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 assert(PL_C_locale_obj);
740 /* Otherwise, we are switching locales. This will generally entail freeing
741 * the current one's space (at the C library's discretion). We need to
742 * stop using that locale before the switch. So switch to a known locale
743 * object that we don't otherwise mess with. This returns the locale
744 * object in effect at the time of the switch. */
745 old_obj = uselocale(PL_C_locale_obj);
749 if (DEBUG_Lv_TEST || debug_initialization) {
750 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
759 if (DEBUG_L_TEST || debug_initialization) {
761 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
772 if (DEBUG_Lv_TEST || debug_initialization) {
773 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
778 /* If we weren't in a thread safe locale, set so that newlocale() below
779 which uses 'old_obj', uses an empty one. Same for our reserved C object.
780 The latter is defensive coding, so that, even if there is some bug, we
781 will never end up trying to modify either of these, as if passed to
782 newlocale(), they can be. */
783 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
784 old_obj = (locale_t) 0;
787 /* Create the new locale (it may actually modify the current one). */
789 # ifndef HAS_QUERYLOCALE
791 if (strEQ(locale, "")) {
793 /* For non-querylocale() systems, we do the setting of "" ourselves to
794 * be sure that we really know what's going on. We follow the Linux
795 * documented behavior (but if that differs from the actual behavior,
796 * this won't work exactly as the OS implements). We go out and
797 * examine the environment based on our understanding of how the system
798 * works, and use that to figure things out */
800 const char * const lc_all = PerlEnv_getenv("LC_ALL");
802 /* Use any "LC_ALL" environment variable, as it overrides everything
804 if (lc_all && strNE(lc_all, "")) {
809 /* Otherwise, we need to dig deeper. Unless overridden, the
810 * default is the LANG environment variable; if it doesn't exist,
813 const char * default_name;
815 /* To minimize other threads messing with the environment, we copy
816 * the variable, making it a temporary. But this doesn't work upon
817 * program initialization before any scopes are created, and at
818 * this time, there's nothing else going on that would interfere.
819 * So skip the copy in that case */
820 if (PL_scopestack_ix == 0) {
821 default_name = PerlEnv_getenv("LANG");
824 default_name = savepv(PerlEnv_getenv("LANG"));
827 if (! default_name || strEQ(default_name, "")) {
830 else if (PL_scopestack_ix != 0) {
831 SAVEFREEPV(default_name);
834 if (category != LC_ALL) {
835 const char * const name = PerlEnv_getenv(category_names[index]);
837 /* Here we are setting a single category. Assume will have the
839 locale = default_name;
841 /* But then look for an overriding environment variable */
842 if (name && strNE(name, "")) {
847 bool did_override = FALSE;
850 /* Here, we are getting LC_ALL. Any categories that don't have
851 * a corresponding environment variable set should be set to
852 * LANG, or to "C" if there is no LANG. If no individual
853 * categories differ from this, we can just set LC_ALL. This
854 * is buggy on systems that have extra categories that we don't
855 * know about. If there is an environment variable that sets
856 * that category, we won't know to look for it, and so our use
857 * of LANG or "C" improperly overrides it. On the other hand,
858 * if we don't do what is done here, and there is no
859 * environment variable, the category's locale should be set to
860 * LANG or "C". So there is no good solution. khw thinks the
861 * best is to look at systems to see what categories they have,
862 * and include them, and then to assume that we know the
865 for (i = 0; i < LC_ALL_INDEX; i++) {
866 const char * const env_override
867 = savepv(PerlEnv_getenv(category_names[i]));
868 const char * this_locale = ( env_override
869 && strNE(env_override, ""))
872 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
874 Safefree(env_override);
878 if (strNE(this_locale, default_name)) {
882 Safefree(env_override);
885 /* If all the categories are the same, we can set LC_ALL to
887 if (! did_override) {
888 locale = default_name;
892 /* Here, LC_ALL is no longer valid, as some individual
893 * categories don't match it. We call ourselves
894 * recursively, as that will execute the code that
895 * generates the proper locale string for this situation.
896 * We don't do the remainder of this function, as that is
897 * to update our records, and we've just done that for the
898 * individual categories in the loop above, and doing so
899 * would cause LC_ALL to be done as well */
900 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
905 else if (strchr(locale, ';')) {
907 /* LC_ALL may actually incude a conglomeration of various categories.
908 * Without querylocale, this code uses the glibc (as of this writing)
909 * syntax for representing that, but that is not a stable API, and
910 * other platforms do it differently, so we have to handle all cases
913 const char * s = locale;
914 const char * e = locale + strlen(locale);
916 const char * category_end;
917 const char * name_start;
918 const char * name_end;
923 /* Parse through the category */
924 while (isWORDCHAR(*p)) {
931 "panic: %s: %d: Unexpected character in locale name '%02X",
932 __FILE__, __LINE__, *(p-1));
935 /* Parse through the locale name */
937 while (p < e && *p != ';') {
940 "panic: %s: %d: Unexpected character in locale name '%02X",
941 __FILE__, __LINE__, *(p-1));
947 /* Space past the semi-colon */
952 /* Find the index of the category name in our lists */
953 for (i = 0; i < LC_ALL_INDEX; i++) {
954 char * individ_locale;
956 /* Keep going if this isn't the index. The strnNE() avoids a
957 * Perl_form(), but would fail if ever a category name could be
958 * a substring of another one, like if there were a
960 if strnNE(s, category_names[i], category_end - s) {
964 /* If this index is for the single category we're changing, we
965 * have found the locale to set it to. */
966 if (category == categories[i]) {
967 locale = Perl_form(aTHX_ "%.*s",
968 (int) (name_end - name_start),
973 assert(category == LC_ALL);
974 individ_locale = Perl_form(aTHX_ "%.*s",
975 (int) (name_end - name_start), name_start);
976 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
985 /* Here we have set all the individual categories by recursive calls.
986 * These collectively should have fixed up LC_ALL, so can just query
987 * what that now is */
988 assert(category == LC_ALL);
990 return do_setlocale_c(LC_ALL, NULL);
995 # endif /* end of ! querylocale */
997 /* Ready to create a new locale by modification of the exising one */
998 new_obj = newlocale(mask, locale, old_obj);
1005 if (DEBUG_L_TEST || debug_initialization) {
1006 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1011 if (! uselocale(old_obj)) {
1015 if (DEBUG_L_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1028 if (DEBUG_Lv_TEST || debug_initialization) {
1029 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1034 /* And switch into it */
1035 if (! uselocale(new_obj)) {
1040 if (DEBUG_L_TEST || debug_initialization) {
1041 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1046 if (! uselocale(old_obj)) {
1050 if (DEBUG_L_TEST || debug_initialization) {
1051 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1057 freelocale(new_obj);
1064 if (DEBUG_Lv_TEST || debug_initialization) {
1065 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1070 /* We are done, except for updating our records (if the system doesn't keep
1071 * them) and in the case of locale "", we don't actually know what the
1072 * locale that got switched to is, as it came from the environment. So
1073 * have to find it */
1075 # ifdef HAS_QUERYLOCALE
1077 if (strEQ(locale, "")) {
1078 locale = querylocale(mask, new_obj);
1083 /* Here, 'locale' is the return value */
1085 /* Without querylocale(), we have to update our records */
1087 if (category == LC_ALL) {
1090 /* For LC_ALL, we change all individual categories to correspond */
1091 /* PL_curlocales is a parallel array, so has same
1092 * length as 'categories' */
1093 for (i = 0; i <= LC_ALL_INDEX; i++) {
1094 Safefree(PL_curlocales[i]);
1095 PL_curlocales[i] = savepv(locale);
1100 /* For a single category, if it's not the same as the one in LC_ALL, we
1103 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1104 Safefree(PL_curlocales[LC_ALL_INDEX]);
1105 PL_curlocales[LC_ALL_INDEX] = NULL;
1108 /* Then update the category's record */
1109 Safefree(PL_curlocales[index]);
1110 PL_curlocales[index] = savepv(locale);
1118 #endif /* USE_POSIX_2008_LOCALE */
1120 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1121 didn't natively support them */
1123 /* The way this would work is that we would keep a per-thread list of the
1124 * correct locale for that thread. Any operation that was locale-sensitive
1125 * would have to be changed so that it would look like this:
1128 * setlocale to the correct locale for this operation
1132 * This leaves the global locale in the most recently used operation's, but it
1133 * was locked long enough to get the result. If that result is static, it
1134 * needs to be copied before the unlock.
1136 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1137 * the setup, but are no-ops when not needed, and similarly,
1138 * END_LOCALE_DEPENDENT_OP for the tear-down
1140 * But every call to a locale-sensitive function would have to be changed, and
1141 * if a module didn't cooperate by using the mutex, things would break.
1143 * This code was abandoned before being completed or tested, and is left as-is
1146 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1147 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1150 S_locking_setlocale(pTHX_
1152 const char * locale,
1154 const bool is_index_valid
1157 /* This function kind of performs a setlocale() on just the current thread;
1158 * thus it is kind of thread-safe. It does this by keeping a thread-level
1159 * array of the current locales for each category. Every time a locale is
1160 * switched to, it does the switch globally, but updates the thread's
1161 * array. A query as to what the current locale is just returns the
1162 * appropriate element from the array, and doesn't actually call the system
1163 * setlocale(). The saving into the array is done in an uninterruptible
1164 * section of code, so is unaffected by whatever any other threads might be
1167 * All locale-sensitive operations must work by first starting a critical
1168 * section, then switching to the thread's locale as kept by this function,
1169 * and then doing the operation, then ending the critical section. Thus,
1170 * each gets done in the appropriate locale. simulating thread-safety.
1172 * This function takes the same parameters, 'category' and 'locale', that
1173 * the regular setlocale() function does, but it also takes two additional
1174 * ones. This is because as described earlier. If we know on input the
1175 * index corresponding to the category into the array where we store the
1176 * current locales, we don't have to calculate it. If the caller knows at
1177 * compile time what the index is, it it can pass it, setting
1178 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1182 /* If the input index might be incorrect, calculate the correct one */
1183 if (! is_index_valid) {
1186 if (DEBUG_Lv_TEST || debug_initialization) {
1187 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1190 for (i = 0; i <= LC_ALL_INDEX; i++) {
1191 if (category == categories[i]) {
1197 /* Here, we don't know about this category, so can't handle it.
1198 * XXX best we can do is to unsafely set this
1201 return my_setlocale(category, locale);
1205 if (DEBUG_Lv_TEST || debug_initialization) {
1206 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1210 /* For a query, just return what's in our records */
1211 if (new_locale == NULL) {
1212 return curlocales[index];
1216 /* Otherwise, we need to do the switch, and save the result, all in a
1217 * critical section */
1219 Safefree(curlocales[[index]]);
1221 /* It might be that this is called from an already-locked section of code.
1222 * We would have to detect and skip the LOCK/UNLOCK if so */
1225 curlocales[index] = savepv(my_setlocale(category, new_locale));
1227 if (strEQ(new_locale, "")) {
1231 /* The locale values come from the environment, and may not all be the
1232 * same, so for LC_ALL, we have to update all the others, while the
1233 * mutex is still locked */
1235 if (category == LC_ALL) {
1237 for (i = 0; i < LC_ALL_INDEX) {
1238 curlocales[i] = my_setlocale(categories[i], NULL);
1247 return curlocales[index];
1253 S_set_numeric_radix(pTHX_ const bool use_locale)
1255 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1256 * TRUE, use the radix character derived from the current locale */
1258 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1259 || defined(HAS_NL_LANGINFO))
1261 const char * radix = (use_locale)
1262 ? my_nl_langinfo(RADIXCHAR, FALSE)
1263 /* FALSE => already in dest locale */
1266 sv_setpv(PL_numeric_radix_sv, radix);
1268 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1269 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1270 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1271 SvCUR(PL_numeric_radix_sv))
1272 && _is_cur_LC_category_utf8(LC_NUMERIC))
1274 SvUTF8_on(PL_numeric_radix_sv);
1279 if (DEBUG_L_TEST || debug_initialization) {
1280 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1281 SvPVX(PL_numeric_radix_sv),
1282 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1286 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1291 S_new_numeric(pTHX_ const char *newnum)
1294 #ifndef USE_LOCALE_NUMERIC
1296 PERL_UNUSED_ARG(newnum);
1300 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1301 * core Perl this and that 'newnum' is the name of the new locale.
1302 * It installs this locale as the current underlying default.
1304 * The default locale and the C locale can be toggled between by use of the
1305 * set_numeric_underlying() and set_numeric_standard() functions, which
1306 * should probably not be called directly, but only via macros like
1307 * SET_NUMERIC_STANDARD() in perl.h.
1309 * The toggling is necessary mainly so that a non-dot radix decimal point
1310 * character can be output, while allowing internal calculations to use a
1313 * This sets several interpreter-level variables:
1314 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1315 * PL_numeric_underlying A boolean indicating if the toggled state is such
1316 * that the current locale is the program's underlying
1318 * PL_numeric_standard An int indicating if the toggled state is such
1319 * that the current locale is the C locale or
1320 * indistinguishable from the C locale. If non-zero, it
1321 * is in C; if > 1, it means it may not be toggled away
1323 * PL_numeric_underlying_is_standard A bool kept by this function
1324 * indicating that the underlying locale and the standard
1325 * C locale are indistinguishable for the purposes of
1326 * LC_NUMERIC. This happens when both of the above two
1327 * variables are true at the same time. (Toggling is a
1328 * no-op under these circumstances.) This variable is
1329 * used to avoid having to recalculate.
1335 Safefree(PL_numeric_name);
1336 PL_numeric_name = NULL;
1337 PL_numeric_standard = TRUE;
1338 PL_numeric_underlying = TRUE;
1339 PL_numeric_underlying_is_standard = TRUE;
1343 save_newnum = stdize_locale(savepv(newnum));
1344 PL_numeric_underlying = TRUE;
1345 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1347 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1349 if (! PL_numeric_standard) {
1350 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1351 FALSE /* Don't toggle locale */ ))
1352 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1355 /* Save the new name if it isn't the same as the previous one, if any */
1356 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1357 Safefree(PL_numeric_name);
1358 PL_numeric_name = save_newnum;
1361 Safefree(save_newnum);
1364 PL_numeric_underlying_is_standard = PL_numeric_standard;
1366 # ifdef HAS_POSIX_2008_LOCALE
1368 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1370 PL_underlying_numeric_obj);
1374 if (DEBUG_L_TEST || debug_initialization) {
1375 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1378 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1379 * have to worry about the radix being a non-dot. (Core operations that
1380 * need the underlying locale change to it temporarily). */
1381 if (PL_numeric_standard) {
1382 set_numeric_radix(0);
1385 set_numeric_standard();
1388 #endif /* USE_LOCALE_NUMERIC */
1393 Perl_set_numeric_standard(pTHX)
1396 #ifdef USE_LOCALE_NUMERIC
1398 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1399 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1400 * macro avoids calling this routine if toggling isn't necessary according
1401 * to our records (which could be wrong if some XS code has changed the
1402 * locale behind our back) */
1404 do_setlocale_c(LC_NUMERIC, "C");
1405 PL_numeric_standard = TRUE;
1406 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1407 set_numeric_radix(0);
1411 if (DEBUG_L_TEST || debug_initialization) {
1412 PerlIO_printf(Perl_debug_log,
1413 "LC_NUMERIC locale now is standard C\n");
1417 #endif /* USE_LOCALE_NUMERIC */
1422 Perl_set_numeric_underlying(pTHX)
1425 #ifdef USE_LOCALE_NUMERIC
1427 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1428 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1429 * instead of calling this directly. The macro avoids calling this routine
1430 * if toggling isn't necessary according to our records (which could be
1431 * wrong if some XS code has changed the locale behind our back) */
1433 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1434 PL_numeric_standard = PL_numeric_underlying_is_standard;
1435 PL_numeric_underlying = TRUE;
1436 set_numeric_radix(! PL_numeric_standard);
1440 if (DEBUG_L_TEST || debug_initialization) {
1441 PerlIO_printf(Perl_debug_log,
1442 "LC_NUMERIC locale now is %s\n",
1447 #endif /* USE_LOCALE_NUMERIC */
1452 * Set up for a new ctype locale.
1455 S_new_ctype(pTHX_ const char *newctype)
1458 #ifndef USE_LOCALE_CTYPE
1460 PERL_ARGS_ASSERT_NEW_CTYPE;
1461 PERL_UNUSED_ARG(newctype);
1462 PERL_UNUSED_CONTEXT;
1466 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1467 * core Perl this and that 'newctype' is the name of the new locale.
1469 * This function sets up the folding arrays for all 256 bytes, assuming
1470 * that tofold() is tolc() since fold case is not a concept in POSIX,
1472 * Any code changing the locale (outside this file) should use
1473 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1474 * this function should be called directly only from this file and from
1475 * POSIX::setlocale() */
1480 /* Don't check for problems if we are suppressing the warnings */
1481 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1483 PERL_ARGS_ASSERT_NEW_CTYPE;
1485 /* We will replace any bad locale warning with 1) nothing if the new one is
1486 * ok; or 2) a new warning for the bad new locale */
1487 if (PL_warn_locale) {
1488 SvREFCNT_dec_NN(PL_warn_locale);
1489 PL_warn_locale = NULL;
1492 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1494 /* A UTF-8 locale gets standard rules. But note that code still has to
1495 * handle this specially because of the three problematic code points */
1496 if (PL_in_utf8_CTYPE_locale) {
1497 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1500 /* We don't populate the other lists if a UTF-8 locale, but do check that
1501 * everything works as expected, unless checking turned off */
1502 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1503 /* Assume enough space for every character being bad. 4 spaces each
1504 * for the 94 printable characters that are output like "'x' "; and 5
1505 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1507 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1508 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1510 unsigned int bad_count = 0; /* Count of bad characters */
1512 for (i = 0; i < 256; i++) {
1513 if (! PL_in_utf8_CTYPE_locale) {
1515 PL_fold_locale[i] = (U8) tolower(i);
1516 else if (islower(i))
1517 PL_fold_locale[i] = (U8) toupper(i);
1519 PL_fold_locale[i] = (U8) i;
1522 /* If checking for locale problems, see if the native ASCII-range
1523 * printables plus \n and \t are in their expected categories in
1524 * the new locale. If not, this could mean big trouble, upending
1525 * Perl's and most programs' assumptions, like having a
1526 * metacharacter with special meaning become a \w. Fortunately,
1527 * it's very rare to find locales that aren't supersets of ASCII
1528 * nowadays. It isn't a problem for most controls to be changed
1529 * into something else; we check only \n and \t, though perhaps \r
1530 * could be an issue as well. */
1531 if ( check_for_problems
1532 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1534 bool is_bad = FALSE;
1535 char name[3] = { '\0' };
1537 /* Convert the name into a string */
1542 else if (i == '\n') {
1543 my_strlcpy(name, "\n", sizeof(name));
1546 my_strlcpy(name, "\t", sizeof(name));
1549 /* Check each possibe class */
1550 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1552 DEBUG_L(PerlIO_printf(Perl_debug_log,
1553 "isalnum('%s') unexpectedly is %d\n",
1554 name, cBOOL(isalnum(i))));
1556 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1558 DEBUG_L(PerlIO_printf(Perl_debug_log,
1559 "isalpha('%s') unexpectedly is %d\n",
1560 name, cBOOL(isalpha(i))));
1562 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1564 DEBUG_L(PerlIO_printf(Perl_debug_log,
1565 "isdigit('%s') unexpectedly is %d\n",
1566 name, cBOOL(isdigit(i))));
1568 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1570 DEBUG_L(PerlIO_printf(Perl_debug_log,
1571 "isgraph('%s') unexpectedly is %d\n",
1572 name, cBOOL(isgraph(i))));
1574 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1576 DEBUG_L(PerlIO_printf(Perl_debug_log,
1577 "islower('%s') unexpectedly is %d\n",
1578 name, cBOOL(islower(i))));
1580 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1582 DEBUG_L(PerlIO_printf(Perl_debug_log,
1583 "isprint('%s') unexpectedly is %d\n",
1584 name, cBOOL(isprint(i))));
1586 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1588 DEBUG_L(PerlIO_printf(Perl_debug_log,
1589 "ispunct('%s') unexpectedly is %d\n",
1590 name, cBOOL(ispunct(i))));
1592 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1594 DEBUG_L(PerlIO_printf(Perl_debug_log,
1595 "isspace('%s') unexpectedly is %d\n",
1596 name, cBOOL(isspace(i))));
1598 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1600 DEBUG_L(PerlIO_printf(Perl_debug_log,
1601 "isupper('%s') unexpectedly is %d\n",
1602 name, cBOOL(isupper(i))));
1604 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1606 DEBUG_L(PerlIO_printf(Perl_debug_log,
1607 "isxdigit('%s') unexpectedly is %d\n",
1608 name, cBOOL(isxdigit(i))));
1610 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1612 DEBUG_L(PerlIO_printf(Perl_debug_log,
1613 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1614 name, tolower(i), (int) toLOWER_A(i)));
1616 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1618 DEBUG_L(PerlIO_printf(Perl_debug_log,
1619 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1620 name, toupper(i), (int) toUPPER_A(i)));
1622 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1624 DEBUG_L(PerlIO_printf(Perl_debug_log,
1625 "'\\n' (=%02X) is not a control\n", (int) i));
1628 /* Add to the list; Separate multiple entries with a blank */
1631 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1633 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1641 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1642 * this locale requires more than one byte, there are going to be
1644 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1645 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1646 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1648 if ( check_for_problems && MB_CUR_MAX > 1
1649 && ! PL_in_utf8_CTYPE_locale
1651 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1652 * locale. Just assume that the implementation for them (plus
1653 * for POSIX) is correct and the > 1 value is spurious. (Since
1654 * these are specially handled to never be considered UTF-8
1655 * locales, as long as this is the only problem, everything
1656 * should work fine */
1657 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1659 multi_byte_locale = TRUE;
1664 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1665 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1666 PL_warn_locale = Perl_newSVpvf(aTHX_
1667 "Locale '%s' contains (at least) the following characters"
1668 " which have\nunexpected meanings: %s\nThe Perl program"
1669 " will use the expected meanings",
1670 newctype, bad_chars_list);
1673 PL_warn_locale = Perl_newSVpvf(aTHX_
1674 "Locale '%s' may not work well.%s%s%s\n",
1677 ? " Some characters in it are not recognized by"
1681 ? "\nThe following characters (and maybe others)"
1682 " may not have the same meaning as the Perl"
1683 " program expects:\n"
1691 # ifdef HAS_NL_LANGINFO
1693 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1694 /* parameter FALSE is a don't care here */
1695 my_nl_langinfo(CODESET, FALSE));
1699 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1701 /* If we are actually in the scope of the locale or are debugging,
1702 * output the message now. If not in that scope, we save the
1703 * message to be output at the first operation using this locale,
1704 * if that actually happens. Most programs don't use locales, so
1705 * they are immune to bad ones. */
1706 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1708 /* The '0' below suppresses a bogus gcc compiler warning */
1709 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1711 if (IN_LC(LC_CTYPE)) {
1712 SvREFCNT_dec_NN(PL_warn_locale);
1713 PL_warn_locale = NULL;
1719 #endif /* USE_LOCALE_CTYPE */
1724 Perl__warn_problematic_locale()
1727 #ifdef USE_LOCALE_CTYPE
1731 /* Internal-to-core function that outputs the message in PL_warn_locale,
1732 * and then NULLS it. Should be called only through the macro
1733 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1735 if (PL_warn_locale) {
1736 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1737 SvPVX(PL_warn_locale),
1738 0 /* dummy to avoid compiler warning */ );
1739 SvREFCNT_dec_NN(PL_warn_locale);
1740 PL_warn_locale = NULL;
1748 S_new_collate(pTHX_ const char *newcoll)
1751 #ifndef USE_LOCALE_COLLATE
1753 PERL_UNUSED_ARG(newcoll);
1754 PERL_UNUSED_CONTEXT;
1758 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1759 * core Perl this and that 'newcoll' is the name of the new locale.
1761 * The design of locale collation is that every locale change is given an
1762 * index 'PL_collation_ix'. The first time a string particpates in an
1763 * operation that requires collation while locale collation is active, it
1764 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1765 * magic includes the collation index, and the transformation of the string
1766 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1767 * instead of the string itself. If a string changes, the magic is
1768 * cleared. The next time the locale changes, the index is incremented,
1769 * and so we know during a comparison that the transformation is not
1770 * necessarily still valid, and so is recomputed. Note that if the locale
1771 * changes enough times, the index could wrap (a U32), and it is possible
1772 * that a transformation would improperly be considered valid, leading to
1773 * an unlikely bug */
1776 if (PL_collation_name) {
1778 Safefree(PL_collation_name);
1779 PL_collation_name = NULL;
1781 PL_collation_standard = TRUE;
1782 is_standard_collation:
1783 PL_collxfrm_base = 0;
1784 PL_collxfrm_mult = 2;
1785 PL_in_utf8_COLLATE_locale = FALSE;
1786 PL_strxfrm_NUL_replacement = '\0';
1787 PL_strxfrm_max_cp = 0;
1791 /* If this is not the same locale as currently, set the new one up */
1792 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1794 Safefree(PL_collation_name);
1795 PL_collation_name = stdize_locale(savepv(newcoll));
1796 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1797 if (PL_collation_standard) {
1798 goto is_standard_collation;
1801 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1802 PL_strxfrm_NUL_replacement = '\0';
1803 PL_strxfrm_max_cp = 0;
1805 /* A locale collation definition includes primary, secondary, tertiary,
1806 * etc. weights for each character. To sort, the primary weights are
1807 * used, and only if they compare equal, then the secondary weights are
1808 * used, and only if they compare equal, then the tertiary, etc.
1810 * strxfrm() works by taking the input string, say ABC, and creating an
1811 * output transformed string consisting of first the primary weights,
1812 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1813 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1814 * may not have weights at every level. In our example, let's say B
1815 * doesn't have a tertiary weight, and A doesn't have a secondary
1816 * weight. The constructed string is then going to be
1817 * A¹B¹C¹ B²C² A³C³ ....
1818 * This has the desired effect that strcmp() will look at the secondary
1819 * or tertiary weights only if the strings compare equal at all higher
1820 * priority weights. The spaces shown here, like in
1822 * are not just for readability. In the general case, these must
1823 * actually be bytes, which we will call here 'separator weights'; and
1824 * they must be smaller than any other weight value, but since these
1825 * are C strings, only the terminating one can be a NUL (some
1826 * implementations may include a non-NUL separator weight just before
1827 * the NUL). Implementations tend to reserve 01 for the separator
1828 * weights. They are needed so that a shorter string's secondary
1829 * weights won't be misconstrued as primary weights of a longer string,
1830 * etc. By making them smaller than any other weight, the shorter
1831 * string will sort first. (Actually, if all secondary weights are
1832 * smaller than all primary ones, there is no need for a separator
1833 * weight between those two levels, etc.)
1835 * The length of the transformed string is roughly a linear function of
1836 * the input string. It's not exactly linear because some characters
1837 * don't have weights at all levels. When we call strxfrm() we have to
1838 * allocate some memory to hold the transformed string. The
1839 * calculations below try to find coefficients 'm' and 'b' for this
1840 * locale so that m*x + b equals how much space we need, given the size
1841 * of the input string in 'x'. If we calculate too small, we increase
1842 * the size as needed, and call strxfrm() again, but it is better to
1843 * get it right the first time to avoid wasted expensive string
1844 * transformations. */
1847 /* We use the string below to find how long the tranformation of it
1848 * is. Almost all locales are supersets of ASCII, or at least the
1849 * ASCII letters. We use all of them, half upper half lower,
1850 * because if we used fewer, we might hit just the ones that are
1851 * outliers in a particular locale. Most of the strings being
1852 * collated will contain a preponderance of letters, and even if
1853 * they are above-ASCII, they are likely to have the same number of
1854 * weight levels as the ASCII ones. It turns out that digits tend
1855 * to have fewer levels, and some punctuation has more, but those
1856 * are relatively sparse in text, and khw believes this gives a
1857 * reasonable result, but it could be changed if experience so
1859 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1860 char * x_longer; /* Transformed 'longer' */
1861 Size_t x_len_longer; /* Length of 'x_longer' */
1863 char * x_shorter; /* We also transform a substring of 'longer' */
1864 Size_t x_len_shorter;
1866 /* _mem_collxfrm() is used get the transformation (though here we
1867 * are interested only in its length). It is used because it has
1868 * the intelligence to handle all cases, but to work, it needs some
1869 * values of 'm' and 'b' to get it started. For the purposes of
1870 * this calculation we use a very conservative estimate of 'm' and
1871 * 'b'. This assumes a weight can be multiple bytes, enough to
1872 * hold any UV on the platform, and there are 5 levels, 4 weight
1873 * bytes, and a trailing NUL. */
1874 PL_collxfrm_base = 5;
1875 PL_collxfrm_mult = 5 * sizeof(UV);
1877 /* Find out how long the transformation really is */
1878 x_longer = _mem_collxfrm(longer,
1882 /* We avoid converting to UTF-8 in the
1883 * called function by telling it the
1884 * string is in UTF-8 if the locale is a
1885 * UTF-8 one. Since the string passed
1886 * here is invariant under UTF-8, we can
1887 * claim it's UTF-8 even though it isn't.
1889 PL_in_utf8_COLLATE_locale);
1892 /* Find out how long the transformation of a substring of 'longer'
1893 * is. Together the lengths of these transformations are
1894 * sufficient to calculate 'm' and 'b'. The substring is all of
1895 * 'longer' except the first character. This minimizes the chances
1896 * of being swayed by outliers */
1897 x_shorter = _mem_collxfrm(longer + 1,
1900 PL_in_utf8_COLLATE_locale);
1901 Safefree(x_shorter);
1903 /* If the results are nonsensical for this simple test, the whole
1904 * locale definition is suspect. Mark it so that locale collation
1905 * is not active at all for it. XXX Should we warn? */
1906 if ( x_len_shorter == 0
1907 || x_len_longer == 0
1908 || x_len_shorter >= x_len_longer)
1910 PL_collxfrm_mult = 0;
1911 PL_collxfrm_base = 0;
1914 SSize_t base; /* Temporary */
1916 /* We have both: m * strlen(longer) + b = x_len_longer
1917 * m * strlen(shorter) + b = x_len_shorter;
1918 * subtracting yields:
1919 * m * (strlen(longer) - strlen(shorter))
1920 * = x_len_longer - x_len_shorter
1921 * But we have set things up so that 'shorter' is 1 byte smaller
1922 * than 'longer'. Hence:
1923 * m = x_len_longer - x_len_shorter
1925 * But if something went wrong, make sure the multiplier is at
1928 if (x_len_longer > x_len_shorter) {
1929 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1932 PL_collxfrm_mult = 1;
1937 * but in case something has gone wrong, make sure it is
1939 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1944 /* Add 1 for the trailing NUL */
1945 PL_collxfrm_base = base + 1;
1950 if (DEBUG_L_TEST || debug_initialization) {
1951 PerlIO_printf(Perl_debug_log,
1952 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1954 " collate multipler=%zu, collate base=%zu\n",
1956 PL_in_utf8_COLLATE_locale,
1957 x_len_shorter, x_len_longer,
1958 PL_collxfrm_mult, PL_collxfrm_base);
1965 #endif /* USE_LOCALE_COLLATE */
1972 S_win32_setlocale(pTHX_ int category, const char* locale)
1974 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1975 * difference between the two unless the input locale is "", which normally
1976 * means on Windows to get the machine default, which is set via the
1977 * computer's "Regional and Language Options" (or its current equivalent).
1978 * In POSIX, it instead means to find the locale from the user's
1979 * environment. This routine changes the Windows behavior to first look in
1980 * the environment, and, if anything is found, use that instead of going to
1981 * the machine default. If there is no environment override, the machine
1982 * default is used, by calling the real setlocale() with "".
1984 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1985 * use the particular category's variable if set; otherwise to use the LANG
1988 bool override_LC_ALL = FALSE;
1992 if (locale && strEQ(locale, "")) {
1996 locale = PerlEnv_getenv("LC_ALL");
1998 if (category == LC_ALL) {
1999 override_LC_ALL = TRUE;
2005 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2006 if (category == categories[i]) {
2007 locale = PerlEnv_getenv(category_names[i]);
2012 locale = PerlEnv_getenv("LANG");
2028 result = setlocale(category, locale);
2029 DEBUG_L(STMT_START {
2031 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2032 setlocale_debug_string(category, locale, result));
2036 if (! override_LC_ALL) {
2040 /* Here the input category was LC_ALL, and we have set it to what is in the
2041 * LANG variable or the system default if there is no LANG. But these have
2042 * lower priority than the other LC_foo variables, so override it for each
2043 * one that is set. (If they are set to "", it means to use the same thing
2044 * we just set LC_ALL to, so can skip) */
2046 for (i = 0; i < LC_ALL_INDEX; i++) {
2047 result = PerlEnv_getenv(category_names[i]);
2048 if (result && strNE(result, "")) {
2049 setlocale(categories[i], result);
2050 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2052 setlocale_debug_string(categories[i], result, "not captured")));
2056 result = setlocale(LC_ALL, NULL);
2057 DEBUG_L(STMT_START {
2059 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2061 setlocale_debug_string(LC_ALL, NULL, result));
2072 =head1 Locale-related functions and macros
2074 =for apidoc Perl_setlocale
2076 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2077 taking the same parameters, and returning the same information, except that it
2078 returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2079 perl keeps that locale category as C<C>, changing it briefly during the
2080 operations where the underlying one is required.
2082 Another reason it isn't completely a drop-in replacement is that it is
2083 declared to return S<C<const char *>>, whereas the system setlocale omits the
2084 C<const>. (If it were being written today, plain setlocale would be declared
2085 const, since it is illegal to change the information it returns; doing so leads
2088 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2089 C<setlocale> can be completely ineffective on some platforms under some
2092 C<Perl_setlocale> should not be used to change the locale except on systems
2093 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2094 the system C<setlocale()> is ineffective, returning the wrong information, and
2095 failing to actually change the locale. C<Perl_setlocale>, however works
2096 properly in all circumstances.
2098 The return points to a per-thread static buffer, which is overwritten the next
2099 time C<Perl_setlocale> is called from the same thread.
2106 Perl_setlocale(const int category, const char * locale)
2108 /* This wraps POSIX::setlocale() */
2110 const char * retval;
2111 const char * newlocale;
2113 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2116 #ifdef USE_LOCALE_NUMERIC
2118 /* A NULL locale means only query what the current one is. We have the
2119 * LC_NUMERIC name saved, because we are normally switched into the C
2120 * locale for it. For an LC_ALL query, switch back to get the correct
2121 * results. All other categories don't require special handling */
2122 if (locale == NULL) {
2123 if (category == LC_NUMERIC) {
2125 /* We don't have to copy this return value, as it is a per-thread
2126 * variable, and won't change until a future setlocale */
2127 return PL_numeric_name;
2132 else if (category == LC_ALL) {
2133 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2142 retval = save_to_buffer(do_setlocale_r(category, locale),
2143 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2146 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2148 if (locale == NULL && category == LC_ALL) {
2149 RESTORE_LC_NUMERIC();
2154 DEBUG_L(PerlIO_printf(Perl_debug_log,
2155 "%s:%d: %s\n", __FILE__, __LINE__,
2156 setlocale_debug_string(category, locale, retval)));
2164 /* If locale == NULL, we are just querying the state */
2165 if (locale == NULL) {
2169 /* Now that have switched locales, we have to update our records to
2174 #ifdef USE_LOCALE_CTYPE
2181 #ifdef USE_LOCALE_COLLATE
2184 new_collate(retval);
2188 #ifdef USE_LOCALE_NUMERIC
2191 new_numeric(retval);
2199 /* LC_ALL updates all the things we care about. The values may not
2200 * be the same as 'retval', as the locale "" may have set things
2203 # ifdef USE_LOCALE_CTYPE
2205 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2206 new_ctype(newlocale);
2208 # endif /* USE_LOCALE_CTYPE */
2209 # ifdef USE_LOCALE_COLLATE
2211 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2212 new_collate(newlocale);
2215 # ifdef USE_LOCALE_NUMERIC
2217 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2218 new_numeric(newlocale);
2220 # endif /* USE_LOCALE_NUMERIC */
2231 PERL_STATIC_INLINE const char *
2232 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2234 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2235 * growing it if necessary */
2239 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2245 string_size = strlen(string) + offset + 1;
2247 if (*buf_size == 0) {
2248 Newx(*buf, string_size, char);
2249 *buf_size = string_size;
2251 else if (string_size > *buf_size) {
2252 Renew(*buf, string_size, char);
2253 *buf_size = string_size;
2256 Copy(string, *buf + offset, string_size - offset, char);
2262 =for apidoc Perl_langinfo
2264 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2265 taking the same C<item> parameter values, and returning the same information.
2266 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2267 of Perl's locale handling from your code, and can be used on systems that lack
2268 a native C<nl_langinfo>.
2276 The reason it isn't quite a drop-in replacement is actually an advantage. The
2277 only difference is that it returns S<C<const char *>>, whereas plain
2278 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2279 forbidden to write into the buffer. By declaring this C<const>, the compiler
2280 enforces this restriction, so if it is violated, you know at compilation time,
2281 rather than getting segfaults at runtime.
2285 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2286 without you having to write extra code. The reason for the extra code would be
2287 because these are from the C<LC_NUMERIC> locale category, which is normally
2288 kept set to the C locale by Perl, no matter what the underlying locale is
2289 supposed to be, and so to get the expected results, you have to temporarily
2290 toggle into the underlying locale, and later toggle back. (You could use plain
2291 C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2292 then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2293 C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2294 radix (decimal point) character to be a dot.)
2298 The system function it replaces can have its static return buffer trashed,
2299 not only by a subesequent call to that function, but by a C<freelocale>,
2300 C<setlocale>, or other locale change. The returned buffer of this function is
2301 not changed until the next call to it, so the buffer is never in a trashed
2306 Its return buffer is per-thread, so it also is never overwritten by a call to
2307 this function from another thread; unlike the function it replaces.
2311 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2312 as Windows, hence makes your code more portable. Of the fifty-some possible
2313 items specified by the POSIX 2008 standard,
2314 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2315 only two are completely unimplemented (though the loss of one of these is
2316 significant). It uses various techniques to recover the other items, including
2317 calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2318 in C89, so should be always be available. Later C<strftime()> versions have
2319 additional capabilities; C<""> is returned for those not available on your
2322 It is important to note that when called with an item that is recovered by
2323 using C<localeconv>, the buffer from any previous explicit call to
2324 C<localeconv> will be overwritten. This means you must save that buffer's
2325 contents if you need to access them after a call to this function. (But note
2326 that you might not want to be using C<localeconv()> directly anyway, because of
2327 issues like the ones listed in the second item of this list (above) for
2328 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2329 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2332 The details for those items which may deviate from what this emulation returns
2333 and what a native C<nl_langinfo()> would return are specified in
2338 When using C<Perl_langinfo> on systems that don't have a native
2339 C<nl_langinfo()>, you must
2341 #include "perl_langinfo.h"
2343 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2344 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2345 C<langinfo.h> would try to import into the namespace for code that doesn't need
2348 The original impetus for C<Perl_langinfo()> was so that code that needs to
2349 find out the current currency symbol, floating point radix character, or digit
2350 grouping separator can use, on all systems, the simpler and more
2351 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2352 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2353 is better to use the methods given in L<perlcall> to call
2354 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2361 #ifdef HAS_NL_LANGINFO
2362 Perl_langinfo(const nl_item item)
2364 Perl_langinfo(const int item)
2367 return my_nl_langinfo(item, TRUE);
2371 #ifdef HAS_NL_LANGINFO
2372 S_my_nl_langinfo(const nl_item item, bool toggle)
2374 S_my_nl_langinfo(const int item, bool toggle)
2378 const char * retval;
2380 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2381 * two items, and only if not already there */
2382 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2383 || PL_numeric_underlying))
2388 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2389 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2390 || ! defined(HAS_POSIX_2008_LOCALE)
2392 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2393 * for those items dependent on it. This must be copied to a buffer before
2394 * switching back, as some systems destroy the buffer when setlocale() is
2398 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2401 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2404 LOCALE_LOCK; /* Prevent interference from another thread executing
2405 this code section (the only call to nl_langinfo in
2409 /* Copy to a per-thread buffer, which is also one that won't be
2410 * destroyed by a subsequent setlocale(), such as the
2411 * RESTORE_LC_NUMERIC may do just below. */
2412 retval = save_to_buffer(nl_langinfo(item),
2413 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2418 RESTORE_LC_NUMERIC();
2422 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2425 bool do_free = FALSE;
2426 locale_t cur = uselocale((locale_t) 0);
2428 if (cur == LC_GLOBAL_LOCALE) {
2429 cur = duplocale(LC_GLOBAL_LOCALE);
2434 if (PL_underlying_numeric_obj) {
2435 cur = PL_underlying_numeric_obj;
2438 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2443 /* We have to save it to a buffer, because the freelocale() just below
2444 * can invalidate the internal one */
2445 retval = save_to_buffer(nl_langinfo_l(item, cur),
2446 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2455 if (strEQ(retval, "")) {
2456 if (item == YESSTR) {
2459 if (item == NOSTR) {
2466 #else /* Below, emulate nl_langinfo as best we can */
2470 # ifdef HAS_LOCALECONV
2472 const struct lconv* lc;
2474 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2477 # ifdef HAS_STRFTIME
2480 bool return_format = FALSE; /* Return the %format, not the value */
2481 const char * format;
2485 /* We copy the results to a per-thread buffer, even if not
2486 * multi-threaded. This is in part to simplify this code, and partly
2487 * because we need a buffer anyway for strftime(), and partly because a
2488 * call of localeconv() could otherwise wipe out the buffer, and the
2489 * programmer would not be expecting this, as this is a nl_langinfo()
2490 * substitute after all, so s/he might be thinking their localeconv()
2491 * is safe until another localeconv() call. */
2496 /* These 2 are unimplemented */
2498 case ERA: /* For use with strftime() %E modifier */
2503 /* We use only an English set, since we don't know any more */
2504 case YESEXPR: return "^[+1yY]";
2505 case YESSTR: return "yes";
2506 case NOEXPR: return "^[-0nN]";
2507 case NOSTR: return "no";
2510 # ifdef HAS_LOCALECONV
2514 /* We don't bother with localeconv_l() because any system that
2515 * has it is likely to also have nl_langinfo() */
2517 LOCALE_LOCK; /* Prevent interference with other threads
2518 using localeconv() */
2522 || ! lc->currency_symbol
2523 || strEQ("", lc->currency_symbol))
2529 /* Leave the first spot empty to be filled in below */
2530 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2531 &PL_langinfo_bufsize, 1);
2532 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2533 { /* khw couldn't figure out how the localedef specifications
2534 would show that the $ should replace the radix; this is
2535 just a guess as to how it might work.*/
2536 PL_langinfo_buf[0] = '.';
2538 else if (lc->p_cs_precedes) {
2539 PL_langinfo_buf[0] = '-';
2542 PL_langinfo_buf[0] = '+';
2552 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2555 LOCALE_LOCK; /* Prevent interference with other threads
2556 using localeconv() */
2563 temp = (item == RADIXCHAR)
2565 : lc->thousands_sep;
2571 retval = save_to_buffer(temp, &PL_langinfo_buf,
2572 &PL_langinfo_bufsize, 0);
2577 RESTORE_LC_NUMERIC();
2583 # ifdef HAS_STRFTIME
2585 /* These are defined by C89, so we assume that strftime supports
2586 * them, and so are returned unconditionally; they may not be what
2587 * the locale actually says, but should give good enough results
2588 * for someone using them as formats (as opposed to trying to parse
2589 * them to figure out what the locale says). The other format
2590 * items are actually tested to verify they work on the platform */
2591 case D_FMT: return "%x";
2592 case T_FMT: return "%X";
2593 case D_T_FMT: return "%c";
2595 /* These formats are only available in later strfmtime's */
2596 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2598 /* The rest can be gotten from most versions of strftime(). */
2599 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2600 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2602 case AM_STR: case PM_STR:
2603 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2604 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2605 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2606 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2607 case DAY_5: case DAY_6: case DAY_7:
2608 case MON_1: case MON_2: case MON_3: case MON_4:
2609 case MON_5: case MON_6: case MON_7: case MON_8:
2610 case MON_9: case MON_10: case MON_11: case MON_12:
2614 init_tm(&tm); /* Precaution against core dumps */
2618 tm.tm_year = 2017 - 1900;
2625 "panic: %s: %d: switch case: %d problem",
2626 __FILE__, __LINE__, item);
2627 NOT_REACHED; /* NOTREACHED */
2629 case PM_STR: tm.tm_hour = 18;
2634 case ABDAY_7: tm.tm_wday++;
2635 case ABDAY_6: tm.tm_wday++;
2636 case ABDAY_5: tm.tm_wday++;
2637 case ABDAY_4: tm.tm_wday++;
2638 case ABDAY_3: tm.tm_wday++;
2639 case ABDAY_2: tm.tm_wday++;
2644 case DAY_7: tm.tm_wday++;
2645 case DAY_6: tm.tm_wday++;
2646 case DAY_5: tm.tm_wday++;
2647 case DAY_4: tm.tm_wday++;
2648 case DAY_3: tm.tm_wday++;
2649 case DAY_2: tm.tm_wday++;
2654 case ABMON_12: tm.tm_mon++;
2655 case ABMON_11: tm.tm_mon++;
2656 case ABMON_10: tm.tm_mon++;
2657 case ABMON_9: tm.tm_mon++;
2658 case ABMON_8: tm.tm_mon++;
2659 case ABMON_7: tm.tm_mon++;
2660 case ABMON_6: tm.tm_mon++;
2661 case ABMON_5: tm.tm_mon++;
2662 case ABMON_4: tm.tm_mon++;
2663 case ABMON_3: tm.tm_mon++;
2664 case ABMON_2: tm.tm_mon++;
2669 case MON_12: tm.tm_mon++;
2670 case MON_11: tm.tm_mon++;
2671 case MON_10: tm.tm_mon++;
2672 case MON_9: tm.tm_mon++;
2673 case MON_8: tm.tm_mon++;
2674 case MON_7: tm.tm_mon++;
2675 case MON_6: tm.tm_mon++;
2676 case MON_5: tm.tm_mon++;
2677 case MON_4: tm.tm_mon++;
2678 case MON_3: tm.tm_mon++;
2679 case MON_2: tm.tm_mon++;
2686 return_format = TRUE;
2691 return_format = TRUE;
2696 return_format = TRUE;
2701 return_format = TRUE;
2706 format = "%Ow"; /* Find the alternate digit for 0 */
2710 /* We can't use my_strftime() because it doesn't look at
2712 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2715 /* A zero return means one of:
2716 * a) there wasn't enough space in PL_langinfo_buf
2717 * b) the format, like a plain %p, returns empty
2718 * c) it was an illegal format, though some
2719 * implementations of strftime will just return the
2720 * illegal format as a plain character sequence.
2722 * To quickly test for case 'b)', try again but precede
2723 * the format with a plain character. If that result is
2724 * still empty, the problem is either 'a)' or 'c)' */
2726 Size_t format_size = strlen(format) + 1;
2727 Size_t mod_size = format_size + 1;
2731 Newx(mod_format, mod_size, char);
2732 Newx(temp_result, PL_langinfo_bufsize, char);
2734 my_strlcpy(mod_format + 1, format, mod_size);
2735 len = strftime(temp_result,
2736 PL_langinfo_bufsize,
2738 Safefree(mod_format);
2739 Safefree(temp_result);
2741 /* If 'len' is non-zero, it means that we had a case like
2742 * %p which means the current locale doesn't use a.m. or
2743 * p.m., and that is valid */
2746 /* Here, still didn't work. If we get well beyond a
2747 * reasonable size, bail out to prevent an infinite
2750 if (PL_langinfo_bufsize > 100 * format_size) {
2751 *PL_langinfo_buf = '\0';
2754 /* Double the buffer size to retry; Add 1 in case
2755 * original was 0, so we aren't stuck at 0. */
2756 PL_langinfo_bufsize *= 2;
2757 PL_langinfo_bufsize++;
2758 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2766 /* Here, we got a result.
2768 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2769 * alternate format for wday 0. If the value is the same as
2770 * the normal 0, there isn't an alternate, so clear the buffer.
2772 if ( item == ALT_DIGITS
2773 && strEQ(PL_langinfo_buf, "0"))
2775 *PL_langinfo_buf = '\0';
2778 /* ALT_DIGITS is problematic. Experiments on it showed that
2779 * strftime() did not always work properly when going from
2780 * alt-9 to alt-10. Only a few locales have this item defined,
2781 * and in all of them on Linux that khw was able to find,
2782 * nl_langinfo() merely returned the alt-0 character, possibly
2783 * doubled. Most Unicode digits are in blocks of 10
2784 * consecutive code points, so that is sufficient information
2785 * for those scripts, as we can infer alt-1, alt-2, .... But
2786 * for a Japanese locale, a CJK ideographic 0 is returned, and
2787 * the CJK digits are not in code point order, so you can't
2788 * really infer anything. The localedef for this locale did
2789 * specify the succeeding digits, so that strftime() works
2790 * properly on them, without needing to infer anything. But
2791 * the nl_langinfo() return did not give sufficient information
2792 * for the caller to understand what's going on. So until
2793 * there is evidence that it should work differently, this
2794 * returns the alt-0 string for ALT_DIGITS.
2796 * wday was chosen because its range is all a single digit.
2797 * Things like tm_sec have two digits as the minimum: '00' */
2801 retval = PL_langinfo_buf;
2803 /* If to return the format, not the value, overwrite the buffer
2804 * with it. But some strftime()s will keep the original format
2805 * if illegal, so change those to "" */
2806 if (return_format) {
2807 if (strEQ(PL_langinfo_buf, format)) {
2808 *PL_langinfo_buf = '\0';
2811 retval = save_to_buffer(format, &PL_langinfo_buf,
2812 &PL_langinfo_bufsize, 0);
2830 * Initialize locale awareness.
2833 Perl_init_i18nl10n(pTHX_ int printwarn)
2837 * 0 if not to output warning when setup locale is bad
2838 * 1 if to output warning based on value of PERL_BADLANG
2839 * >1 if to output regardless of PERL_BADLANG
2842 * 1 = set ok or not applicable,
2843 * 0 = fallback to a locale of lower priority
2844 * -1 = fallback to all locales failed, not even to the C locale
2846 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2847 * set, debugging information is output.
2849 * This looks more complicated than it is, mainly due to the #ifdefs.
2851 * We try to set LC_ALL to the value determined by the environment. If
2852 * there is no LC_ALL on this platform, we try the individual categories we
2853 * know about. If this works, we are done.
2855 * But if it doesn't work, we have to do something else. We search the
2856 * environment variables ourselves instead of relying on the system to do
2857 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2858 * think there is one), and the ultimate fallback "C". This is all done in
2859 * the same loop as above to avoid duplicating code, but it makes things
2860 * more complex. The 'trial_locales' array is initialized with just one
2861 * element; it causes the behavior described in the paragraph above this to
2862 * happen. If that fails, we add elements to 'trial_locales', and do extra
2863 * loop iterations to cause the behavior described in this paragraph.
2865 * On Ultrix, the locale MUST come from the environment, so there is
2866 * preliminary code to set it. I (khw) am not sure that it is necessary,
2867 * and that this couldn't be folded into the loop, but barring any real
2868 * platforms to test on, it's staying as-is
2870 * A slight complication is that in embedded Perls, the locale may already
2871 * be set-up, and we don't want to get it from the normal environment
2872 * variables. This is handled by having a special environment variable
2873 * indicate we're in this situation. We simply set setlocale's 2nd
2874 * parameter to be a NULL instead of "". That indicates to setlocale that
2875 * it is not to change anything, but to return the current value,
2876 * effectively initializing perl's db to what the locale already is.
2878 * We play the same trick with NULL if a LC_ALL succeeds. We call
2879 * setlocale() on the individual categores with NULL to get their existing
2880 * values for our db, instead of trying to change them.
2887 PERL_UNUSED_ARG(printwarn);
2889 #else /* USE_LOCALE */
2892 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
2896 /* NULL uses the existing already set up locale */
2897 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2900 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2901 unsigned int trial_locales_count;
2902 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2903 const char * const lang = savepv(PerlEnv_getenv("LANG"));
2904 bool setlocale_failure = FALSE;
2907 /* A later getenv() could zap this, so only use here */
2908 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2910 const bool locwarn = (printwarn > 1
2912 && ( ! bad_lang_use_once
2914 /* disallow with "" or "0" */
2916 && strNE("0", bad_lang_use_once)))));
2918 /* setlocale() return vals; not copied so must be looked at immediately */
2919 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
2921 /* current locale for given category; should have been copied so aren't
2923 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
2927 /* In some systems you can find out the system default locale
2928 * and use that as the fallback locale. */
2929 # define SYSTEM_DEFAULT_LOCALE
2931 # ifdef SYSTEM_DEFAULT_LOCALE
2933 const char *system_default_locale = NULL;
2938 # define DEBUG_LOCALE_INIT(a,b,c)
2941 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
2943 # define DEBUG_LOCALE_INIT(category, locale, result) \
2945 if (debug_initialization) { \
2946 PerlIO_printf(Perl_debug_log, \
2948 __FILE__, __LINE__, \
2949 setlocale_debug_string(category, \
2955 /* Make sure the parallel arrays are properly set up */
2956 # ifdef USE_LOCALE_NUMERIC
2957 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2958 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
2959 # ifdef USE_POSIX_2008_LOCALE
2960 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
2963 # ifdef USE_LOCALE_CTYPE
2964 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2965 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
2966 # ifdef USE_POSIX_2008_LOCALE
2967 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
2970 # ifdef USE_LOCALE_COLLATE
2971 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2972 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
2973 # ifdef USE_POSIX_2008_LOCALE
2974 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
2977 # ifdef USE_LOCALE_TIME
2978 assert(categories[LC_TIME_INDEX] == LC_TIME);
2979 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2980 # ifdef USE_POSIX_2008_LOCALE
2981 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
2984 # ifdef USE_LOCALE_MESSAGES
2985 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2986 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
2987 # ifdef USE_POSIX_2008_LOCALE
2988 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
2991 # ifdef USE_LOCALE_MONETARY
2992 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2993 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
2994 # ifdef USE_POSIX_2008_LOCALE
2995 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
2998 # ifdef USE_LOCALE_ADDRESS
2999 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3000 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3001 # ifdef USE_POSIX_2008_LOCALE
3002 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3005 # ifdef USE_LOCALE_IDENTIFICATION
3006 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3007 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3008 # ifdef USE_POSIX_2008_LOCALE
3009 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3012 # ifdef USE_LOCALE_MEASUREMENT
3013 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3014 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3015 # ifdef USE_POSIX_2008_LOCALE
3016 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3019 # ifdef USE_LOCALE_PAPER
3020 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3021 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3022 # ifdef USE_POSIX_2008_LOCALE
3023 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3026 # ifdef USE_LOCALE_TELEPHONE
3027 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3028 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3029 # ifdef USE_POSIX_2008_LOCALE
3030 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3034 assert(categories[LC_ALL_INDEX] == LC_ALL);
3035 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3036 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3037 # ifdef USE_POSIX_2008_LOCALE
3038 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3041 # endif /* DEBUGGING */
3043 /* Initialize the cache of the program's UTF-8ness for the always known
3044 * locales C and POSIX */
3045 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3046 sizeof(PL_locale_utf8ness));
3048 # ifdef USE_THREAD_SAFE_LOCALE
3051 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3055 # if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3057 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3058 if (! PL_C_locale_obj) {
3059 Perl_croak_nocontext(
3060 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3062 if (DEBUG_Lv_TEST || debug_initialization) {
3063 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3068 PL_numeric_radix_sv = newSVpvs(".");
3070 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3072 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3073 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3076 # ifdef LOCALE_ENVIRON_REQUIRED
3079 * Ultrix setlocale(..., "") fails if there are no environment
3080 * variables from which to get a locale name.
3084 # error Ultrix without LC_ALL not implemented
3090 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3091 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3092 if (sl_result[LC_ALL_INDEX])
3095 setlocale_failure = TRUE;
3097 if (! setlocale_failure) {
3098 const char * locale_param;
3099 for (i = 0; i < LC_ALL_INDEX; i++) {
3100 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3103 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3104 if (! sl_result[i]) {
3105 setlocale_failure = TRUE;
3107 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3112 # endif /* LC_ALL */
3113 # endif /* LOCALE_ENVIRON_REQUIRED */
3115 /* We try each locale in the list until we get one that works, or exhaust
3116 * the list. Normally the loop is executed just once. But if setting the
3117 * locale fails, inside the loop we add fallback trials to the array and so
3118 * will execute the loop multiple times */
3119 trial_locales[0] = setlocale_init;
3120 trial_locales_count = 1;
3122 for (i= 0; i < trial_locales_count; i++) {
3123 const char * trial_locale = trial_locales[i];
3127 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3128 * when i==0, but I (khw) don't think that behavior makes much
3130 setlocale_failure = FALSE;
3132 # ifdef SYSTEM_DEFAULT_LOCALE
3133 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3135 /* On Windows machines, an entry of "" after the 0th means to use
3136 * the system default locale, which we now proceed to get. */
3137 if (strEQ(trial_locale, "")) {
3140 /* Note that this may change the locale, but we are going to do
3141 * that anyway just below */
3142 system_default_locale = do_setlocale_c(LC_ALL, "");
3143 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3145 /* Skip if invalid or if it's already on the list of locales to
3147 if (! system_default_locale) {
3148 goto next_iteration;
3150 for (j = 0; j < trial_locales_count; j++) {
3151 if (strEQ(system_default_locale, trial_locales[j])) {
3152 goto next_iteration;
3156 trial_locale = system_default_locale;
3159 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3161 # endif /* SYSTEM_DEFAULT_LOCALE */
3167 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3168 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3169 if (! sl_result[LC_ALL_INDEX]) {
3170 setlocale_failure = TRUE;
3173 /* Since LC_ALL succeeded, it should have changed all the other
3174 * categories it can to its value; so we massage things so that the
3175 * setlocales below just return their category's current values.
3176 * This adequately handles the case in NetBSD where LC_COLLATE may
3177 * not be defined for a locale, and setting it individually will
3178 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3179 * the POSIX locale. */
3180 trial_locale = NULL;
3183 # endif /* LC_ALL */
3185 if (! setlocale_failure) {
3187 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3189 = savepv(do_setlocale_r(categories[j], trial_locale));
3190 if (! curlocales[j]) {
3191 setlocale_failure = TRUE;
3193 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3196 if (! setlocale_failure) { /* All succeeded */
3197 break; /* Exit trial_locales loop */
3201 /* Here, something failed; will need to try a fallback. */
3207 if (locwarn) { /* Output failure info only on the first one */
3211 PerlIO_printf(Perl_error_log,
3212 "perl: warning: Setting locale failed.\n");
3214 # else /* !LC_ALL */
3216 PerlIO_printf(Perl_error_log,
3217 "perl: warning: Setting locale failed for the categories:\n\t");
3219 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3220 if (! curlocales[j]) {
3221 PerlIO_printf(Perl_error_log, category_names[j]);
3224 Safefree(curlocales[j]);
3228 # endif /* LC_ALL */
3230 PerlIO_printf(Perl_error_log,
3231 "perl: warning: Please check that your locale settings:\n");
3235 PerlIO_printf(Perl_error_log,
3236 "\tLANGUAGE = %c%s%c,\n",
3237 language ? '"' : '(',
3238 language ? language : "unset",
3239 language ? '"' : ')');
3242 PerlIO_printf(Perl_error_log,
3243 "\tLC_ALL = %c%s%c,\n",
3245 lc_all ? lc_all : "unset",
3246 lc_all ? '"' : ')');
3248 # if defined(USE_ENVIRON_ARRAY)
3253 /* Look through the environment for any variables of the
3254 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3255 * already handled above. These are assumed to be locale
3256 * settings. Output them and their values. */
3257 for (e = environ; *e; e++) {
3258 const STRLEN prefix_len = sizeof("LC_") - 1;
3261 if ( strBEGINs(*e, "LC_")
3262 && ! strBEGINs(*e, "LC_ALL=")
3263 && (uppers_len = strspn(*e + prefix_len,
3264 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3265 && ((*e)[prefix_len + uppers_len] == '='))
3267 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3268 (int) (prefix_len + uppers_len), *e,
3269 *e + prefix_len + uppers_len + 1);
3276 PerlIO_printf(Perl_error_log,
3277 "\t(possibly more locale environment variables)\n");
3281 PerlIO_printf(Perl_error_log,
3282 "\tLANG = %c%s%c\n",
3284 lang ? lang : "unset",
3287 PerlIO_printf(Perl_error_log,
3288 " are supported and installed on your system.\n");
3291 /* Calculate what fallback locales to try. We have avoided this
3292 * until we have to, because failure is quite unlikely. This will
3293 * usually change the upper bound of the loop we are in.
3295 * Since the system's default way of setting the locale has not
3296 * found one that works, We use Perl's defined ordering: LC_ALL,
3297 * LANG, and the C locale. We don't try the same locale twice, so
3298 * don't add to the list if already there. (On POSIX systems, the
3299 * LC_ALL element will likely be a repeat of the 0th element "",
3300 * but there's no harm done by doing it explicitly.
3302 * Note that this tries the LC_ALL environment variable even on
3303 * systems which have no LC_ALL locale setting. This may or may
3304 * not have been originally intentional, but there's no real need
3305 * to change the behavior. */
3307 for (j = 0; j < trial_locales_count; j++) {
3308 if (strEQ(lc_all, trial_locales[j])) {
3312 trial_locales[trial_locales_count++] = lc_all;
3317 for (j = 0; j < trial_locales_count; j++) {
3318 if (strEQ(lang, trial_locales[j])) {
3322 trial_locales[trial_locales_count++] = lang;
3326 # if defined(WIN32) && defined(LC_ALL)
3328 /* For Windows, we also try the system default locale before "C".
3329 * (If there exists a Windows without LC_ALL we skip this because
3330 * it gets too complicated. For those, the "C" is the next
3331 * fallback possibility). The "" is the same as the 0th element of
3332 * the array, but the code at the loop above knows to treat it
3333 * differently when not the 0th */
3334 trial_locales[trial_locales_count++] = "";
3338 for (j = 0; j < trial_locales_count; j++) {
3339 if (strEQ("C", trial_locales[j])) {
3343 trial_locales[trial_locales_count++] = "C";
3346 } /* end of first time through the loop */
3354 } /* end of looping through the trial locales */
3356 if (ok < 1) { /* If we tried to fallback */
3358 if (! setlocale_failure) { /* fallback succeeded */
3359 msg = "Falling back to";
3361 else { /* fallback failed */
3364 /* We dropped off the end of the loop, so have to decrement i to
3365 * get back to the value the last time through */
3369 msg = "Failed to fall back to";
3371 /* To continue, we should use whatever values we've got */
3373 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3374 Safefree(curlocales[j]);
3375 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3376 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3381 const char * description;
3382 const char * name = "";
3383 if (strEQ(trial_locales[i], "C")) {
3384 description = "the standard locale";
3388 # ifdef SYSTEM_DEFAULT_LOCALE
3390 else if (strEQ(trial_locales[i], "")) {
3391 description = "the system default locale";
3392 if (system_default_locale) {
3393 name = system_default_locale;
3397 # endif /* SYSTEM_DEFAULT_LOCALE */
3400 description = "a fallback locale";
3401 name = trial_locales[i];
3403 if (name && strNE(name, "")) {
3404 PerlIO_printf(Perl_error_log,
3405 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3408 PerlIO_printf(Perl_error_log,
3409 "perl: warning: %s %s.\n", msg, description);
3412 } /* End of tried to fallback */
3414 /* Done with finding the locales; update our records */
3416 # ifdef USE_LOCALE_CTYPE
3418 new_ctype(curlocales[LC_CTYPE_INDEX]);
3421 # ifdef USE_LOCALE_COLLATE
3423 new_collate(curlocales[LC_COLLATE_INDEX]);
3426 # ifdef USE_LOCALE_NUMERIC
3428 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3432 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3434 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3436 /* This caches whether each category's locale is UTF-8 or not. This
3437 * may involve changing the locale. It is ok to do this at
3438 * initialization time before any threads have started, but not later
3439 * unless thread-safe operations are used.
3440 * Caching means that if the program heeds our dictate not to change
3441 * locales in threaded applications, this data will remain valid, and
3442 * it may get queried without having to change locales. If the
3443 * environment is such that all categories have the same locale, this
3444 * isn't needed, as the code will not change the locale; but this
3445 * handles the uncommon case where the environment has disparate
3446 * locales for the categories */
3447 (void) _is_cur_LC_category_utf8(categories[i]);
3451 Safefree(curlocales[i]);
3454 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3456 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3457 * locale is UTF-8. The call to new_ctype() just above has already
3458 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3459 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3460 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3461 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3462 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3464 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3465 This is an alternative to using the -C command line switch
3466 (the -C if present will override this). */
3468 const char *p = PerlEnv_getenv("PERL_UNICODE");
3469 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3470 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3484 #endif /* USE_LOCALE */
3487 /* So won't continue to output stuff */
3488 DEBUG_INITIALIZATION_set(FALSE);
3495 #ifdef USE_LOCALE_COLLATE
3498 Perl__mem_collxfrm(pTHX_ const char *input_string,
3499 STRLEN len, /* Length of 'input_string' */
3500 STRLEN *xlen, /* Set to length of returned string
3501 (not including the collation index
3503 bool utf8 /* Is the input in UTF-8? */
3507 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3508 * differences. First, it handles embedded NULs. Second, it allocates a bit
3509 * more memory than needed for the transformed data itself. The real
3510 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3511 * the length of that, and doesn't include the collation index size.
3512 * Please see sv_collxfrm() to see how this is used. */
3514 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3516 char * s = (char *) input_string;
3517 STRLEN s_strlen = strlen(input_string);
3519 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3520 STRLEN length_in_chars;
3521 bool first_time = TRUE; /* Cleared after first loop iteration */
3523 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3525 /* Must be NUL-terminated */
3526 assert(*(input_string + len) == '\0');
3528 /* If this locale has defective collation, skip */
3529 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3530 DEBUG_L(PerlIO_printf(Perl_debug_log,
3531 "_mem_collxfrm: locale's collation is defective\n"));
3535 /* Replace any embedded NULs with the control that sorts before any others.
3536 * This will give as good as possible results on strings that don't
3537 * otherwise contain that character, but otherwise there may be
3538 * less-than-perfect results with that character and NUL. This is
3539 * unavoidable unless we replace strxfrm with our own implementation. */
3540 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3544 STRLEN sans_nuls_len;
3545 int try_non_controls;
3546 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3547 making sure 2nd byte is NUL.
3549 STRLEN this_replacement_len;
3551 /* If we don't know what non-NUL control character sorts lowest for
3552 * this locale, find it */
3553 if (PL_strxfrm_NUL_replacement == '\0') {
3555 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3556 includes the collation index
3559 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3561 /* Unlikely, but it may be that no control will work to replace
3562 * NUL, in which case we instead look for any character. Controls
3563 * are preferred because collation order is, in general, context
3564 * sensitive, with adjoining characters affecting the order, and
3565 * controls are less likely to have such interactions, allowing the
3566 * NUL-replacement to stand on its own. (Another way to look at it
3567 * is to imagine what would happen if the NUL were replaced by a
3568 * combining character; it wouldn't work out all that well.) */
3569 for (try_non_controls = 0;
3570 try_non_controls < 2;
3573 /* Look through all legal code points (NUL isn't) */
3574 for (j = 1; j < 256; j++) {
3575 char * x; /* j's xfrm plus collation index */
3576 STRLEN x_len; /* length of 'x' */
3577 STRLEN trial_len = 1;
3578 char cur_source[] = { '\0', '\0' };
3580 /* Skip non-controls the first time through the loop. The
3581 * controls in a UTF-8 locale are the L1 ones */
3582 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3589 /* Create a 1-char string of the current code point */
3590 cur_source[0] = (char) j;
3592 /* Then transform it */
3593 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3594 0 /* The string is not in UTF-8 */);
3596 /* Ignore any character that didn't successfully transform.
3602 /* If this character's transformation is lower than
3603 * the current lowest, this one becomes the lowest */
3604 if ( cur_min_x == NULL
3605 || strLT(x + COLLXFRM_HDR_LEN,
3606 cur_min_x + COLLXFRM_HDR_LEN))
3608 PL_strxfrm_NUL_replacement = j;
3614 } /* end of loop through all 255 characters */
3616 /* Stop looking if found */
3621 /* Unlikely, but possible, if there aren't any controls that
3622 * work in the locale, repeat the loop, looking for any
3623 * character that works */
3624 DEBUG_L(PerlIO_printf(Perl_debug_log,
3625 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3626 } /* End of loop to try first the controls, then any char */
3629 DEBUG_L(PerlIO_printf(Perl_debug_log,
3630 "_mem_collxfrm: Couldn't find any character to replace"
3631 " embedded NULs in locale %s with", PL_collation_name));
3635 DEBUG_L(PerlIO_printf(Perl_debug_log,
3636 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3637 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3639 Safefree(cur_min_x);
3640 } /* End of determining the character that is to replace NULs */
3642 /* If the replacement is variant under UTF-8, it must match the
3643 * UTF8-ness of the original */
3644 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3645 this_replacement_char[0] =
3646 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3647 this_replacement_char[1] =
3648 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3649 this_replacement_len = 2;
3652 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3653 /* this_replacement_char[1] = '\0' was done at initialization */
3654 this_replacement_len = 1;
3657 /* The worst case length for the replaced string would be if every
3658 * character in it is NUL. Multiply that by the length of each
3659 * replacement, and allow for a trailing NUL */
3660 sans_nuls_len = (len * this_replacement_len) + 1;
3661 Newx(sans_nuls, sans_nuls_len, char);
3664 /* Replace each NUL with the lowest collating control. Loop until have
3665 * exhausted all the NULs */
3666 while (s + s_strlen < e) {
3667 my_strlcat(sans_nuls, s, sans_nuls_len);
3669 /* Do the actual replacement */
3670 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3672 /* Move past the input NUL */
3674 s_strlen = strlen(s);
3677 /* And add anything that trails the final NUL */
3678 my_strlcat(sans_nuls, s, sans_nuls_len);
3680 /* Switch so below we transform this modified string */
3683 } /* End of replacing NULs */
3685 /* Make sure the UTF8ness of the string and locale match */
3686 if (utf8 != PL_in_utf8_COLLATE_locale) {
3687 /* XXX convert above Unicode to 10FFFF? */
3688 const char * const t = s; /* Temporary so we can later find where the
3691 /* Here they don't match. Change the string's to be what the locale is
3694 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3695 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3698 else { /* locale is not UTF-8; but input is; downgrade the input */
3700 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3702 /* If the downgrade was successful we are done, but if the input
3703 * contains things that require UTF-8 to represent, have to do
3704 * damage control ... */
3705 if (UNLIKELY(utf8)) {
3707 /* What we do is construct a non-UTF-8 string with
3708 * 1) the characters representable by a single byte converted
3709 * to be so (if necessary);
3710 * 2) and the rest converted to collate the same as the
3711 * highest collating representable character. That makes
3712 * them collate at the end. This is similar to how we
3713 * handle embedded NULs, but we use the highest collating
3714 * code point instead of the smallest. Like the NUL case,
3715 * this isn't perfect, but is the best we can reasonably
3716 * do. Every above-255 code point will sort the same as
3717 * the highest-sorting 0-255 code point. If that code
3718 * point can combine in a sequence with some other code
3719 * points for weight calculations, us changing something to
3720 * be it can adversely affect the results. But in most
3721 * cases, it should work reasonably. And note that this is
3722 * really an illegal situation: using code points above 255
3723 * on a locale where only 0-255 are valid. If two strings
3724 * sort entirely equal, then the sort order for the
3725 * above-255 code points will be in code point order. */
3729 /* If we haven't calculated the code point with the maximum
3730 * collating order for this locale, do so now */
3731 if (! PL_strxfrm_max_cp) {
3734 /* The current transformed string that collates the
3735 * highest (except it also includes the prefixed collation
3737 char * cur_max_x = NULL;
3739 /* Look through all legal code points (NUL isn't) */
3740 for (j = 1; j < 256; j++) {
3743 char cur_source[] = { '\0', '\0' };
3745 /* Create a 1-char string of the current code point */
3746 cur_source[0] = (char) j;
3748 /* Then transform it */
3749 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3751 /* If something went wrong (which it shouldn't), just
3752 * ignore this code point */
3757 /* If this character's transformation is higher than
3758 * the current highest, this one becomes the highest */
3759 if ( cur_max_x == NULL
3760 || strGT(x + COLLXFRM_HDR_LEN,
3761 cur_max_x + COLLXFRM_HDR_LEN))
3763 PL_strxfrm_max_cp = j;
3772 DEBUG_L(PerlIO_printf(Perl_debug_log,
3773 "_mem_collxfrm: Couldn't find any character to"
3774 " replace above-Latin1 chars in locale %s with",
3775 PL_collation_name));
3779 DEBUG_L(PerlIO_printf(Perl_debug_log,
3780 "_mem_collxfrm: highest 1-byte collating character"
3781 " in locale %s is 0x%02X\n",
3783 PL_strxfrm_max_cp));
3785 Safefree(cur_max_x);
3788 /* Here we know which legal code point collates the highest.
3789 * We are ready to construct the non-UTF-8 string. The length
3790 * will be at least 1 byte smaller than the input string
3791 * (because we changed at least one 2-byte character into a
3792 * single byte), but that is eaten up by the trailing NUL */
3798 char * e = (char *) t + len;
3800 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3802 if (UTF8_IS_INVARIANT(cur_char)) {
3805 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
3806 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3808 else { /* Replace illegal cp with highest collating
3810 s[d++] = PL_strxfrm_max_cp;
3814 Renew(s, d, char); /* Free up unused space */
3819 /* Here, we have constructed a modified version of the input. It could
3820 * be that we already had a modified copy before we did this version.
3821 * If so, that copy is no longer needed */
3822 if (t != input_string) {
3827 length_in_chars = (utf8)
3828 ? utf8_length((U8 *) s, (U8 *) s + len)
3831 /* The first element in the output is the collation id, used by
3832 * sv_collxfrm(); then comes the space for the transformed string. The
3833 * equation should give us a good estimate as to how much is needed */
3834 xAlloc = COLLXFRM_HDR_LEN
3836 + (PL_collxfrm_mult * length_in_chars);
3837 Newx(xbuf, xAlloc, char);
3838 if (UNLIKELY(! xbuf)) {
3839 DEBUG_L(PerlIO_printf(Perl_debug_log,
3840 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
3844 /* Store the collation id */
3845 *(U32*)xbuf = PL_collation_ix;
3847 /* Then the transformation of the input. We loop until successful, or we
3851 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
3853 /* If the transformed string occupies less space than we told strxfrm()
3854 * was available, it means it successfully transformed the whole
3856 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
3858 /* Some systems include a trailing NUL in the returned length.
3859 * Ignore it, using a loop in case multiple trailing NULs are
3862 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3867 /* If the first try didn't get it, it means our prediction was low.
3868 * Modify the coefficients so that we predict a larger value in any
3869 * future transformations */
3871 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3872 STRLEN computed_guess = PL_collxfrm_base
3873 + (PL_collxfrm_mult * length_in_chars);
3875 /* On zero-length input, just keep current slope instead of
3877 const STRLEN new_m = (length_in_chars != 0)
3878 ? needed / length_in_chars
3881 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3882 "%s: %d: initial size of %zu bytes for a length "
3883 "%zu string was insufficient, %zu needed\n",
3885 computed_guess, length_in_chars, needed));
3887 /* If slope increased, use it, but discard this result for
3888 * length 1 strings, as we can't be sure that it's a real slope
3890 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
3894 STRLEN old_m = PL_collxfrm_mult;
3895 STRLEN old_b = PL_collxfrm_base;
3899 PL_collxfrm_mult = new_m;
3900 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3901 computed_guess = PL_collxfrm_base
3902 + (PL_collxfrm_mult * length_in_chars);
3903 if (computed_guess < needed) {
3904 PL_collxfrm_base += needed - computed_guess;
3907 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3908 "%s: %d: slope is now %zu; was %zu, base "
3909 "is now %zu; was %zu\n",
3911 PL_collxfrm_mult, old_m,
3912 PL_collxfrm_base, old_b));
3914 else { /* Slope didn't change, but 'b' did */
3915 const STRLEN new_b = needed
3918 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3919 "%s: %d: base is now %zu; was %zu\n",
3921 new_b, PL_collxfrm_base));
3922 PL_collxfrm_base = new_b;
3929 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3930 DEBUG_L(PerlIO_printf(Perl_debug_log,
3931 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3932 *xlen, PERL_INT_MAX));
3936 /* A well-behaved strxfrm() returns exactly how much space it needs
3937 * (usually not including the trailing NUL) when it fails due to not
3938 * enough space being provided. Assume that this is the case unless
3939 * it's been proven otherwise */
3940 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
3941 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
3943 else { /* Here, either:
3944 * 1) The strxfrm() has previously shown bad behavior; or
3945 * 2) It isn't the first time through the loop, which means
3946 * that the strxfrm() is now showing bad behavior, because
3947 * we gave it what it said was needed in the previous
3948 * iteration, and it came back saying it needed still more.
3949 * (Many versions of cygwin fit this. When the buffer size
3950 * isn't sufficient, they return the input size instead of
3951 * how much is needed.)
3952 * Increase the buffer size by a fixed percentage and try again.
3954 xAlloc += (xAlloc / 4) + 1;
3955 PL_strxfrm_is_behaved = FALSE;
3959 if (DEBUG_Lv_TEST || debug_initialization) {
3960 PerlIO_printf(Perl_debug_log,
3961 "_mem_collxfrm required more space than previously calculated"
3962 " for locale %s, trying again with new guess=%d+%zu\n",
3963 PL_collation_name, (int) COLLXFRM_HDR_LEN,
3964 xAlloc - COLLXFRM_HDR_LEN);
3971 Renew(xbuf, xAlloc, char);
3972 if (UNLIKELY(! xbuf)) {
3973 DEBUG_L(PerlIO_printf(Perl_debug_log,
3974 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
3984 if (DEBUG_Lv_TEST || debug_initialization) {
3986 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
3987 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
3988 PerlIO_printf(Perl_debug_log, "%s\n",
3989 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
3995 /* Free up unneeded space; retain ehough for trailing NUL */
3996 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
3998 if (s != input_string) {
4006 if (s != input_string) {
4013 if (DEBUG_Lv_TEST || debug_initialization) {
4014 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4025 S_print_collxfrm_input_and_return(pTHX_
4026 const char * const s,
4027 const char * const e,
4028 const STRLEN * const xlen,
4032 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4034 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4035 (UV)PL_collation_ix);
4037 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4040 PerlIO_printf(Perl_debug_log, "NULL");
4042 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4044 print_bytes_for_locale(s, e, is_utf8);
4046 PerlIO_printf(Perl_debug_log, "'\n");
4050 S_print_bytes_for_locale(pTHX_
4051 const char * const s,
4052 const char * const e,
4056 bool prev_was_printable = TRUE;
4057 bool first_time = TRUE;
4059 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4063 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4066 if (! prev_was_printable) {
4067 PerlIO_printf(Perl_debug_log, " ");
4069 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4070 prev_was_printable = TRUE;
4074 PerlIO_printf(Perl_debug_log, " ");
4076 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4077 prev_was_printable = FALSE;
4079 t += (is_utf8) ? UTF8SKIP(t) : 1;
4084 # endif /* #ifdef DEBUGGING */
4085 #endif /* USE_LOCALE_COLLATE */
4090 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4092 /* Changes the locale for LC_'switch_category" to that of
4093 * LC_'template_category', if they aren't already the same. If not NULL,
4094 * 'template_locale' is the locale that 'template_category' is in.
4096 * Returns a copy of the name of the original locale for 'switch_category'
4097 * so can be switched back to with the companion function
4098 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4100 char * restore_to_locale = NULL;
4102 if (switch_category == template_category) { /* No changes needed */
4106 /* Find the original locale of the category we may need to change, so that
4107 * it can be restored to later */
4108 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4110 if (! restore_to_locale) {
4112 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4113 __FILE__, __LINE__, category_name(switch_category), errno);
4116 /* If the locale of the template category wasn't passed in, find it now */
4117 if (template_locale == NULL) {
4118 template_locale = do_setlocale_r(template_category, NULL);
4119 if (! template_locale) {
4121 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4122 __FILE__, __LINE__, category_name(template_category), errno);
4126 /* It the locales are the same, there's nothing to do */
4127 if (strEQ(restore_to_locale, template_locale)) {
4128 Safefree(restore_to_locale);
4130 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4131 category_name(switch_category), restore_to_locale));
4136 /* Finally, change the locale to the template one */
4137 if (! do_setlocale_r(switch_category, template_locale)) {
4139 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4140 __FILE__, __LINE__, category_name(switch_category),
4141 template_locale, errno);
4144 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4145 category_name(switch_category), template_locale));
4147 return restore_to_locale;
4151 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4153 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4154 * copy that will be freed by this function), or do nothing if the latter
4155 * parameter is NULL */
4157 if (original_locale == NULL) {
4161 if (! do_setlocale_r(category, original_locale)) {
4163 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4165 category_name(category), original_locale, errno);
4168 Safefree(original_locale);
4172 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4174 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4175 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4176 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4177 * could give the wrong result. The result will very likely be correct for
4178 * languages that have commonly used non-ASCII characters, but for notably
4179 * English, it comes down to if the locale's name ends in something like
4180 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4182 * If the platform is early C89, not containing mbtowc(), or we are
4183 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4184 * These work very well for non-Latin locales or those whose currency
4185 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4186 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4187 * on the name containing UTF-8 or not. */
4189 /* Name of current locale corresponding to the input category */
4190 const char *save_input_locale = NULL;
4192 bool is_utf8 = FALSE; /* The return value */
4194 /* The variables below are for the cache of previous lookups using this
4195 * function. The cache is a C string, described at the definition for
4196 * 'C_and_POSIX_utf8ness'.
4198 * The first part of the cache is fixed, for the C and POSIX locales. The
4199 * varying part starts just after them. */
4200 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4202 Size_t utf8ness_cache_size; /* Size of the varying portion */
4203 Size_t input_name_len; /* Length in bytes of save_input_locale */
4204 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4205 the name in the cache */
4206 char * delimited; /* The name plus the delimiters used to store
4208 char * name_pos; /* position of 'delimited' in the cache, or 0
4214 assert(category != LC_ALL);
4218 /* Get the desired category's locale */
4219 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4220 if (! save_input_locale) {
4222 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4223 __FILE__, __LINE__, category_name(category), errno);
4226 DEBUG_L(PerlIO_printf(Perl_debug_log,
4227 "Current locale for %s is %s\n",
4228 category_name(category), save_input_locale));
4230 input_name_len = strlen(save_input_locale);
4232 /* In our cache, each name is accompanied by two delimiters and a single
4234 input_name_len_with_overhead = input_name_len + 3;
4236 /* Allocate and populate space for a copy of the name surrounded by the
4238 Newx(delimited, input_name_len_with_overhead, char);
4239 delimited[0] = UTF8NESS_SEP[0];
4240 Copy(save_input_locale, delimited + 1, input_name_len, char);
4241 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4242 delimited[input_name_len+2] = '\0';
4244 /* And see if that is in the cache */
4245 name_pos = instr(PL_locale_utf8ness, delimited);
4247 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4251 if (DEBUG_Lv_TEST || debug_initialization) {
4252 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4253 save_input_locale, is_utf8);
4258 /* And, if not already in that position, move it to the beginning of
4259 * the non-constant portion of the list, since it is the most recently
4260 * used. (We don't have to worry about overflow, since just moving
4261 * existing names around) */
4262 if (name_pos > utf8ness_cache) {
4263 Move(utf8ness_cache,
4264 utf8ness_cache + input_name_len_with_overhead,
4265 name_pos - utf8ness_cache, char);
4268 input_name_len_with_overhead - 1, char);
4269 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4272 Safefree(delimited);
4273 Safefree(save_input_locale);
4277 /* Here we don't have stored the utf8ness for the input locale. We have to
4280 # if defined(USE_LOCALE_CTYPE) \
4281 && ( defined(HAS_NL_LANGINFO) \
4282 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4285 const char *original_ctype_locale
4286 = switch_category_locale_to_template(LC_CTYPE,
4290 /* Here the current LC_CTYPE is set to the locale of the category whose
4291 * information is desired. This means that nl_langinfo() and mbtowc()
4292 * should give the correct results */
4294 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4295 calling the functions if we have this */
4297 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4298 * Unicode code point. */
4300 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4301 __FILE__, __LINE__, (int) MB_CUR_MAX));
4302 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
https://perl5.git.perl.org / perl5.git / blob