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)) {
1005 if (DEBUG_L_TEST || debug_initialization) {
1006 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1018 if (DEBUG_Lv_TEST || debug_initialization) {
1019 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1024 /* And switch into it */
1025 if (! uselocale(new_obj)) {
1030 if (DEBUG_L_TEST || debug_initialization) {
1031 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1036 if (! uselocale(old_obj)) {
1040 if (DEBUG_L_TEST || debug_initialization) {
1041 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1047 freelocale(new_obj);
1054 if (DEBUG_Lv_TEST || debug_initialization) {
1055 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1060 /* We are done, except for updating our records (if the system doesn't keep
1061 * them) and in the case of locale "", we don't actually know what the
1062 * locale that got switched to is, as it came from the environment. So
1063 * have to find it */
1065 # ifdef HAS_QUERYLOCALE
1067 if (strEQ(locale, "")) {
1068 locale = querylocale(mask, new_obj);
1073 /* Here, 'locale' is the return value */
1075 /* Without querylocale(), we have to update our records */
1077 if (category == LC_ALL) {
1080 /* For LC_ALL, we change all individual categories to correspond */
1081 /* PL_curlocales is a parallel array, so has same
1082 * length as 'categories' */
1083 for (i = 0; i <= LC_ALL_INDEX; i++) {
1084 Safefree(PL_curlocales[i]);
1085 PL_curlocales[i] = savepv(locale);
1090 /* For a single category, if it's not the same as the one in LC_ALL, we
1093 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1094 Safefree(PL_curlocales[LC_ALL_INDEX]);
1095 PL_curlocales[LC_ALL_INDEX] = NULL;
1098 /* Then update the category's record */
1099 Safefree(PL_curlocales[index]);
1100 PL_curlocales[index] = savepv(locale);
1108 #endif /* USE_POSIX_2008_LOCALE */
1110 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1111 didn't natively support them */
1113 /* The way this would work is that we would keep a per-thread list of the
1114 * correct locale for that thread. Any operation that was locale-sensitive
1115 * would have to be changed so that it would look like this:
1118 * setlocale to the correct locale for this operation
1122 * This leaves the global locale in the most recently used operation's, but it
1123 * was locked long enough to get the result. If that result is static, it
1124 * needs to be copied before the unlock.
1126 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1127 * the setup, but are no-ops when not needed, and similarly,
1128 * END_LOCALE_DEPENDENT_OP for the tear-down
1130 * But every call to a locale-sensitive function would have to be changed, and
1131 * if a module didn't cooperate by using the mutex, things would break.
1133 * This code was abandoned before being completed or tested, and is left as-is
1136 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1137 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1140 S_locking_setlocale(pTHX_
1142 const char * locale,
1144 const bool is_index_valid
1147 /* This function kind of performs a setlocale() on just the current thread;
1148 * thus it is kind of thread-safe. It does this by keeping a thread-level
1149 * array of the current locales for each category. Every time a locale is
1150 * switched to, it does the switch globally, but updates the thread's
1151 * array. A query as to what the current locale is just returns the
1152 * appropriate element from the array, and doesn't actually call the system
1153 * setlocale(). The saving into the array is done in an uninterruptible
1154 * section of code, so is unaffected by whatever any other threads might be
1157 * All locale-sensitive operations must work by first starting a critical
1158 * section, then switching to the thread's locale as kept by this function,
1159 * and then doing the operation, then ending the critical section. Thus,
1160 * each gets done in the appropriate locale. simulating thread-safety.
1162 * This function takes the same parameters, 'category' and 'locale', that
1163 * the regular setlocale() function does, but it also takes two additional
1164 * ones. This is because as described earlier. If we know on input the
1165 * index corresponding to the category into the array where we store the
1166 * current locales, we don't have to calculate it. If the caller knows at
1167 * compile time what the index is, it it can pass it, setting
1168 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1172 /* If the input index might be incorrect, calculate the correct one */
1173 if (! is_index_valid) {
1176 if (DEBUG_Lv_TEST || debug_initialization) {
1177 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1180 for (i = 0; i <= LC_ALL_INDEX; i++) {
1181 if (category == categories[i]) {
1187 /* Here, we don't know about this category, so can't handle it.
1188 * XXX best we can do is to unsafely set this
1191 return my_setlocale(category, locale);
1195 if (DEBUG_Lv_TEST || debug_initialization) {
1196 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1200 /* For a query, just return what's in our records */
1201 if (new_locale == NULL) {
1202 return curlocales[index];
1206 /* Otherwise, we need to do the switch, and save the result, all in a
1207 * critical section */
1209 Safefree(curlocales[[index]]);
1211 /* It might be that this is called from an already-locked section of code.
1212 * We would have to detect and skip the LOCK/UNLOCK if so */
1215 curlocales[index] = savepv(my_setlocale(category, new_locale));
1217 if (strEQ(new_locale, "")) {
1221 /* The locale values come from the environment, and may not all be the
1222 * same, so for LC_ALL, we have to update all the others, while the
1223 * mutex is still locked */
1225 if (category == LC_ALL) {
1227 for (i = 0; i < LC_ALL_INDEX) {
1228 curlocales[i] = my_setlocale(categories[i], NULL);
1237 return curlocales[index];
1243 S_set_numeric_radix(pTHX_ const bool use_locale)
1245 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1246 * TRUE, use the radix character derived from the current locale */
1248 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1249 || defined(HAS_NL_LANGINFO))
1251 const char * radix = (use_locale)
1252 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1253 /* FALSE => already in dest locale */
1256 sv_setpv(PL_numeric_radix_sv, radix);
1258 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1259 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1260 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1261 SvCUR(PL_numeric_radix_sv))
1262 && _is_cur_LC_category_utf8(LC_NUMERIC))
1264 SvUTF8_on(PL_numeric_radix_sv);
1269 if (DEBUG_L_TEST || debug_initialization) {
1270 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1271 SvPVX(PL_numeric_radix_sv),
1272 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1276 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1281 S_new_numeric(pTHX_ const char *newnum)
1284 #ifndef USE_LOCALE_NUMERIC
1286 PERL_UNUSED_ARG(newnum);
1290 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1291 * core Perl this and that 'newnum' is the name of the new locale.
1292 * It installs this locale as the current underlying default.
1294 * The default locale and the C locale can be toggled between by use of the
1295 * set_numeric_underlying() and set_numeric_standard() functions, which
1296 * should probably not be called directly, but only via macros like
1297 * SET_NUMERIC_STANDARD() in perl.h.
1299 * The toggling is necessary mainly so that a non-dot radix decimal point
1300 * character can be output, while allowing internal calculations to use a
1303 * This sets several interpreter-level variables:
1304 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1305 * PL_numeric_underlying A boolean indicating if the toggled state is such
1306 * that the current locale is the program's underlying
1308 * PL_numeric_standard An int indicating if the toggled state is such
1309 * that the current locale is the C locale or
1310 * indistinguishable from the C locale. If non-zero, it
1311 * is in C; if > 1, it means it may not be toggled away
1313 * PL_numeric_underlying_is_standard A bool kept by this function
1314 * indicating that the underlying locale and the standard
1315 * C locale are indistinguishable for the purposes of
1316 * LC_NUMERIC. This happens when both of the above two
1317 * variables are true at the same time. (Toggling is a
1318 * no-op under these circumstances.) This variable is
1319 * used to avoid having to recalculate.
1325 Safefree(PL_numeric_name);
1326 PL_numeric_name = NULL;
1327 PL_numeric_standard = TRUE;
1328 PL_numeric_underlying = TRUE;
1329 PL_numeric_underlying_is_standard = TRUE;
1333 save_newnum = stdize_locale(savepv(newnum));
1334 PL_numeric_underlying = TRUE;
1335 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1337 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1339 if (! PL_numeric_standard) {
1340 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1341 FALSE /* Don't toggle locale */ ))
1342 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1346 /* Save the new name if it isn't the same as the previous one, if any */
1347 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1348 Safefree(PL_numeric_name);
1349 PL_numeric_name = save_newnum;
1352 Safefree(save_newnum);
1355 PL_numeric_underlying_is_standard = PL_numeric_standard;
1357 # ifdef HAS_POSIX_2008_LOCALE
1359 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1361 PL_underlying_numeric_obj);
1365 if (DEBUG_L_TEST || debug_initialization) {
1366 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1369 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1370 * have to worry about the radix being a non-dot. (Core operations that
1371 * need the underlying locale change to it temporarily). */
1372 if (PL_numeric_standard) {
1373 set_numeric_radix(0);
1376 set_numeric_standard();
1379 #endif /* USE_LOCALE_NUMERIC */
1384 Perl_set_numeric_standard(pTHX)
1387 #ifdef USE_LOCALE_NUMERIC
1389 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1390 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1391 * macro avoids calling this routine if toggling isn't necessary according
1392 * to our records (which could be wrong if some XS code has changed the
1393 * locale behind our back) */
1395 do_setlocale_c(LC_NUMERIC, "C");
1396 PL_numeric_standard = TRUE;
1397 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1398 set_numeric_radix(0);
1402 if (DEBUG_L_TEST || debug_initialization) {
1403 PerlIO_printf(Perl_debug_log,
1404 "LC_NUMERIC locale now is standard C\n");
1408 #endif /* USE_LOCALE_NUMERIC */
1413 Perl_set_numeric_underlying(pTHX)
1416 #ifdef USE_LOCALE_NUMERIC
1418 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1419 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1420 * instead of calling this directly. The macro avoids calling this routine
1421 * if toggling isn't necessary according to our records (which could be
1422 * wrong if some XS code has changed the locale behind our back) */
1424 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1425 PL_numeric_standard = PL_numeric_underlying_is_standard;
1426 PL_numeric_underlying = TRUE;
1427 set_numeric_radix(! PL_numeric_standard);
1431 if (DEBUG_L_TEST || debug_initialization) {
1432 PerlIO_printf(Perl_debug_log,
1433 "LC_NUMERIC locale now is %s\n",
1438 #endif /* USE_LOCALE_NUMERIC */
1443 * Set up for a new ctype locale.
1446 S_new_ctype(pTHX_ const char *newctype)
1449 #ifndef USE_LOCALE_CTYPE
1451 PERL_ARGS_ASSERT_NEW_CTYPE;
1452 PERL_UNUSED_ARG(newctype);
1453 PERL_UNUSED_CONTEXT;
1457 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1458 * core Perl this and that 'newctype' is the name of the new locale.
1460 * This function sets up the folding arrays for all 256 bytes, assuming
1461 * that tofold() is tolc() since fold case is not a concept in POSIX,
1463 * Any code changing the locale (outside this file) should use
1464 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1465 * this function should be called directly only from this file and from
1466 * POSIX::setlocale() */
1471 /* Don't check for problems if we are suppressing the warnings */
1472 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1474 PERL_ARGS_ASSERT_NEW_CTYPE;
1476 /* We will replace any bad locale warning with 1) nothing if the new one is
1477 * ok; or 2) a new warning for the bad new locale */
1478 if (PL_warn_locale) {
1479 SvREFCNT_dec_NN(PL_warn_locale);
1480 PL_warn_locale = NULL;
1483 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1485 /* A UTF-8 locale gets standard rules. But note that code still has to
1486 * handle this specially because of the three problematic code points */
1487 if (PL_in_utf8_CTYPE_locale) {
1488 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1491 /* We don't populate the other lists if a UTF-8 locale, but do check that
1492 * everything works as expected, unless checking turned off */
1493 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1494 /* Assume enough space for every character being bad. 4 spaces each
1495 * for the 94 printable characters that are output like "'x' "; and 5
1496 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1498 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1499 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1501 unsigned int bad_count = 0; /* Count of bad characters */
1503 for (i = 0; i < 256; i++) {
1504 if (! PL_in_utf8_CTYPE_locale) {
1506 PL_fold_locale[i] = (U8) tolower(i);
1507 else if (islower(i))
1508 PL_fold_locale[i] = (U8) toupper(i);
1510 PL_fold_locale[i] = (U8) i;
1513 /* If checking for locale problems, see if the native ASCII-range
1514 * printables plus \n and \t are in their expected categories in
1515 * the new locale. If not, this could mean big trouble, upending
1516 * Perl's and most programs' assumptions, like having a
1517 * metacharacter with special meaning become a \w. Fortunately,
1518 * it's very rare to find locales that aren't supersets of ASCII
1519 * nowadays. It isn't a problem for most controls to be changed
1520 * into something else; we check only \n and \t, though perhaps \r
1521 * could be an issue as well. */
1522 if ( check_for_problems
1523 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1525 bool is_bad = FALSE;
1526 char name[3] = { '\0' };
1528 /* Convert the name into a string */
1533 else if (i == '\n') {
1534 my_strlcpy(name, "\n", sizeof(name));
1537 my_strlcpy(name, "\t", sizeof(name));
1540 /* Check each possibe class */
1541 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1543 DEBUG_L(PerlIO_printf(Perl_debug_log,
1544 "isalnum('%s') unexpectedly is %d\n",
1545 name, cBOOL(isalnum(i))));
1547 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1549 DEBUG_L(PerlIO_printf(Perl_debug_log,
1550 "isalpha('%s') unexpectedly is %d\n",
1551 name, cBOOL(isalpha(i))));
1553 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1555 DEBUG_L(PerlIO_printf(Perl_debug_log,
1556 "isdigit('%s') unexpectedly is %d\n",
1557 name, cBOOL(isdigit(i))));
1559 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1561 DEBUG_L(PerlIO_printf(Perl_debug_log,
1562 "isgraph('%s') unexpectedly is %d\n",
1563 name, cBOOL(isgraph(i))));
1565 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1567 DEBUG_L(PerlIO_printf(Perl_debug_log,
1568 "islower('%s') unexpectedly is %d\n",
1569 name, cBOOL(islower(i))));
1571 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1573 DEBUG_L(PerlIO_printf(Perl_debug_log,
1574 "isprint('%s') unexpectedly is %d\n",
1575 name, cBOOL(isprint(i))));
1577 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1579 DEBUG_L(PerlIO_printf(Perl_debug_log,
1580 "ispunct('%s') unexpectedly is %d\n",
1581 name, cBOOL(ispunct(i))));
1583 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1585 DEBUG_L(PerlIO_printf(Perl_debug_log,
1586 "isspace('%s') unexpectedly is %d\n",
1587 name, cBOOL(isspace(i))));
1589 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1591 DEBUG_L(PerlIO_printf(Perl_debug_log,
1592 "isupper('%s') unexpectedly is %d\n",
1593 name, cBOOL(isupper(i))));
1595 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1597 DEBUG_L(PerlIO_printf(Perl_debug_log,
1598 "isxdigit('%s') unexpectedly is %d\n",
1599 name, cBOOL(isxdigit(i))));
1601 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1603 DEBUG_L(PerlIO_printf(Perl_debug_log,
1604 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1605 name, tolower(i), (int) toLOWER_A(i)));
1607 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1609 DEBUG_L(PerlIO_printf(Perl_debug_log,
1610 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1611 name, toupper(i), (int) toUPPER_A(i)));
1613 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1615 DEBUG_L(PerlIO_printf(Perl_debug_log,
1616 "'\\n' (=%02X) is not a control\n", (int) i));
1619 /* Add to the list; Separate multiple entries with a blank */
1622 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1624 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1632 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1633 * this locale requires more than one byte, there are going to be
1635 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1636 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1637 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1639 if ( check_for_problems && MB_CUR_MAX > 1
1640 && ! PL_in_utf8_CTYPE_locale
1642 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1643 * locale. Just assume that the implementation for them (plus
1644 * for POSIX) is correct and the > 1 value is spurious. (Since
1645 * these are specially handled to never be considered UTF-8
1646 * locales, as long as this is the only problem, everything
1647 * should work fine */
1648 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1650 multi_byte_locale = TRUE;
1655 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1656 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1657 PL_warn_locale = Perl_newSVpvf(aTHX_
1658 "Locale '%s' contains (at least) the following characters"
1659 " which have\nunexpected meanings: %s\nThe Perl program"
1660 " will use the expected meanings",
1661 newctype, bad_chars_list);
1664 PL_warn_locale = Perl_newSVpvf(aTHX_
1665 "Locale '%s' may not work well.%s%s%s\n",
1668 ? " Some characters in it are not recognized by"
1672 ? "\nThe following characters (and maybe others)"
1673 " may not have the same meaning as the Perl"
1674 " program expects:\n"
1682 # ifdef HAS_NL_LANGINFO
1684 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1685 /* parameter FALSE is a don't care here */
1686 my_nl_langinfo(PERL_CODESET, FALSE));
1690 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1692 /* If we are actually in the scope of the locale or are debugging,
1693 * output the message now. If not in that scope, we save the
1694 * message to be output at the first operation using this locale,
1695 * if that actually happens. Most programs don't use locales, so
1696 * they are immune to bad ones. */
1697 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1699 /* The '0' below suppresses a bogus gcc compiler warning */
1700 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1702 if (IN_LC(LC_CTYPE)) {
1703 SvREFCNT_dec_NN(PL_warn_locale);
1704 PL_warn_locale = NULL;
1710 #endif /* USE_LOCALE_CTYPE */
1715 Perl__warn_problematic_locale()
1718 #ifdef USE_LOCALE_CTYPE
1722 /* Internal-to-core function that outputs the message in PL_warn_locale,
1723 * and then NULLS it. Should be called only through the macro
1724 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1726 if (PL_warn_locale) {
1727 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1728 SvPVX(PL_warn_locale),
1729 0 /* dummy to avoid compiler warning */ );
1730 SvREFCNT_dec_NN(PL_warn_locale);
1731 PL_warn_locale = NULL;
1739 S_new_collate(pTHX_ const char *newcoll)
1742 #ifndef USE_LOCALE_COLLATE
1744 PERL_UNUSED_ARG(newcoll);
1745 PERL_UNUSED_CONTEXT;
1749 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1750 * core Perl this and that 'newcoll' is the name of the new locale.
1752 * The design of locale collation is that every locale change is given an
1753 * index 'PL_collation_ix'. The first time a string particpates in an
1754 * operation that requires collation while locale collation is active, it
1755 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1756 * magic includes the collation index, and the transformation of the string
1757 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1758 * instead of the string itself. If a string changes, the magic is
1759 * cleared. The next time the locale changes, the index is incremented,
1760 * and so we know during a comparison that the transformation is not
1761 * necessarily still valid, and so is recomputed. Note that if the locale
1762 * changes enough times, the index could wrap (a U32), and it is possible
1763 * that a transformation would improperly be considered valid, leading to
1764 * an unlikely bug */
1767 if (PL_collation_name) {
1769 Safefree(PL_collation_name);
1770 PL_collation_name = NULL;
1772 PL_collation_standard = TRUE;
1773 is_standard_collation:
1774 PL_collxfrm_base = 0;
1775 PL_collxfrm_mult = 2;
1776 PL_in_utf8_COLLATE_locale = FALSE;
1777 PL_strxfrm_NUL_replacement = '\0';
1778 PL_strxfrm_max_cp = 0;
1782 /* If this is not the same locale as currently, set the new one up */
1783 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1785 Safefree(PL_collation_name);
1786 PL_collation_name = stdize_locale(savepv(newcoll));
1787 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1788 if (PL_collation_standard) {
1789 goto is_standard_collation;
1792 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1793 PL_strxfrm_NUL_replacement = '\0';
1794 PL_strxfrm_max_cp = 0;
1796 /* A locale collation definition includes primary, secondary, tertiary,
1797 * etc. weights for each character. To sort, the primary weights are
1798 * used, and only if they compare equal, then the secondary weights are
1799 * used, and only if they compare equal, then the tertiary, etc.
1801 * strxfrm() works by taking the input string, say ABC, and creating an
1802 * output transformed string consisting of first the primary weights,
1803 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1804 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1805 * may not have weights at every level. In our example, let's say B
1806 * doesn't have a tertiary weight, and A doesn't have a secondary
1807 * weight. The constructed string is then going to be
1808 * A¹B¹C¹ B²C² A³C³ ....
1809 * This has the desired effect that strcmp() will look at the secondary
1810 * or tertiary weights only if the strings compare equal at all higher
1811 * priority weights. The spaces shown here, like in
1813 * are not just for readability. In the general case, these must
1814 * actually be bytes, which we will call here 'separator weights'; and
1815 * they must be smaller than any other weight value, but since these
1816 * are C strings, only the terminating one can be a NUL (some
1817 * implementations may include a non-NUL separator weight just before
1818 * the NUL). Implementations tend to reserve 01 for the separator
1819 * weights. They are needed so that a shorter string's secondary
1820 * weights won't be misconstrued as primary weights of a longer string,
1821 * etc. By making them smaller than any other weight, the shorter
1822 * string will sort first. (Actually, if all secondary weights are
1823 * smaller than all primary ones, there is no need for a separator
1824 * weight between those two levels, etc.)
1826 * The length of the transformed string is roughly a linear function of
1827 * the input string. It's not exactly linear because some characters
1828 * don't have weights at all levels. When we call strxfrm() we have to
1829 * allocate some memory to hold the transformed string. The
1830 * calculations below try to find coefficients 'm' and 'b' for this
1831 * locale so that m*x + b equals how much space we need, given the size
1832 * of the input string in 'x'. If we calculate too small, we increase
1833 * the size as needed, and call strxfrm() again, but it is better to
1834 * get it right the first time to avoid wasted expensive string
1835 * transformations. */
1838 /* We use the string below to find how long the tranformation of it
1839 * is. Almost all locales are supersets of ASCII, or at least the
1840 * ASCII letters. We use all of them, half upper half lower,
1841 * because if we used fewer, we might hit just the ones that are
1842 * outliers in a particular locale. Most of the strings being
1843 * collated will contain a preponderance of letters, and even if
1844 * they are above-ASCII, they are likely to have the same number of
1845 * weight levels as the ASCII ones. It turns out that digits tend
1846 * to have fewer levels, and some punctuation has more, but those
1847 * are relatively sparse in text, and khw believes this gives a
1848 * reasonable result, but it could be changed if experience so
1850 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1851 char * x_longer; /* Transformed 'longer' */
1852 Size_t x_len_longer; /* Length of 'x_longer' */
1854 char * x_shorter; /* We also transform a substring of 'longer' */
1855 Size_t x_len_shorter;
1857 /* _mem_collxfrm() is used get the transformation (though here we
1858 * are interested only in its length). It is used because it has
1859 * the intelligence to handle all cases, but to work, it needs some
1860 * values of 'm' and 'b' to get it started. For the purposes of
1861 * this calculation we use a very conservative estimate of 'm' and
1862 * 'b'. This assumes a weight can be multiple bytes, enough to
1863 * hold any UV on the platform, and there are 5 levels, 4 weight
1864 * bytes, and a trailing NUL. */
1865 PL_collxfrm_base = 5;
1866 PL_collxfrm_mult = 5 * sizeof(UV);
1868 /* Find out how long the transformation really is */
1869 x_longer = _mem_collxfrm(longer,
1873 /* We avoid converting to UTF-8 in the
1874 * called function by telling it the
1875 * string is in UTF-8 if the locale is a
1876 * UTF-8 one. Since the string passed
1877 * here is invariant under UTF-8, we can
1878 * claim it's UTF-8 even though it isn't.
1880 PL_in_utf8_COLLATE_locale);
1883 /* Find out how long the transformation of a substring of 'longer'
1884 * is. Together the lengths of these transformations are
1885 * sufficient to calculate 'm' and 'b'. The substring is all of
1886 * 'longer' except the first character. This minimizes the chances
1887 * of being swayed by outliers */
1888 x_shorter = _mem_collxfrm(longer + 1,
1891 PL_in_utf8_COLLATE_locale);
1892 Safefree(x_shorter);
1894 /* If the results are nonsensical for this simple test, the whole
1895 * locale definition is suspect. Mark it so that locale collation
1896 * is not active at all for it. XXX Should we warn? */
1897 if ( x_len_shorter == 0
1898 || x_len_longer == 0
1899 || x_len_shorter >= x_len_longer)
1901 PL_collxfrm_mult = 0;
1902 PL_collxfrm_base = 0;
1905 SSize_t base; /* Temporary */
1907 /* We have both: m * strlen(longer) + b = x_len_longer
1908 * m * strlen(shorter) + b = x_len_shorter;
1909 * subtracting yields:
1910 * m * (strlen(longer) - strlen(shorter))
1911 * = x_len_longer - x_len_shorter
1912 * But we have set things up so that 'shorter' is 1 byte smaller
1913 * than 'longer'. Hence:
1914 * m = x_len_longer - x_len_shorter
1916 * But if something went wrong, make sure the multiplier is at
1919 if (x_len_longer > x_len_shorter) {
1920 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1923 PL_collxfrm_mult = 1;
1928 * but in case something has gone wrong, make sure it is
1930 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1935 /* Add 1 for the trailing NUL */
1936 PL_collxfrm_base = base + 1;
1941 if (DEBUG_L_TEST || debug_initialization) {
1942 PerlIO_printf(Perl_debug_log,
1943 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1945 " collate multipler=%zu, collate base=%zu\n",
1947 PL_in_utf8_COLLATE_locale,
1948 x_len_shorter, x_len_longer,
1949 PL_collxfrm_mult, PL_collxfrm_base);
1956 #endif /* USE_LOCALE_COLLATE */
1963 S_win32_setlocale(pTHX_ int category, const char* locale)
1965 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1966 * difference between the two unless the input locale is "", which normally
1967 * means on Windows to get the machine default, which is set via the
1968 * computer's "Regional and Language Options" (or its current equivalent).
1969 * In POSIX, it instead means to find the locale from the user's
1970 * environment. This routine changes the Windows behavior to first look in
1971 * the environment, and, if anything is found, use that instead of going to
1972 * the machine default. If there is no environment override, the machine
1973 * default is used, by calling the real setlocale() with "".
1975 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1976 * use the particular category's variable if set; otherwise to use the LANG
1979 bool override_LC_ALL = FALSE;
1983 if (locale && strEQ(locale, "")) {
1987 locale = PerlEnv_getenv("LC_ALL");
1989 if (category == LC_ALL) {
1990 override_LC_ALL = TRUE;
1996 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1997 if (category == categories[i]) {
1998 locale = PerlEnv_getenv(category_names[i]);
2003 locale = PerlEnv_getenv("LANG");
2019 result = setlocale(category, locale);
2020 DEBUG_L(STMT_START {
2022 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2023 setlocale_debug_string(category, locale, result));
2027 if (! override_LC_ALL) {
2031 /* Here the input category was LC_ALL, and we have set it to what is in the
2032 * LANG variable or the system default if there is no LANG. But these have
2033 * lower priority than the other LC_foo variables, so override it for each
2034 * one that is set. (If they are set to "", it means to use the same thing
2035 * we just set LC_ALL to, so can skip) */
2037 for (i = 0; i < LC_ALL_INDEX; i++) {
2038 result = PerlEnv_getenv(category_names[i]);
2039 if (result && strNE(result, "")) {
2040 setlocale(categories[i], result);
2041 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2043 setlocale_debug_string(categories[i], result, "not captured")));
2047 result = setlocale(LC_ALL, NULL);
2048 DEBUG_L(STMT_START {
2050 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2052 setlocale_debug_string(LC_ALL, NULL, result));
2063 =head1 Locale-related functions and macros
2065 =for apidoc Perl_setlocale
2067 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2068 taking the same parameters, and returning the same information, except that it
2069 returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2070 perl keeps that locale category as C<C>, changing it briefly during the
2071 operations where the underlying one is required.
2073 Another reason it isn't completely a drop-in replacement is that it is
2074 declared to return S<C<const char *>>, whereas the system setlocale omits the
2075 C<const>. (If it were being written today, plain setlocale would be declared
2076 const, since it is illegal to change the information it returns; doing so leads
2079 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2080 C<setlocale> can be completely ineffective on some platforms under some
2083 C<Perl_setlocale> should not be used to change the locale except on systems
2084 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2085 the system C<setlocale()> is ineffective, returning the wrong information, and
2086 failing to actually change the locale. C<Perl_setlocale>, however works
2087 properly in all circumstances.
2089 The return points to a per-thread static buffer, which is overwritten the next
2090 time C<Perl_setlocale> is called from the same thread.
2097 Perl_setlocale(const int category, const char * locale)
2099 /* This wraps POSIX::setlocale() */
2101 const char * retval;
2102 const char * newlocale;
2104 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2107 #ifdef USE_LOCALE_NUMERIC
2109 /* A NULL locale means only query what the current one is. We have the
2110 * LC_NUMERIC name saved, because we are normally switched into the C
2111 * locale for it. For an LC_ALL query, switch back to get the correct
2112 * results. All other categories don't require special handling */
2113 if (locale == NULL) {
2114 if (category == LC_NUMERIC) {
2116 /* We don't have to copy this return value, as it is a per-thread
2117 * variable, and won't change until a future setlocale */
2118 return PL_numeric_name;
2123 else if (category == LC_ALL) {
2124 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2133 retval = save_to_buffer(do_setlocale_r(category, locale),
2134 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2137 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2139 if (locale == NULL && category == LC_ALL) {
2140 RESTORE_LC_NUMERIC();
2145 DEBUG_L(PerlIO_printf(Perl_debug_log,
2146 "%s:%d: %s\n", __FILE__, __LINE__,
2147 setlocale_debug_string(category, locale, retval)));
2155 /* If locale == NULL, we are just querying the state */
2156 if (locale == NULL) {
2160 /* Now that have switched locales, we have to update our records to
2165 #ifdef USE_LOCALE_CTYPE
2172 #ifdef USE_LOCALE_COLLATE
2175 new_collate(retval);
2179 #ifdef USE_LOCALE_NUMERIC
2182 new_numeric(retval);
2190 /* LC_ALL updates all the things we care about. The values may not
2191 * be the same as 'retval', as the locale "" may have set things
2194 # ifdef USE_LOCALE_CTYPE
2196 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2197 new_ctype(newlocale);
2199 # endif /* USE_LOCALE_CTYPE */
2200 # ifdef USE_LOCALE_COLLATE
2202 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2203 new_collate(newlocale);
2206 # ifdef USE_LOCALE_NUMERIC
2208 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2209 new_numeric(newlocale);
2211 # endif /* USE_LOCALE_NUMERIC */
2222 PERL_STATIC_INLINE const char *
2223 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2225 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2226 * growing it if necessary */
2230 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2236 string_size = strlen(string) + offset + 1;
2238 if (*buf_size == 0) {
2239 Newx(*buf, string_size, char);
2240 *buf_size = string_size;
2242 else if (string_size > *buf_size) {
2243 Renew(*buf, string_size, char);
2244 *buf_size = string_size;
2247 Copy(string, *buf + offset, string_size - offset, char);
2253 =for apidoc Perl_langinfo
2255 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2256 taking the same C<item> parameter values, and returning the same information.
2257 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2258 of Perl's locale handling from your code, and can be used on systems that lack
2259 a native C<nl_langinfo>.
2267 The reason it isn't quite a drop-in replacement is actually an advantage. The
2268 only difference is that it returns S<C<const char *>>, whereas plain
2269 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2270 forbidden to write into the buffer. By declaring this C<const>, the compiler
2271 enforces this restriction, so if it is violated, you know at compilation time,
2272 rather than getting segfaults at runtime.
2276 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2277 without you having to write extra code. The reason for the extra code would be
2278 because these are from the C<LC_NUMERIC> locale category, which is normally
2279 kept set to the C locale by Perl, no matter what the underlying locale is
2280 supposed to be, and so to get the expected results, you have to temporarily
2281 toggle into the underlying locale, and later toggle back. (You could use plain
2282 C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2283 then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2284 C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2285 radix (decimal point) character to be a dot.)
2289 The system function it replaces can have its static return buffer trashed,
2290 not only by a subesequent call to that function, but by a C<freelocale>,
2291 C<setlocale>, or other locale change. The returned buffer of this function is
2292 not changed until the next call to it, so the buffer is never in a trashed
2297 Its return buffer is per-thread, so it also is never overwritten by a call to
2298 this function from another thread; unlike the function it replaces.
2302 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2303 as Windows, hence makes your code more portable. Of the fifty-some possible
2304 items specified by the POSIX 2008 standard,
2305 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2306 only two are completely unimplemented (though the loss of one of these is
2307 significant). It uses various techniques to recover the other items, including
2308 calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2309 in C89, so should be always be available. Later C<strftime()> versions have
2310 additional capabilities; C<""> is returned for those not available on your
2313 It is important to note that when called with an item that is recovered by
2314 using C<localeconv>, the buffer from any previous explicit call to
2315 C<localeconv> will be overwritten. This means you must save that buffer's
2316 contents if you need to access them after a call to this function.
2318 The details for those items which may differ from what this emulation returns
2319 and what a native C<nl_langinfo()> would return are:
2327 Unimplemented, so returns C<"">.
2337 Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
2338 removed from POSIX 2008, and are retained here for backwards compatibility.
2339 Your platform's C<nl_langinfo> may not support them.
2343 Always evaluates to C<%x>, the locale's appropriate date representation.
2347 Always evaluates to C<%X>, the locale's appropriate time representation.
2351 Always evaluates to C<%c>, the locale's appropriate date and time
2356 The return may be incorrect for those rare locales where the currency symbol
2357 replaces the radix character.
2358 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2359 to work differently.
2363 Currently this gives the same results as Linux does.
2364 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2365 to work differently.
2371 =item C<ERA_D_T_FMT>
2375 These are derived by using C<strftime()>, and not all versions of that function
2376 know about them. C<""> is returned for these on such systems.
2382 When using C<Perl_langinfo> on systems that don't have a native
2383 C<nl_langinfo()>, you must
2385 #include "perl_langinfo.h"
2387 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2388 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2389 C<langinfo.h> imports into the namespace for code that doesn't need it.)
2391 You also should not use the bare C<langinfo.h> item names, but should preface
2392 them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2393 The C<PERL_I<foo>> versions will also work for this function on systems that do
2394 have a native C<nl_langinfo>.
2396 The original impetus for C<Perl_langinfo()> was so that code that needs to
2397 find out the current currency symbol, floating point radix character, or digit
2398 grouping separator can use, on all systems, the simpler and more
2399 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2400 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2401 is better to use the methods given in L<perlcall> to call
2402 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2409 #ifdef HAS_NL_LANGINFO
2410 Perl_langinfo(const nl_item item)
2412 Perl_langinfo(const int item)
2415 return my_nl_langinfo(item, TRUE);
2419 #ifdef HAS_NL_LANGINFO
2420 S_my_nl_langinfo(const nl_item item, bool toggle)
2422 S_my_nl_langinfo(const int item, bool toggle)
2426 const char * retval;
2428 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2429 * two items, and only if not already there */
2430 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2431 || PL_numeric_underlying))
2436 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2437 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2438 || ! defined(HAS_POSIX_2008_LOCALE)
2440 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2441 * for those items dependent on it. This must be copied to a buffer before
2442 * switching back, as some systems destroy the buffer when setlocale() is
2446 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2449 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2452 LOCALE_LOCK; /* Prevent interference from another thread executing
2453 this code section (the only call to nl_langinfo in
2457 /* Copy to a per-thread buffer, which is also one that won't be
2458 * destroyed by a subsequent setlocale(), such as the
2459 * RESTORE_LC_NUMERIC may do just below. */
2460 retval = save_to_buffer(nl_langinfo(item),
2461 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2466 RESTORE_LC_NUMERIC();
2470 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2473 bool do_free = FALSE;
2474 locale_t cur = uselocale((locale_t) 0);
2476 if (cur == LC_GLOBAL_LOCALE) {
2477 cur = duplocale(LC_GLOBAL_LOCALE);
2482 if (PL_underlying_numeric_obj) {
2483 cur = PL_underlying_numeric_obj;
2486 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2491 /* We have to save it to a buffer, because the freelocale() just below
2492 * can invalidate the internal one */
2493 retval = save_to_buffer(nl_langinfo_l(item, cur),
2494 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2503 if (strEQ(retval, "")) {
2504 if (item == PERL_YESSTR) {
2507 if (item == PERL_NOSTR) {
2514 #else /* Below, emulate nl_langinfo as best we can */
2518 # ifdef HAS_LOCALECONV
2520 const struct lconv* lc;
2522 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2525 # ifdef HAS_STRFTIME
2528 bool return_format = FALSE; /* Return the %format, not the value */
2529 const char * format;
2533 /* We copy the results to a per-thread buffer, even if not
2534 * multi-threaded. This is in part to simplify this code, and partly
2535 * because we need a buffer anyway for strftime(), and partly because a
2536 * call of localeconv() could otherwise wipe out the buffer, and the
2537 * programmer would not be expecting this, as this is a nl_langinfo()
2538 * substitute after all, so s/he might be thinking their localeconv()
2539 * is safe until another localeconv() call. */
2544 /* These 2 are unimplemented */
2546 case PERL_ERA: /* For use with strftime() %E modifier */
2551 /* We use only an English set, since we don't know any more */
2552 case PERL_YESEXPR: return "^[+1yY]";
2553 case PERL_YESSTR: return "yes";
2554 case PERL_NOEXPR: return "^[-0nN]";
2555 case PERL_NOSTR: return "no";
2557 # ifdef HAS_LOCALECONV
2561 /* We don't bother with localeconv_l() because any system that
2562 * has it is likely to also have nl_langinfo() */
2564 LOCALE_LOCK; /* Prevent interference with other threads
2565 using localeconv() */
2569 || ! lc->currency_symbol
2570 || strEQ("", lc->currency_symbol))
2576 /* Leave the first spot empty to be filled in below */
2577 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2578 &PL_langinfo_bufsize, 1);
2579 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2580 { /* khw couldn't figure out how the localedef specifications
2581 would show that the $ should replace the radix; this is
2582 just a guess as to how it might work.*/
2583 PL_langinfo_buf[0] = '.';
2585 else if (lc->p_cs_precedes) {
2586 PL_langinfo_buf[0] = '-';
2589 PL_langinfo_buf[0] = '+';
2595 case PERL_RADIXCHAR:
2599 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2602 LOCALE_LOCK; /* Prevent interference with other threads
2603 using localeconv() */
2610 temp = (item == PERL_RADIXCHAR)
2612 : lc->thousands_sep;
2618 retval = save_to_buffer(temp, &PL_langinfo_buf,
2619 &PL_langinfo_bufsize, 0);
2624 RESTORE_LC_NUMERIC();
2630 # ifdef HAS_STRFTIME
2632 /* These are defined by C89, so we assume that strftime supports
2633 * them, and so are returned unconditionally; they may not be what
2634 * the locale actually says, but should give good enough results
2635 * for someone using them as formats (as opposed to trying to parse
2636 * them to figure out what the locale says). The other format
2637 * items are actually tested to verify they work on the platform */
2638 case PERL_D_FMT: return "%x";
2639 case PERL_T_FMT: return "%X";
2640 case PERL_D_T_FMT: return "%c";
2642 /* These formats are only available in later strfmtime's */
2643 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2644 case PERL_T_FMT_AMPM:
2646 /* The rest can be gotten from most versions of strftime(). */
2647 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2648 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2650 case PERL_ALT_DIGITS:
2651 case PERL_AM_STR: case PERL_PM_STR:
2652 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2653 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2654 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2655 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2656 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2657 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2658 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2659 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2660 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2665 init_tm(&tm); /* Precaution against core dumps */
2669 tm.tm_year = 2017 - 1900;
2676 "panic: %s: %d: switch case: %d problem",
2677 __FILE__, __LINE__, item);
2678 NOT_REACHED; /* NOTREACHED */
2680 case PERL_PM_STR: tm.tm_hour = 18;
2685 case PERL_ABDAY_7: tm.tm_wday++;
2686 case PERL_ABDAY_6: tm.tm_wday++;
2687 case PERL_ABDAY_5: tm.tm_wday++;
2688 case PERL_ABDAY_4: tm.tm_wday++;
2689 case PERL_ABDAY_3: tm.tm_wday++;
2690 case PERL_ABDAY_2: tm.tm_wday++;
2695 case PERL_DAY_7: tm.tm_wday++;
2696 case PERL_DAY_6: tm.tm_wday++;
2697 case PERL_DAY_5: tm.tm_wday++;
2698 case PERL_DAY_4: tm.tm_wday++;
2699 case PERL_DAY_3: tm.tm_wday++;
2700 case PERL_DAY_2: tm.tm_wday++;
2705 case PERL_ABMON_12: tm.tm_mon++;
2706 case PERL_ABMON_11: tm.tm_mon++;
2707 case PERL_ABMON_10: tm.tm_mon++;
2708 case PERL_ABMON_9: tm.tm_mon++;
2709 case PERL_ABMON_8: tm.tm_mon++;
2710 case PERL_ABMON_7: tm.tm_mon++;
2711 case PERL_ABMON_6: tm.tm_mon++;
2712 case PERL_ABMON_5: tm.tm_mon++;
2713 case PERL_ABMON_4: tm.tm_mon++;
2714 case PERL_ABMON_3: tm.tm_mon++;
2715 case PERL_ABMON_2: tm.tm_mon++;
2720 case PERL_MON_12: tm.tm_mon++;
2721 case PERL_MON_11: tm.tm_mon++;
2722 case PERL_MON_10: tm.tm_mon++;
2723 case PERL_MON_9: tm.tm_mon++;
2724 case PERL_MON_8: tm.tm_mon++;
2725 case PERL_MON_7: tm.tm_mon++;
2726 case PERL_MON_6: tm.tm_mon++;
2727 case PERL_MON_5: tm.tm_mon++;
2728 case PERL_MON_4: tm.tm_mon++;
2729 case PERL_MON_3: tm.tm_mon++;
2730 case PERL_MON_2: tm.tm_mon++;
2735 case PERL_T_FMT_AMPM:
2737 return_format = TRUE;
2740 case PERL_ERA_D_FMT:
2742 return_format = TRUE;
2745 case PERL_ERA_T_FMT:
2747 return_format = TRUE;
2750 case PERL_ERA_D_T_FMT:
2752 return_format = TRUE;
2755 case PERL_ALT_DIGITS:
2757 format = "%Ow"; /* Find the alternate digit for 0 */
2761 /* We can't use my_strftime() because it doesn't look at
2763 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2766 /* A zero return means one of:
2767 * a) there wasn't enough space in PL_langinfo_buf
2768 * b) the format, like a plain %p, returns empty
2769 * c) it was an illegal format, though some
2770 * implementations of strftime will just return the
2771 * illegal format as a plain character sequence.
2773 * To quickly test for case 'b)', try again but precede
2774 * the format with a plain character. If that result is
2775 * still empty, the problem is either 'a)' or 'c)' */
2777 Size_t format_size = strlen(format) + 1;
2778 Size_t mod_size = format_size + 1;
2782 Newx(mod_format, mod_size, char);
2783 Newx(temp_result, PL_langinfo_bufsize, char);
2785 my_strlcpy(mod_format + 1, format, mod_size);
2786 len = strftime(temp_result,
2787 PL_langinfo_bufsize,
2789 Safefree(mod_format);
2790 Safefree(temp_result);
2792 /* If 'len' is non-zero, it means that we had a case like
2793 * %p which means the current locale doesn't use a.m. or
2794 * p.m., and that is valid */
2797 /* Here, still didn't work. If we get well beyond a
2798 * reasonable size, bail out to prevent an infinite
2801 if (PL_langinfo_bufsize > 100 * format_size) {
2802 *PL_langinfo_buf = '\0';
2805 /* Double the buffer size to retry; Add 1 in case
2806 * original was 0, so we aren't stuck at 0. */
2807 PL_langinfo_bufsize *= 2;
2808 PL_langinfo_bufsize++;
2809 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2817 /* Here, we got a result.
2819 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2820 * alternate format for wday 0. If the value is the same as
2821 * the normal 0, there isn't an alternate, so clear the buffer.
2823 if ( item == PERL_ALT_DIGITS
2824 && strEQ(PL_langinfo_buf, "0"))
2826 *PL_langinfo_buf = '\0';
2829 /* ALT_DIGITS is problematic. Experiments on it showed that
2830 * strftime() did not always work properly when going from
2831 * alt-9 to alt-10. Only a few locales have this item defined,
2832 * and in all of them on Linux that khw was able to find,
2833 * nl_langinfo() merely returned the alt-0 character, possibly
2834 * doubled. Most Unicode digits are in blocks of 10
2835 * consecutive code points, so that is sufficient information
2836 * for those scripts, as we can infer alt-1, alt-2, .... But
2837 * for a Japanese locale, a CJK ideographic 0 is returned, and
2838 * the CJK digits are not in code point order, so you can't
2839 * really infer anything. The localedef for this locale did
2840 * specify the succeeding digits, so that strftime() works
2841 * properly on them, without needing to infer anything. But
2842 * the nl_langinfo() return did not give sufficient information
2843 * for the caller to understand what's going on. So until
2844 * there is evidence that it should work differently, this
2845 * returns the alt-0 string for ALT_DIGITS.
2847 * wday was chosen because its range is all a single digit.
2848 * Things like tm_sec have two digits as the minimum: '00' */
2852 retval = PL_langinfo_buf;
2854 /* If to return the format, not the value, overwrite the buffer
2855 * with it. But some strftime()s will keep the original format
2856 * if illegal, so change those to "" */
2857 if (return_format) {
2858 if (strEQ(PL_langinfo_buf, format)) {
2859 *PL_langinfo_buf = '\0';
2862 retval = save_to_buffer(format, &PL_langinfo_buf,
2863 &PL_langinfo_bufsize, 0);
2881 * Initialize locale awareness.
2884 Perl_init_i18nl10n(pTHX_ int printwarn)
2888 * 0 if not to output warning when setup locale is bad
2889 * 1 if to output warning based on value of PERL_BADLANG
2890 * >1 if to output regardless of PERL_BADLANG
2893 * 1 = set ok or not applicable,
2894 * 0 = fallback to a locale of lower priority
2895 * -1 = fallback to all locales failed, not even to the C locale
2897 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2898 * set, debugging information is output.
2900 * This looks more complicated than it is, mainly due to the #ifdefs.
2902 * We try to set LC_ALL to the value determined by the environment. If
2903 * there is no LC_ALL on this platform, we try the individual categories we
2904 * know about. If this works, we are done.
2906 * But if it doesn't work, we have to do something else. We search the
2907 * environment variables ourselves instead of relying on the system to do
2908 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2909 * think there is one), and the ultimate fallback "C". This is all done in
2910 * the same loop as above to avoid duplicating code, but it makes things
2911 * more complex. The 'trial_locales' array is initialized with just one
2912 * element; it causes the behavior described in the paragraph above this to
2913 * happen. If that fails, we add elements to 'trial_locales', and do extra
2914 * loop iterations to cause the behavior described in this paragraph.
2916 * On Ultrix, the locale MUST come from the environment, so there is
2917 * preliminary code to set it. I (khw) am not sure that it is necessary,
2918 * and that this couldn't be folded into the loop, but barring any real
2919 * platforms to test on, it's staying as-is
2921 * A slight complication is that in embedded Perls, the locale may already
2922 * be set-up, and we don't want to get it from the normal environment
2923 * variables. This is handled by having a special environment variable
2924 * indicate we're in this situation. We simply set setlocale's 2nd
2925 * parameter to be a NULL instead of "". That indicates to setlocale that
2926 * it is not to change anything, but to return the current value,
2927 * effectively initializing perl's db to what the locale already is.
2929 * We play the same trick with NULL if a LC_ALL succeeds. We call
2930 * setlocale() on the individual categores with NULL to get their existing
2931 * values for our db, instead of trying to change them.
2938 PERL_UNUSED_ARG(printwarn);
2940 #else /* USE_LOCALE */
2943 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
2947 /* NULL uses the existing already set up locale */
2948 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2951 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2952 unsigned int trial_locales_count;
2953 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2954 const char * const lang = savepv(PerlEnv_getenv("LANG"));
2955 bool setlocale_failure = FALSE;
2958 /* A later getenv() could zap this, so only use here */
2959 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2961 const bool locwarn = (printwarn > 1
2963 && ( ! bad_lang_use_once
2965 /* disallow with "" or "0" */
2967 && strNE("0", bad_lang_use_once)))));
2969 /* setlocale() return vals; not copied so must be looked at immediately */
2970 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
2972 /* current locale for given category; should have been copied so aren't
2974 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
2978 /* In some systems you can find out the system default locale
2979 * and use that as the fallback locale. */
2980 # define SYSTEM_DEFAULT_LOCALE
2982 # ifdef SYSTEM_DEFAULT_LOCALE
2984 const char *system_default_locale = NULL;
2989 # define DEBUG_LOCALE_INIT(a,b,c)
2992 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
2994 # define DEBUG_LOCALE_INIT(category, locale, result) \
2996 if (debug_initialization) { \
2997 PerlIO_printf(Perl_debug_log, \
2999 __FILE__, __LINE__, \
3000 setlocale_debug_string(category, \
3006 /* Make sure the parallel arrays are properly set up */
3007 # ifdef USE_LOCALE_NUMERIC
3008 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3009 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3010 # ifdef USE_POSIX_2008_LOCALE
3011 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3014 # ifdef USE_LOCALE_CTYPE
3015 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3016 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3017 # ifdef USE_POSIX_2008_LOCALE
3018 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3021 # ifdef USE_LOCALE_COLLATE
3022 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3023 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3024 # ifdef USE_POSIX_2008_LOCALE
3025 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3028 # ifdef USE_LOCALE_TIME
3029 assert(categories[LC_TIME_INDEX] == LC_TIME);
3030 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3031 # ifdef USE_POSIX_2008_LOCALE
3032 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3035 # ifdef USE_LOCALE_MESSAGES
3036 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3037 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3038 # ifdef USE_POSIX_2008_LOCALE
3039 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3042 # ifdef USE_LOCALE_MONETARY
3043 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3044 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3045 # ifdef USE_POSIX_2008_LOCALE
3046 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3049 # ifdef USE_LOCALE_ADDRESS
3050 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3051 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3052 # ifdef USE_POSIX_2008_LOCALE
3053 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3056 # ifdef USE_LOCALE_IDENTIFICATION
3057 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3058 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3059 # ifdef USE_POSIX_2008_LOCALE
3060 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3063 # ifdef USE_LOCALE_MEASUREMENT
3064 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3065 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3066 # ifdef USE_POSIX_2008_LOCALE
3067 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3070 # ifdef USE_LOCALE_PAPER
3071 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3072 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3073 # ifdef USE_POSIX_2008_LOCALE
3074 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3077 # ifdef USE_LOCALE_TELEPHONE
3078 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3079 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3080 # ifdef USE_POSIX_2008_LOCALE
3081 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3085 assert(categories[LC_ALL_INDEX] == LC_ALL);
3086 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3087 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3088 # ifdef USE_POSIX_2008_LOCALE
3089 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3092 # endif /* DEBUGGING */
3094 /* Initialize the cache of the program's UTF-8ness for the always known
3095 * locales C and POSIX */
3096 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3097 sizeof(PL_locale_utf8ness));
3099 # ifdef USE_THREAD_SAFE_LOCALE
3102 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3106 # if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3108 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3109 if (! PL_C_locale_obj) {
3110 Perl_croak_nocontext(
3111 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3113 if (DEBUG_Lv_TEST || debug_initialization) {
3114 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3119 PL_numeric_radix_sv = newSVpvs(".");
3121 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3123 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3124 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3127 # ifdef LOCALE_ENVIRON_REQUIRED
3130 * Ultrix setlocale(..., "") fails if there are no environment
3131 * variables from which to get a locale name.
3135 # error Ultrix without LC_ALL not implemented
3141 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3142 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3143 if (sl_result[LC_ALL_INDEX])
3146 setlocale_failure = TRUE;
3148 if (! setlocale_failure) {
3149 const char * locale_param;
3150 for (i = 0; i < LC_ALL_INDEX; i++) {
3151 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3154 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3155 if (! sl_result[i]) {
3156 setlocale_failure = TRUE;
3158 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3163 # endif /* LC_ALL */
3164 # endif /* LOCALE_ENVIRON_REQUIRED */
3166 /* We try each locale in the list until we get one that works, or exhaust
3167 * the list. Normally the loop is executed just once. But if setting the
3168 * locale fails, inside the loop we add fallback trials to the array and so
3169 * will execute the loop multiple times */
3170 trial_locales[0] = setlocale_init;
3171 trial_locales_count = 1;
3173 for (i= 0; i < trial_locales_count; i++) {
3174 const char * trial_locale = trial_locales[i];
3178 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3179 * when i==0, but I (khw) don't think that behavior makes much
3181 setlocale_failure = FALSE;
3183 # ifdef SYSTEM_DEFAULT_LOCALE
3184 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3186 /* On Windows machines, an entry of "" after the 0th means to use
3187 * the system default locale, which we now proceed to get. */
3188 if (strEQ(trial_locale, "")) {
3191 /* Note that this may change the locale, but we are going to do
3192 * that anyway just below */
3193 system_default_locale = do_setlocale_c(LC_ALL, "");
3194 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3196 /* Skip if invalid or if it's already on the list of locales to
3198 if (! system_default_locale) {
3199 goto next_iteration;
3201 for (j = 0; j < trial_locales_count; j++) {
3202 if (strEQ(system_default_locale, trial_locales[j])) {
3203 goto next_iteration;
3207 trial_locale = system_default_locale;
3210 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3212 # endif /* SYSTEM_DEFAULT_LOCALE */
3218 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3219 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3220 if (! sl_result[LC_ALL_INDEX]) {
3221 setlocale_failure = TRUE;
3224 /* Since LC_ALL succeeded, it should have changed all the other
3225 * categories it can to its value; so we massage things so that the
3226 * setlocales below just return their category's current values.
3227 * This adequately handles the case in NetBSD where LC_COLLATE may
3228 * not be defined for a locale, and setting it individually will
3229 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3230 * the POSIX locale. */
3231 trial_locale = NULL;
3234 # endif /* LC_ALL */
3236 if (! setlocale_failure) {
3238 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3240 = savepv(do_setlocale_r(categories[j], trial_locale));
3241 if (! curlocales[j]) {
3242 setlocale_failure = TRUE;
3244 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3247 if (! setlocale_failure) { /* All succeeded */
3248 break; /* Exit trial_locales loop */
3252 /* Here, something failed; will need to try a fallback. */
3258 if (locwarn) { /* Output failure info only on the first one */
3262 PerlIO_printf(Perl_error_log,
3263 "perl: warning: Setting locale failed.\n");
3265 # else /* !LC_ALL */
3267 PerlIO_printf(Perl_error_log,
3268 "perl: warning: Setting locale failed for the categories:\n\t");
3270 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3271 if (! curlocales[j]) {
3272 PerlIO_printf(Perl_error_log, category_names[j]);
3275 Safefree(curlocales[j]);
3279 # endif /* LC_ALL */
3281 PerlIO_printf(Perl_error_log,
3282 "perl: warning: Please check that your locale settings:\n");
3286 PerlIO_printf(Perl_error_log,
3287 "\tLANGUAGE = %c%s%c,\n",
3288 language ? '"' : '(',
3289 language ? language : "unset",
3290 language ? '"' : ')');
3293 PerlIO_printf(Perl_error_log,
3294 "\tLC_ALL = %c%s%c,\n",
3296 lc_all ? lc_all : "unset",
3297 lc_all ? '"' : ')');
3299 # if defined(USE_ENVIRON_ARRAY)
3304 /* Look through the environment for any variables of the
3305 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3306 * already handled above. These are assumed to be locale
3307 * settings. Output them and their values. */
3308 for (e = environ; *e; e++) {
3309 const STRLEN prefix_len = sizeof("LC_") - 1;
3312 if ( strBEGINs(*e, "LC_")
3313 && ! strBEGINs(*e, "LC_ALL=")
3314 && (uppers_len = strspn(*e + prefix_len,
3315 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3316 && ((*e)[prefix_len + uppers_len] == '='))
3318 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3319 (int) (prefix_len + uppers_len), *e,
3320 *e + prefix_len + uppers_len + 1);
3327 PerlIO_printf(Perl_error_log,
3328 "\t(possibly more locale environment variables)\n");
3332 PerlIO_printf(Perl_error_log,
3333 "\tLANG = %c%s%c\n",
3335 lang ? lang : "unset",
3338 PerlIO_printf(Perl_error_log,
3339 " are supported and installed on your system.\n");
3342 /* Calculate what fallback locales to try. We have avoided this
3343 * until we have to, because failure is quite unlikely. This will
3344 * usually change the upper bound of the loop we are in.
3346 * Since the system's default way of setting the locale has not
3347 * found one that works, We use Perl's defined ordering: LC_ALL,
3348 * LANG, and the C locale. We don't try the same locale twice, so
3349 * don't add to the list if already there. (On POSIX systems, the
3350 * LC_ALL element will likely be a repeat of the 0th element "",
3351 * but there's no harm done by doing it explicitly.
3353 * Note that this tries the LC_ALL environment variable even on
3354 * systems which have no LC_ALL locale setting. This may or may
3355 * not have been originally intentional, but there's no real need
3356 * to change the behavior. */
3358 for (j = 0; j < trial_locales_count; j++) {
3359 if (strEQ(lc_all, trial_locales[j])) {
3363 trial_locales[trial_locales_count++] = lc_all;
3368 for (j = 0; j < trial_locales_count; j++) {
3369 if (strEQ(lang, trial_locales[j])) {
3373 trial_locales[trial_locales_count++] = lang;
3377 # if defined(WIN32) && defined(LC_ALL)
3379 /* For Windows, we also try the system default locale before "C".
3380 * (If there exists a Windows without LC_ALL we skip this because
3381 * it gets too complicated. For those, the "C" is the next
3382 * fallback possibility). The "" is the same as the 0th element of
3383 * the array, but the code at the loop above knows to treat it
3384 * differently when not the 0th */
3385 trial_locales[trial_locales_count++] = "";
3389 for (j = 0; j < trial_locales_count; j++) {
3390 if (strEQ("C", trial_locales[j])) {
3394 trial_locales[trial_locales_count++] = "C";
3397 } /* end of first time through the loop */
3405 } /* end of looping through the trial locales */
3407 if (ok < 1) { /* If we tried to fallback */
3409 if (! setlocale_failure) { /* fallback succeeded */
3410 msg = "Falling back to";
3412 else { /* fallback failed */
3415 /* We dropped off the end of the loop, so have to decrement i to
3416 * get back to the value the last time through */
3420 msg = "Failed to fall back to";
3422 /* To continue, we should use whatever values we've got */
3424 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3425 Safefree(curlocales[j]);
3426 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3427 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3432 const char * description;
3433 const char * name = "";
3434 if (strEQ(trial_locales[i], "C")) {
3435 description = "the standard locale";
3439 # ifdef SYSTEM_DEFAULT_LOCALE
3441 else if (strEQ(trial_locales[i], "")) {
3442 description = "the system default locale";
3443 if (system_default_locale) {
3444 name = system_default_locale;
3448 # endif /* SYSTEM_DEFAULT_LOCALE */
3451 description = "a fallback locale";
3452 name = trial_locales[i];
3454 if (name && strNE(name, "")) {
3455 PerlIO_printf(Perl_error_log,
3456 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3459 PerlIO_printf(Perl_error_log,
3460 "perl: warning: %s %s.\n", msg, description);
3463 } /* End of tried to fallback */
3465 /* Done with finding the locales; update our records */
3467 # ifdef USE_LOCALE_CTYPE
3469 new_ctype(curlocales[LC_CTYPE_INDEX]);
3472 # ifdef USE_LOCALE_COLLATE
3474 new_collate(curlocales[LC_COLLATE_INDEX]);
3477 # ifdef USE_LOCALE_NUMERIC
3479 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3483 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3485 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3487 /* This caches whether each category's locale is UTF-8 or not. This
3488 * may involve changing the locale. It is ok to do this at
3489 * initialization time before any threads have started, but not later
3490 * unless thread-safe operations are used.
3491 * Caching means that if the program heeds our dictate not to change
3492 * locales in threaded applications, this data will remain valid, and
3493 * it may get queried without having to change locales. If the
3494 * environment is such that all categories have the same locale, this
3495 * isn't needed, as the code will not change the locale; but this
3496 * handles the uncommon case where the environment has disparate
3497 * locales for the categories */
3498 (void) _is_cur_LC_category_utf8(categories[i]);
3502 Safefree(curlocales[i]);
3505 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3507 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3508 * locale is UTF-8. The call to new_ctype() just above has already
3509 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3510 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3511 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3512 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3513 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3515 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3516 This is an alternative to using the -C command line switch
3517 (the -C if present will override this). */
3519 const char *p = PerlEnv_getenv("PERL_UNICODE");
3520 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3521 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3535 #endif /* USE_LOCALE */
3538 /* So won't continue to output stuff */
3539 DEBUG_INITIALIZATION_set(FALSE);
3546 #ifdef USE_LOCALE_COLLATE
3549 Perl__mem_collxfrm(pTHX_ const char *input_string,
3550 STRLEN len, /* Length of 'input_string' */
3551 STRLEN *xlen, /* Set to length of returned string
3552 (not including the collation index
3554 bool utf8 /* Is the input in UTF-8? */
3558 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3559 * differences. First, it handles embedded NULs. Second, it allocates a bit
3560 * more memory than needed for the transformed data itself. The real
3561 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3562 * the length of that, and doesn't include the collation index size.
3563 * Please see sv_collxfrm() to see how this is used. */
3565 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3567 char * s = (char *) input_string;
3568 STRLEN s_strlen = strlen(input_string);
3570 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3571 STRLEN length_in_chars;
3572 bool first_time = TRUE; /* Cleared after first loop iteration */
3574 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3576 /* Must be NUL-terminated */
3577 assert(*(input_string + len) == '\0');
3579 /* If this locale has defective collation, skip */
3580 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3581 DEBUG_L(PerlIO_printf(Perl_debug_log,
3582 "_mem_collxfrm: locale's collation is defective\n"));
3586 /* Replace any embedded NULs with the control that sorts before any others.
3587 * This will give as good as possible results on strings that don't
3588 * otherwise contain that character, but otherwise there may be
3589 * less-than-perfect results with that character and NUL. This is
3590 * unavoidable unless we replace strxfrm with our own implementation. */
3591 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3595 STRLEN sans_nuls_len;
3596 int try_non_controls;
3597 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3598 making sure 2nd byte is NUL.
3600 STRLEN this_replacement_len;
3602 /* If we don't know what non-NUL control character sorts lowest for
3603 * this locale, find it */
3604 if (PL_strxfrm_NUL_replacement == '\0') {
3606 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3607 includes the collation index
3610 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3612 /* Unlikely, but it may be that no control will work to replace
3613 * NUL, in which case we instead look for any character. Controls
3614 * are preferred because collation order is, in general, context
3615 * sensitive, with adjoining characters affecting the order, and
3616 * controls are less likely to have such interactions, allowing the
3617 * NUL-replacement to stand on its own. (Another way to look at it
3618 * is to imagine what would happen if the NUL were replaced by a
3619 * combining character; it wouldn't work out all that well.) */
3620 for (try_non_controls = 0;
3621 try_non_controls < 2;
3624 /* Look through all legal code points (NUL isn't) */
3625 for (j = 1; j < 256; j++) {
3626 char * x; /* j's xfrm plus collation index */
3627 STRLEN x_len; /* length of 'x' */
3628 STRLEN trial_len = 1;
3629 char cur_source[] = { '\0', '\0' };
3631 /* Skip non-controls the first time through the loop. The
3632 * controls in a UTF-8 locale are the L1 ones */
3633 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3640 /* Create a 1-char string of the current code point */
3641 cur_source[0] = (char) j;
3643 /* Then transform it */
3644 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3645 0 /* The string is not in UTF-8 */);
3647 /* Ignore any character that didn't successfully transform.
3653 /* If this character's transformation is lower than
3654 * the current lowest, this one becomes the lowest */
3655 if ( cur_min_x == NULL
3656 || strLT(x + COLLXFRM_HDR_LEN,
3657 cur_min_x + COLLXFRM_HDR_LEN))
3659 PL_strxfrm_NUL_replacement = j;
3665 } /* end of loop through all 255 characters */
3667 /* Stop looking if found */
3672 /* Unlikely, but possible, if there aren't any controls that
3673 * work in the locale, repeat the loop, looking for any
3674 * character that works */
3675 DEBUG_L(PerlIO_printf(Perl_debug_log,
3676 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3677 } /* End of loop to try first the controls, then any char */
3680 DEBUG_L(PerlIO_printf(Perl_debug_log,
3681 "_mem_collxfrm: Couldn't find any character to replace"
3682 " embedded NULs in locale %s with", PL_collation_name));
3686 DEBUG_L(PerlIO_printf(Perl_debug_log,
3687 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3688 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3690 Safefree(cur_min_x);
3691 } /* End of determining the character that is to replace NULs */
3693 /* If the replacement is variant under UTF-8, it must match the
3694 * UTF8-ness of the original */
3695 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3696 this_replacement_char[0] =
3697 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3698 this_replacement_char[1] =
3699 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3700 this_replacement_len = 2;
3703 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3704 /* this_replacement_char[1] = '\0' was done at initialization */
3705 this_replacement_len = 1;
3708 /* The worst case length for the replaced string would be if every
3709 * character in it is NUL. Multiply that by the length of each
3710 * replacement, and allow for a trailing NUL */
3711 sans_nuls_len = (len * this_replacement_len) + 1;
3712 Newx(sans_nuls, sans_nuls_len, char);
3715 /* Replace each NUL with the lowest collating control. Loop until have
3716 * exhausted all the NULs */
3717 while (s + s_strlen < e) {
3718 my_strlcat(sans_nuls, s, sans_nuls_len);
3720 /* Do the actual replacement */
3721 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3723 /* Move past the input NUL */
3725 s_strlen = strlen(s);
3728 /* And add anything that trails the final NUL */
3729 my_strlcat(sans_nuls, s, sans_nuls_len);
3731 /* Switch so below we transform this modified string */
3734 } /* End of replacing NULs */
3736 /* Make sure the UTF8ness of the string and locale match */
3737 if (utf8 != PL_in_utf8_COLLATE_locale) {
3738 /* XXX convert above Unicode to 10FFFF? */
3739 const char * const t = s; /* Temporary so we can later find where the
3742 /* Here they don't match. Change the string's to be what the locale is
3745 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3746 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3749 else { /* locale is not UTF-8; but input is; downgrade the input */
3751 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3753 /* If the downgrade was successful we are done, but if the input
3754 * contains things that require UTF-8 to represent, have to do
3755 * damage control ... */
3756 if (UNLIKELY(utf8)) {
3758 /* What we do is construct a non-UTF-8 string with
3759 * 1) the characters representable by a single byte converted
3760 * to be so (if necessary);
3761 * 2) and the rest converted to collate the same as the
3762 * highest collating representable character. That makes
3763 * them collate at the end. This is similar to how we
3764 * handle embedded NULs, but we use the highest collating
3765 * code point instead of the smallest. Like the NUL case,
3766 * this isn't perfect, but is the best we can reasonably
3767 * do. Every above-255 code point will sort the same as
3768 * the highest-sorting 0-255 code point. If that code
3769 * point can combine in a sequence with some other code
3770 * points for weight calculations, us changing something to
3771 * be it can adversely affect the results. But in most
3772 * cases, it should work reasonably. And note that this is
3773 * really an illegal situation: using code points above 255
3774 * on a locale where only 0-255 are valid. If two strings
3775 * sort entirely equal, then the sort order for the
3776 * above-255 code points will be in code point order. */
3780 /* If we haven't calculated the code point with the maximum
3781 * collating order for this locale, do so now */
3782 if (! PL_strxfrm_max_cp) {
3785 /* The current transformed string that collates the
3786 * highest (except it also includes the prefixed collation
3788 char * cur_max_x = NULL;
3790 /* Look through all legal code points (NUL isn't) */
3791 for (j = 1; j < 256; j++) {
3794 char cur_source[] = { '\0', '\0' };
3796 /* Create a 1-char string of the current code point */
3797 cur_source[0] = (char) j;
3799 /* Then transform it */
3800 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3802 /* If something went wrong (which it shouldn't), just
3803 * ignore this code point */
3808 /* If this character's transformation is higher than
3809 * the current highest, this one becomes the highest */
3810 if ( cur_max_x == NULL
3811 || strGT(x + COLLXFRM_HDR_LEN,
3812 cur_max_x + COLLXFRM_HDR_LEN))
3814 PL_strxfrm_max_cp = j;
3823 DEBUG_L(PerlIO_printf(Perl_debug_log,
3824 "_mem_collxfrm: Couldn't find any character to"
3825 " replace above-Latin1 chars in locale %s with",
3826 PL_collation_name));
3830 DEBUG_L(PerlIO_printf(Perl_debug_log,
3831 "_mem_collxfrm: highest 1-byte collating character"
3832 " in locale %s is 0x%02X\n",
3834 PL_strxfrm_max_cp));
3836 Safefree(cur_max_x);
3839 /* Here we know which legal code point collates the highest.
3840 * We are ready to construct the non-UTF-8 string. The length
3841 * will be at least 1 byte smaller than the input string
3842 * (because we changed at least one 2-byte character into a
3843 * single byte), but that is eaten up by the trailing NUL */
3849 char * e = (char *) t + len;
3851 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3853 if (UTF8_IS_INVARIANT(cur_char)) {
3856 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
3857 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3859 else { /* Replace illegal cp with highest collating
3861 s[d++] = PL_strxfrm_max_cp;
3865 Renew(s, d, char); /* Free up unused space */
3870 /* Here, we have constructed a modified version of the input. It could
3871 * be that we already had a modified copy before we did this version.
3872 * If so, that copy is no longer needed */
3873 if (t != input_string) {
3878 length_in_chars = (utf8)
3879 ? utf8_length((U8 *) s, (U8 *) s + len)
3882 /* The first element in the output is the collation id, used by
3883 * sv_collxfrm(); then comes the space for the transformed string. The
3884 * equation should give us a good estimate as to how much is needed */
3885 xAlloc = COLLXFRM_HDR_LEN
3887 + (PL_collxfrm_mult * length_in_chars);
3888 Newx(xbuf, xAlloc, char);
3889 if (UNLIKELY(! xbuf)) {
3890 DEBUG_L(PerlIO_printf(Perl_debug_log,
3891 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
3895 /* Store the collation id */
3896 *(U32*)xbuf = PL_collation_ix;
3898 /* Then the transformation of the input. We loop until successful, or we
3902 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
3904 /* If the transformed string occupies less space than we told strxfrm()
3905 * was available, it means it successfully transformed the whole
3907 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
3909 /* Some systems include a trailing NUL in the returned length.
3910 * Ignore it, using a loop in case multiple trailing NULs are
3913 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3918 /* If the first try didn't get it, it means our prediction was low.
3919 * Modify the coefficients so that we predict a larger value in any
3920 * future transformations */
3922 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3923 STRLEN computed_guess = PL_collxfrm_base
3924 + (PL_collxfrm_mult * length_in_chars);
3926 /* On zero-length input, just keep current slope instead of
3928 const STRLEN new_m = (length_in_chars != 0)
3929 ? needed / length_in_chars
3932 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3933 "%s: %d: initial size of %zu bytes for a length "
3934 "%zu string was insufficient, %zu needed\n",
3936 computed_guess, length_in_chars, needed));
3938 /* If slope increased, use it, but discard this result for
3939 * length 1 strings, as we can't be sure that it's a real slope
3941 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
3945 STRLEN old_m = PL_collxfrm_mult;
3946 STRLEN old_b = PL_collxfrm_base;
3950 PL_collxfrm_mult = new_m;
3951 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3952 computed_guess = PL_collxfrm_base
3953 + (PL_collxfrm_mult * length_in_chars);
3954 if (computed_guess < needed) {
3955 PL_collxfrm_base += needed - computed_guess;
3958 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3959 "%s: %d: slope is now %zu; was %zu, base "
3960 "is now %zu; was %zu\n",
3962 PL_collxfrm_mult, old_m,
3963 PL_collxfrm_base, old_b));
3965 else { /* Slope didn't change, but 'b' did */
3966 const STRLEN new_b = needed
3969 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3970 "%s: %d: base is now %zu; was %zu\n",
3972 new_b, PL_collxfrm_base));
3973 PL_collxfrm_base = new_b;
3980 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3981 DEBUG_L(PerlIO_printf(Perl_debug_log,
3982 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3983 *xlen, PERL_INT_MAX));
3987 /* A well-behaved strxfrm() returns exactly how much space it needs
3988 * (usually not including the trailing NUL) when it fails due to not
3989 * enough space being provided. Assume that this is the case unless
3990 * it's been proven otherwise */
3991 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
3992 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
3994 else { /* Here, either:
3995 * 1) The strxfrm() has previously shown bad behavior; or
3996 * 2) It isn't the first time through the loop, which means
3997 * that the strxfrm() is now showing bad behavior, because
3998 * we gave it what it said was needed in the previous
3999 * iteration, and it came back saying it needed still more.
4000 * (Many versions of cygwin fit this. When the buffer size
4001 * isn't sufficient, they return the input size instead of
4002 * how much is needed.)
4003 * Increase the buffer size by a fixed percentage and try again.
4005 xAlloc += (xAlloc / 4) + 1;
4006 PL_strxfrm_is_behaved = FALSE;
4010 if (DEBUG_Lv_TEST || debug_initialization) {
4011 PerlIO_printf(Perl_debug_log,
4012 "_mem_collxfrm required more space than previously calculated"
4013 " for locale %s, trying again with new guess=%d+%zu\n",
4014 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4015 xAlloc - COLLXFRM_HDR_LEN);
4022 Renew(xbuf, xAlloc, char);
4023 if (UNLIKELY(! xbuf)) {
4024 DEBUG_L(PerlIO_printf(Perl_debug_log,
4025 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4035 if (DEBUG_Lv_TEST || debug_initialization) {
4037 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4038 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4039 PerlIO_printf(Perl_debug_log, "%s\n",
4040 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4046 /* Free up unneeded space; retain ehough for trailing NUL */
4047 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4049 if (s != input_string) {
4057 if (s != input_string) {
4064 if (DEBUG_Lv_TEST || debug_initialization) {
4065 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4076 S_print_collxfrm_input_and_return(pTHX_
4077 const char * const s,
4078 const char * const e,
4079 const STRLEN * const xlen,
4083 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4085 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4086 (UV)PL_collation_ix);
4088 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4091 PerlIO_printf(Perl_debug_log, "NULL");
4093 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4095 print_bytes_for_locale(s, e, is_utf8);
4097 PerlIO_printf(Perl_debug_log, "'\n");
4101 S_print_bytes_for_locale(pTHX_
4102 const char * const s,
4103 const char * const e,
4107 bool prev_was_printable = TRUE;
4108 bool first_time = TRUE;
4110 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4114 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4117 if (! prev_was_printable) {
4118 PerlIO_printf(Perl_debug_log, " ");
4120 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4121 prev_was_printable = TRUE;
4125 PerlIO_printf(Perl_debug_log, " ");
4127 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4128 prev_was_printable = FALSE;
4130 t += (is_utf8) ? UTF8SKIP(t) : 1;
4135 # endif /* #ifdef DEBUGGING */
4136 #endif /* USE_LOCALE_COLLATE */
4141 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4143 /* Changes the locale for LC_'switch_category" to that of
4144 * LC_'template_category', if they aren't already the same. If not NULL,
4145 * 'template_locale' is the locale that 'template_category' is in.
4147 * Returns a copy of the name of the original locale for 'switch_category'
4148 * so can be switched back to with the companion function
4149 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4151 char * restore_to_locale = NULL;
4153 if (switch_category == template_category) { /* No changes needed */
4157 /* Find the original locale of the category we may need to change, so that
4158 * it can be restored to later */
4159 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4161 if (! restore_to_locale) {
4163 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4164 __FILE__, __LINE__, category_name(switch_category), errno);
4167 /* If the locale of the template category wasn't passed in, find it now */
4168 if (template_locale == NULL) {
4169 template_locale = do_setlocale_r(template_category, NULL);
4170 if (! template_locale) {
4172 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4173 __FILE__, __LINE__, category_name(template_category), errno);
4177 /* It the locales are the same, there's nothing to do */
4178 if (strEQ(restore_to_locale, template_locale)) {
4179 Safefree(restore_to_locale);
4181 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4182 category_name(switch_category), restore_to_locale));
4187 /* Finally, change the locale to the template one */
4188 if (! do_setlocale_r(switch_category, template_locale)) {
4190 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4191 __FILE__, __LINE__, category_name(switch_category),
4192 template_locale, errno);
4195 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4196 category_name(switch_category), template_locale));
4198 return restore_to_locale;
4202 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4204 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4205 * copy that will be freed by this function), or do nothing if the latter
4206 * parameter is NULL */
4208 if (original_locale == NULL) {
4212 if (! do_setlocale_r(category, original_locale)) {
4214 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4216 category_name(category), original_locale, errno);
4219 Safefree(original_locale);
4223 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4225 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4226 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4227 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4228 * could give the wrong result. The result will very likely be correct for
4229 * languages that have commonly used non-ASCII characters, but for notably
4230 * English, it comes down to if the locale's name ends in something like
4231 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4233 * If the platform is early C89, not containing mbtowc(), or we are
4234 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4235 * These work very well for non-Latin locales or those whose currency
4236 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4237 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4238 * on the name containing UTF-8 or not. */
4240 /* Name of current locale corresponding to the input category */
4241 const char *save_input_locale = NULL;
4243 bool is_utf8 = FALSE; /* The return value */
4245 /* The variables below are for the cache of previous lookups using this
4246 * function. The cache is a C string, described at the definition for
4247 * 'C_and_POSIX_utf8ness'.
4249 * The first part of the cache is fixed, for the C and POSIX locales. The
4250 * varying part starts just after them. */
4251 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4253 Size_t utf8ness_cache_size; /* Size of the varying portion */
4254 Size_t input_name_len; /* Length in bytes of save_input_locale */
4255 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4256 the name in the cache */
4257 char * delimited; /* The name plus the delimiters used to store
4259 char * name_pos; /* position of 'delimited' in the cache, or 0
4265 assert(category != LC_ALL);
4269 /* Get the desired category's locale */
4270 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4271 if (! save_input_locale) {
4273 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4274 __FILE__, __LINE__, category_name(category), errno);
4277 DEBUG_L(PerlIO_printf(Perl_debug_log,
4278 "Current locale for %s is %s\n",
4279 category_name(category), save_input_locale));
4281 input_name_len = strlen(save_input_locale);
4283 /* In our cache, each name is accompanied by two delimiters and a single
4285 input_name_len_with_overhead = input_name_len + 3;
4287 /* Allocate and populate space for a copy of the name surrounded by the
4289 Newx(delimited, input_name_len_with_overhead, char);
4290 delimited[0] = UTF8NESS_SEP[0];
4291 Copy(save_input_locale, delimited + 1, input_name_len, char);
4292 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4293 delimited[input_name_len+2] = '\0';
4295 /* And see if that is in the cache */
4296 name_pos = instr(PL_locale_utf8ness, delimited);
4298 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4302 if (DEBUG_Lv_TEST || debug_initialization) {
4303 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4304 save_input_locale, is_utf8);
4309 /* And, if not already in that position, move it to the beginning of
4310 * the non-constant portion of the list, since it is the most recently
4311 * used. (We don't have to worry about overflow, since just moving
4312 * existing names around) */
4313 if (name_pos > utf8ness_cache) {
4314 Move(utf8ness_cache,
4315 utf8ness_cache + input_name_len_with_overhead,
4316 name_pos - utf8ness_cache, char);
4319 input_name_len_with_overhead - 1, char);
4320 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';