often that will be in the native script, and if the symbol
4636 * isn't in UTF-8, we know that the locale isn't. If it is non-ASCII
4637 * UTF-8, we infer that the locale is too, as the odds of a non-UTF8
4638 * string being valid UTF-8 are quite small */
4640 # ifdef USE_LOCALE_MONETARY
4642 /* If have LC_MONETARY, we can look at the currency symbol. Often that
4643 * will be in the native script. We do this one first because there is
4644 * just one string to examine, so potentially avoids work */
4647 const char *original_monetary_locale
4648 = switch_category_locale_to_template(LC_MONETARY,
4651 bool only_ascii = FALSE;
4652 const U8 * currency_string
4653 = (const U8 *) my_nl_langinfo(CRNCYSTR, FALSE);
4654 /* 2nd param not relevant for this item */
4655 const U8 * first_variant;
4657 assert( *currency_string == '-'
4658 || *currency_string == '+'
4659 || *currency_string == '.');
4663 if (is_utf8_invariant_string_loc(currency_string, 0, &first_variant))
4665 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));
4669 is_utf8 = is_strict_utf8_string(first_variant, 0);
4672 restore_switched_locale(LC_MONETARY, original_monetary_locale);
4676 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
4677 * otherwise assume the locale is UTF-8 if and only if the symbol
4678 * is non-ascii UTF-8. */
4679 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
4680 save_input_locale, is_utf8));
4681 goto finish_and_return;
4685 # endif /* USE_LOCALE_MONETARY */
4686 # if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
4688 /* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
4689 * the names of the months and weekdays, timezone, and am/pm indicator */
4691 const char *original_time_locale
4692 = switch_category_locale_to_template(LC_TIME,
4696 bool is_dst = FALSE;
4700 char * formatted_time;
4702 /* Here the current LC_TIME is set to the locale of the category
4703 * whose information is desired. Look at all the days of the week and
4704 * month names, and the timezone and am/pm indicator for UTF-8 variant
4705 * characters. The first such a one found will tell us if the locale
4706 * is UTF-8 or not */
4708 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
4709 formatted_time = my_strftime("%A %B %Z %p",
4710 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
4711 if ( ! formatted_time
4712 || is_utf8_invariant_string((U8 *) formatted_time, 0))
4715 /* Here, we didn't find a non-ASCII. Try the next time through
4716 * with the complemented dst and am/pm, and try with the next
4717 * weekday. After we have gotten all weekdays, try the next
4720 hour = (hour + 12) % 24;
4728 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
4729 * false otherwise. But first, restore LC_TIME to its original
4730 * locale if we changed it */
4731 restore_switched_locale(LC_TIME, original_time_locale);
4733 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
4735 is_utf8_string((U8 *) formatted_time, 0)));
4736 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
4737 goto finish_and_return;
4740 /* Falling off the end of the loop indicates all the names were just
4741 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
4742 * to its original locale */
4743 restore_switched_locale(LC_TIME, original_time_locale);
4744 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));
4749 # if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
4751 /* This code is ifdefd out because it was found to not be necessary in testing
4752 * on our dromedary test machine, which has over 700 locales. There, this
4753 * added no value to looking at the currency symbol and the time strings. I
4754 * left it in so as to avoid rewriting it if real-world experience indicates
4755 * that dromedary is an outlier. Essentially, instead of returning abpve if we
4756 * haven't found illegal utf8, we continue on and examine all the strerror()
4757 * messages on the platform for utf8ness. If all are ASCII, we still don't
4758 * know the answer; but otherwise we have a pretty good indication of the
4759 * utf8ness. The reason this doesn't help much is that the messages may not
4760 * have been translated into the locale. The currency symbol and time strings
4761 * are much more likely to have been translated. */
4764 bool non_ascii = FALSE;
4765 const char *original_messages_locale
4766 = switch_category_locale_to_template(LC_MESSAGES,
4769 const char * errmsg = NULL;
4771 /* Here the current LC_MESSAGES is set to the locale of the category
4772 * whose information is desired. Look through all the messages. We
4773 * can't use Strerror() here because it may expand to code that
4774 * segfaults in miniperl */
4776 for (e = 0; e <= sys_nerr; e++) {
4778 errmsg = sys_errlist[e];
4779 if (errno || !errmsg) {
4782 errmsg = savepv(errmsg);
4783 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
4785 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
4791 restore_switched_locale(LC_MESSAGES, original_messages_locale);
4795 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
4796 * any non-ascii means it is one; otherwise we assume it isn't */
4797 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
4800 goto finish_and_return;
4803 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));
4807 # ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
4810 /* As a last resort, look at the locale name to see if it matches
4811 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
4812 * return of setlocale(), is actually defined to be opaque, so we can't
4813 * really rely on the absence of various substrings in the name to indicate
4814 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
4815 * be a UTF-8 locale. Similarly for the other common names */
4818 const Size_t final_pos = strlen(save_input_locale) - 1;
4820 if (final_pos >= 3) {
4821 const char *name = save_input_locale;
4823 /* Find next 'U' or 'u' and look from there */
4824 while ((name += strcspn(name, "Uu") + 1)
4825 <= save_input_locale + final_pos - 2)
4827 if ( isALPHA_FOLD_NE(*name, 't')
4828 || isALPHA_FOLD_NE(*(name + 1), 'f'))
4833 if (*(name) == '-') {
4834 if ((name > save_input_locale + final_pos - 1)) {
4839 if (*(name) == '8') {
4840 DEBUG_L(PerlIO_printf(Perl_debug_log,
4841 "Locale %s ends with UTF-8 in name\n",
4842 save_input_locale));
4844 goto finish_and_return;
4847 DEBUG_L(PerlIO_printf(Perl_debug_log,
4848 "Locale %s doesn't end with UTF-8 in name\n",
4849 save_input_locale));
4854 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
4855 if (memENDs(save_input_locale, final_pos, "65001")) {
4856 DEBUG_L(PerlIO_printf(Perl_debug_log,
4857 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
4858 save_input_locale));
4860 goto finish_and_return;
4867 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
4868 * since we are about to return FALSE anyway, there is no point in doing
4869 * this extra work */
4872 if (instr(save_input_locale, "8859")) {
4873 DEBUG_L(PerlIO_printf(Perl_debug_log,
4874 "Locale %s has 8859 in name, not UTF-8 locale\n",
4875 save_input_locale));
4877 goto finish_and_return;
4881 DEBUG_L(PerlIO_printf(Perl_debug_log,
4882 "Assuming locale %s is not a UTF-8 locale\n",
4883 save_input_locale));
4886 # endif /* the code that is compiled when no modern LC_CTYPE */
4890 /* Cache this result so we don't have to go through all this next time. */
4891 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
4892 - (utf8ness_cache - PL_locale_utf8ness);
4894 /* But we can't save it if it is too large for the total space available */
4895 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
4896 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
4898 /* Here it can fit, but we may need to clear out the oldest cached
4899 * result(s) to do so. Check */
4900 if (utf8ness_cache_len + input_name_len_with_overhead
4901 >= utf8ness_cache_size)
4903 /* Here we have to clear something out to make room for this.
4904 * Start looking at the rightmost place where it could fit and find
4905 * the beginning of the entry that extends past that. */
4906 char * cutoff = (char *) my_memrchr(utf8ness_cache,
4909 - input_name_len_with_overhead);
4912 assert(cutoff >= utf8ness_cache);
4914 /* This and all subsequent entries must be removed */
4916 utf8ness_cache_len = strlen(utf8ness_cache);
4919 /* Make space for the new entry */
4920 Move(utf8ness_cache,
4921 utf8ness_cache + input_name_len_with_overhead,
4922 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
4925 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
4926 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4928 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
4929 & (PERL_UINTMAX_T) ~1) != '0')
4932 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%zu,"
4933 " inserted_name=%s, its_len=%zu\n",
4935 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
4936 delimited, input_name_len_with_overhead);
4942 if (DEBUG_Lv_TEST) {
4943 const char * s = PL_locale_utf8ness;
4945 /* Audit the structure */
4946 while (s < PL_locale_utf8ness + strlen(PL_locale_utf8ness)) {
4949 if (*s != UTF8NESS_SEP[0]) {
4951 "panic: %s: %d: Corrupt utf8ness_cache: missing"
4952 " separator %.*s<-- HERE %s\n",
4954 (int) (s - PL_locale_utf8ness), PL_locale_utf8ness,
4958 e = strchr(s, UTF8NESS_PREFIX[0]);
4961 "panic: %s: %d: Corrupt utf8ness_cache: missing"
4962 " separator %.*s<-- HERE %s\n",
4964 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
4968 if (*e != '0' && *e != '1') {
4970 "panic: %s: %d: Corrupt utf8ness_cache: utf8ness"
4971 " must be [01] %.*s<-- HERE %s\n",
4973 (int) (e + 1 - PL_locale_utf8ness),
4974 PL_locale_utf8ness, e + 1);
4976 if (ninstr(PL_locale_utf8ness, s, s-1, e)) {
4978 "panic: %s: %d: Corrupt utf8ness_cache: entry"
4979 " has duplicate %.*s<-- HERE %s\n",
4981 (int) (e - PL_locale_utf8ness), PL_locale_utf8ness,
4988 if (DEBUG_Lv_TEST || debug_initialization) {
4990 PerlIO_printf(Perl_debug_log,
4991 "PL_locale_utf8ness is now %s; returning %d\n",
4992 PL_locale_utf8ness, is_utf8);
4997 Safefree(delimited);
4998 Safefree(save_input_locale);
5005 Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
5008 /* Internal function which returns if we are in the scope of a pragma that
5009 * enables the locale category 'category'. 'compiling' should indicate if
5010 * this is during the compilation phase (TRUE) or not (FALSE). */
5012 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
5014 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
5015 if (! categories || categories == &PL_sv_placeholder) {
5019 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
5020 * a valid unsigned */
5021 assert(category >= -1);
5022 return cBOOL(SvUV(categories) & (1U << (category + 1)));
5026 Perl_my_strerror(pTHX_ const int errnum)
5028 /* Returns a mortalized copy of the text of the error message associated
5029 * with 'errnum'. It uses the current locale's text unless the platform
5030 * doesn't have the LC_MESSAGES category or we are not being called from
5031 * within the scope of 'use locale'. In the former case, it uses whatever
5032 * strerror returns; in the latter case it uses the text from the C locale.
5034 * The function just calls strerror(), but temporarily switches, if needed,
5035 * to the C locale */
5040 #ifndef USE_LOCALE_MESSAGES
5042 /* If platform doesn't have messages category, we don't do any switching to
5043 * the C locale; we just use whatever strerror() returns */
5045 errstr = savepv(Strerror(errnum));
5047 #else /* Has locale messages */
5049 const bool within_locale_scope = IN_LC(LC_MESSAGES);
5051 # ifndef USE_ITHREADS
5053 /* This function is trivial without threads. */
5054 if (within_locale_scope) {
5055 errstr = savepv(strerror(errnum));
5058 const char * save_locale = savepv(do_setlocale_c(LC_MESSAGES, NULL));
5060 do_setlocale_c(LC_MESSAGES, "C");
5061 errstr = savepv(strerror(errnum));
5062 do_setlocale_c(LC_MESSAGES, save_locale);
5063 Safefree(save_locale);
5066 # elif defined(HAS_POSIX_2008_LOCALE) \
5067 && defined(HAS_STRERROR_L) \
5068 && defined(HAS_DUPLOCALE)
5070 /* This function is also trivial if we don't have to worry about thread
5071 * safety and have strerror_l(), as it handles the switch of locales so we
5072 * don't have to deal with that. We don't have to worry about thread
5073 * safety if strerror_r() is also available. Both it and strerror_l() are
5074 * thread-safe. Plain strerror() isn't thread safe. But on threaded
5075 * builds when strerror_r() is available, the apparent call to strerror()
5076 * below is actually a macro that behind-the-scenes calls strerror_r(). */
5078 # ifdef HAS_STRERROR_R
5080 if (within_locale_scope) {
5081 errstr = savepv(strerror(errnum));
5084 errstr = savepv(strerror_l(errnum, PL_C_locale_obj));
5089 /* Here we have strerror_l(), but not strerror_r() and we are on a
5090 * threaded-build. We use strerror_l() for everything, constructing a
5091 * locale to pass to it if necessary */
5093 bool do_free = FALSE;
5094 locale_t locale_to_use;
5096 if (within_locale_scope) {
5097 locale_to_use = uselocale((locale_t) 0);
5098 if (locale_to_use == LC_GLOBAL_LOCALE) {
5099 locale_to_use = duplocale(LC_GLOBAL_LOCALE);
5103 else { /* Use C locale if not within 'use locale' scope */
5104 locale_to_use = PL_C_locale_obj;
5107 errstr = savepv(strerror_l(errnum, locale_to_use));
5110 freelocale(locale_to_use);
5114 # else /* Doesn't have strerror_l() */
5116 const char * save_locale = NULL;
5117 bool locale_is_C = FALSE;
5119 /* We have a critical section to prevent another thread from executing this
5120 * same code at the same time. (On thread-safe perls, the LOCK is a
5121 * no-op.) Since this is the only place in core that changes LC_MESSAGES
5122 * (unless the user has called setlocale(), this works to prevent races. */
5125 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5126 "my_strerror called with errnum %d\n", errnum));
5127 if (! within_locale_scope) {
5128 save_locale = do_setlocale_c(LC_MESSAGES, NULL);
5129 if (! save_locale) {
5131 "panic: %s: %d: Could not find current LC_MESSAGES locale,"
5132 " errno=%d\n", __FILE__, __LINE__, errno);
5135 locale_is_C = isNAME_C_OR_POSIX(save_locale);
5137 /* Switch to the C locale if not already in it */
5138 if (! locale_is_C) {
5140 /* The setlocale() just below likely will zap 'save_locale', so
5142 save_locale = savepv(save_locale);
5143 do_setlocale_c(LC_MESSAGES, "C");
5146 } /* end of ! within_locale_scope */
5148 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s: %d: WITHIN locale scope\n",
5149 __FILE__, __LINE__));
5152 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5153 "Any locale change has been done; about to call Strerror\n"));
5154 errstr = savepv(Strerror(errnum));
5156 if (! within_locale_scope) {
5157 if (save_locale && ! locale_is_C) {
5158 if (! do_setlocale_c(LC_MESSAGES, save_locale)) {
5160 "panic: %s: %d: setlocale restore failed, errno=%d\n",
5161 __FILE__, __LINE__, errno);
5163 Safefree(save_locale);
5169 # endif /* End of doesn't have strerror_l */
5170 #endif /* End of does have locale messages */
5174 if (DEBUG_Lv_TEST) {
5175 PerlIO_printf(Perl_debug_log, "Strerror returned; saving a copy: '");
5176 print_bytes_for_locale(errstr, errstr + strlen(errstr), 0);
5177 PerlIO_printf(Perl_debug_log, "'\n");
5188 =for apidoc switch_to_global_locale
5190 On systems without locale support, or on single-threaded builds, or on
5191 platforms that do not support per-thread locale operations, this function does
5192 nothing. On such systems that do have locale support, only a locale global to
5193 the whole program is available.
5195 On multi-threaded builds on systems that do have per-thread locale operations,
5196 this function converts the thread it is running in to use the global locale.
5197 This is for code that has not yet or cannot be updated to handle multi-threaded
5198 locale operation. As long as only a single thread is so-converted, everything
5199 works fine, as all the other threads continue to ignore the global one, so only
5200 this thread looks at it.
5202 However, on Windows systems this isn't quite true prior to Visual Studio 15,
5203 at which point Microsoft fixed a bug. A race can occur if you use the
5204 following operations on earlier Windows platforms:
5208 =item L<POSIX::localeconv|POSIX/localeconv>
5210 =item L<I18N::Langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5212 =item L<perlapi/Perl_langinfo>, items C<CRNCYSTR> and C<THOUSEP>
5216 The first item is not fixable (except by upgrading to a later Visual Studio
5217 release), but it would be possible to work around the latter two items by using
5218 the Windows API functions C<GetNumberFormat> and C<GetCurrencyFormat>; patches
5221 Without this function call, threads that use the L<C<setlocale(3)>> system
5222 function will not work properly, as all the locale-sensitive functions will
5223 look at the per-thread locale, and C<setlocale> will have no effect on this
5226 Perl code should convert to either call
5227 L<C<Perl_setlocale>|perlapi/Perl_setlocale> (which is a drop-in for the system
5228 C<setlocale>) or use the methods given in L<perlcall> to call
5229 L<C<POSIX::setlocale>|POSIX/setlocale>. Either one will transparently properly
5230 handle all cases of single- vs multi-thread, POSIX 2008-supported or not.
5232 Non-Perl libraries, such as C<gtk>, that call the system C<setlocale> can
5233 continue to work if this function is called before transferring control to the
5236 Upon return from the code that needs to use the global locale,
5237 L<C<sync_locale()>|perlapi/sync_locale> should be called to restore the safe
5238 multi-thread operation.
5244 Perl_switch_to_global_locale()
5247 #ifdef USE_THREAD_SAFE_LOCALE
5250 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
5253 # ifdef HAS_QUERYLOCALE
5255 setlocale(LC_ALL, querylocale(LC_ALL_MASK, uselocale((locale_t) 0)));
5262 for (i = 0; i < LC_ALL_INDEX; i++) {
5263 setlocale(categories[i], do_setlocale_r(categories[i], NULL));
5269 uselocale(LC_GLOBAL_LOCALE);
5278 =for apidoc sync_locale
5280 L<C<Perl_setlocale>|perlapi/Perl_setlocale> can be used at any time to query or
5281 change the locale (though changing the locale is antisocial and dangerous on
5282 multi-threaded systems that don't have multi-thread safe locale operations.
5283 (See L<perllocale/Multi-threaded operation>). Using the system
5284 L<C<setlocale(3)>> should be avoided. Nevertheless, certain non-Perl libraries
5285 called from XS, such as C<Gtk> do so, and this can't be changed. When the
5286 locale is changed by XS code that didn't use
5287 L<C<Perl_setlocale>|perlapi/Perl_setlocale>, Perl needs to be told that the
5288 locale has changed. Use this function to do so, before returning to Perl.
5290 The return value is a boolean: TRUE if the global locale at the time of call
5291 was in effect; and FALSE if a per-thread locale was in effect. This can be
5292 used by the caller that needs to restore things as-they-were to decide whether
5294 L<C<Perl_switch_to_global_locale>|perlapi/switch_to_global_locale>.
5302 const char * newlocale;
5305 #ifdef USE_POSIX_2008_LOCALE
5307 bool was_in_global_locale = FALSE;
5308 locale_t cur_obj = uselocale((locale_t) 0);
5310 /* On Windows, unless the foreign code has turned off the thread-safe
5311 * locale setting, any plain setlocale() will have affected what we see, so
5312 * no need to worry. Otherwise, If the foreign code has done a plain
5313 * setlocale(), it will only affect the global locale on POSIX systems, but
5314 * will affect the */
5315 if (cur_obj == LC_GLOBAL_LOCALE) {
5317 # ifdef HAS_QUERY_LOCALE
5319 do_setlocale_c(LC_ALL, setlocale(LC_ALL, NULL));
5325 /* We can't trust that we can read the LC_ALL format on the
5326 * platform, so do them individually */
5327 for (i = 0; i < LC_ALL_INDEX; i++) {
5328 do_setlocale_r(categories[i], setlocale(categories[i], NULL));
5333 was_in_global_locale = TRUE;
5338 bool was_in_global_locale = TRUE;
5341 #ifdef USE_LOCALE_CTYPE
5343 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
5344 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5345 "%s:%d: %s\n", __FILE__, __LINE__,
5346 setlocale_debug_string(LC_CTYPE, NULL, newlocale)));
5347 new_ctype(newlocale);
5348 Safefree(newlocale);
5350 #endif /* USE_LOCALE_CTYPE */
5351 #ifdef USE_LOCALE_COLLATE
5353 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
5354 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5355 "%s:%d: %s\n", __FILE__, __LINE__,
5356 setlocale_debug_string(LC_COLLATE, NULL, newlocale)));
5357 new_collate(newlocale);
5358 Safefree(newlocale);
5361 #ifdef USE_LOCALE_NUMERIC
5363 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
5364 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
5365 "%s:%d: %s\n", __FILE__, __LINE__,
5366 setlocale_debug_string(LC_NUMERIC, NULL, newlocale)));
5367 new_numeric(newlocale);
5368 Safefree(newlocale);
5370 #endif /* USE_LOCALE_NUMERIC */
5372 return was_in_global_locale;
5375 #if defined(DEBUGGING) && defined(USE_LOCALE)
5378 S_setlocale_debug_string(const int category, /* category number,
5380 const char* const locale, /* locale name */
5382 /* return value from setlocale() when attempting to
5383 * set 'category' to 'locale' */
5384 const char* const retval)
5386 /* Returns a pointer to a NUL-terminated string in static storage with
5387 * added text about the info passed in. This is not thread safe and will
5388 * be overwritten by the next call, so this should be used just to
5389 * formulate a string to immediately print or savepv() on. */
5391 /* initialise to a non-null value to keep it out of BSS and so keep
5392 * -DPERL_GLOBAL_STRUCT_PRIVATE happy */
5393 static char ret[256] = "If you can read this, thank your buggy C"
5394 " library strlcpy(), and change your hints file"
5397 my_strlcpy(ret, "setlocale(", sizeof(ret));
5398 my_strlcat(ret, category_name(category), sizeof(ret));
5399 my_strlcat(ret, ", ", sizeof(ret));
5402 my_strlcat(ret, "\"", sizeof(ret));
5403 my_strlcat(ret, locale, sizeof(ret));
5404 my_strlcat(ret, "\"", sizeof(ret));
5407 my_strlcat(ret, "NULL", sizeof(ret));
5410 my_strlcat(ret, ") returned ", sizeof(ret));
5413 my_strlcat(ret, "\"", sizeof(ret));
5414 my_strlcat(ret, retval, sizeof(ret));
5415 my_strlcat(ret, "\"", sizeof(ret));
5418 my_strlcat(ret, "NULL", sizeof(ret));
5421 assert(strlen(ret) < sizeof(ret));
5429 Perl_thread_locale_init()
5431 /* Called from a thread on startup*/
5433 #ifdef USE_THREAD_SAFE_LOCALE
5437 /* C starts the new thread in the global C locale. If we are thread-safe,
5438 * we want to not be in the global locale */
5440 DEBUG_L(PerlIO_printf(Perl_debug_log,
5441 "%s:%d: new thread, initial locale is %s; calling setlocale\n",
5442 __FILE__, __LINE__, setlocale(LC_ALL, NULL)));
5446 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
5450 Perl_setlocale(LC_ALL, "C");
5458 Perl_thread_locale_term()
5460 /* Called from a thread as it gets ready to terminate */
5462 #ifdef USE_THREAD_SAFE_LOCALE
5464 /* C starts the new thread in the global C locale. If we are thread-safe,
5465 * we want to not be in the global locale */
5470 locale_t cur_obj = uselocale(LC_GLOBAL_LOCALE);
5471 if (cur_obj != LC_GLOBAL_LOCALE) {
5472 freelocale(cur_obj);
5482 * ex: set ts=8 sts=4 sw=4 et: