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 emulate_setlocale(categories[i], this_locale, i, TRUE);
874 if (strNE(this_locale, default_name)) {
878 Safefree(env_override);
881 /* If all the categories are the same, we can set LC_ALL to
883 if (! did_override) {
884 locale = default_name;
888 /* Here, LC_ALL is no longer valid, as some individual
889 * categories don't match it. We call ourselves
890 * recursively, as that will execute the code that
891 * generates the proper locale string for this situation.
892 * We don't do the remainder of this function, as that is
893 * to update our records, and we've just done that for the
894 * individual categories in the loop above, and doing so
895 * would cause LC_ALL to be done as well */
896 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
901 else if (strchr(locale, ';')) {
903 /* LC_ALL may actually incude a conglomeration of various categories.
904 * Without querylocale, this code uses the glibc (as of this writing)
905 * syntax for representing that, but that is not a stable API, and
906 * other platforms do it differently, so we have to handle all cases
909 const char * s = locale;
910 const char * e = locale + strlen(locale);
912 const char * category_end;
913 const char * name_start;
914 const char * name_end;
919 /* Parse through the category */
920 while (isWORDCHAR(*p)) {
927 "panic: %s: %d: Unexpected character in locale name '%02X",
928 __FILE__, __LINE__, *(p-1));
931 /* Parse through the locale name */
933 while (isGRAPH(*p) && *p != ';') {
940 "panic: %s: %d: Unexpected character in locale name '%02X",
941 __FILE__, __LINE__, *(p-1));
944 /* Find the index of the category name in our lists */
945 for (i = 0; i < LC_ALL_INDEX; i++) {
947 /* Keep going if this isn't the index. The strnNE() avoids a
948 * Perl_form(), but would fail if ever a category name could be
949 * a substring of another one, like if there were a
951 if strnNE(s, category_names[i], category_end - s) {
955 /* If this index is for the single category we're changing, we
956 * have found the locale to set it to. */
957 if (category == categories[i]) {
958 locale = Perl_form(aTHX_ "%.*s",
959 (int) (name_end - name_start),
964 if (category == LC_ALL) {
965 char * individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
966 emulate_setlocale(categories[i], individ_locale, i, TRUE);
967 Safefree(individ_locale);
974 /* Here we have set all the individual categories by recursive calls.
975 * These collectively should have fixed up LC_ALL, so can just query
976 * what that now is */
977 assert(category == LC_ALL);
979 return do_setlocale_c(LC_ALL, NULL);
984 # endif /* end of ! querylocale */
986 /* Ready to create a new locale by modification of the exising one */
987 new_obj = newlocale(mask, locale, old_obj);
994 if (DEBUG_L_TEST || debug_initialization) {
995 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1000 if (! uselocale(old_obj)) {
1004 if (DEBUG_L_TEST || debug_initialization) {
1005 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1017 if (DEBUG_Lv_TEST || debug_initialization) {
1018 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1023 /* And switch into it */
1024 if (! uselocale(new_obj)) {
1029 if (DEBUG_L_TEST || debug_initialization) {
1030 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1035 if (! uselocale(old_obj)) {
1039 if (DEBUG_L_TEST || debug_initialization) {
1040 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1046 freelocale(new_obj);
1053 if (DEBUG_Lv_TEST || debug_initialization) {
1054 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1059 /* We are done, except for updating our records (if the system doesn't keep
1060 * them) and in the case of locale "", we don't actually know what the
1061 * locale that got switched to is, as it came from the environment. So
1062 * have to find it */
1064 # ifdef HAS_QUERYLOCALE
1066 if (strEQ(locale, "")) {
1067 locale = querylocale(mask, new_obj);
1072 /* Here, 'locale' is the return value */
1074 /* Without querylocale(), we have to update our records */
1076 if (category == LC_ALL) {
1079 /* For LC_ALL, we change all individual categories to correspond */
1080 /* PL_curlocales is a parallel array, so has same
1081 * length as 'categories' */
1082 for (i = 0; i <= LC_ALL_INDEX; i++) {
1083 Safefree(PL_curlocales[i]);
1084 PL_curlocales[i] = savepv(locale);
1089 /* For a single category, if it's not the same as the one in LC_ALL, we
1092 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1093 Safefree(PL_curlocales[LC_ALL_INDEX]);
1094 PL_curlocales[LC_ALL_INDEX] = NULL;
1097 /* Then update the category's record */
1098 Safefree(PL_curlocales[index]);
1099 PL_curlocales[index] = savepv(locale);
1107 #endif /* USE_POSIX_2008_LOCALE */
1109 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1110 didn't natively support them */
1112 /* The way this would work is that we would keep a per-thread list of the
1113 * correct locale for that thread. Any operation that was locale-sensitive
1114 * would have to be changed so that it would look like this:
1117 * setlocale to the correct locale for this operation
1121 * This leaves the global locale in the most recently used operation's, but it
1122 * was locked long enough to get the result. If that result is static, it
1123 * needs to be copied before the unlock.
1125 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1126 * the setup, but are no-ops when not needed, and similarly,
1127 * END_LOCALE_DEPENDENT_OP for the tear-down
1129 * But every call to a locale-sensitive function would have to be changed, and
1130 * if a module didn't cooperate by using the mutex, things would break.
1132 * This code was abandoned before being completed or tested, and is left as-is
1135 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1136 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1139 S_locking_setlocale(pTHX_
1141 const char * locale,
1143 const bool is_index_valid
1146 /* This function kind of performs a setlocale() on just the current thread;
1147 * thus it is kind of thread-safe. It does this by keeping a thread-level
1148 * array of the current locales for each category. Every time a locale is
1149 * switched to, it does the switch globally, but updates the thread's
1150 * array. A query as to what the current locale is just returns the
1151 * appropriate element from the array, and doesn't actually call the system
1152 * setlocale(). The saving into the array is done in an uninterruptible
1153 * section of code, so is unaffected by whatever any other threads might be
1156 * All locale-sensitive operations must work by first starting a critical
1157 * section, then switching to the thread's locale as kept by this function,
1158 * and then doing the operation, then ending the critical section. Thus,
1159 * each gets done in the appropriate locale. simulating thread-safety.
1161 * This function takes the same parameters, 'category' and 'locale', that
1162 * the regular setlocale() function does, but it also takes two additional
1163 * ones. This is because as described earlier. If we know on input the
1164 * index corresponding to the category into the array where we store the
1165 * current locales, we don't have to calculate it. If the caller knows at
1166 * compile time what the index is, it it can pass it, setting
1167 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1171 /* If the input index might be incorrect, calculate the correct one */
1172 if (! is_index_valid) {
1175 if (DEBUG_Lv_TEST || debug_initialization) {
1176 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1179 for (i = 0; i <= LC_ALL_INDEX; i++) {
1180 if (category == categories[i]) {
1186 /* Here, we don't know about this category, so can't handle it.
1187 * XXX best we can do is to unsafely set this
1190 return my_setlocale(category, locale);
1194 if (DEBUG_Lv_TEST || debug_initialization) {
1195 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1199 /* For a query, just return what's in our records */
1200 if (new_locale == NULL) {
1201 return curlocales[index];
1205 /* Otherwise, we need to do the switch, and save the result, all in a
1206 * critical section */
1208 Safefree(curlocales[[index]]);
1210 /* It might be that this is called from an already-locked section of code.
1211 * We would have to detect and skip the LOCK/UNLOCK if so */
1214 curlocales[index] = savepv(my_setlocale(category, new_locale));
1216 if (strEQ(new_locale, "")) {
1220 /* The locale values come from the environment, and may not all be the
1221 * same, so for LC_ALL, we have to update all the others, while the
1222 * mutex is still locked */
1224 if (category == LC_ALL) {
1226 for (i = 0; i < LC_ALL_INDEX) {
1227 curlocales[i] = my_setlocale(categories[i], NULL);
1236 return curlocales[index];
1242 S_set_numeric_radix(pTHX_ const bool use_locale)
1244 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1245 * TRUE, use the radix character derived from the current locale */
1247 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1248 || defined(HAS_NL_LANGINFO))
1250 const char * radix = (use_locale)
1251 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1252 /* FALSE => already in dest locale */
1255 sv_setpv(PL_numeric_radix_sv, radix);
1257 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1258 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1259 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1260 SvCUR(PL_numeric_radix_sv))
1261 && _is_cur_LC_category_utf8(LC_NUMERIC))
1263 SvUTF8_on(PL_numeric_radix_sv);
1268 if (DEBUG_L_TEST || debug_initialization) {
1269 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1270 SvPVX(PL_numeric_radix_sv),
1271 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1275 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1280 S_new_numeric(pTHX_ const char *newnum)
1283 #ifndef USE_LOCALE_NUMERIC
1285 PERL_UNUSED_ARG(newnum);
1289 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1290 * core Perl this and that 'newnum' is the name of the new locale.
1291 * It installs this locale as the current underlying default.
1293 * The default locale and the C locale can be toggled between by use of the
1294 * set_numeric_underlying() and set_numeric_standard() functions, which
1295 * should probably not be called directly, but only via macros like
1296 * SET_NUMERIC_STANDARD() in perl.h.
1298 * The toggling is necessary mainly so that a non-dot radix decimal point
1299 * character can be output, while allowing internal calculations to use a
1302 * This sets several interpreter-level variables:
1303 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1304 * PL_numeric_underlying A boolean indicating if the toggled state is such
1305 * that the current locale is the program's underlying
1307 * PL_numeric_standard An int indicating if the toggled state is such
1308 * that the current locale is the C locale or
1309 * indistinguishable from the C locale. If non-zero, it
1310 * is in C; if > 1, it means it may not be toggled away
1312 * PL_numeric_underlying_is_standard A bool kept by this function
1313 * indicating that the underlying locale and the standard
1314 * C locale are indistinguishable for the purposes of
1315 * LC_NUMERIC. This happens when both of the above two
1316 * variables are true at the same time. (Toggling is a
1317 * no-op under these circumstances.) This variable is
1318 * used to avoid having to recalculate.
1324 Safefree(PL_numeric_name);
1325 PL_numeric_name = NULL;
1326 PL_numeric_standard = TRUE;
1327 PL_numeric_underlying = TRUE;
1328 PL_numeric_underlying_is_standard = TRUE;
1332 save_newnum = stdize_locale(savepv(newnum));
1333 PL_numeric_underlying = TRUE;
1334 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1336 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1338 if (! PL_numeric_standard) {
1339 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1340 FALSE /* Don't toggle locale */ ))
1341 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1345 /* Save the new name if it isn't the same as the previous one, if any */
1346 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1347 Safefree(PL_numeric_name);
1348 PL_numeric_name = save_newnum;
1351 Safefree(save_newnum);
1354 PL_numeric_underlying_is_standard = PL_numeric_standard;
1356 # ifdef HAS_POSIX_2008_LOCALE
1358 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1360 PL_underlying_numeric_obj);
1364 if (DEBUG_L_TEST || debug_initialization) {
1365 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1368 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1369 * have to worry about the radix being a non-dot. (Core operations that
1370 * need the underlying locale change to it temporarily). */
1371 if (PL_numeric_standard) {
1372 set_numeric_radix(0);
1375 set_numeric_standard();
1378 #endif /* USE_LOCALE_NUMERIC */
1383 Perl_set_numeric_standard(pTHX)
1386 #ifdef USE_LOCALE_NUMERIC
1388 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1389 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1390 * macro avoids calling this routine if toggling isn't necessary according
1391 * to our records (which could be wrong if some XS code has changed the
1392 * locale behind our back) */
1394 do_setlocale_c(LC_NUMERIC, "C");
1395 PL_numeric_standard = TRUE;
1396 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1397 set_numeric_radix(0);
1401 if (DEBUG_L_TEST || debug_initialization) {
1402 PerlIO_printf(Perl_debug_log,
1403 "LC_NUMERIC locale now is standard C\n");
1407 #endif /* USE_LOCALE_NUMERIC */
1412 Perl_set_numeric_underlying(pTHX)
1415 #ifdef USE_LOCALE_NUMERIC
1417 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1418 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1419 * instead of calling this directly. The macro avoids calling this routine
1420 * if toggling isn't necessary according to our records (which could be
1421 * wrong if some XS code has changed the locale behind our back) */
1423 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1424 PL_numeric_standard = PL_numeric_underlying_is_standard;
1425 PL_numeric_underlying = TRUE;
1426 set_numeric_radix(! PL_numeric_standard);
1430 if (DEBUG_L_TEST || debug_initialization) {
1431 PerlIO_printf(Perl_debug_log,
1432 "LC_NUMERIC locale now is %s\n",
1437 #endif /* USE_LOCALE_NUMERIC */
1442 * Set up for a new ctype locale.
1445 S_new_ctype(pTHX_ const char *newctype)
1448 #ifndef USE_LOCALE_CTYPE
1450 PERL_ARGS_ASSERT_NEW_CTYPE;
1451 PERL_UNUSED_ARG(newctype);
1452 PERL_UNUSED_CONTEXT;
1456 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1457 * core Perl this and that 'newctype' is the name of the new locale.
1459 * This function sets up the folding arrays for all 256 bytes, assuming
1460 * that tofold() is tolc() since fold case is not a concept in POSIX,
1462 * Any code changing the locale (outside this file) should use
1463 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1464 * this function should be called directly only from this file and from
1465 * POSIX::setlocale() */
1470 /* Don't check for problems if we are suppressing the warnings */
1471 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1473 PERL_ARGS_ASSERT_NEW_CTYPE;
1475 /* We will replace any bad locale warning with 1) nothing if the new one is
1476 * ok; or 2) a new warning for the bad new locale */
1477 if (PL_warn_locale) {
1478 SvREFCNT_dec_NN(PL_warn_locale);
1479 PL_warn_locale = NULL;
1482 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1484 /* A UTF-8 locale gets standard rules. But note that code still has to
1485 * handle this specially because of the three problematic code points */
1486 if (PL_in_utf8_CTYPE_locale) {
1487 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1490 /* We don't populate the other lists if a UTF-8 locale, but do check that
1491 * everything works as expected, unless checking turned off */
1492 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1493 /* Assume enough space for every character being bad. 4 spaces each
1494 * for the 94 printable characters that are output like "'x' "; and 5
1495 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1497 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1498 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1500 unsigned int bad_count = 0; /* Count of bad characters */
1502 for (i = 0; i < 256; i++) {
1503 if (! PL_in_utf8_CTYPE_locale) {
1505 PL_fold_locale[i] = (U8) tolower(i);
1506 else if (islower(i))
1507 PL_fold_locale[i] = (U8) toupper(i);
1509 PL_fold_locale[i] = (U8) i;
1512 /* If checking for locale problems, see if the native ASCII-range
1513 * printables plus \n and \t are in their expected categories in
1514 * the new locale. If not, this could mean big trouble, upending
1515 * Perl's and most programs' assumptions, like having a
1516 * metacharacter with special meaning become a \w. Fortunately,
1517 * it's very rare to find locales that aren't supersets of ASCII
1518 * nowadays. It isn't a problem for most controls to be changed
1519 * into something else; we check only \n and \t, though perhaps \r
1520 * could be an issue as well. */
1521 if ( check_for_problems
1522 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1524 bool is_bad = FALSE;
1525 char name[3] = { '\0' };
1527 /* Convert the name into a string */
1532 else if (i == '\n') {
1533 my_strlcpy(name, "\n", sizeof(name));
1536 my_strlcpy(name, "\t", sizeof(name));
1539 /* Check each possibe class */
1540 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1542 DEBUG_L(PerlIO_printf(Perl_debug_log,
1543 "isalnum('%s') unexpectedly is %d\n",
1544 name, cBOOL(isalnum(i))));
1546 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1548 DEBUG_L(PerlIO_printf(Perl_debug_log,
1549 "isalpha('%s') unexpectedly is %d\n",
1550 name, cBOOL(isalpha(i))));
1552 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1554 DEBUG_L(PerlIO_printf(Perl_debug_log,
1555 "isdigit('%s') unexpectedly is %d\n",
1556 name, cBOOL(isdigit(i))));
1558 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1560 DEBUG_L(PerlIO_printf(Perl_debug_log,
1561 "isgraph('%s') unexpectedly is %d\n",
1562 name, cBOOL(isgraph(i))));
1564 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1566 DEBUG_L(PerlIO_printf(Perl_debug_log,
1567 "islower('%s') unexpectedly is %d\n",
1568 name, cBOOL(islower(i))));
1570 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1572 DEBUG_L(PerlIO_printf(Perl_debug_log,
1573 "isprint('%s') unexpectedly is %d\n",
1574 name, cBOOL(isprint(i))));
1576 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1578 DEBUG_L(PerlIO_printf(Perl_debug_log,
1579 "ispunct('%s') unexpectedly is %d\n",
1580 name, cBOOL(ispunct(i))));
1582 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isspace('%s') unexpectedly is %d\n",
1586 name, cBOOL(isspace(i))));
1588 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isupper('%s') unexpectedly is %d\n",
1592 name, cBOOL(isupper(i))));
1594 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isxdigit('%s') unexpectedly is %d\n",
1598 name, cBOOL(isxdigit(i))));
1600 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1604 name, tolower(i), (int) toLOWER_A(i)));
1606 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1610 name, toupper(i), (int) toUPPER_A(i)));
1612 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "'\\n' (=%02X) is not a control\n", (int) i));
1618 /* Add to the list; Separate multiple entries with a blank */
1621 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1623 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1631 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1632 * this locale requires more than one byte, there are going to be
1634 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1635 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1636 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1638 if ( check_for_problems && MB_CUR_MAX > 1
1639 && ! PL_in_utf8_CTYPE_locale
1641 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1642 * locale. Just assume that the implementation for them (plus
1643 * for POSIX) is correct and the > 1 value is spurious. (Since
1644 * these are specially handled to never be considered UTF-8
1645 * locales, as long as this is the only problem, everything
1646 * should work fine */
1647 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1649 multi_byte_locale = TRUE;
1654 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1655 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1656 PL_warn_locale = Perl_newSVpvf(aTHX_
1657 "Locale '%s' contains (at least) the following characters"
1658 " which have\nunexpected meanings: %s\nThe Perl program"
1659 " will use the expected meanings",
1660 newctype, bad_chars_list);
1663 PL_warn_locale = Perl_newSVpvf(aTHX_
1664 "Locale '%s' may not work well.%s%s%s\n",
1667 ? " Some characters in it are not recognized by"
1671 ? "\nThe following characters (and maybe others)"
1672 " may not have the same meaning as the Perl"
1673 " program expects:\n"
1681 # ifdef HAS_NL_LANGINFO
1683 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1684 /* parameter FALSE is a don't care here */
1685 my_nl_langinfo(PERL_CODESET, FALSE));
1689 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1691 /* If we are actually in the scope of the locale or are debugging,
1692 * output the message now. If not in that scope, we save the
1693 * message to be output at the first operation using this locale,
1694 * if that actually happens. Most programs don't use locales, so
1695 * they are immune to bad ones. */
1696 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1698 /* The '0' below suppresses a bogus gcc compiler warning */
1699 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1701 if (IN_LC(LC_CTYPE)) {
1702 SvREFCNT_dec_NN(PL_warn_locale);
1703 PL_warn_locale = NULL;
1709 #endif /* USE_LOCALE_CTYPE */
1714 Perl__warn_problematic_locale()
1717 #ifdef USE_LOCALE_CTYPE
1721 /* Internal-to-core function that outputs the message in PL_warn_locale,
1722 * and then NULLS it. Should be called only through the macro
1723 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1725 if (PL_warn_locale) {
1726 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1727 SvPVX(PL_warn_locale),
1728 0 /* dummy to avoid compiler warning */ );
1729 SvREFCNT_dec_NN(PL_warn_locale);
1730 PL_warn_locale = NULL;
1738 S_new_collate(pTHX_ const char *newcoll)
1741 #ifndef USE_LOCALE_COLLATE
1743 PERL_UNUSED_ARG(newcoll);
1744 PERL_UNUSED_CONTEXT;
1748 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1749 * core Perl this and that 'newcoll' is the name of the new locale.
1751 * The design of locale collation is that every locale change is given an
1752 * index 'PL_collation_ix'. The first time a string particpates in an
1753 * operation that requires collation while locale collation is active, it
1754 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1755 * magic includes the collation index, and the transformation of the string
1756 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1757 * instead of the string itself. If a string changes, the magic is
1758 * cleared. The next time the locale changes, the index is incremented,
1759 * and so we know during a comparison that the transformation is not
1760 * necessarily still valid, and so is recomputed. Note that if the locale
1761 * changes enough times, the index could wrap (a U32), and it is possible
1762 * that a transformation would improperly be considered valid, leading to
1763 * an unlikely bug */
1766 if (PL_collation_name) {
1768 Safefree(PL_collation_name);
1769 PL_collation_name = NULL;
1771 PL_collation_standard = TRUE;
1772 is_standard_collation:
1773 PL_collxfrm_base = 0;
1774 PL_collxfrm_mult = 2;
1775 PL_in_utf8_COLLATE_locale = FALSE;
1776 PL_strxfrm_NUL_replacement = '\0';
1777 PL_strxfrm_max_cp = 0;
1781 /* If this is not the same locale as currently, set the new one up */
1782 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1784 Safefree(PL_collation_name);
1785 PL_collation_name = stdize_locale(savepv(newcoll));
1786 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1787 if (PL_collation_standard) {
1788 goto is_standard_collation;
1791 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1792 PL_strxfrm_NUL_replacement = '\0';
1793 PL_strxfrm_max_cp = 0;
1795 /* A locale collation definition includes primary, secondary, tertiary,
1796 * etc. weights for each character. To sort, the primary weights are
1797 * used, and only if they compare equal, then the secondary weights are
1798 * used, and only if they compare equal, then the tertiary, etc.
1800 * strxfrm() works by taking the input string, say ABC, and creating an
1801 * output transformed string consisting of first the primary weights,
1802 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1803 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1804 * may not have weights at every level. In our example, let's say B
1805 * doesn't have a tertiary weight, and A doesn't have a secondary
1806 * weight. The constructed string is then going to be
1807 * A¹B¹C¹ B²C² A³C³ ....
1808 * This has the desired effect that strcmp() will look at the secondary
1809 * or tertiary weights only if the strings compare equal at all higher
1810 * priority weights. The spaces shown here, like in
1812 * are not just for readability. In the general case, these must
1813 * actually be bytes, which we will call here 'separator weights'; and
1814 * they must be smaller than any other weight value, but since these
1815 * are C strings, only the terminating one can be a NUL (some
1816 * implementations may include a non-NUL separator weight just before
1817 * the NUL). Implementations tend to reserve 01 for the separator
1818 * weights. They are needed so that a shorter string's secondary
1819 * weights won't be misconstrued as primary weights of a longer string,
1820 * etc. By making them smaller than any other weight, the shorter
1821 * string will sort first. (Actually, if all secondary weights are
1822 * smaller than all primary ones, there is no need for a separator
1823 * weight between those two levels, etc.)
1825 * The length of the transformed string is roughly a linear function of
1826 * the input string. It's not exactly linear because some characters
1827 * don't have weights at all levels. When we call strxfrm() we have to
1828 * allocate some memory to hold the transformed string. The
1829 * calculations below try to find coefficients 'm' and 'b' for this
1830 * locale so that m*x + b equals how much space we need, given the size
1831 * of the input string in 'x'. If we calculate too small, we increase
1832 * the size as needed, and call strxfrm() again, but it is better to
1833 * get it right the first time to avoid wasted expensive string
1834 * transformations. */
1837 /* We use the string below to find how long the tranformation of it
1838 * is. Almost all locales are supersets of ASCII, or at least the
1839 * ASCII letters. We use all of them, half upper half lower,
1840 * because if we used fewer, we might hit just the ones that are
1841 * outliers in a particular locale. Most of the strings being
1842 * collated will contain a preponderance of letters, and even if
1843 * they are above-ASCII, they are likely to have the same number of
1844 * weight levels as the ASCII ones. It turns out that digits tend
1845 * to have fewer levels, and some punctuation has more, but those
1846 * are relatively sparse in text, and khw believes this gives a
1847 * reasonable result, but it could be changed if experience so
1849 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1850 char * x_longer; /* Transformed 'longer' */
1851 Size_t x_len_longer; /* Length of 'x_longer' */
1853 char * x_shorter; /* We also transform a substring of 'longer' */
1854 Size_t x_len_shorter;
1856 /* _mem_collxfrm() is used get the transformation (though here we
1857 * are interested only in its length). It is used because it has
1858 * the intelligence to handle all cases, but to work, it needs some
1859 * values of 'm' and 'b' to get it started. For the purposes of
1860 * this calculation we use a very conservative estimate of 'm' and
1861 * 'b'. This assumes a weight can be multiple bytes, enough to
1862 * hold any UV on the platform, and there are 5 levels, 4 weight
1863 * bytes, and a trailing NUL. */
1864 PL_collxfrm_base = 5;
1865 PL_collxfrm_mult = 5 * sizeof(UV);
1867 /* Find out how long the transformation really is */
1868 x_longer = _mem_collxfrm(longer,
1872 /* We avoid converting to UTF-8 in the
1873 * called function by telling it the
1874 * string is in UTF-8 if the locale is a
1875 * UTF-8 one. Since the string passed
1876 * here is invariant under UTF-8, we can
1877 * claim it's UTF-8 even though it isn't.
1879 PL_in_utf8_COLLATE_locale);
1882 /* Find out how long the transformation of a substring of 'longer'
1883 * is. Together the lengths of these transformations are
1884 * sufficient to calculate 'm' and 'b'. The substring is all of
1885 * 'longer' except the first character. This minimizes the chances
1886 * of being swayed by outliers */
1887 x_shorter = _mem_collxfrm(longer + 1,
1890 PL_in_utf8_COLLATE_locale);
1891 Safefree(x_shorter);
1893 /* If the results are nonsensical for this simple test, the whole
1894 * locale definition is suspect. Mark it so that locale collation
1895 * is not active at all for it. XXX Should we warn? */
1896 if ( x_len_shorter == 0
1897 || x_len_longer == 0
1898 || x_len_shorter >= x_len_longer)
1900 PL_collxfrm_mult = 0;
1901 PL_collxfrm_base = 0;
1904 SSize_t base; /* Temporary */
1906 /* We have both: m * strlen(longer) + b = x_len_longer
1907 * m * strlen(shorter) + b = x_len_shorter;
1908 * subtracting yields:
1909 * m * (strlen(longer) - strlen(shorter))
1910 * = x_len_longer - x_len_shorter
1911 * But we have set things up so that 'shorter' is 1 byte smaller
1912 * than 'longer'. Hence:
1913 * m = x_len_longer - x_len_shorter
1915 * But if something went wrong, make sure the multiplier is at
1918 if (x_len_longer > x_len_shorter) {
1919 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1922 PL_collxfrm_mult = 1;
1927 * but in case something has gone wrong, make sure it is
1929 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1934 /* Add 1 for the trailing NUL */
1935 PL_collxfrm_base = base + 1;
1940 if (DEBUG_L_TEST || debug_initialization) {
1941 PerlIO_printf(Perl_debug_log,
1942 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1944 " collate multipler=%zu, collate base=%zu\n",
1946 PL_in_utf8_COLLATE_locale,
1947 x_len_shorter, x_len_longer,
1948 PL_collxfrm_mult, PL_collxfrm_base);
1955 #endif /* USE_LOCALE_COLLATE */
1962 S_win32_setlocale(pTHX_ int category, const char* locale)
1964 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1965 * difference between the two unless the input locale is "", which normally
1966 * means on Windows to get the machine default, which is set via the
1967 * computer's "Regional and Language Options" (or its current equivalent).
1968 * In POSIX, it instead means to find the locale from the user's
1969 * environment. This routine changes the Windows behavior to first look in
1970 * the environment, and, if anything is found, use that instead of going to
1971 * the machine default. If there is no environment override, the machine
1972 * default is used, by calling the real setlocale() with "".
1974 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1975 * use the particular category's variable if set; otherwise to use the LANG
1978 bool override_LC_ALL = FALSE;
1982 if (locale && strEQ(locale, "")) {
1986 locale = PerlEnv_getenv("LC_ALL");
1988 if (category == LC_ALL) {
1989 override_LC_ALL = TRUE;
1995 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1996 if (category == categories[i]) {
1997 locale = PerlEnv_getenv(category_names[i]);
2002 locale = PerlEnv_getenv("LANG");
2018 result = setlocale(category, locale);
2019 DEBUG_L(STMT_START {
2021 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2022 setlocale_debug_string(category, locale, result));
2026 if (! override_LC_ALL) {
2030 /* Here the input category was LC_ALL, and we have set it to what is in the
2031 * LANG variable or the system default if there is no LANG. But these have
2032 * lower priority than the other LC_foo variables, so override it for each
2033 * one that is set. (If they are set to "", it means to use the same thing
2034 * we just set LC_ALL to, so can skip) */
2036 for (i = 0; i < LC_ALL_INDEX; i++) {
2037 result = PerlEnv_getenv(category_names[i]);
2038 if (result && strNE(result, "")) {
2039 setlocale(categories[i], result);
2040 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2042 setlocale_debug_string(categories[i], result, "not captured")));
2046 result = setlocale(LC_ALL, NULL);
2047 DEBUG_L(STMT_START {
2049 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2051 setlocale_debug_string(LC_ALL, NULL, result));
2062 =head1 Locale-related functions and macros
2064 =for apidoc Perl_setlocale
2066 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2067 taking the same parameters, and returning the same information, except that it
2068 returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2069 perl keeps that locale category as C<C>, changing it briefly during the
2070 operations where the underlying one is required.
2072 Another reason it isn't completely a drop-in replacement is that it is
2073 declared to return S<C<const char *>>, whereas the system setlocale omits the
2074 C<const>. (If it were being written today, plain setlocale would be declared
2075 const, since it is illegal to change the information it returns; doing so leads
2078 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2079 C<setlocale> can be completely ineffective on some platforms under some
2082 C<Perl_setlocale> should not be used to change the locale except on systems
2083 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2084 the system C<setlocale()> is ineffective, returning the wrong information, and
2085 failing to actually change the locale. C<Perl_setlocale>, however works
2086 properly in all circumstances.
2088 The return points to a per-thread static buffer, which is overwritten the next
2089 time C<Perl_setlocale> is called from the same thread.
2096 Perl_setlocale(const int category, const char * locale)
2098 /* This wraps POSIX::setlocale() */
2100 const char * retval;
2101 const char * newlocale;
2103 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2106 #ifdef USE_LOCALE_NUMERIC
2108 /* A NULL locale means only query what the current one is. We have the
2109 * LC_NUMERIC name saved, because we are normally switched into the C
2110 * locale for it. For an LC_ALL query, switch back to get the correct
2111 * results. All other categories don't require special handling */
2112 if (locale == NULL) {
2113 if (category == LC_NUMERIC) {
2115 /* We don't have to copy this return value, as it is a per-thread
2116 * variable, and won't change until a future setlocale */
2117 return PL_numeric_name;
2122 else if (category == LC_ALL) {
2123 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2132 retval = save_to_buffer(do_setlocale_r(category, locale),
2133 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2136 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2138 if (locale == NULL && category == LC_ALL) {
2139 RESTORE_LC_NUMERIC();
2144 DEBUG_L(PerlIO_printf(Perl_debug_log,
2145 "%s:%d: %s\n", __FILE__, __LINE__,
2146 setlocale_debug_string(category, locale, retval)));
2154 /* If locale == NULL, we are just querying the state */
2155 if (locale == NULL) {
2159 /* Now that have switched locales, we have to update our records to
2164 #ifdef USE_LOCALE_CTYPE
2171 #ifdef USE_LOCALE_COLLATE
2174 new_collate(retval);
2178 #ifdef USE_LOCALE_NUMERIC
2181 new_numeric(retval);
2189 /* LC_ALL updates all the things we care about. The values may not
2190 * be the same as 'retval', as the locale "" may have set things
2193 # ifdef USE_LOCALE_CTYPE
2195 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2196 new_ctype(newlocale);
2198 # endif /* USE_LOCALE_CTYPE */
2199 # ifdef USE_LOCALE_COLLATE
2201 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2202 new_collate(newlocale);
2205 # ifdef USE_LOCALE_NUMERIC
2207 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2208 new_numeric(newlocale);
2210 # endif /* USE_LOCALE_NUMERIC */
2221 PERL_STATIC_INLINE const char *
2222 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2224 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2225 * growing it if necessary */
2229 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2235 string_size = strlen(string) + offset + 1;
2237 if (*buf_size == 0) {
2238 Newx(*buf, string_size, char);
2239 *buf_size = string_size;
2241 else if (string_size > *buf_size) {
2242 Renew(*buf, string_size, char);
2243 *buf_size = string_size;
2246 Copy(string, *buf + offset, string_size - offset, char);
2252 =for apidoc Perl_langinfo
2254 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2255 taking the same C<item> parameter values, and returning the same information.
2256 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2257 of Perl's locale handling from your code, and can be used on systems that lack
2258 a native C<nl_langinfo>.
2266 The reason it isn't quite a drop-in replacement is actually an advantage. The
2267 only difference is that it returns S<C<const char *>>, whereas plain
2268 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2269 forbidden to write into the buffer. By declaring this C<const>, the compiler
2270 enforces this restriction, so if it is violated, you know at compilation time,
2271 rather than getting segfaults at runtime.
2275 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2276 without you having to write extra code. The reason for the extra code would be
2277 because these are from the C<LC_NUMERIC> locale category, which is normally
2278 kept set to the C locale by Perl, no matter what the underlying locale is
2279 supposed to be, and so to get the expected results, you have to temporarily
2280 toggle into the underlying locale, and later toggle back. (You could use plain
2281 C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2282 then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2283 C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2284 radix (decimal point) character to be a dot.)
2288 The system function it replaces can have its static return buffer trashed,
2289 not only by a subesequent call to that function, but by a C<freelocale>,
2290 C<setlocale>, or other locale change. The returned buffer of this function is
2291 not changed until the next call to it, so the buffer is never in a trashed
2296 Its return buffer is per-thread, so it also is never overwritten by a call to
2297 this function from another thread; unlike the function it replaces.
2301 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2302 as Windows, hence makes your code more portable. Of the fifty-some possible
2303 items specified by the POSIX 2008 standard,
2304 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2305 only two are completely unimplemented (though the loss of one of these is
2306 significant). It uses various techniques to recover the other items, including
2307 calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2308 in C89, so should be always be available. Later C<strftime()> versions have
2309 additional capabilities; C<""> is returned for those not available on your
2312 It is important to note that when called with an item that is recovered by
2313 using C<localeconv>, the buffer from any previous explicit call to
2314 C<localeconv> will be overwritten. This means you must save that buffer's
2315 contents if you need to access them after a call to this function.
2317 The details for those items which may differ from what this emulation returns
2318 and what a native C<nl_langinfo()> would return are:
2326 Unimplemented, so returns C<"">.
2336 Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
2337 removed from POSIX 2008, and are retained here for backwards compatibility.
2338 Your platform's C<nl_langinfo> may not support them.
2342 Always evaluates to C<%x>, the locale's appropriate date representation.
2346 Always evaluates to C<%X>, the locale's appropriate time representation.
2350 Always evaluates to C<%c>, the locale's appropriate date and time
2355 The return may be incorrect for those rare locales where the currency symbol
2356 replaces the radix character.
2357 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2358 to work differently.
2362 Currently this gives the same results as Linux does.
2363 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2364 to work differently.
2370 =item C<ERA_D_T_FMT>
2374 These are derived by using C<strftime()>, and not all versions of that function
2375 know about them. C<""> is returned for these on such systems.
2381 When using C<Perl_langinfo> on systems that don't have a native
2382 C<nl_langinfo()>, you must
2384 #include "perl_langinfo.h"
2386 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2387 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2388 C<langinfo.h> imports into the namespace for code that doesn't need it.)
2390 You also should not use the bare C<langinfo.h> item names, but should preface
2391 them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2392 The C<PERL_I<foo>> versions will also work for this function on systems that do
2393 have a native C<nl_langinfo>.
2395 The original impetus for C<Perl_langinfo()> was so that code that needs to
2396 find out the current currency symbol, floating point radix character, or digit
2397 grouping separator can use, on all systems, the simpler and more
2398 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2399 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2400 is better to use the methods given in L<perlcall> to call
2401 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2408 #ifdef HAS_NL_LANGINFO
2409 Perl_langinfo(const nl_item item)
2411 Perl_langinfo(const int item)
2414 return my_nl_langinfo(item, TRUE);
2418 #ifdef HAS_NL_LANGINFO
2419 S_my_nl_langinfo(const nl_item item, bool toggle)
2421 S_my_nl_langinfo(const int item, bool toggle)
2425 const char * retval;
2427 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2428 * two items, and only if not already there */
2429 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2430 || PL_numeric_underlying))
2435 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2436 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2437 || ! defined(HAS_POSIX_2008_LOCALE)
2439 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2440 * for those items dependent on it. This must be copied to a buffer before
2441 * switching back, as some systems destroy the buffer when setlocale() is
2445 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2448 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2451 LOCALE_LOCK; /* Prevent interference from another thread executing
2452 this code section (the only call to nl_langinfo in
2456 /* Copy to a per-thread buffer, which is also one that won't be
2457 * destroyed by a subsequent setlocale(), such as the
2458 * RESTORE_LC_NUMERIC may do just below. */
2459 retval = save_to_buffer(nl_langinfo(item),
2460 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2465 RESTORE_LC_NUMERIC();
2469 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2472 bool do_free = FALSE;
2473 locale_t cur = uselocale((locale_t) 0);
2475 if (cur == LC_GLOBAL_LOCALE) {
2476 cur = duplocale(LC_GLOBAL_LOCALE);
2481 if (PL_underlying_numeric_obj) {
2482 cur = PL_underlying_numeric_obj;
2485 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2490 /* We have to save it to a buffer, because the freelocale() just below
2491 * can invalidate the internal one */
2492 retval = save_to_buffer(nl_langinfo_l(item, cur),
2493 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2502 if (strEQ(retval, "")) {
2503 if (item == PERL_YESSTR) {
2506 if (item == PERL_NOSTR) {
2513 #else /* Below, emulate nl_langinfo as best we can */
2517 # ifdef HAS_LOCALECONV
2519 const struct lconv* lc;
2521 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2524 # ifdef HAS_STRFTIME
2527 bool return_format = FALSE; /* Return the %format, not the value */
2528 const char * format;
2532 /* We copy the results to a per-thread buffer, even if not
2533 * multi-threaded. This is in part to simplify this code, and partly
2534 * because we need a buffer anyway for strftime(), and partly because a
2535 * call of localeconv() could otherwise wipe out the buffer, and the
2536 * programmer would not be expecting this, as this is a nl_langinfo()
2537 * substitute after all, so s/he might be thinking their localeconv()
2538 * is safe until another localeconv() call. */
2543 /* These 2 are unimplemented */
2545 case PERL_ERA: /* For use with strftime() %E modifier */
2550 /* We use only an English set, since we don't know any more */
2551 case PERL_YESEXPR: return "^[+1yY]";
2552 case PERL_YESSTR: return "yes";
2553 case PERL_NOEXPR: return "^[-0nN]";
2554 case PERL_NOSTR: return "no";
2556 # ifdef HAS_LOCALECONV
2560 /* We don't bother with localeconv_l() because any system that
2561 * has it is likely to also have nl_langinfo() */
2563 LOCALE_LOCK; /* Prevent interference with other threads
2564 using localeconv() */
2568 || ! lc->currency_symbol
2569 || strEQ("", lc->currency_symbol))
2575 /* Leave the first spot empty to be filled in below */
2576 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2577 &PL_langinfo_bufsize, 1);
2578 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2579 { /* khw couldn't figure out how the localedef specifications
2580 would show that the $ should replace the radix; this is
2581 just a guess as to how it might work.*/
2582 PL_langinfo_buf[0] = '.';
2584 else if (lc->p_cs_precedes) {
2585 PL_langinfo_buf[0] = '-';
2588 PL_langinfo_buf[0] = '+';
2594 case PERL_RADIXCHAR:
2598 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2601 LOCALE_LOCK; /* Prevent interference with other threads
2602 using localeconv() */
2609 temp = (item == PERL_RADIXCHAR)
2611 : lc->thousands_sep;
2617 retval = save_to_buffer(temp, &PL_langinfo_buf,
2618 &PL_langinfo_bufsize, 0);
2623 RESTORE_LC_NUMERIC();
2629 # ifdef HAS_STRFTIME
2631 /* These are defined by C89, so we assume that strftime supports
2632 * them, and so are returned unconditionally; they may not be what
2633 * the locale actually says, but should give good enough results
2634 * for someone using them as formats (as opposed to trying to parse
2635 * them to figure out what the locale says). The other format
2636 * items are actually tested to verify they work on the platform */
2637 case PERL_D_FMT: return "%x";
2638 case PERL_T_FMT: return "%X";
2639 case PERL_D_T_FMT: return "%c";
2641 /* These formats are only available in later strfmtime's */
2642 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2643 case PERL_T_FMT_AMPM:
2645 /* The rest can be gotten from most versions of strftime(). */
2646 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2647 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2649 case PERL_ALT_DIGITS:
2650 case PERL_AM_STR: case PERL_PM_STR:
2651 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2652 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2653 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2654 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2655 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2656 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2657 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2658 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2659 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2664 init_tm(&tm); /* Precaution against core dumps */
2668 tm.tm_year = 2017 - 1900;
2675 "panic: %s: %d: switch case: %d problem",
2676 __FILE__, __LINE__, item);
2677 NOT_REACHED; /* NOTREACHED */
2679 case PERL_PM_STR: tm.tm_hour = 18;
2684 case PERL_ABDAY_7: tm.tm_wday++;
2685 case PERL_ABDAY_6: tm.tm_wday++;
2686 case PERL_ABDAY_5: tm.tm_wday++;
2687 case PERL_ABDAY_4: tm.tm_wday++;
2688 case PERL_ABDAY_3: tm.tm_wday++;
2689 case PERL_ABDAY_2: tm.tm_wday++;
2694 case PERL_DAY_7: tm.tm_wday++;
2695 case PERL_DAY_6: tm.tm_wday++;
2696 case PERL_DAY_5: tm.tm_wday++;
2697 case PERL_DAY_4: tm.tm_wday++;
2698 case PERL_DAY_3: tm.tm_wday++;
2699 case PERL_DAY_2: tm.tm_wday++;
2704 case PERL_ABMON_12: tm.tm_mon++;
2705 case PERL_ABMON_11: tm.tm_mon++;
2706 case PERL_ABMON_10: tm.tm_mon++;
2707 case PERL_ABMON_9: tm.tm_mon++;
2708 case PERL_ABMON_8: tm.tm_mon++;
2709 case PERL_ABMON_7: tm.tm_mon++;
2710 case PERL_ABMON_6: tm.tm_mon++;
2711 case PERL_ABMON_5: tm.tm_mon++;
2712 case PERL_ABMON_4: tm.tm_mon++;
2713 case PERL_ABMON_3: tm.tm_mon++;
2714 case PERL_ABMON_2: tm.tm_mon++;
2719 case PERL_MON_12: tm.tm_mon++;
2720 case PERL_MON_11: tm.tm_mon++;
2721 case PERL_MON_10: tm.tm_mon++;
2722 case PERL_MON_9: tm.tm_mon++;
2723 case PERL_MON_8: tm.tm_mon++;
2724 case PERL_MON_7: tm.tm_mon++;
2725 case PERL_MON_6: tm.tm_mon++;
2726 case PERL_MON_5: tm.tm_mon++;
2727 case PERL_MON_4: tm.tm_mon++;
2728 case PERL_MON_3: tm.tm_mon++;
2729 case PERL_MON_2: tm.tm_mon++;
2734 case PERL_T_FMT_AMPM:
2736 return_format = TRUE;
2739 case PERL_ERA_D_FMT:
2741 return_format = TRUE;
2744 case PERL_ERA_T_FMT:
2746 return_format = TRUE;
2749 case PERL_ERA_D_T_FMT:
2751 return_format = TRUE;
2754 case PERL_ALT_DIGITS:
2756 format = "%Ow"; /* Find the alternate digit for 0 */
2760 /* We can't use my_strftime() because it doesn't look at
2762 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2765 /* A zero return means one of:
2766 * a) there wasn't enough space in PL_langinfo_buf
2767 * b) the format, like a plain %p, returns empty
2768 * c) it was an illegal format, though some
2769 * implementations of strftime will just return the
2770 * illegal format as a plain character sequence.
2772 * To quickly test for case 'b)', try again but precede
2773 * the format with a plain character. If that result is
2774 * still empty, the problem is either 'a)' or 'c)' */
2776 Size_t format_size = strlen(format) + 1;
2777 Size_t mod_size = format_size + 1;
2781 Newx(mod_format, mod_size, char);
2782 Newx(temp_result, PL_langinfo_bufsize, char);
2784 my_strlcpy(mod_format + 1, format, mod_size);
2785 len = strftime(temp_result,
2786 PL_langinfo_bufsize,
2788 Safefree(mod_format);
2789 Safefree(temp_result);
2791 /* If 'len' is non-zero, it means that we had a case like
2792 * %p which means the current locale doesn't use a.m. or
2793 * p.m., and that is valid */
2796 /* Here, still didn't work. If we get well beyond a
2797 * reasonable size, bail out to prevent an infinite
2800 if (PL_langinfo_bufsize > 100 * format_size) {
2801 *PL_langinfo_buf = '\0';
2804 /* Double the buffer size to retry; Add 1 in case
2805 * original was 0, so we aren't stuck at 0. */
2806 PL_langinfo_bufsize *= 2;
2807 PL_langinfo_bufsize++;
2808 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2816 /* Here, we got a result.
2818 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2819 * alternate format for wday 0. If the value is the same as
2820 * the normal 0, there isn't an alternate, so clear the buffer.
2822 if ( item == PERL_ALT_DIGITS
2823 && strEQ(PL_langinfo_buf, "0"))
2825 *PL_langinfo_buf = '\0';
2828 /* ALT_DIGITS is problematic. Experiments on it showed that
2829 * strftime() did not always work properly when going from
2830 * alt-9 to alt-10. Only a few locales have this item defined,
2831 * and in all of them on Linux that khw was able to find,
2832 * nl_langinfo() merely returned the alt-0 character, possibly
2833 * doubled. Most Unicode digits are in blocks of 10
2834 * consecutive code points, so that is sufficient information
2835 * for those scripts, as we can infer alt-1, alt-2, .... But
2836 * for a Japanese locale, a CJK ideographic 0 is returned, and
2837 * the CJK digits are not in code point order, so you can't
2838 * really infer anything. The localedef for this locale did
2839 * specify the succeeding digits, so that strftime() works
2840 * properly on them, without needing to infer anything. But
2841 * the nl_langinfo() return did not give sufficient information
2842 * for the caller to understand what's going on. So until
2843 * there is evidence that it should work differently, this
2844 * returns the alt-0 string for ALT_DIGITS.
2846 * wday was chosen because its range is all a single digit.
2847 * Things like tm_sec have two digits as the minimum: '00' */
2851 retval = PL_langinfo_buf;
2853 /* If to return the format, not the value, overwrite the buffer
2854 * with it. But some strftime()s will keep the original format
2855 * if illegal, so change those to "" */
2856 if (return_format) {
2857 if (strEQ(PL_langinfo_buf, format)) {
2858 *PL_langinfo_buf = '\0';
2861 retval = save_to_buffer(format, &PL_langinfo_buf,
2862 &PL_langinfo_bufsize, 0);
2880 * Initialize locale awareness.
2883 Perl_init_i18nl10n(pTHX_ int printwarn)
2887 * 0 if not to output warning when setup locale is bad
2888 * 1 if to output warning based on value of PERL_BADLANG
2889 * >1 if to output regardless of PERL_BADLANG
2892 * 1 = set ok or not applicable,
2893 * 0 = fallback to a locale of lower priority
2894 * -1 = fallback to all locales failed, not even to the C locale
2896 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2897 * set, debugging information is output.
2899 * This looks more complicated than it is, mainly due to the #ifdefs.
2901 * We try to set LC_ALL to the value determined by the environment. If
2902 * there is no LC_ALL on this platform, we try the individual categories we
2903 * know about. If this works, we are done.
2905 * But if it doesn't work, we have to do something else. We search the
2906 * environment variables ourselves instead of relying on the system to do
2907 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2908 * think there is one), and the ultimate fallback "C". This is all done in
2909 * the same loop as above to avoid duplicating code, but it makes things
2910 * more complex. The 'trial_locales' array is initialized with just one
2911 * element; it causes the behavior described in the paragraph above this to
2912 * happen. If that fails, we add elements to 'trial_locales', and do extra
2913 * loop iterations to cause the behavior described in this paragraph.
2915 * On Ultrix, the locale MUST come from the environment, so there is
2916 * preliminary code to set it. I (khw) am not sure that it is necessary,
2917 * and that this couldn't be folded into the loop, but barring any real
2918 * platforms to test on, it's staying as-is
2920 * A slight complication is that in embedded Perls, the locale may already
2921 * be set-up, and we don't want to get it from the normal environment
2922 * variables. This is handled by having a special environment variable
2923 * indicate we're in this situation. We simply set setlocale's 2nd
2924 * parameter to be a NULL instead of "". That indicates to setlocale that
2925 * it is not to change anything, but to return the current value,
2926 * effectively initializing perl's db to what the locale already is.
2928 * We play the same trick with NULL if a LC_ALL succeeds. We call
2929 * setlocale() on the individual categores with NULL to get their existing
2930 * values for our db, instead of trying to change them.
2937 PERL_UNUSED_ARG(printwarn);
2939 #else /* USE_LOCALE */
2942 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
2946 /* NULL uses the existing already set up locale */
2947 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2950 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2951 unsigned int trial_locales_count;
2952 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2953 const char * const lang = savepv(PerlEnv_getenv("LANG"));
2954 bool setlocale_failure = FALSE;
2957 /* A later getenv() could zap this, so only use here */
2958 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2960 const bool locwarn = (printwarn > 1
2962 && ( ! bad_lang_use_once
2964 /* disallow with "" or "0" */
2966 && strNE("0", bad_lang_use_once)))));
2968 /* setlocale() return vals; not copied so must be looked at immediately */
2969 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
2971 /* current locale for given category; should have been copied so aren't
2973 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
2977 /* In some systems you can find out the system default locale
2978 * and use that as the fallback locale. */
2979 # define SYSTEM_DEFAULT_LOCALE
2981 # ifdef SYSTEM_DEFAULT_LOCALE
2983 const char *system_default_locale = NULL;
2988 # define DEBUG_LOCALE_INIT(a,b,c)
2991 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
2993 # define DEBUG_LOCALE_INIT(category, locale, result) \
2995 if (debug_initialization) { \
2996 PerlIO_printf(Perl_debug_log, \
2998 __FILE__, __LINE__, \
2999 setlocale_debug_string(category, \
3005 /* Make sure the parallel arrays are properly set up */
3006 # ifdef USE_LOCALE_NUMERIC
3007 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3008 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3009 # ifdef USE_POSIX_2008_LOCALE
3010 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3013 # ifdef USE_LOCALE_CTYPE
3014 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3015 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3016 # ifdef USE_POSIX_2008_LOCALE
3017 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3020 # ifdef USE_LOCALE_COLLATE
3021 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3022 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3023 # ifdef USE_POSIX_2008_LOCALE
3024 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3027 # ifdef USE_LOCALE_TIME
3028 assert(categories[LC_TIME_INDEX] == LC_TIME);
3029 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3030 # ifdef USE_POSIX_2008_LOCALE
3031 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3034 # ifdef USE_LOCALE_MESSAGES
3035 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3036 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3037 # ifdef USE_POSIX_2008_LOCALE
3038 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3041 # ifdef USE_LOCALE_MONETARY
3042 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3043 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3044 # ifdef USE_POSIX_2008_LOCALE
3045 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3048 # ifdef USE_LOCALE_ADDRESS
3049 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3050 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3051 # ifdef USE_POSIX_2008_LOCALE
3052 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3055 # ifdef USE_LOCALE_IDENTIFICATION
3056 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3057 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3058 # ifdef USE_POSIX_2008_LOCALE
3059 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3062 # ifdef USE_LOCALE_MEASUREMENT
3063 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3064 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3065 # ifdef USE_POSIX_2008_LOCALE
3066 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3069 # ifdef USE_LOCALE_PAPER
3070 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3071 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3072 # ifdef USE_POSIX_2008_LOCALE
3073 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3076 # ifdef USE_LOCALE_TELEPHONE
3077 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3078 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3079 # ifdef USE_POSIX_2008_LOCALE
3080 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3084 assert(categories[LC_ALL_INDEX] == LC_ALL);
3085 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3086 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3087 # ifdef USE_POSIX_2008_LOCALE
3088 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3091 # endif /* DEBUGGING */
3093 /* Initialize the cache of the program's UTF-8ness for the always known
3094 * locales C and POSIX */
3095 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3096 sizeof(PL_locale_utf8ness));
3098 # ifdef USE_THREAD_SAFE_LOCALE
3101 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3105 # if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3107 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3108 if (! PL_C_locale_obj) {
3109 Perl_croak_nocontext(
3110 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3112 if (DEBUG_Lv_TEST || debug_initialization) {
3113 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3118 PL_numeric_radix_sv = newSVpvs(".");
3120 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3122 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3123 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3126 # ifdef LOCALE_ENVIRON_REQUIRED
3129 * Ultrix setlocale(..., "") fails if there are no environment
3130 * variables from which to get a locale name.
3134 # error Ultrix without LC_ALL not implemented
3140 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3141 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3142 if (sl_result[LC_ALL_INDEX])
3145 setlocale_failure = TRUE;
3147 if (! setlocale_failure) {
3148 const char * locale_param;
3149 for (i = 0; i < LC_ALL_INDEX; i++) {
3150 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3153 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3154 if (! sl_result[i]) {
3155 setlocale_failure = TRUE;
3157 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3162 # endif /* LC_ALL */
3163 # endif /* LOCALE_ENVIRON_REQUIRED */
3165 /* We try each locale in the list until we get one that works, or exhaust
3166 * the list. Normally the loop is executed just once. But if setting the
3167 * locale fails, inside the loop we add fallback trials to the array and so
3168 * will execute the loop multiple times */
3169 trial_locales[0] = setlocale_init;
3170 trial_locales_count = 1;
3172 for (i= 0; i < trial_locales_count; i++) {
3173 const char * trial_locale = trial_locales[i];
3177 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3178 * when i==0, but I (khw) don't think that behavior makes much
3180 setlocale_failure = FALSE;
3182 # ifdef SYSTEM_DEFAULT_LOCALE
3183 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3185 /* On Windows machines, an entry of "" after the 0th means to use
3186 * the system default locale, which we now proceed to get. */
3187 if (strEQ(trial_locale, "")) {
3190 /* Note that this may change the locale, but we are going to do
3191 * that anyway just below */
3192 system_default_locale = do_setlocale_c(LC_ALL, "");
3193 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3195 /* Skip if invalid or if it's already on the list of locales to
3197 if (! system_default_locale) {
3198 goto next_iteration;
3200 for (j = 0; j < trial_locales_count; j++) {
3201 if (strEQ(system_default_locale, trial_locales[j])) {
3202 goto next_iteration;
3206 trial_locale = system_default_locale;
3209 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3211 # endif /* SYSTEM_DEFAULT_LOCALE */
3217 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3218 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3219 if (! sl_result[LC_ALL_INDEX]) {
3220 setlocale_failure = TRUE;
3223 /* Since LC_ALL succeeded, it should have changed all the other
3224 * categories it can to its value; so we massage things so that the
3225 * setlocales below just return their category's current values.
3226 * This adequately handles the case in NetBSD where LC_COLLATE may
3227 * not be defined for a locale, and setting it individually will
3228 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3229 * the POSIX locale. */
3230 trial_locale = NULL;
3233 # endif /* LC_ALL */
3235 if (! setlocale_failure) {
3237 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3239 = savepv(do_setlocale_r(categories[j], trial_locale));
3240 if (! curlocales[j]) {
3241 setlocale_failure = TRUE;
3243 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3246 if (! setlocale_failure) { /* All succeeded */
3247 break; /* Exit trial_locales loop */
3251 /* Here, something failed; will need to try a fallback. */
3257 if (locwarn) { /* Output failure info only on the first one */
3261 PerlIO_printf(Perl_error_log,
3262 "perl: warning: Setting locale failed.\n");
3264 # else /* !LC_ALL */
3266 PerlIO_printf(Perl_error_log,
3267 "perl: warning: Setting locale failed for the categories:\n\t");
3269 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3270 if (! curlocales[j]) {
3271 PerlIO_printf(Perl_error_log, category_names[j]);
3274 Safefree(curlocales[j]);
3278 # endif /* LC_ALL */
3280 PerlIO_printf(Perl_error_log,
3281 "perl: warning: Please check that your locale settings:\n");
3285 PerlIO_printf(Perl_error_log,
3286 "\tLANGUAGE = %c%s%c,\n",
3287 language ? '"' : '(',
3288 language ? language : "unset",
3289 language ? '"' : ')');
3292 PerlIO_printf(Perl_error_log,
3293 "\tLC_ALL = %c%s%c,\n",
3295 lc_all ? lc_all : "unset",
3296 lc_all ? '"' : ')');
3298 # if defined(USE_ENVIRON_ARRAY)
3303 /* Look through the environment for any variables of the
3304 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3305 * already handled above. These are assumed to be locale
3306 * settings. Output them and their values. */
3307 for (e = environ; *e; e++) {
3308 const STRLEN prefix_len = sizeof("LC_") - 1;
3311 if ( strBEGINs(*e, "LC_")
3312 && ! strBEGINs(*e, "LC_ALL=")
3313 && (uppers_len = strspn(*e + prefix_len,
3314 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3315 && ((*e)[prefix_len + uppers_len] == '='))
3317 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3318 (int) (prefix_len + uppers_len), *e,
3319 *e + prefix_len + uppers_len + 1);
3326 PerlIO_printf(Perl_error_log,
3327 "\t(possibly more locale environment variables)\n");
3331 PerlIO_printf(Perl_error_log,
3332 "\tLANG = %c%s%c\n",
3334 lang ? lang : "unset",
3337 PerlIO_printf(Perl_error_log,
3338 " are supported and installed on your system.\n");
3341 /* Calculate what fallback locales to try. We have avoided this
3342 * until we have to, because failure is quite unlikely. This will
3343 * usually change the upper bound of the loop we are in.
3345 * Since the system's default way of setting the locale has not
3346 * found one that works, We use Perl's defined ordering: LC_ALL,
3347 * LANG, and the C locale. We don't try the same locale twice, so
3348 * don't add to the list if already there. (On POSIX systems, the
3349 * LC_ALL element will likely be a repeat of the 0th element "",
3350 * but there's no harm done by doing it explicitly.
3352 * Note that this tries the LC_ALL environment variable even on
3353 * systems which have no LC_ALL locale setting. This may or may
3354 * not have been originally intentional, but there's no real need
3355 * to change the behavior. */
3357 for (j = 0; j < trial_locales_count; j++) {
3358 if (strEQ(lc_all, trial_locales[j])) {
3362 trial_locales[trial_locales_count++] = lc_all;
3367 for (j = 0; j < trial_locales_count; j++) {
3368 if (strEQ(lang, trial_locales[j])) {
3372 trial_locales[trial_locales_count++] = lang;
3376 # if defined(WIN32) && defined(LC_ALL)
3378 /* For Windows, we also try the system default locale before "C".
3379 * (If there exists a Windows without LC_ALL we skip this because
3380 * it gets too complicated. For those, the "C" is the next
3381 * fallback possibility). The "" is the same as the 0th element of
3382 * the array, but the code at the loop above knows to treat it
3383 * differently when not the 0th */
3384 trial_locales[trial_locales_count++] = "";
3388 for (j = 0; j < trial_locales_count; j++) {
3389 if (strEQ("C", trial_locales[j])) {
3393 trial_locales[trial_locales_count++] = "C";
3396 } /* end of first time through the loop */
3404 } /* end of looping through the trial locales */
3406 if (ok < 1) { /* If we tried to fallback */
3408 if (! setlocale_failure) { /* fallback succeeded */
3409 msg = "Falling back to";
3411 else { /* fallback failed */
3414 /* We dropped off the end of the loop, so have to decrement i to
3415 * get back to the value the last time through */
3419 msg = "Failed to fall back to";
3421 /* To continue, we should use whatever values we've got */
3423 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3424 Safefree(curlocales[j]);
3425 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3426 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3431 const char * description;
3432 const char * name = "";
3433 if (strEQ(trial_locales[i], "C")) {
3434 description = "the standard locale";
3438 # ifdef SYSTEM_DEFAULT_LOCALE
3440 else if (strEQ(trial_locales[i], "")) {
3441 description = "the system default locale";
3442 if (system_default_locale) {
3443 name = system_default_locale;
3447 # endif /* SYSTEM_DEFAULT_LOCALE */
3450 description = "a fallback locale";
3451 name = trial_locales[i];
3453 if (name && strNE(name, "")) {
3454 PerlIO_printf(Perl_error_log,
3455 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3458 PerlIO_printf(Perl_error_log,
3459 "perl: warning: %s %s.\n", msg, description);
3462 } /* End of tried to fallback */
3464 /* Done with finding the locales; update our records */
3466 # ifdef USE_LOCALE_CTYPE
3468 new_ctype(curlocales[LC_CTYPE_INDEX]);
3471 # ifdef USE_LOCALE_COLLATE
3473 new_collate(curlocales[LC_COLLATE_INDEX]);
3476 # ifdef USE_LOCALE_NUMERIC
3478 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3482 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3484 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3486 /* This caches whether each category's locale is UTF-8 or not. This
3487 * may involve changing the locale. It is ok to do this at
3488 * initialization time before any threads have started, but not later
3489 * unless thread-safe operations are used.
3490 * Caching means that if the program heeds our dictate not to change
3491 * locales in threaded applications, this data will remain valid, and
3492 * it may get queried without having to change locales. If the
3493 * environment is such that all categories have the same locale, this
3494 * isn't needed, as the code will not change the locale; but this
3495 * handles the uncommon case where the environment has disparate
3496 * locales for the categories */
3497 (void) _is_cur_LC_category_utf8(categories[i]);
3501 Safefree(curlocales[i]);
3504 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3506 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3507 * locale is UTF-8. The call to new_ctype() just above has already
3508 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3509 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3510 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3511 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3512 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3514 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3515 This is an alternative to using the -C command line switch
3516 (the -C if present will override this). */
3518 const char *p = PerlEnv_getenv("PERL_UNICODE");
3519 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3520 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3534 #endif /* USE_LOCALE */
3537 /* So won't continue to output stuff */
3538 DEBUG_INITIALIZATION_set(FALSE);
3545 #ifdef USE_LOCALE_COLLATE
3548 Perl__mem_collxfrm(pTHX_ const char *input_string,
3549 STRLEN len, /* Length of 'input_string' */
3550 STRLEN *xlen, /* Set to length of returned string
3551 (not including the collation index
3553 bool utf8 /* Is the input in UTF-8? */
3557 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3558 * differences. First, it handles embedded NULs. Second, it allocates a bit
3559 * more memory than needed for the transformed data itself. The real
3560 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3561 * the length of that, and doesn't include the collation index size.
3562 * Please see sv_collxfrm() to see how this is used. */
3564 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3566 char * s = (char *) input_string;
3567 STRLEN s_strlen = strlen(input_string);
3569 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3570 STRLEN length_in_chars;
3571 bool first_time = TRUE; /* Cleared after first loop iteration */
3573 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3575 /* Must be NUL-terminated */
3576 assert(*(input_string + len) == '\0');
3578 /* If this locale has defective collation, skip */
3579 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3580 DEBUG_L(PerlIO_printf(Perl_debug_log,
3581 "_mem_collxfrm: locale's collation is defective\n"));
3585 /* Replace any embedded NULs with the control that sorts before any others.
3586 * This will give as good as possible results on strings that don't
3587 * otherwise contain that character, but otherwise there may be
3588 * less-than-perfect results with that character and NUL. This is
3589 * unavoidable unless we replace strxfrm with our own implementation. */
3590 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3594 STRLEN sans_nuls_len;
3595 int try_non_controls;
3596 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3597 making sure 2nd byte is NUL.
3599 STRLEN this_replacement_len;
3601 /* If we don't know what non-NUL control character sorts lowest for
3602 * this locale, find it */
3603 if (PL_strxfrm_NUL_replacement == '\0') {
3605 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3606 includes the collation index
3609 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3611 /* Unlikely, but it may be that no control will work to replace
3612 * NUL, in which case we instead look for any character. Controls
3613 * are preferred because collation order is, in general, context
3614 * sensitive, with adjoining characters affecting the order, and
3615 * controls are less likely to have such interactions, allowing the
3616 * NUL-replacement to stand on its own. (Another way to look at it
3617 * is to imagine what would happen if the NUL were replaced by a
3618 * combining character; it wouldn't work out all that well.) */
3619 for (try_non_controls = 0;
3620 try_non_controls < 2;
3623 /* Look through all legal code points (NUL isn't) */
3624 for (j = 1; j < 256; j++) {
3625 char * x; /* j's xfrm plus collation index */
3626 STRLEN x_len; /* length of 'x' */
3627 STRLEN trial_len = 1;
3628 char cur_source[] = { '\0', '\0' };
3630 /* Skip non-controls the first time through the loop. The
3631 * controls in a UTF-8 locale are the L1 ones */
3632 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3639 /* Create a 1-char string of the current code point */
3640 cur_source[0] = (char) j;
3642 /* Then transform it */
3643 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3644 0 /* The string is not in UTF-8 */);
3646 /* Ignore any character that didn't successfully transform.
3652 /* If this character's transformation is lower than
3653 * the current lowest, this one becomes the lowest */
3654 if ( cur_min_x == NULL
3655 || strLT(x + COLLXFRM_HDR_LEN,
3656 cur_min_x + COLLXFRM_HDR_LEN))
3658 PL_strxfrm_NUL_replacement = j;
3664 } /* end of loop through all 255 characters */
3666 /* Stop looking if found */
3671 /* Unlikely, but possible, if there aren't any controls that
3672 * work in the locale, repeat the loop, looking for any
3673 * character that works */
3674 DEBUG_L(PerlIO_printf(Perl_debug_log,
3675 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3676 } /* End of loop to try first the controls, then any char */
3679 DEBUG_L(PerlIO_printf(Perl_debug_log,
3680 "_mem_collxfrm: Couldn't find any character to replace"
3681 " embedded NULs in locale %s with", PL_collation_name));
3685 DEBUG_L(PerlIO_printf(Perl_debug_log,
3686 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3687 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3689 Safefree(cur_min_x);
3690 } /* End of determining the character that is to replace NULs */
3692 /* If the replacement is variant under UTF-8, it must match the
3693 * UTF8-ness of the original */
3694 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3695 this_replacement_char[0] =
3696 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3697 this_replacement_char[1] =
3698 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3699 this_replacement_len = 2;
3702 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3703 /* this_replacement_char[1] = '\0' was done at initialization */
3704 this_replacement_len = 1;
3707 /* The worst case length for the replaced string would be if every
3708 * character in it is NUL. Multiply that by the length of each
3709 * replacement, and allow for a trailing NUL */
3710 sans_nuls_len = (len * this_replacement_len) + 1;
3711 Newx(sans_nuls, sans_nuls_len, char);
3714 /* Replace each NUL with the lowest collating control. Loop until have
3715 * exhausted all the NULs */
3716 while (s + s_strlen < e) {
3717 my_strlcat(sans_nuls, s, sans_nuls_len);
3719 /* Do the actual replacement */
3720 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3722 /* Move past the input NUL */
3724 s_strlen = strlen(s);
3727 /* And add anything that trails the final NUL */
3728 my_strlcat(sans_nuls, s, sans_nuls_len);
3730 /* Switch so below we transform this modified string */
3733 } /* End of replacing NULs */
3735 /* Make sure the UTF8ness of the string and locale match */
3736 if (utf8 != PL_in_utf8_COLLATE_locale) {
3737 /* XXX convert above Unicode to 10FFFF? */
3738 const char * const t = s; /* Temporary so we can later find where the
3741 /* Here they don't match. Change the string's to be what the locale is
3744 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3745 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3748 else { /* locale is not UTF-8; but input is; downgrade the input */
3750 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3752 /* If the downgrade was successful we are done, but if the input
3753 * contains things that require UTF-8 to represent, have to do
3754 * damage control ... */
3755 if (UNLIKELY(utf8)) {
3757 /* What we do is construct a non-UTF-8 string with
3758 * 1) the characters representable by a single byte converted
3759 * to be so (if necessary);
3760 * 2) and the rest converted to collate the same as the
3761 * highest collating representable character. That makes
3762 * them collate at the end. This is similar to how we
3763 * handle embedded NULs, but we use the highest collating
3764 * code point instead of the smallest. Like the NUL case,
3765 * this isn't perfect, but is the best we can reasonably
3766 * do. Every above-255 code point will sort the same as
3767 * the highest-sorting 0-255 code point. If that code
3768 * point can combine in a sequence with some other code
3769 * points for weight calculations, us changing something to
3770 * be it can adversely affect the results. But in most
3771 * cases, it should work reasonably. And note that this is
3772 * really an illegal situation: using code points above 255
3773 * on a locale where only 0-255 are valid. If two strings
3774 * sort entirely equal, then the sort order for the
3775 * above-255 code points will be in code point order. */
3779 /* If we haven't calculated the code point with the maximum
3780 * collating order for this locale, do so now */
3781 if (! PL_strxfrm_max_cp) {
3784 /* The current transformed string that collates the
3785 * highest (except it also includes the prefixed collation
3787 char * cur_max_x = NULL;
3789 /* Look through all legal code points (NUL isn't) */
3790 for (j = 1; j < 256; j++) {
3793 char cur_source[] = { '\0', '\0' };
3795 /* Create a 1-char string of the current code point */
3796 cur_source[0] = (char) j;
3798 /* Then transform it */
3799 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3801 /* If something went wrong (which it shouldn't), just
3802 * ignore this code point */
3807 /* If this character's transformation is higher than
3808 * the current highest, this one becomes the highest */
3809 if ( cur_max_x == NULL
3810 || strGT(x + COLLXFRM_HDR_LEN,
3811 cur_max_x + COLLXFRM_HDR_LEN))
3813 PL_strxfrm_max_cp = j;
3822 DEBUG_L(PerlIO_printf(Perl_debug_log,
3823 "_mem_collxfrm: Couldn't find any character to"
3824 " replace above-Latin1 chars in locale %s with",
3825 PL_collation_name));
3829 DEBUG_L(PerlIO_printf(Perl_debug_log,
3830 "_mem_collxfrm: highest 1-byte collating character"
3831 " in locale %s is 0x%02X\n",
3833 PL_strxfrm_max_cp));
3835 Safefree(cur_max_x);
3838 /* Here we know which legal code point collates the highest.
3839 * We are ready to construct the non-UTF-8 string. The length
3840 * will be at least 1 byte smaller than the input string
3841 * (because we changed at least one 2-byte character into a
3842 * single byte), but that is eaten up by the trailing NUL */
3848 char * e = (char *) t + len;
3850 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3852 if (UTF8_IS_INVARIANT(cur_char)) {
3855 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
3856 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3858 else { /* Replace illegal cp with highest collating
3860 s[d++] = PL_strxfrm_max_cp;
3864 Renew(s, d, char); /* Free up unused space */
3869 /* Here, we have constructed a modified version of the input. It could
3870 * be that we already had a modified copy before we did this version.
3871 * If so, that copy is no longer needed */
3872 if (t != input_string) {
3877 length_in_chars = (utf8)
3878 ? utf8_length((U8 *) s, (U8 *) s + len)
3881 /* The first element in the output is the collation id, used by
3882 * sv_collxfrm(); then comes the space for the transformed string. The
3883 * equation should give us a good estimate as to how much is needed */
3884 xAlloc = COLLXFRM_HDR_LEN
3886 + (PL_collxfrm_mult * length_in_chars);
3887 Newx(xbuf, xAlloc, char);
3888 if (UNLIKELY(! xbuf)) {
3889 DEBUG_L(PerlIO_printf(Perl_debug_log,
3890 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
3894 /* Store the collation id */
3895 *(U32*)xbuf = PL_collation_ix;
3897 /* Then the transformation of the input. We loop until successful, or we
3901 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
3903 /* If the transformed string occupies less space than we told strxfrm()
3904 * was available, it means it successfully transformed the whole
3906 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
3908 /* Some systems include a trailing NUL in the returned length.
3909 * Ignore it, using a loop in case multiple trailing NULs are
3912 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3917 /* If the first try didn't get it, it means our prediction was low.
3918 * Modify the coefficients so that we predict a larger value in any
3919 * future transformations */
3921 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3922 STRLEN computed_guess = PL_collxfrm_base
3923 + (PL_collxfrm_mult * length_in_chars);
3925 /* On zero-length input, just keep current slope instead of
3927 const STRLEN new_m = (length_in_chars != 0)
3928 ? needed / length_in_chars
3931 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3932 "%s: %d: initial size of %zu bytes for a length "
3933 "%zu string was insufficient, %zu needed\n",
3935 computed_guess, length_in_chars, needed));
3937 /* If slope increased, use it, but discard this result for
3938 * length 1 strings, as we can't be sure that it's a real slope
3940 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
3944 STRLEN old_m = PL_collxfrm_mult;
3945 STRLEN old_b = PL_collxfrm_base;
3949 PL_collxfrm_mult = new_m;
3950 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3951 computed_guess = PL_collxfrm_base
3952 + (PL_collxfrm_mult * length_in_chars);
3953 if (computed_guess < needed) {
3954 PL_collxfrm_base += needed - computed_guess;
3957 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3958 "%s: %d: slope is now %zu; was %zu, base "
3959 "is now %zu; was %zu\n",
3961 PL_collxfrm_mult, old_m,
3962 PL_collxfrm_base, old_b));
3964 else { /* Slope didn't change, but 'b' did */
3965 const STRLEN new_b = needed
3968 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3969 "%s: %d: base is now %zu; was %zu\n",
3971 new_b, PL_collxfrm_base));
3972 PL_collxfrm_base = new_b;
3979 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3980 DEBUG_L(PerlIO_printf(Perl_debug_log,
3981 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3982 *xlen, PERL_INT_MAX));
3986 /* A well-behaved strxfrm() returns exactly how much space it needs
3987 * (usually not including the trailing NUL) when it fails due to not
3988 * enough space being provided. Assume that this is the case unless
3989 * it's been proven otherwise */
3990 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
3991 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
3993 else { /* Here, either:
3994 * 1) The strxfrm() has previously shown bad behavior; or
3995 * 2) It isn't the first time through the loop, which means
3996 * that the strxfrm() is now showing bad behavior, because
3997 * we gave it what it said was needed in the previous
3998 * iteration, and it came back saying it needed still more.
3999 * (Many versions of cygwin fit this. When the buffer size
4000 * isn't sufficient, they return the input size instead of
4001 * how much is needed.)
4002 * Increase the buffer size by a fixed percentage and try again.
4004 xAlloc += (xAlloc / 4) + 1;
4005 PL_strxfrm_is_behaved = FALSE;
4009 if (DEBUG_Lv_TEST || debug_initialization) {
4010 PerlIO_printf(Perl_debug_log,
4011 "_mem_collxfrm required more space than previously calculated"
4012 " for locale %s, trying again with new guess=%d+%zu\n",
4013 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4014 xAlloc - COLLXFRM_HDR_LEN);
4021 Renew(xbuf, xAlloc, char);
4022 if (UNLIKELY(! xbuf)) {
4023 DEBUG_L(PerlIO_printf(Perl_debug_log,
4024 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4034 if (DEBUG_Lv_TEST || debug_initialization) {
4036 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4037 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4038 PerlIO_printf(Perl_debug_log, "%s\n",
4039 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4045 /* Free up unneeded space; retain ehough for trailing NUL */
4046 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4048 if (s != input_string) {
4056 if (s != input_string) {
4063 if (DEBUG_Lv_TEST || debug_initialization) {
4064 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4075 S_print_collxfrm_input_and_return(pTHX_
4076 const char * const s,
4077 const char * const e,
4078 const STRLEN * const xlen,
4082 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4084 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4085 (UV)PL_collation_ix);
4087 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4090 PerlIO_printf(Perl_debug_log, "NULL");
4092 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4094 print_bytes_for_locale(s, e, is_utf8);
4096 PerlIO_printf(Perl_debug_log, "'\n");
4100 S_print_bytes_for_locale(pTHX_
4101 const char * const s,
4102 const char * const e,
4106 bool prev_was_printable = TRUE;
4107 bool first_time = TRUE;
4109 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4113 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4116 if (! prev_was_printable) {
4117 PerlIO_printf(Perl_debug_log, " ");
4119 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4120 prev_was_printable = TRUE;
4124 PerlIO_printf(Perl_debug_log, " ");
4126 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4127 prev_was_printable = FALSE;
4129 t += (is_utf8) ? UTF8SKIP(t) : 1;
4134 # endif /* #ifdef DEBUGGING */
4135 #endif /* USE_LOCALE_COLLATE */
4140 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4142 /* Changes the locale for LC_'switch_category" to that of
4143 * LC_'template_category', if they aren't already the same. If not NULL,
4144 * 'template_locale' is the locale that 'template_category' is in.
4146 * Returns a copy of the name of the original locale for 'switch_category'
4147 * so can be switched back to with the companion function
4148 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4150 char * restore_to_locale = NULL;
4152 if (switch_category == template_category) { /* No changes needed */
4156 /* Find the original locale of the category we may need to change, so that
4157 * it can be restored to later */
4158 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4160 if (! restore_to_locale) {
4162 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4163 __FILE__, __LINE__, category_name(switch_category), errno);
4166 /* If the locale of the template category wasn't passed in, find it now */
4167 if (template_locale == NULL) {
4168 template_locale = do_setlocale_r(template_category, NULL);
4169 if (! template_locale) {
4171 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4172 __FILE__, __LINE__, category_name(template_category), errno);
4176 /* It the locales are the same, there's nothing to do */
4177 if (strEQ(restore_to_locale, template_locale)) {
4178 Safefree(restore_to_locale);
4180 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4181 category_name(switch_category), restore_to_locale));
4186 /* Finally, change the locale to the template one */
4187 if (! do_setlocale_r(switch_category, template_locale)) {
4189 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4190 __FILE__, __LINE__, category_name(switch_category),
4191 template_locale, errno);
4194 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4195 category_name(switch_category), template_locale));
4197 return restore_to_locale;
4201 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4203 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4204 * copy that will be freed by this function), or do nothing if the latter
4205 * parameter is NULL */
4207 if (original_locale == NULL) {
4211 if (! do_setlocale_r(category, original_locale)) {
4213 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4215 category_name(category), original_locale, errno);
4218 Safefree(original_locale);
4222 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4224 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4225 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4226 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4227 * could give the wrong result. The result will very likely be correct for
4228 * languages that have commonly used non-ASCII characters, but for notably
4229 * English, it comes down to if the locale's name ends in something like
4230 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4232 * If the platform is early C89, not containing mbtowc(), or we are
4233 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4234 * These work very well for non-Latin locales or those whose currency
4235 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4236 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4237 * on the name containing UTF-8 or not. */
4239 /* Name of current locale corresponding to the input category */
4240 const char *save_input_locale = NULL;
4242 bool is_utf8 = FALSE; /* The return value */
4244 /* The variables below are for the cache of previous lookups using this
4245 * function. The cache is a C string, described at the definition for
4246 * 'C_and_POSIX_utf8ness'.
4248 * The first part of the cache is fixed, for the C and POSIX locales. The
4249 * varying part starts just after them. */
4250 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4252 Size_t utf8ness_cache_size; /* Size of the varying portion */
4253 Size_t input_name_len; /* Length in bytes of save_input_locale */
4254 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4255 the name in the cache */
4256 char * delimited; /* The name plus the delimiters used to store
4258 char * name_pos; /* position of 'delimited' in the cache, or 0
4264 assert(category != LC_ALL);
4268 /* Get the desired category's locale */
4269 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4270 if (! save_input_locale) {
4272 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4273 __FILE__, __LINE__, category_name(category), errno);
4276 DEBUG_L(PerlIO_printf(Perl_debug_log,
4277 "Current locale for %s is %s\n",
4278 category_name(category), save_input_locale));
4280 input_name_len = strlen(save_input_locale);
4282 /* In our cache, each name is accompanied by two delimiters and a single
4284 input_name_len_with_overhead = input_name_len + 3;
4286 /* Allocate and populate space for a copy of the name surrounded by the
4288 Newx(delimited, input_name_len_with_overhead, char);
4289 delimited[0] = UTF8NESS_SEP[0];
4290 Copy(save_input_locale, delimited + 1, input_name_len, char);
4291 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4292 delimited[input_name_len+2] = '\0';
4294 /* And see if that is in the cache */
4295 name_pos = instr(PL_locale_utf8ness, delimited);
4297 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4301 if (DEBUG_Lv_TEST || debug_initialization) {
4302 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4303 save_input_locale, is_utf8);
4308 /* And, if not already in that position, move it to the beginning of
4309 * the non-constant portion of the list, since it is the most recently
4310 * used. (We don't have to worry about overflow, since just moving
4311 * existing names around) */
4312 if (name_pos > utf8ness_cache) {
4313 Move(utf8ness_cache,
4314 utf8ness_cache + input_name_len_with_overhead,
4315 name_pos - utf8ness_cache, char);
4318 input_name_len_with_overhead - 1, char);
4319 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4322 Safefree(delimited);