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 #ifndef TS_W32_BROKEN_LOCALECONV
1349 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1350 * them. But on broken Windows systems calling my_nl_langinfo() for
1351 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1352 * and just always change the locale if not C nor POSIX on those systems */
1353 if (! PL_numeric_standard) {
1354 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1355 FALSE /* Don't toggle locale */ ))
1356 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1361 /* Save the new name if it isn't the same as the previous one, if any */
1362 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1363 Safefree(PL_numeric_name);
1364 PL_numeric_name = save_newnum;
1367 Safefree(save_newnum);
1370 PL_numeric_underlying_is_standard = PL_numeric_standard;
1372 # ifdef HAS_POSIX_2008_LOCALE
1374 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1376 PL_underlying_numeric_obj);
1380 if (DEBUG_L_TEST || debug_initialization) {
1381 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1384 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1385 * have to worry about the radix being a non-dot. (Core operations that
1386 * need the underlying locale change to it temporarily). */
1387 if (PL_numeric_standard) {
1388 set_numeric_radix(0);
1391 set_numeric_standard();
1394 #endif /* USE_LOCALE_NUMERIC */
1399 Perl_set_numeric_standard(pTHX)
1402 #ifdef USE_LOCALE_NUMERIC
1404 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1405 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1406 * macro avoids calling this routine if toggling isn't necessary according
1407 * to our records (which could be wrong if some XS code has changed the
1408 * locale behind our back) */
1410 do_setlocale_c(LC_NUMERIC, "C");
1411 PL_numeric_standard = TRUE;
1412 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1413 set_numeric_radix(0);
1417 if (DEBUG_L_TEST || debug_initialization) {
1418 PerlIO_printf(Perl_debug_log,
1419 "LC_NUMERIC locale now is standard C\n");
1423 #endif /* USE_LOCALE_NUMERIC */
1428 Perl_set_numeric_underlying(pTHX)
1431 #ifdef USE_LOCALE_NUMERIC
1433 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1434 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1435 * instead of calling this directly. The macro avoids calling this routine
1436 * if toggling isn't necessary according to our records (which could be
1437 * wrong if some XS code has changed the locale behind our back) */
1439 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1440 PL_numeric_standard = PL_numeric_underlying_is_standard;
1441 PL_numeric_underlying = TRUE;
1442 set_numeric_radix(! PL_numeric_standard);
1446 if (DEBUG_L_TEST || debug_initialization) {
1447 PerlIO_printf(Perl_debug_log,
1448 "LC_NUMERIC locale now is %s\n",
1453 #endif /* USE_LOCALE_NUMERIC */
1458 * Set up for a new ctype locale.
1461 S_new_ctype(pTHX_ const char *newctype)
1464 #ifndef USE_LOCALE_CTYPE
1466 PERL_ARGS_ASSERT_NEW_CTYPE;
1467 PERL_UNUSED_ARG(newctype);
1468 PERL_UNUSED_CONTEXT;
1472 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1473 * core Perl this and that 'newctype' is the name of the new locale.
1475 * This function sets up the folding arrays for all 256 bytes, assuming
1476 * that tofold() is tolc() since fold case is not a concept in POSIX,
1478 * Any code changing the locale (outside this file) should use
1479 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1480 * this function should be called directly only from this file and from
1481 * POSIX::setlocale() */
1486 /* Don't check for problems if we are suppressing the warnings */
1487 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1489 PERL_ARGS_ASSERT_NEW_CTYPE;
1491 /* We will replace any bad locale warning with 1) nothing if the new one is
1492 * ok; or 2) a new warning for the bad new locale */
1493 if (PL_warn_locale) {
1494 SvREFCNT_dec_NN(PL_warn_locale);
1495 PL_warn_locale = NULL;
1498 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1500 /* A UTF-8 locale gets standard rules. But note that code still has to
1501 * handle this specially because of the three problematic code points */
1502 if (PL_in_utf8_CTYPE_locale) {
1503 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1506 /* We don't populate the other lists if a UTF-8 locale, but do check that
1507 * everything works as expected, unless checking turned off */
1508 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1509 /* Assume enough space for every character being bad. 4 spaces each
1510 * for the 94 printable characters that are output like "'x' "; and 5
1511 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1513 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1514 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1516 unsigned int bad_count = 0; /* Count of bad characters */
1518 for (i = 0; i < 256; i++) {
1519 if (! PL_in_utf8_CTYPE_locale) {
1521 PL_fold_locale[i] = (U8) tolower(i);
1522 else if (islower(i))
1523 PL_fold_locale[i] = (U8) toupper(i);
1525 PL_fold_locale[i] = (U8) i;
1528 /* If checking for locale problems, see if the native ASCII-range
1529 * printables plus \n and \t are in their expected categories in
1530 * the new locale. If not, this could mean big trouble, upending
1531 * Perl's and most programs' assumptions, like having a
1532 * metacharacter with special meaning become a \w. Fortunately,
1533 * it's very rare to find locales that aren't supersets of ASCII
1534 * nowadays. It isn't a problem for most controls to be changed
1535 * into something else; we check only \n and \t, though perhaps \r
1536 * could be an issue as well. */
1537 if ( check_for_problems
1538 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1540 bool is_bad = FALSE;
1541 char name[3] = { '\0' };
1543 /* Convert the name into a string */
1548 else if (i == '\n') {
1549 my_strlcpy(name, "\n", sizeof(name));
1552 my_strlcpy(name, "\t", sizeof(name));
1555 /* Check each possibe class */
1556 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1558 DEBUG_L(PerlIO_printf(Perl_debug_log,
1559 "isalnum('%s') unexpectedly is %d\n",
1560 name, cBOOL(isalnum(i))));
1562 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1564 DEBUG_L(PerlIO_printf(Perl_debug_log,
1565 "isalpha('%s') unexpectedly is %d\n",
1566 name, cBOOL(isalpha(i))));
1568 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1570 DEBUG_L(PerlIO_printf(Perl_debug_log,
1571 "isdigit('%s') unexpectedly is %d\n",
1572 name, cBOOL(isdigit(i))));
1574 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1576 DEBUG_L(PerlIO_printf(Perl_debug_log,
1577 "isgraph('%s') unexpectedly is %d\n",
1578 name, cBOOL(isgraph(i))));
1580 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1582 DEBUG_L(PerlIO_printf(Perl_debug_log,
1583 "islower('%s') unexpectedly is %d\n",
1584 name, cBOOL(islower(i))));
1586 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1588 DEBUG_L(PerlIO_printf(Perl_debug_log,
1589 "isprint('%s') unexpectedly is %d\n",
1590 name, cBOOL(isprint(i))));
1592 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1594 DEBUG_L(PerlIO_printf(Perl_debug_log,
1595 "ispunct('%s') unexpectedly is %d\n",
1596 name, cBOOL(ispunct(i))));
1598 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1600 DEBUG_L(PerlIO_printf(Perl_debug_log,
1601 "isspace('%s') unexpectedly is %d\n",
1602 name, cBOOL(isspace(i))));
1604 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1606 DEBUG_L(PerlIO_printf(Perl_debug_log,
1607 "isupper('%s') unexpectedly is %d\n",
1608 name, cBOOL(isupper(i))));
1610 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1612 DEBUG_L(PerlIO_printf(Perl_debug_log,
1613 "isxdigit('%s') unexpectedly is %d\n",
1614 name, cBOOL(isxdigit(i))));
1616 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1618 DEBUG_L(PerlIO_printf(Perl_debug_log,
1619 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1620 name, tolower(i), (int) toLOWER_A(i)));
1622 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1624 DEBUG_L(PerlIO_printf(Perl_debug_log,
1625 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1626 name, toupper(i), (int) toUPPER_A(i)));
1628 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1630 DEBUG_L(PerlIO_printf(Perl_debug_log,
1631 "'\\n' (=%02X) is not a control\n", (int) i));
1634 /* Add to the list; Separate multiple entries with a blank */
1637 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1639 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1647 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1648 * this locale requires more than one byte, there are going to be
1650 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1651 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1652 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1654 if ( check_for_problems && MB_CUR_MAX > 1
1655 && ! PL_in_utf8_CTYPE_locale
1657 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1658 * locale. Just assume that the implementation for them (plus
1659 * for POSIX) is correct and the > 1 value is spurious. (Since
1660 * these are specially handled to never be considered UTF-8
1661 * locales, as long as this is the only problem, everything
1662 * should work fine */
1663 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1665 multi_byte_locale = TRUE;
1670 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1671 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1672 PL_warn_locale = Perl_newSVpvf(aTHX_
1673 "Locale '%s' contains (at least) the following characters"
1674 " which have\nunexpected meanings: %s\nThe Perl program"
1675 " will use the expected meanings",
1676 newctype, bad_chars_list);
1679 PL_warn_locale = Perl_newSVpvf(aTHX_
1680 "Locale '%s' may not work well.%s%s%s\n",
1683 ? " Some characters in it are not recognized by"
1687 ? "\nThe following characters (and maybe others)"
1688 " may not have the same meaning as the Perl"
1689 " program expects:\n"
1697 # ifdef HAS_NL_LANGINFO
1699 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1700 /* parameter FALSE is a don't care here */
1701 my_nl_langinfo(CODESET, FALSE));
1705 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1707 /* If we are actually in the scope of the locale or are debugging,
1708 * output the message now. If not in that scope, we save the
1709 * message to be output at the first operation using this locale,
1710 * if that actually happens. Most programs don't use locales, so
1711 * they are immune to bad ones. */
1712 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1714 /* The '0' below suppresses a bogus gcc compiler warning */
1715 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1717 if (IN_LC(LC_CTYPE)) {
1718 SvREFCNT_dec_NN(PL_warn_locale);
1719 PL_warn_locale = NULL;
1725 #endif /* USE_LOCALE_CTYPE */
1730 Perl__warn_problematic_locale()
1733 #ifdef USE_LOCALE_CTYPE
1737 /* Internal-to-core function that outputs the message in PL_warn_locale,
1738 * and then NULLS it. Should be called only through the macro
1739 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1741 if (PL_warn_locale) {
1742 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1743 SvPVX(PL_warn_locale),
1744 0 /* dummy to avoid compiler warning */ );
1745 SvREFCNT_dec_NN(PL_warn_locale);
1746 PL_warn_locale = NULL;
1754 S_new_collate(pTHX_ const char *newcoll)
1757 #ifndef USE_LOCALE_COLLATE
1759 PERL_UNUSED_ARG(newcoll);
1760 PERL_UNUSED_CONTEXT;
1764 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1765 * core Perl this and that 'newcoll' is the name of the new locale.
1767 * The design of locale collation is that every locale change is given an
1768 * index 'PL_collation_ix'. The first time a string particpates in an
1769 * operation that requires collation while locale collation is active, it
1770 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1771 * magic includes the collation index, and the transformation of the string
1772 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1773 * instead of the string itself. If a string changes, the magic is
1774 * cleared. The next time the locale changes, the index is incremented,
1775 * and so we know during a comparison that the transformation is not
1776 * necessarily still valid, and so is recomputed. Note that if the locale
1777 * changes enough times, the index could wrap (a U32), and it is possible
1778 * that a transformation would improperly be considered valid, leading to
1779 * an unlikely bug */
1782 if (PL_collation_name) {
1784 Safefree(PL_collation_name);
1785 PL_collation_name = NULL;
1787 PL_collation_standard = TRUE;
1788 is_standard_collation:
1789 PL_collxfrm_base = 0;
1790 PL_collxfrm_mult = 2;
1791 PL_in_utf8_COLLATE_locale = FALSE;
1792 PL_strxfrm_NUL_replacement = '\0';
1793 PL_strxfrm_max_cp = 0;
1797 /* If this is not the same locale as currently, set the new one up */
1798 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1800 Safefree(PL_collation_name);
1801 PL_collation_name = stdize_locale(savepv(newcoll));
1802 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1803 if (PL_collation_standard) {
1804 goto is_standard_collation;
1807 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1808 PL_strxfrm_NUL_replacement = '\0';
1809 PL_strxfrm_max_cp = 0;
1811 /* A locale collation definition includes primary, secondary, tertiary,
1812 * etc. weights for each character. To sort, the primary weights are
1813 * used, and only if they compare equal, then the secondary weights are
1814 * used, and only if they compare equal, then the tertiary, etc.
1816 * strxfrm() works by taking the input string, say ABC, and creating an
1817 * output transformed string consisting of first the primary weights,
1818 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1819 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1820 * may not have weights at every level. In our example, let's say B
1821 * doesn't have a tertiary weight, and A doesn't have a secondary
1822 * weight. The constructed string is then going to be
1823 * A¹B¹C¹ B²C² A³C³ ....
1824 * This has the desired effect that strcmp() will look at the secondary
1825 * or tertiary weights only if the strings compare equal at all higher
1826 * priority weights. The spaces shown here, like in
1828 * are not just for readability. In the general case, these must
1829 * actually be bytes, which we will call here 'separator weights'; and
1830 * they must be smaller than any other weight value, but since these
1831 * are C strings, only the terminating one can be a NUL (some
1832 * implementations may include a non-NUL separator weight just before
1833 * the NUL). Implementations tend to reserve 01 for the separator
1834 * weights. They are needed so that a shorter string's secondary
1835 * weights won't be misconstrued as primary weights of a longer string,
1836 * etc. By making them smaller than any other weight, the shorter
1837 * string will sort first. (Actually, if all secondary weights are
1838 * smaller than all primary ones, there is no need for a separator
1839 * weight between those two levels, etc.)
1841 * The length of the transformed string is roughly a linear function of
1842 * the input string. It's not exactly linear because some characters
1843 * don't have weights at all levels. When we call strxfrm() we have to
1844 * allocate some memory to hold the transformed string. The
1845 * calculations below try to find coefficients 'm' and 'b' for this
1846 * locale so that m*x + b equals how much space we need, given the size
1847 * of the input string in 'x'. If we calculate too small, we increase
1848 * the size as needed, and call strxfrm() again, but it is better to
1849 * get it right the first time to avoid wasted expensive string
1850 * transformations. */
1853 /* We use the string below to find how long the tranformation of it
1854 * is. Almost all locales are supersets of ASCII, or at least the
1855 * ASCII letters. We use all of them, half upper half lower,
1856 * because if we used fewer, we might hit just the ones that are
1857 * outliers in a particular locale. Most of the strings being
1858 * collated will contain a preponderance of letters, and even if
1859 * they are above-ASCII, they are likely to have the same number of
1860 * weight levels as the ASCII ones. It turns out that digits tend
1861 * to have fewer levels, and some punctuation has more, but those
1862 * are relatively sparse in text, and khw believes this gives a
1863 * reasonable result, but it could be changed if experience so
1865 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1866 char * x_longer; /* Transformed 'longer' */
1867 Size_t x_len_longer; /* Length of 'x_longer' */
1869 char * x_shorter; /* We also transform a substring of 'longer' */
1870 Size_t x_len_shorter;
1872 /* _mem_collxfrm() is used get the transformation (though here we
1873 * are interested only in its length). It is used because it has
1874 * the intelligence to handle all cases, but to work, it needs some
1875 * values of 'm' and 'b' to get it started. For the purposes of
1876 * this calculation we use a very conservative estimate of 'm' and
1877 * 'b'. This assumes a weight can be multiple bytes, enough to
1878 * hold any UV on the platform, and there are 5 levels, 4 weight
1879 * bytes, and a trailing NUL. */
1880 PL_collxfrm_base = 5;
1881 PL_collxfrm_mult = 5 * sizeof(UV);
1883 /* Find out how long the transformation really is */
1884 x_longer = _mem_collxfrm(longer,
1888 /* We avoid converting to UTF-8 in the
1889 * called function by telling it the
1890 * string is in UTF-8 if the locale is a
1891 * UTF-8 one. Since the string passed
1892 * here is invariant under UTF-8, we can
1893 * claim it's UTF-8 even though it isn't.
1895 PL_in_utf8_COLLATE_locale);
1898 /* Find out how long the transformation of a substring of 'longer'
1899 * is. Together the lengths of these transformations are
1900 * sufficient to calculate 'm' and 'b'. The substring is all of
1901 * 'longer' except the first character. This minimizes the chances
1902 * of being swayed by outliers */
1903 x_shorter = _mem_collxfrm(longer + 1,
1906 PL_in_utf8_COLLATE_locale);
1907 Safefree(x_shorter);
1909 /* If the results are nonsensical for this simple test, the whole
1910 * locale definition is suspect. Mark it so that locale collation
1911 * is not active at all for it. XXX Should we warn? */
1912 if ( x_len_shorter == 0
1913 || x_len_longer == 0
1914 || x_len_shorter >= x_len_longer)
1916 PL_collxfrm_mult = 0;
1917 PL_collxfrm_base = 0;
1920 SSize_t base; /* Temporary */
1922 /* We have both: m * strlen(longer) + b = x_len_longer
1923 * m * strlen(shorter) + b = x_len_shorter;
1924 * subtracting yields:
1925 * m * (strlen(longer) - strlen(shorter))
1926 * = x_len_longer - x_len_shorter
1927 * But we have set things up so that 'shorter' is 1 byte smaller
1928 * than 'longer'. Hence:
1929 * m = x_len_longer - x_len_shorter
1931 * But if something went wrong, make sure the multiplier is at
1934 if (x_len_longer > x_len_shorter) {
1935 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1938 PL_collxfrm_mult = 1;
1943 * but in case something has gone wrong, make sure it is
1945 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1950 /* Add 1 for the trailing NUL */
1951 PL_collxfrm_base = base + 1;
1956 if (DEBUG_L_TEST || debug_initialization) {
1957 PerlIO_printf(Perl_debug_log,
1958 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1960 " collate multipler=%zu, collate base=%zu\n",
1962 PL_in_utf8_COLLATE_locale,
1963 x_len_shorter, x_len_longer,
1964 PL_collxfrm_mult, PL_collxfrm_base);
1971 #endif /* USE_LOCALE_COLLATE */
1978 S_win32_setlocale(pTHX_ int category, const char* locale)
1980 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1981 * difference between the two unless the input locale is "", which normally
1982 * means on Windows to get the machine default, which is set via the
1983 * computer's "Regional and Language Options" (or its current equivalent).
1984 * In POSIX, it instead means to find the locale from the user's
1985 * environment. This routine changes the Windows behavior to first look in
1986 * the environment, and, if anything is found, use that instead of going to
1987 * the machine default. If there is no environment override, the machine
1988 * default is used, by calling the real setlocale() with "".
1990 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1991 * use the particular category's variable if set; otherwise to use the LANG
1994 bool override_LC_ALL = FALSE;
1998 if (locale && strEQ(locale, "")) {
2002 locale = PerlEnv_getenv("LC_ALL");
2004 if (category == LC_ALL) {
2005 override_LC_ALL = TRUE;
2011 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2012 if (category == categories[i]) {
2013 locale = PerlEnv_getenv(category_names[i]);
2018 locale = PerlEnv_getenv("LANG");
2034 result = setlocale(category, locale);
2035 DEBUG_L(STMT_START {
2037 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2038 setlocale_debug_string(category, locale, result));
2042 if (! override_LC_ALL) {
2046 /* Here the input category was LC_ALL, and we have set it to what is in the
2047 * LANG variable or the system default if there is no LANG. But these have
2048 * lower priority than the other LC_foo variables, so override it for each
2049 * one that is set. (If they are set to "", it means to use the same thing
2050 * we just set LC_ALL to, so can skip) */
2052 for (i = 0; i < LC_ALL_INDEX; i++) {
2053 result = PerlEnv_getenv(category_names[i]);
2054 if (result && strNE(result, "")) {
2055 setlocale(categories[i], result);
2056 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2058 setlocale_debug_string(categories[i], result, "not captured")));
2062 result = setlocale(LC_ALL, NULL);
2063 DEBUG_L(STMT_START {
2065 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2067 setlocale_debug_string(LC_ALL, NULL, result));
2078 =head1 Locale-related functions and macros
2080 =for apidoc Perl_setlocale
2082 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2083 taking the same parameters, and returning the same information, except that it
2084 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2085 instead return C<C> if the underlying locale has a non-dot decimal point
2086 character, or a non-empty thousands separator for displaying floating point
2087 numbers. This is because perl keeps that locale category such that it has a
2088 dot and empty separator, changing the locale briefly during the operations
2089 where the underlying one is required. C<Perl_setlocale> knows about this, and
2090 compensates; regular C<setlocale> doesn't.
2092 Another reason it isn't completely a drop-in replacement is that it is
2093 declared to return S<C<const char *>>, whereas the system setlocale omits the
2094 C<const> (presumably because its API was specified long ago, and can't be
2095 updated; it is illegal to change the information C<setlocale> returns; doing
2096 so leads to segfaults.)
2098 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2099 C<setlocale> can be completely ineffective on some platforms under some
2102 C<Perl_setlocale> should not be used to change the locale except on systems
2103 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2104 the system C<setlocale()> is ineffective, returning the wrong information, and
2105 failing to actually change the locale. C<Perl_setlocale>, however works
2106 properly in all circumstances.
2108 The return points to a per-thread static buffer, which is overwritten the next
2109 time C<Perl_setlocale> is called from the same thread.
2116 Perl_setlocale(const int category, const char * locale)
2118 /* This wraps POSIX::setlocale() */
2120 const char * retval;
2121 const char * newlocale;
2123 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2126 #ifdef USE_LOCALE_NUMERIC
2128 /* A NULL locale means only query what the current one is. We have the
2129 * LC_NUMERIC name saved, because we are normally switched into the C
2130 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2131 * the correct results. All other categories don't require special
2133 if (locale == NULL) {
2134 if (category == LC_NUMERIC) {
2136 /* We don't have to copy this return value, as it is a per-thread
2137 * variable, and won't change until a future setlocale */
2138 return PL_numeric_name;
2143 else if (category == LC_ALL) {
2144 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2153 retval = save_to_buffer(do_setlocale_r(category, locale),
2154 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2157 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2159 if (locale == NULL && category == LC_ALL) {
2160 RESTORE_LC_NUMERIC();
2165 DEBUG_L(PerlIO_printf(Perl_debug_log,
2166 "%s:%d: %s\n", __FILE__, __LINE__,
2167 setlocale_debug_string(category, locale, retval)));
2175 /* If locale == NULL, we are just querying the state */
2176 if (locale == NULL) {
2180 /* Now that have switched locales, we have to update our records to
2185 #ifdef USE_LOCALE_CTYPE
2192 #ifdef USE_LOCALE_COLLATE
2195 new_collate(retval);
2199 #ifdef USE_LOCALE_NUMERIC
2202 new_numeric(retval);
2210 /* LC_ALL updates all the things we care about. The values may not
2211 * be the same as 'retval', as the locale "" may have set things
2214 # ifdef USE_LOCALE_CTYPE
2216 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2217 new_ctype(newlocale);
2219 # endif /* USE_LOCALE_CTYPE */
2220 # ifdef USE_LOCALE_COLLATE
2222 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2223 new_collate(newlocale);
2226 # ifdef USE_LOCALE_NUMERIC
2228 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2229 new_numeric(newlocale);
2231 # endif /* USE_LOCALE_NUMERIC */
2242 PERL_STATIC_INLINE const char *
2243 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2245 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2246 * growing it if necessary */
2250 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2256 string_size = strlen(string) + offset + 1;
2258 if (*buf_size == 0) {
2259 Newx(*buf, string_size, char);
2260 *buf_size = string_size;
2262 else if (string_size > *buf_size) {
2263 Renew(*buf, string_size, char);
2264 *buf_size = string_size;
2267 Copy(string, *buf + offset, string_size - offset, char);
2273 =for apidoc Perl_langinfo
2275 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2276 taking the same C<item> parameter values, and returning the same information.
2277 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2278 of Perl's locale handling from your code, and can be used on systems that lack
2279 a native C<nl_langinfo>.
2287 The reason it isn't quite a drop-in replacement is actually an advantage. The
2288 only difference is that it returns S<C<const char *>>, whereas plain
2289 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2290 forbidden to write into the buffer. By declaring this C<const>, the compiler
2291 enforces this restriction, so if it is violated, you know at compilation time,
2292 rather than getting segfaults at runtime.
2296 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2297 without you having to write extra code. The reason for the extra code would be
2298 because these are from the C<LC_NUMERIC> locale category, which is normally
2299 kept set by Perl so that the radix is a dot, and the separator is the empty
2300 string, no matter what the underlying locale is supposed to be, and so to get
2301 the expected results, you have to temporarily toggle into the underlying
2302 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2303 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2304 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2305 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2306 (decimal point) character to be a dot.)
2310 The system function it replaces can have its static return buffer trashed,
2311 not only by a subesequent call to that function, but by a C<freelocale>,
2312 C<setlocale>, or other locale change. The returned buffer of this function is
2313 not changed until the next call to it, so the buffer is never in a trashed
2318 Its return buffer is per-thread, so it also is never overwritten by a call to
2319 this function from another thread; unlike the function it replaces.
2323 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2324 as Windows, hence makes your code more portable. Of the fifty-some possible
2325 items specified by the POSIX 2008 standard,
2326 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2327 only one is completely unimplemented, though on non-Windows platforms, another
2328 significant one is also not implemented). It uses various techniques to
2329 recover the other items, including calling C<L<localeconv(3)>>, and
2330 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2331 available. Later C<strftime()> versions have additional capabilities; C<""> is
2332 returned for those not available on your system.
2334 It is important to note that when called with an item that is recovered by
2335 using C<localeconv>, the buffer from any previous explicit call to
2336 C<localeconv> will be overwritten. This means you must save that buffer's
2337 contents if you need to access them after a call to this function. (But note
2338 that you might not want to be using C<localeconv()> directly anyway, because of
2339 issues like the ones listed in the second item of this list (above) for
2340 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2341 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2344 The details for those items which may deviate from what this emulation returns
2345 and what a native C<nl_langinfo()> would return are specified in
2350 When using C<Perl_langinfo> on systems that don't have a native
2351 C<nl_langinfo()>, you must
2353 #include "perl_langinfo.h"
2355 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2356 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2357 C<langinfo.h> would try to import into the namespace for code that doesn't need
2360 The original impetus for C<Perl_langinfo()> was so that code that needs to
2361 find out the current currency symbol, floating point radix character, or digit
2362 grouping separator can use, on all systems, the simpler and more
2363 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2364 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2365 is better to use the methods given in L<perlcall> to call
2366 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2373 #ifdef HAS_NL_LANGINFO
2374 Perl_langinfo(const nl_item item)
2376 Perl_langinfo(const int item)
2379 return my_nl_langinfo(item, TRUE);
2383 #ifdef HAS_NL_LANGINFO
2384 S_my_nl_langinfo(const nl_item item, bool toggle)
2386 S_my_nl_langinfo(const int item, bool toggle)
2390 const char * retval;
2392 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2393 * two items, and only if not already there */
2394 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2395 || PL_numeric_underlying))
2400 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2401 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2402 || ! defined(HAS_POSIX_2008_LOCALE) \
2403 || ! defined(DUPLOCALE)
2405 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2406 * for those items dependent on it. This must be copied to a buffer before
2407 * switching back, as some systems destroy the buffer when setlocale() is
2411 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2414 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2417 LOCALE_LOCK; /* Prevent interference from another thread executing
2418 this code section (the only call to nl_langinfo in
2422 /* Copy to a per-thread buffer, which is also one that won't be
2423 * destroyed by a subsequent setlocale(), such as the
2424 * RESTORE_LC_NUMERIC may do just below. */
2425 retval = save_to_buffer(nl_langinfo(item),
2426 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2431 RESTORE_LC_NUMERIC();
2435 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2438 bool do_free = FALSE;
2439 locale_t cur = uselocale((locale_t) 0);
2441 if (cur == LC_GLOBAL_LOCALE) {
2442 cur = duplocale(LC_GLOBAL_LOCALE);
2447 if (PL_underlying_numeric_obj) {
2448 cur = PL_underlying_numeric_obj;
2451 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2456 /* We have to save it to a buffer, because the freelocale() just below
2457 * can invalidate the internal one */
2458 retval = save_to_buffer(nl_langinfo_l(item, cur),
2459 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2468 if (strEQ(retval, "")) {
2469 if (item == YESSTR) {
2472 if (item == NOSTR) {
2479 #else /* Below, emulate nl_langinfo as best we can */
2483 # ifdef HAS_LOCALECONV
2485 const struct lconv* lc;
2487 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2489 # ifdef TS_W32_BROKEN_LOCALECONV
2491 const char * save_global;
2492 const char * save_thread;
2500 # ifdef HAS_STRFTIME
2503 bool return_format = FALSE; /* Return the %format, not the value */
2504 const char * format;
2508 /* We copy the results to a per-thread buffer, even if not
2509 * multi-threaded. This is in part to simplify this code, and partly
2510 * because we need a buffer anyway for strftime(), and partly because a
2511 * call of localeconv() could otherwise wipe out the buffer, and the
2512 * programmer would not be expecting this, as this is a nl_langinfo()
2513 * substitute after all, so s/he might be thinking their localeconv()
2514 * is safe until another localeconv() call. */
2519 /* This is unimplemented */
2520 case ERA: /* For use with strftime() %E modifier */
2525 /* We use only an English set, since we don't know any more */
2526 case YESEXPR: return "^[+1yY]";
2527 case YESSTR: return "yes";
2528 case NOEXPR: return "^[-0nN]";
2529 case NOSTR: return "no";
2535 /* On non-windows, this is unimplemented, in part because of
2536 * inconsistencies between vendors. The Darwin native
2537 * nl_langinfo() implementation simply looks at everything past
2538 * any dot in the name, but that doesn't work for other
2539 * vendors. Many Linux locales that don't have UTF-8 in their
2540 * names really are UTF-8, for example; z/OS locales that do
2541 * have UTF-8 in their names, aren't really UTF-8 */
2546 { /* But on Windows, the name does seem to be consistent, so
2551 const char * name = my_setlocale(LC_CTYPE, NULL);
2553 if (isNAME_C_OR_POSIX(name)) {
2554 return "ANSI_X3.4-1968";
2557 /* Find the dot in the locale name */
2558 first = (const char *) strchr(name, '.');
2564 /* Look at everything past the dot */
2569 if (! isDIGIT(*p)) {
2576 /* Here everything past the dot is a digit. Treat it as a
2578 save_to_buffer("CP", &PL_langinfo_buf,
2579 &PL_langinfo_bufsize, 0);
2580 offset = STRLENs("CP");
2584 retval = save_to_buffer(first, &PL_langinfo_buf,
2585 &PL_langinfo_bufsize, offset);
2591 # ifdef HAS_LOCALECONV
2595 /* We don't bother with localeconv_l() because any system that
2596 * has it is likely to also have nl_langinfo() */
2598 LOCALE_LOCK_V; /* Prevent interference with other threads
2599 using localeconv() */
2601 # ifdef TS_W32_BROKEN_LOCALECONV
2603 /* This is a workaround for a Windows bug prior to VS 15.
2604 * What we do here is, while locked, switch to the global
2605 * locale so localeconv() works; then switch back just before
2606 * the unlock. This can screw things up if some thread is
2607 * already using the global locale while assuming no other is.
2608 * A different workaround would be to call GetCurrencyFormat on
2609 * a known value, and parse it; patches welcome
2611 * We have to use LC_ALL instead of LC_MONETARY because of
2612 * another bug in Windows */
2614 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2615 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2616 save_global= savepv(my_setlocale(LC_ALL, NULL));
2617 my_setlocale(LC_ALL, save_thread);
2623 || ! lc->currency_symbol
2624 || strEQ("", lc->currency_symbol))
2630 /* Leave the first spot empty to be filled in below */
2631 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2632 &PL_langinfo_bufsize, 1);
2633 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2634 { /* khw couldn't figure out how the localedef specifications
2635 would show that the $ should replace the radix; this is
2636 just a guess as to how it might work.*/
2637 PL_langinfo_buf[0] = '.';
2639 else if (lc->p_cs_precedes) {
2640 PL_langinfo_buf[0] = '-';
2643 PL_langinfo_buf[0] = '+';
2646 # ifdef TS_W32_BROKEN_LOCALECONV
2648 my_setlocale(LC_ALL, save_global);
2649 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2650 my_setlocale(LC_ALL, save_thread);
2651 Safefree(save_global);
2652 Safefree(save_thread);
2659 # ifdef TS_W32_BROKEN_LOCALECONV
2663 /* For this, we output a known simple floating point number to
2664 * a buffer, and parse it, looking for the radix */
2667 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2670 if (PL_langinfo_bufsize < 10) {
2671 PL_langinfo_bufsize = 10;
2672 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2675 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2677 if (needed_size >= (int) PL_langinfo_bufsize) {
2678 PL_langinfo_bufsize = needed_size + 1;
2679 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2680 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2682 assert(needed_size < (int) PL_langinfo_bufsize);
2685 ptr = PL_langinfo_buf;
2686 e = PL_langinfo_buf + PL_langinfo_bufsize;
2689 while (ptr < e && *ptr != '1') {
2696 while (ptr < e && *ptr != '5') {
2700 /* Everything in between is the radix string */
2702 PL_langinfo_buf[0] = '?';
2703 PL_langinfo_buf[1] = '\0';
2707 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2711 RESTORE_LC_NUMERIC();
2714 retval = PL_langinfo_buf;
2719 case RADIXCHAR: /* No special handling needed */
2726 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2729 LOCALE_LOCK_V; /* Prevent interference with other threads
2730 using localeconv() */
2732 # ifdef TS_W32_BROKEN_LOCALECONV
2734 /* This should only be for the thousands separator. A
2735 * different work around would be to use GetNumberFormat on a
2736 * known value and parse the result to find the separator */
2737 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2738 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2739 save_global = savepv(my_setlocale(LC_ALL, NULL));
2740 my_setlocale(LC_ALL, save_thread);
2742 /* This is the start of code that for broken Windows replaces
2743 * the above and below code, and instead calls
2744 * GetNumberFormat() and then would parse that to find the
2745 * thousands separator. It needs to handle UTF-16 vs -8
2748 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2749 DEBUG_L(PerlIO_printf(Perl_debug_log,
2750 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2751 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2761 temp = (item == RADIXCHAR)
2763 : lc->thousands_sep;
2769 retval = save_to_buffer(temp, &PL_langinfo_buf,
2770 &PL_langinfo_bufsize, 0);
2772 # ifdef TS_W32_BROKEN_LOCALECONV
2774 my_setlocale(LC_ALL, save_global);
2775 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2776 my_setlocale(LC_ALL, save_thread);
2777 Safefree(save_global);
2778 Safefree(save_thread);
2785 RESTORE_LC_NUMERIC();
2791 # ifdef HAS_STRFTIME
2793 /* These are defined by C89, so we assume that strftime supports
2794 * them, and so are returned unconditionally; they may not be what
2795 * the locale actually says, but should give good enough results
2796 * for someone using them as formats (as opposed to trying to parse
2797 * them to figure out what the locale says). The other format
2798 * items are actually tested to verify they work on the platform */
2799 case D_FMT: return "%x";
2800 case T_FMT: return "%X";
2801 case D_T_FMT: return "%c";
2803 /* These formats are only available in later strfmtime's */
2804 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2806 /* The rest can be gotten from most versions of strftime(). */
2807 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2808 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2810 case AM_STR: case PM_STR:
2811 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2812 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2813 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2814 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2815 case DAY_5: case DAY_6: case DAY_7:
2816 case MON_1: case MON_2: case MON_3: case MON_4:
2817 case MON_5: case MON_6: case MON_7: case MON_8:
2818 case MON_9: case MON_10: case MON_11: case MON_12:
2822 init_tm(&tm); /* Precaution against core dumps */
2826 tm.tm_year = 2017 - 1900;
2833 "panic: %s: %d: switch case: %d problem",
2834 __FILE__, __LINE__, item);
2835 NOT_REACHED; /* NOTREACHED */
2837 case PM_STR: tm.tm_hour = 18;
2842 case ABDAY_7: tm.tm_wday++;
2843 case ABDAY_6: tm.tm_wday++;
2844 case ABDAY_5: tm.tm_wday++;
2845 case ABDAY_4: tm.tm_wday++;
2846 case ABDAY_3: tm.tm_wday++;
2847 case ABDAY_2: tm.tm_wday++;
2852 case DAY_7: tm.tm_wday++;
2853 case DAY_6: tm.tm_wday++;
2854 case DAY_5: tm.tm_wday++;
2855 case DAY_4: tm.tm_wday++;
2856 case DAY_3: tm.tm_wday++;
2857 case DAY_2: tm.tm_wday++;
2862 case ABMON_12: tm.tm_mon++;
2863 case ABMON_11: tm.tm_mon++;
2864 case ABMON_10: tm.tm_mon++;
2865 case ABMON_9: tm.tm_mon++;
2866 case ABMON_8: tm.tm_mon++;
2867 case ABMON_7: tm.tm_mon++;
2868 case ABMON_6: tm.tm_mon++;
2869 case ABMON_5: tm.tm_mon++;
2870 case ABMON_4: tm.tm_mon++;
2871 case ABMON_3: tm.tm_mon++;
2872 case ABMON_2: tm.tm_mon++;
2877 case MON_12: tm.tm_mon++;
2878 case MON_11: tm.tm_mon++;
2879 case MON_10: tm.tm_mon++;
2880 case MON_9: tm.tm_mon++;
2881 case MON_8: tm.tm_mon++;
2882 case MON_7: tm.tm_mon++;
2883 case MON_6: tm.tm_mon++;
2884 case MON_5: tm.tm_mon++;
2885 case MON_4: tm.tm_mon++;
2886 case MON_3: tm.tm_mon++;
2887 case MON_2: tm.tm_mon++;
2894 return_format = TRUE;
2899 return_format = TRUE;
2904 return_format = TRUE;
2909 return_format = TRUE;
2914 format = "%Ow"; /* Find the alternate digit for 0 */
2918 /* We can't use my_strftime() because it doesn't look at
2920 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2923 /* A zero return means one of:
2924 * a) there wasn't enough space in PL_langinfo_buf
2925 * b) the format, like a plain %p, returns empty
2926 * c) it was an illegal format, though some
2927 * implementations of strftime will just return the
2928 * illegal format as a plain character sequence.
2930 * To quickly test for case 'b)', try again but precede
2931 * the format with a plain character. If that result is
2932 * still empty, the problem is either 'a)' or 'c)' */
2934 Size_t format_size = strlen(format) + 1;
2935 Size_t mod_size = format_size + 1;
2939 Newx(mod_format, mod_size, char);
2940 Newx(temp_result, PL_langinfo_bufsize, char);
2942 my_strlcpy(mod_format + 1, format, mod_size);
2943 len = strftime(temp_result,
2944 PL_langinfo_bufsize,
2946 Safefree(mod_format);
2947 Safefree(temp_result);
2949 /* If 'len' is non-zero, it means that we had a case like
2950 * %p which means the current locale doesn't use a.m. or
2951 * p.m., and that is valid */
2954 /* Here, still didn't work. If we get well beyond a
2955 * reasonable size, bail out to prevent an infinite
2958 if (PL_langinfo_bufsize > 100 * format_size) {
2959 *PL_langinfo_buf = '\0';
2962 /* Double the buffer size to retry; Add 1 in case
2963 * original was 0, so we aren't stuck at 0. */
2964 PL_langinfo_bufsize *= 2;
2965 PL_langinfo_bufsize++;
2966 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2974 /* Here, we got a result.
2976 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2977 * alternate format for wday 0. If the value is the same as
2978 * the normal 0, there isn't an alternate, so clear the buffer.
2980 if ( item == ALT_DIGITS
2981 && strEQ(PL_langinfo_buf, "0"))
2983 *PL_langinfo_buf = '\0';
2986 /* ALT_DIGITS is problematic. Experiments on it showed that
2987 * strftime() did not always work properly when going from
2988 * alt-9 to alt-10. Only a few locales have this item defined,
2989 * and in all of them on Linux that khw was able to find,
2990 * nl_langinfo() merely returned the alt-0 character, possibly
2991 * doubled. Most Unicode digits are in blocks of 10
2992 * consecutive code points, so that is sufficient information
2993 * for those scripts, as we can infer alt-1, alt-2, .... But
2994 * for a Japanese locale, a CJK ideographic 0 is returned, and
2995 * the CJK digits are not in code point order, so you can't
2996 * really infer anything. The localedef for this locale did
2997 * specify the succeeding digits, so that strftime() works
2998 * properly on them, without needing to infer anything. But
2999 * the nl_langinfo() return did not give sufficient information
3000 * for the caller to understand what's going on. So until
3001 * there is evidence that it should work differently, this
3002 * returns the alt-0 string for ALT_DIGITS.
3004 * wday was chosen because its range is all a single digit.
3005 * Things like tm_sec have two digits as the minimum: '00' */
3009 retval = PL_langinfo_buf;
3011 /* If to return the format, not the value, overwrite the buffer
3012 * with it. But some strftime()s will keep the original format
3013 * if illegal, so change those to "" */
3014 if (return_format) {
3015 if (strEQ(PL_langinfo_buf, format)) {
3016 *PL_langinfo_buf = '\0';
3019 retval = save_to_buffer(format, &PL_langinfo_buf,
3020 &PL_langinfo_bufsize, 0);
3038 * Initialize locale awareness.
3041 Perl_init_i18nl10n(pTHX_ int printwarn)
3045 * 0 if not to output warning when setup locale is bad
3046 * 1 if to output warning based on value of PERL_BADLANG
3047 * >1 if to output regardless of PERL_BADLANG
3050 * 1 = set ok or not applicable,
3051 * 0 = fallback to a locale of lower priority
3052 * -1 = fallback to all locales failed, not even to the C locale
3054 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3055 * set, debugging information is output.
3057 * This looks more complicated than it is, mainly due to the #ifdefs.
3059 * We try to set LC_ALL to the value determined by the environment. If
3060 * there is no LC_ALL on this platform, we try the individual categories we
3061 * know about. If this works, we are done.
3063 * But if it doesn't work, we have to do something else. We search the
3064 * environment variables ourselves instead of relying on the system to do
3065 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3066 * think there is one), and the ultimate fallback "C". This is all done in
3067 * the same loop as above to avoid duplicating code, but it makes things
3068 * more complex. The 'trial_locales' array is initialized with just one
3069 * element; it causes the behavior described in the paragraph above this to
3070 * happen. If that fails, we add elements to 'trial_locales', and do extra
3071 * loop iterations to cause the behavior described in this paragraph.
3073 * On Ultrix, the locale MUST come from the environment, so there is
3074 * preliminary code to set it. I (khw) am not sure that it is necessary,
3075 * and that this couldn't be folded into the loop, but barring any real
3076 * platforms to test on, it's staying as-is
3078 * A slight complication is that in embedded Perls, the locale may already
3079 * be set-up, and we don't want to get it from the normal environment
3080 * variables. This is handled by having a special environment variable
3081 * indicate we're in this situation. We simply set setlocale's 2nd
3082 * parameter to be a NULL instead of "". That indicates to setlocale that
3083 * it is not to change anything, but to return the current value,
3084 * effectively initializing perl's db to what the locale already is.
3086 * We play the same trick with NULL if a LC_ALL succeeds. We call
3087 * setlocale() on the individual categores with NULL to get their existing
3088 * values for our db, instead of trying to change them.
3095 PERL_UNUSED_ARG(printwarn);
3097 #else /* USE_LOCALE */
3100 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3104 /* NULL uses the existing already set up locale */
3105 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3108 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3109 unsigned int trial_locales_count;
3110 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3111 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3112 bool setlocale_failure = FALSE;
3115 /* A later getenv() could zap this, so only use here */
3116 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3118 const bool locwarn = (printwarn > 1
3120 && ( ! bad_lang_use_once
3122 /* disallow with "" or "0" */
3124 && strNE("0", bad_lang_use_once)))));
3126 /* setlocale() return vals; not copied so must be looked at immediately */
3127 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3129 /* current locale for given category; should have been copied so aren't
3131 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3135 /* In some systems you can find out the system default locale
3136 * and use that as the fallback locale. */
3137 # define SYSTEM_DEFAULT_LOCALE
3139 # ifdef SYSTEM_DEFAULT_LOCALE
3141 const char *system_default_locale = NULL;
3146 # define DEBUG_LOCALE_INIT(a,b,c)
3149 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3151 # define DEBUG_LOCALE_INIT(category, locale, result) \
3153 if (debug_initialization) { \
3154 PerlIO_printf(Perl_debug_log, \
3156 __FILE__, __LINE__, \
3157 setlocale_debug_string(category, \
3163 /* Make sure the parallel arrays are properly set up */
3164 # ifdef USE_LOCALE_NUMERIC
3165 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3166 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3167 # ifdef USE_POSIX_2008_LOCALE
3168 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3171 # ifdef USE_LOCALE_CTYPE
3172 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3173 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3174 # ifdef USE_POSIX_2008_LOCALE
3175 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3178 # ifdef USE_LOCALE_COLLATE
3179 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3180 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3181 # ifdef USE_POSIX_2008_LOCALE
3182 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3185 # ifdef USE_LOCALE_TIME
3186 assert(categories[LC_TIME_INDEX] == LC_TIME);
3187 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3188 # ifdef USE_POSIX_2008_LOCALE
3189 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3192 # ifdef USE_LOCALE_MESSAGES
3193 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3194 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3195 # ifdef USE_POSIX_2008_LOCALE
3196 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3199 # ifdef USE_LOCALE_MONETARY
3200 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3201 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3202 # ifdef USE_POSIX_2008_LOCALE
3203 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3206 # ifdef USE_LOCALE_ADDRESS
3207 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3208 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3209 # ifdef USE_POSIX_2008_LOCALE
3210 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3213 # ifdef USE_LOCALE_IDENTIFICATION
3214 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3215 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3216 # ifdef USE_POSIX_2008_LOCALE
3217 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3220 # ifdef USE_LOCALE_MEASUREMENT
3221 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3222 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3223 # ifdef USE_POSIX_2008_LOCALE
3224 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3227 # ifdef USE_LOCALE_PAPER
3228 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3229 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3230 # ifdef USE_POSIX_2008_LOCALE
3231 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3234 # ifdef USE_LOCALE_TELEPHONE
3235 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3236 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3237 # ifdef USE_POSIX_2008_LOCALE
3238 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3242 assert(categories[LC_ALL_INDEX] == LC_ALL);
3243 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3244 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3245 # ifdef USE_POSIX_2008_LOCALE
3246 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3249 # endif /* DEBUGGING */
3251 /* Initialize the cache of the program's UTF-8ness for the always known
3252 * locales C and POSIX */
3253 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3254 sizeof(PL_locale_utf8ness));
3256 # ifdef USE_THREAD_SAFE_LOCALE
3259 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3263 # ifdef USE_POSIX_2008_LOCALE
3265 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3266 if (! PL_C_locale_obj) {
3267 Perl_croak_nocontext(
3268 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3270 if (DEBUG_Lv_TEST || debug_initialization) {
3271 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3276 PL_numeric_radix_sv = newSVpvs(".");
3278 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3280 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3281 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3284 # ifdef LOCALE_ENVIRON_REQUIRED
3287 * Ultrix setlocale(..., "") fails if there are no environment
3288 * variables from which to get a locale name.
3292 # error Ultrix without LC_ALL not implemented
3298 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3299 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3300 if (sl_result[LC_ALL_INDEX])
3303 setlocale_failure = TRUE;
3305 if (! setlocale_failure) {
3306 const char * locale_param;
3307 for (i = 0; i < LC_ALL_INDEX; i++) {
3308 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3311 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3312 if (! sl_result[i]) {
3313 setlocale_failure = TRUE;
3315 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3320 # endif /* LC_ALL */
3321 # endif /* LOCALE_ENVIRON_REQUIRED */
3323 /* We try each locale in the list until we get one that works, or exhaust
3324 * the list. Normally the loop is executed just once. But if setting the
3325 * locale fails, inside the loop we add fallback trials to the array and so
3326 * will execute the loop multiple times */
3327 trial_locales[0] = setlocale_init;
3328 trial_locales_count = 1;
3330 for (i= 0; i < trial_locales_count; i++) {
3331 const char * trial_locale = trial_locales[i];
3335 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3336 * when i==0, but I (khw) don't think that behavior makes much
3338 setlocale_failure = FALSE;
3340 # ifdef SYSTEM_DEFAULT_LOCALE
3341 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3343 /* On Windows machines, an entry of "" after the 0th means to use
3344 * the system default locale, which we now proceed to get. */
3345 if (strEQ(trial_locale, "")) {
3348 /* Note that this may change the locale, but we are going to do
3349 * that anyway just below */
3350 system_default_locale = do_setlocale_c(LC_ALL, "");
3351 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3353 /* Skip if invalid or if it's already on the list of locales to
3355 if (! system_default_locale) {
3356 goto next_iteration;
3358 for (j = 0; j < trial_locales_count; j++) {
3359 if (strEQ(system_default_locale, trial_locales[j])) {
3360 goto next_iteration;
3364 trial_locale = system_default_locale;
3367 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3369 # endif /* SYSTEM_DEFAULT_LOCALE */
3375 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3376 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3377 if (! sl_result[LC_ALL_INDEX]) {
3378 setlocale_failure = TRUE;
3381 /* Since LC_ALL succeeded, it should have changed all the other
3382 * categories it can to its value; so we massage things so that the
3383 * setlocales below just return their category's current values.
3384 * This adequately handles the case in NetBSD where LC_COLLATE may
3385 * not be defined for a locale, and setting it individually will
3386 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3387 * the POSIX locale. */
3388 trial_locale = NULL;
3391 # endif /* LC_ALL */
3393 if (! setlocale_failure) {
3395 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3397 = savepv(do_setlocale_r(categories[j], trial_locale));
3398 if (! curlocales[j]) {
3399 setlocale_failure = TRUE;
3401 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3404 if (! setlocale_failure) { /* All succeeded */
3405 break; /* Exit trial_locales loop */
3409 /* Here, something failed; will need to try a fallback. */
3415 if (locwarn) { /* Output failure info only on the first one */
3419 PerlIO_printf(Perl_error_log,
3420 "perl: warning: Setting locale failed.\n");
3422 # else /* !LC_ALL */
3424 PerlIO_printf(Perl_error_log,
3425 "perl: warning: Setting locale failed for the categories:\n\t");
3427 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3428 if (! curlocales[j]) {
3429 PerlIO_printf(Perl_error_log, category_names[j]);
3432 Safefree(curlocales[j]);
3436 # endif /* LC_ALL */
3438 PerlIO_printf(Perl_error_log,
3439 "perl: warning: Please check that your locale settings:\n");
3443 PerlIO_printf(Perl_error_log,
3444 "\tLANGUAGE = %c%s%c,\n",
3445 language ? '"' : '(',
3446 language ? language : "unset",
3447 language ? '"' : ')');
3450 PerlIO_printf(Perl_error_log,
3451 "\tLC_ALL = %c%s%c,\n",
3453 lc_all ? lc_all : "unset",
3454 lc_all ? '"' : ')');
3456 # if defined(USE_ENVIRON_ARRAY)
3461 /* Look through the environment for any variables of the
3462 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3463 * already handled above. These are assumed to be locale
3464 * settings. Output them and their values. */
3465 for (e = environ; *e; e++) {
3466 const STRLEN prefix_len = sizeof("LC_") - 1;
3469 if ( strBEGINs(*e, "LC_")
3470 && ! strBEGINs(*e, "LC_ALL=")
3471 && (uppers_len = strspn(*e + prefix_len,
3472 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3473 && ((*e)[prefix_len + uppers_len] == '='))
3475 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3476 (int) (prefix_len + uppers_len), *e,
3477 *e + prefix_len + uppers_len + 1);
3484 PerlIO_printf(Perl_error_log,
3485 "\t(possibly more locale environment variables)\n");
3489 PerlIO_printf(Perl_error_log,
3490 "\tLANG = %c%s%c\n",
3492 lang ? lang : "unset",
3495 PerlIO_printf(Perl_error_log,
3496 " are supported and installed on your system.\n");
3499 /* Calculate what fallback locales to try. We have avoided this
3500 * until we have to, because failure is quite unlikely. This will
3501 * usually change the upper bound of the loop we are in.
3503 * Since the system's default way of setting the locale has not
3504 * found one that works, We use Perl's defined ordering: LC_ALL,
3505 * LANG, and the C locale. We don't try the same locale twice, so
3506 * don't add to the list if already there. (On POSIX systems, the
3507 * LC_ALL element will likely be a repeat of the 0th element "",
3508 * but there's no harm done by doing it explicitly.
3510 * Note that this tries the LC_ALL environment variable even on
3511 * systems which have no LC_ALL locale setting. This may or may
3512 * not have been originally intentional, but there's no real need
3513 * to change the behavior. */
3515 for (j = 0; j < trial_locales_count; j++) {
3516 if (strEQ(lc_all, trial_locales[j])) {
3520 trial_locales[trial_locales_count++] = lc_all;
3525 for (j = 0; j < trial_locales_count; j++) {
3526 if (strEQ(lang, trial_locales[j])) {
3530 trial_locales[trial_locales_count++] = lang;
3534 # if defined(WIN32) && defined(LC_ALL)
3536 /* For Windows, we also try the system default locale before "C".
3537 * (If there exists a Windows without LC_ALL we skip this because
3538 * it gets too complicated. For those, the "C" is the next
3539 * fallback possibility). The "" is the same as the 0th element of
3540 * the array, but the code at the loop above knows to treat it
3541 * differently when not the 0th */
3542 trial_locales[trial_locales_count++] = "";
3546 for (j = 0; j < trial_locales_count; j++) {
3547 if (strEQ("C", trial_locales[j])) {
3551 trial_locales[trial_locales_count++] = "C";
3554 } /* end of first time through the loop */
3562 } /* end of looping through the trial locales */
3564 if (ok < 1) { /* If we tried to fallback */
3566 if (! setlocale_failure) { /* fallback succeeded */
3567 msg = "Falling back to";
3569 else { /* fallback failed */
3572 /* We dropped off the end of the loop, so have to decrement i to
3573 * get back to the value the last time through */
3577 msg = "Failed to fall back to";
3579 /* To continue, we should use whatever values we've got */
3581 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3582 Safefree(curlocales[j]);
3583 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3584 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3589 const char * description;
3590 const char * name = "";
3591 if (strEQ(trial_locales[i], "C")) {
3592 description = "the standard locale";
3596 # ifdef SYSTEM_DEFAULT_LOCALE
3598 else if (strEQ(trial_locales[i], "")) {
3599 description = "the system default locale";
3600 if (system_default_locale) {
3601 name = system_default_locale;
3605 # endif /* SYSTEM_DEFAULT_LOCALE */
3608 description = "a fallback locale";
3609 name = trial_locales[i];
3611 if (name && strNE(name, "")) {
3612 PerlIO_printf(Perl_error_log,
3613 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3616 PerlIO_printf(Perl_error_log,
3617 "perl: warning: %s %s.\n", msg, description);
3620 } /* End of tried to fallback */
3622 /* Done with finding the locales; update our records */
3624 # ifdef USE_LOCALE_CTYPE
3626 new_ctype(curlocales[LC_CTYPE_INDEX]);
3629 # ifdef USE_LOCALE_COLLATE
3631 new_collate(curlocales[LC_COLLATE_INDEX]);
3634 # ifdef USE_LOCALE_NUMERIC
3636 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3640 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3642 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3644 /* This caches whether each category's locale is UTF-8 or not. This
3645 * may involve changing the locale. It is ok to do this at
3646 * initialization time before any threads have started, but not later
3647 * unless thread-safe operations are used.
3648 * Caching means that if the program heeds our dictate not to change
3649 * locales in threaded applications, this data will remain valid, and
3650 * it may get queried without having to change locales. If the
3651 * environment is such that all categories have the same locale, this
3652 * isn't needed, as the code will not change the locale; but this
3653 * handles the uncommon case where the environment has disparate
3654 * locales for the categories */
3655 (void) _is_cur_LC_category_utf8(categories[i]);
3659 Safefree(curlocales[i]);
3662 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3664 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3665 * locale is UTF-8. The call to new_ctype() just above has already
3666 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3667 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3668 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3669 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3670 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3672 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3673 This is an alternative to using the -C command line switch
3674 (the -C if present will override this). */
3676 const char *p = PerlEnv_getenv("PERL_UNICODE");
3677 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3678 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3692 #endif /* USE_LOCALE */
3695 /* So won't continue to output stuff */
3696 DEBUG_INITIALIZATION_set(FALSE);
3703 #ifdef USE_LOCALE_COLLATE
3706 Perl__mem_collxfrm(pTHX_ const char *input_string,
3707 STRLEN len, /* Length of 'input_string' */
3708 STRLEN *xlen, /* Set to length of returned string
3709 (not including the collation index
3711 bool utf8 /* Is the input in UTF-8? */
3715 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3716 * differences. First, it handles embedded NULs. Second, it allocates a bit
3717 * more memory than needed for the transformed data itself. The real
3718 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3719 * the length of that, and doesn't include the collation index size.
3720 * Please see sv_collxfrm() to see how this is used. */
3722 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3724 char * s = (char *) input_string;
3725 STRLEN s_strlen = strlen(input_string);
3727 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3728 STRLEN length_in_chars;
3729 bool first_time = TRUE; /* Cleared after first loop iteration */
3731 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3733 /* Must be NUL-terminated */
3734 assert(*(input_string + len) == '\0');
3736 /* If this locale has defective collation, skip */
3737 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3738 DEBUG_L(PerlIO_printf(Perl_debug_log,
3739 "_mem_collxfrm: locale's collation is defective\n"));
3743 /* Replace any embedded NULs with the control that sorts before any others.
3744 * This will give as good as possible results on strings that don't
3745 * otherwise contain that character, but otherwise there may be
3746 * less-than-perfect results with that character and NUL. This is
3747 * unavoidable unless we replace strxfrm with our own implementation. */
3748 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3752 STRLEN sans_nuls_len;
3753 int try_non_controls;
3754 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3755 making sure 2nd byte is NUL.
3757 STRLEN this_replacement_len;
3759 /* If we don't know what non-NUL control character sorts lowest for
3760 * this locale, find it */
3761 if (PL_strxfrm_NUL_replacement == '\0') {
3763 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3764 includes the collation index
3767 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3769 /* Unlikely, but it may be that no control will work to replace
3770 * NUL, in which case we instead look for any character. Controls
3771 * are preferred because collation order is, in general, context
3772 * sensitive, with adjoining characters affecting the order, and
3773 * controls are less likely to have such interactions, allowing the
3774 * NUL-replacement to stand on its own. (Another way to look at it
3775 * is to imagine what would happen if the NUL were replaced by a
3776 * combining character; it wouldn't work out all that well.) */
3777 for (try_non_controls = 0;
3778 try_non_controls < 2;
3781 /* Look through all legal code points (NUL isn't) */
3782 for (j = 1; j < 256; j++) {
3783 char * x; /* j's xfrm plus collation index */
3784 STRLEN x_len; /* length of 'x' */
3785 STRLEN trial_len = 1;
3786 char cur_source[] = { '\0', '\0' };
3788 /* Skip non-controls the first time through the loop. The
3789 * controls in a UTF-8 locale are the L1 ones */
3790 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3797 /* Create a 1-char string of the current code point */
3798 cur_source[0] = (char) j;
3800 /* Then transform it */
3801 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3802 0 /* The string is not in UTF-8 */);
3804 /* Ignore any character that didn't successfully transform.
3810 /* If this character's transformation is lower than
3811 * the current lowest, this one becomes the lowest */
3812 if ( cur_min_x == NULL
3813 || strLT(x + COLLXFRM_HDR_LEN,
3814 cur_min_x + COLLXFRM_HDR_LEN))
3816 PL_strxfrm_NUL_replacement = j;
3822 } /* end of loop through all 255 characters */
3824 /* Stop looking if found */
3829 /* Unlikely, but possible, if there aren't any controls that
3830 * work in the locale, repeat the loop, looking for any
3831 * character that works */
3832 DEBUG_L(PerlIO_printf(Perl_debug_log,
3833 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3834 } /* End of loop to try first the controls, then any char */
3837 DEBUG_L(PerlIO_printf(Perl_debug_log,
3838 "_mem_collxfrm: Couldn't find any character to replace"
3839 " embedded NULs in locale %s with", PL_collation_name));
3843 DEBUG_L(PerlIO_printf(Perl_debug_log,
3844 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3845 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3847 Safefree(cur_min_x);
3848 } /* End of determining the character that is to replace NULs */
3850 /* If the replacement is variant under UTF-8, it must match the
3851 * UTF8-ness of the original */
3852 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3853 this_replacement_char[0] =
3854 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3855 this_replacement_char[1] =
3856 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3857 this_replacement_len = 2;
3860 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3861 /* this_replacement_char[1] = '\0' was done at initialization */
3862 this_replacement_len = 1;
3865 /* The worst case length for the replaced string would be if every
3866 * character in it is NUL. Multiply that by the length of each
3867 * replacement, and allow for a trailing NUL */
3868 sans_nuls_len = (len * this_replacement_len) + 1;
3869 Newx(sans_nuls, sans_nuls_len, char);
3872 /* Replace each NUL with the lowest collating control. Loop until have
3873 * exhausted all the NULs */
3874 while (s + s_strlen < e) {
3875 my_strlcat(sans_nuls, s, sans_nuls_len);
3877 /* Do the actual replacement */
3878 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3880 /* Move past the input NUL */
3882 s_strlen = strlen(s);
3885 /* And add anything that trails the final NUL */
3886 my_strlcat(sans_nuls, s, sans_nuls_len);
3888 /* Switch so below we transform this modified string */
3891 } /* End of replacing NULs */
3893 /* Make sure the UTF8ness of the string and locale match */
3894 if (utf8 != PL_in_utf8_COLLATE_locale) {
3895 /* XXX convert above Unicode to 10FFFF? */
3896 const char * const t = s; /* Temporary so we can later find where the
3899 /* Here they don't match. Change the string's to be what the locale is
3902 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3903 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3906 else { /* locale is not UTF-8; but input is; downgrade the input */
3908 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3910 /* If the downgrade was successful we are done, but if the input
3911 * contains things that require UTF-8 to represent, have to do
3912 * damage control ... */
3913 if (UNLIKELY(utf8)) {
3915 /* What we do is construct a non-UTF-8 string with
3916 * 1) the characters representable by a single byte converted
3917 * to be so (if necessary);
3918 * 2) and the rest converted to collate the same as the
3919 * highest collating representable character. That makes
3920 * them collate at the end. This is similar to how we
3921 * handle embedded NULs, but we use the highest collating
3922 * code point instead of the smallest. Like the NUL case,
3923 * this isn't perfect, but is the best we can reasonably
3924 * do. Every above-255 code point will sort the same as
3925 * the highest-sorting 0-255 code point. If that code
3926 * point can combine in a sequence with some other code
3927 * points for weight calculations, us changing something to
3928 * be it can adversely affect the results. But in most
3929 * cases, it should work reasonably. And note that this is
3930 * really an illegal situation: using code points above 255
3931 * on a locale where only 0-255 are valid. If two strings
3932 * sort entirely equal, then the sort order for the
3933 * above-255 code points will be in code point order. */
3937 /* If we haven't calculated the code point with the maximum
3938 * collating order for this locale, do so now */
3939 if (! PL_strxfrm_max_cp) {
3942 /* The current transformed string that collates the
3943 * highest (except it also includes the prefixed collation
3945 char * cur_max_x = NULL;
3947 /* Look through all legal code points (NUL isn't) */
3948 for (j = 1; j < 256; j++) {
3951 char cur_source[] = { '\0', '\0' };
3953 /* Create a 1-char string of the current code point */
3954 cur_source[0] = (char) j;
3956 /* Then transform it */
3957 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3959 /* If something went wrong (which it shouldn't), just
3960 * ignore this code point */
3965 /* If this character's transformation is higher than
3966 * the current highest, this one becomes the highest */
3967 if ( cur_max_x == NULL
3968 || strGT(x + COLLXFRM_HDR_LEN,
3969 cur_max_x + COLLXFRM_HDR_LEN))
3971 PL_strxfrm_max_cp = j;
3980 DEBUG_L(PerlIO_printf(Perl_debug_log,
3981 "_mem_collxfrm: Couldn't find any character to"
3982 " replace above-Latin1 chars in locale %s with",
3983 PL_collation_name));
3987 DEBUG_L(PerlIO_printf(Perl_debug_log,
3988 "_mem_collxfrm: highest 1-byte collating character"
3989 " in locale %s is 0x%02X\n",
3991 PL_strxfrm_max_cp));
3993 Safefree(cur_max_x);
3996 /* Here we know which legal code point collates the highest.
3997 * We are ready to construct the non-UTF-8 string. The length
3998 * will be at least 1 byte smaller than the input string
3999 * (because we changed at least one 2-byte character into a
4000 * single byte), but that is eaten up by the trailing NUL */
4006 char * e = (char *) t + len;
4008 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4010 if (UTF8_IS_INVARIANT(cur_char)) {
4013 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4014 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4016 else { /* Replace illegal cp with highest collating
4018 s[d++] = PL_strxfrm_max_cp;
4022 Renew(s, d, char); /* Free up unused space */
4027 /* Here, we have constructed a modified version of the input. It could
4028 * be that we already had a modified copy before we did this version.
4029 * If so, that copy is no longer needed */
4030 if (t != input_string) {
4035 length_in_chars = (utf8)
4036 ? utf8_length((U8 *) s, (U8 *) s + len)
4039 /* The first element in the output is the collation id, used by
4040 * sv_collxfrm(); then comes the space for the transformed string. The
4041 * equation should give us a good estimate as to how much is needed */
4042 xAlloc = COLLXFRM_HDR_LEN
4044 + (PL_collxfrm_mult * length_in_chars);
4045 Newx(xbuf, xAlloc, char);
4046 if (UNLIKELY(! xbuf)) {
4047 DEBUG_L(PerlIO_printf(Perl_debug_log,
4048 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4052 /* Store the collation id */
4053 *(U32*)xbuf = PL_collation_ix;
4055 /* Then the transformation of the input. We loop until successful, or we
4059 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4061 /* If the transformed string occupies less space than we told strxfrm()
4062 * was available, it means it successfully transformed the whole
4064 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4066 /* Some systems include a trailing NUL in the returned length.
4067 * Ignore it, using a loop in case multiple trailing NULs are
4070 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4075 /* If the first try didn't get it, it means our prediction was low.
4076 * Modify the coefficients so that we predict a larger value in any
4077 * future transformations */
4079 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4080 STRLEN computed_guess = PL_collxfrm_base
4081 + (PL_collxfrm_mult * length_in_chars);
4083 /* On zero-length input, just keep current slope instead of
4085 const STRLEN new_m = (length_in_chars != 0)
4086 ? needed / length_in_chars
4089 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4090 "%s: %d: initial size of %zu bytes for a length "
4091 "%zu string was insufficient, %zu needed\n",
4093 computed_guess, length_in_chars, needed));
4095 /* If slope increased, use it, but discard this result for
4096 * length 1 strings, as we can't be sure that it's a real slope
4098 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4102 STRLEN old_m = PL_collxfrm_mult;
4103 STRLEN old_b = PL_collxfrm_base;
4107 PL_collxfrm_mult = new_m;
4108 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4109 computed_guess = PL_collxfrm_base
4110 + (PL_collxfrm_mult * length_in_chars);
4111 if (computed_guess < needed) {
4112 PL_collxfrm_base += needed - computed_guess;
4115 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4116 "%s: %d: slope is now %zu; was %zu, base "
4117 "is now %zu; was %zu\n",
4119 PL_collxfrm_mult, old_m,
4120 PL_collxfrm_base, old_b));
4122 else { /* Slope didn't change, but 'b' did */
4123 const STRLEN new_b = needed
4126 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4127 "%s: %d: base is now %zu; was %zu\n",
4129 new_b, PL_collxfrm_base));
4130 PL_collxfrm_base = new_b;
4137 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4138 DEBUG_L(PerlIO_printf(Perl_debug_log,
4139 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4140 *xlen, PERL_INT_MAX));
4144 /* A well-behaved strxfrm() returns exactly how much space it needs
4145 * (usually not including the trailing NUL) when it fails due to not
4146 * enough space being provided. Assume that this is the case unless
4147 * it's been proven otherwise */
4148 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4149 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4151 else { /* Here, either:
4152 * 1) The strxfrm() has previously shown bad behavior; or
4153 * 2) It isn't the first time through the loop, which means
4154 * that the strxfrm() is now showing bad behavior, because
4155 * we gave it what it said was needed in the previous
4156 * iteration, and it came back saying it needed still more.
4157 * (Many versions of cygwin fit this. When the buffer size
4158 * isn't sufficient, they return the input size instead of
4159 * how much is needed.)
4160 * Increase the buffer size by a fixed percentage and try again.
4162 xAlloc += (xAlloc / 4) + 1;
4163 PL_strxfrm_is_behaved = FALSE;
4167 if (DEBUG_Lv_TEST || debug_initialization) {
4168 PerlIO_printf(Perl_debug_log,
4169 "_mem_collxfrm required more space than previously calculated"
4170 " for locale %s, trying again with new guess=%d+%zu\n",
4171 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4172 xAlloc - COLLXFRM_HDR_LEN);
4179 Renew(xbuf, xAlloc, char);
4180 if (UNLIKELY(! xbuf)) {
4181 DEBUG_L(PerlIO_printf(Perl_debug_log,
4182 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4192 if (DEBUG_Lv_TEST || debug_initialization) {
4194 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4195 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4196 PerlIO_printf(Perl_debug_log, "%s\n",
4197 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4203 /* Free up unneeded space; retain ehough for trailing NUL */
4204 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4206 if (s != input_string) {
4214 if (s != input_string) {
4221 if (DEBUG_Lv_TEST || debug_initialization) {
4222 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4233 S_print_collxfrm_input_and_return(pTHX_
4234 const char * const s,
4235 const char * const e,
4236 const STRLEN * const xlen,
4240 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4242 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4243 (UV)PL_collation_ix);
4245 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4248 PerlIO_printf(Perl_debug_log, "NULL");
4250 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4252 print_bytes_for_locale(s, e, is_utf8);
4254 PerlIO_printf(Perl_debug_log, "'\n");
4258 S_print_bytes_for_locale(pTHX_
4259 const char * const s,
4260 const char * const e,
4264 bool prev_was_printable = TRUE;
4265 bool first_time = TRUE;
4267 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4271 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4274 if (! prev_was_printable) {
4275 PerlIO_printf(Perl_debug_log, " ");
4277 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4278 prev_was_printable = TRUE;
4282 PerlIO_printf(Perl_debug_log, " ");
4284 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4285 prev_was_printable = FALSE;
4287 t += (is_utf8) ? UTF8SKIP(t) : 1;
4292 # endif /* #ifdef DEBUGGING */
4293 #endif /* USE_LOCALE_COLLATE */
https://perl5.git.perl.org / perl5.git / blob