3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * A Elbereth Gilthoniel,
13 * silivren penna míriel
14 * o menel aglar elenath!
15 * Na-chaered palan-díriel
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
47 #define PERL_IN_LOCALE_C
48 #include "perl_langinfo.h"
57 /* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
60 #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61 # define debug_initialization 0
62 # define DEBUG_INITIALIZATION_set(v)
64 static bool debug_initialization = FALSE;
65 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
69 /* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
72 #define GET_ERRNO saved_errno
74 /* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
77 #define STRLENs(s) (sizeof("" s "") - 1)
79 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89 #define isNAME_C_OR_POSIX(name) \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
96 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
99 #define UTF8NESS_SEP "\v"
100 #define UTF8NESS_PREFIX "\f"
102 /* So, the string looks like:
104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
109 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
112 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
115 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
120 S_stdize_locale(pTHX_ char *locs)
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
136 const char * const s = strchr(locs, '=');
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
142 const char * const t = strchr(s, '.');
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
161 /* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
165 const int categories[] = {
167 # ifdef USE_LOCALE_NUMERIC
170 # ifdef USE_LOCALE_CTYPE
173 # ifdef USE_LOCALE_COLLATE
176 # ifdef USE_LOCALE_TIME
179 # ifdef USE_LOCALE_MESSAGES
182 # ifdef USE_LOCALE_MONETARY
185 # ifdef USE_LOCALE_ADDRESS
188 # ifdef USE_LOCALE_IDENTIFICATION
191 # ifdef USE_LOCALE_MEASUREMENT
194 # ifdef USE_LOCALE_PAPER
197 # ifdef USE_LOCALE_TELEPHONE
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
208 /* The top-most real element is LC_ALL */
210 const char * category_names[] = {
212 # ifdef USE_LOCALE_NUMERIC
215 # ifdef USE_LOCALE_CTYPE
218 # ifdef USE_LOCALE_COLLATE
221 # ifdef USE_LOCALE_TIME
224 # ifdef USE_LOCALE_MESSAGES
227 # ifdef USE_LOCALE_MONETARY
230 # ifdef USE_LOCALE_ADDRESS
233 # ifdef USE_LOCALE_IDENTIFICATION
236 # ifdef USE_LOCALE_MEASUREMENT
239 # ifdef USE_LOCALE_PAPER
242 # ifdef USE_LOCALE_TELEPHONE
248 NULL /* Placeholder */
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
265 /* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
274 S_category_name(const int category)
280 if (category == LC_ALL) {
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
293 const char suffix[] = " (unknown)";
295 Size_t length = sizeof(suffix) + 1;
304 /* Calculate the number of digits */
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
317 /* Now create LC_foo_INDEX #defines for just those categories on this system */
318 # ifdef USE_LOCALE_NUMERIC
319 # define LC_NUMERIC_INDEX 0
320 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
322 # define _DUMMY_NUMERIC -1
324 # ifdef USE_LOCALE_CTYPE
325 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326 # define _DUMMY_CTYPE LC_CTYPE_INDEX
328 # define _DUMMY_CTYPE _DUMMY_NUMERIC
330 # ifdef USE_LOCALE_COLLATE
331 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332 # define _DUMMY_COLLATE LC_COLLATE_INDEX
334 # define _DUMMY_COLLATE _DUMMY_COLLATE
336 # ifdef USE_LOCALE_TIME
337 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
338 # define _DUMMY_TIME LC_TIME_INDEX
340 # define _DUMMY_TIME _DUMMY_COLLATE
342 # ifdef USE_LOCALE_MESSAGES
343 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
346 # define _DUMMY_MESSAGES _DUMMY_TIME
348 # ifdef USE_LOCALE_MONETARY
349 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350 # define _DUMMY_MONETARY LC_MONETARY_INDEX
352 # define _DUMMY_MONETARY _DUMMY_MESSAGES
354 # ifdef USE_LOCALE_ADDRESS
355 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
358 # define _DUMMY_ADDRESS _DUMMY_MONETARY
360 # ifdef USE_LOCALE_IDENTIFICATION
361 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
364 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
366 # ifdef USE_LOCALE_MEASUREMENT
367 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
370 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
372 # ifdef USE_LOCALE_PAPER
373 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374 # define _DUMMY_PAPER LC_PAPER_INDEX
376 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
378 # ifdef USE_LOCALE_TELEPHONE
379 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
382 # define _DUMMY_TELEPHONE _DUMMY_PAPER
385 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
387 #endif /* ifdef USE_LOCALE */
389 /* Windows requres a customized base-level setlocale() */
391 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
393 # define my_setlocale(cat, locale) setlocale(cat, locale)
396 #ifndef USE_POSIX_2008_LOCALE
398 /* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
403 #else /* Below uses POSIX 2008 */
405 /* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411 # define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
415 /* A third array, parallel to the ones above to map from category to its
417 const int category_masks[] = {
418 # ifdef USE_LOCALE_NUMERIC
421 # ifdef USE_LOCALE_CTYPE
424 # ifdef USE_LOCALE_COLLATE
427 # ifdef USE_LOCALE_TIME
430 # ifdef USE_LOCALE_MESSAGES
433 # ifdef USE_LOCALE_MONETARY
436 # ifdef USE_LOCALE_ADDRESS
439 # ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
442 # ifdef USE_LOCALE_MEASUREMENT
445 # ifdef USE_LOCALE_PAPER
448 # ifdef USE_LOCALE_TELEPHONE
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
460 S_emulate_setlocale(const int category,
463 const bool is_index_valid
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
554 mask = category_masks[index];
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
580 # ifdef HAS_QUERYLOCALE
582 return (char *) querylocale(mask, cur_obj);
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
589 # if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
608 # ifdef USE_C_BACKTRACE
610 dump_c_backtrace(Perl_debug_log, 20, 1);
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
626 /* Without querylocale(), we have to use our record-keeping we've
629 if (category != LC_ALL) {
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
639 return PL_curlocales[index];
641 else { /* For LC_ALL */
643 Size_t names_len = 0;
645 bool are_all_categories_the_same_locale = TRUE;
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
658 return PL_curlocales[LC_ALL_INDEX];
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
666 for (i = 0; i < LC_ALL_INDEX; i++) {
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
676 names_len += strlen(category_names[i])
678 + strlen(PL_curlocales[i])
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
728 SETERRNO(EINVAL, LIB_INVARG);
738 assert(PL_C_locale_obj);
740 /* Otherwise, we are switching locales. This will generally entail freeing
741 * the current one's space (at the C library's discretion). We need to
742 * stop using that locale before the switch. So switch to a known locale
743 * object that we don't otherwise mess with. This returns the locale
744 * object in effect at the time of the switch. */
745 old_obj = uselocale(PL_C_locale_obj);
749 if (DEBUG_Lv_TEST || debug_initialization) {
750 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
759 if (DEBUG_L_TEST || debug_initialization) {
761 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
772 if (DEBUG_Lv_TEST || debug_initialization) {
773 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
778 /* If we weren't in a thread safe locale, set so that newlocale() below
779 which uses 'old_obj', uses an empty one. Same for our reserved C object.
780 The latter is defensive coding, so that, even if there is some bug, we
781 will never end up trying to modify either of these, as if passed to
782 newlocale(), they can be. */
783 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
784 old_obj = (locale_t) 0;
787 /* Create the new locale (it may actually modify the current one). */
789 # ifndef HAS_QUERYLOCALE
791 if (strEQ(locale, "")) {
793 /* For non-querylocale() systems, we do the setting of "" ourselves to
794 * be sure that we really know what's going on. We follow the Linux
795 * documented behavior (but if that differs from the actual behavior,
796 * this won't work exactly as the OS implements). We go out and
797 * examine the environment based on our understanding of how the system
798 * works, and use that to figure things out */
800 const char * const lc_all = PerlEnv_getenv("LC_ALL");
802 /* Use any "LC_ALL" environment variable, as it overrides everything
804 if (lc_all && strNE(lc_all, "")) {
809 /* Otherwise, we need to dig deeper. Unless overridden, the
810 * default is the LANG environment variable; if it doesn't exist,
813 const char * default_name;
815 /* To minimize other threads messing with the environment, we copy
816 * the variable, making it a temporary. But this doesn't work upon
817 * program initialization before any scopes are created, and at
818 * this time, there's nothing else going on that would interfere.
819 * So skip the copy in that case */
820 if (PL_scopestack_ix == 0) {
821 default_name = PerlEnv_getenv("LANG");
824 default_name = savepv(PerlEnv_getenv("LANG"));
827 if (! default_name || strEQ(default_name, "")) {
830 else if (PL_scopestack_ix != 0) {
831 SAVEFREEPV(default_name);
834 if (category != LC_ALL) {
835 const char * const name = PerlEnv_getenv(category_names[index]);
837 /* Here we are setting a single category. Assume will have the
839 locale = default_name;
841 /* But then look for an overriding environment variable */
842 if (name && strNE(name, "")) {
847 bool did_override = FALSE;
850 /* Here, we are getting LC_ALL. Any categories that don't have
851 * a corresponding environment variable set should be set to
852 * LANG, or to "C" if there is no LANG. If no individual
853 * categories differ from this, we can just set LC_ALL. This
854 * is buggy on systems that have extra categories that we don't
855 * know about. If there is an environment variable that sets
856 * that category, we won't know to look for it, and so our use
857 * of LANG or "C" improperly overrides it. On the other hand,
858 * if we don't do what is done here, and there is no
859 * environment variable, the category's locale should be set to
860 * LANG or "C". So there is no good solution. khw thinks the
861 * best is to look at systems to see what categories they have,
862 * and include them, and then to assume that we know the
865 for (i = 0; i < LC_ALL_INDEX; i++) {
866 const char * const env_override
867 = savepv(PerlEnv_getenv(category_names[i]));
868 const char * this_locale = ( env_override
869 && strNE(env_override, ""))
872 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
874 Safefree(env_override);
878 if (strNE(this_locale, default_name)) {
882 Safefree(env_override);
885 /* If all the categories are the same, we can set LC_ALL to
887 if (! did_override) {
888 locale = default_name;
892 /* Here, LC_ALL is no longer valid, as some individual
893 * categories don't match it. We call ourselves
894 * recursively, as that will execute the code that
895 * generates the proper locale string for this situation.
896 * We don't do the remainder of this function, as that is
897 * to update our records, and we've just done that for the
898 * individual categories in the loop above, and doing so
899 * would cause LC_ALL to be done as well */
900 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
905 else if (strchr(locale, ';')) {
907 /* LC_ALL may actually incude a conglomeration of various categories.
908 * Without querylocale, this code uses the glibc (as of this writing)
909 * syntax for representing that, but that is not a stable API, and
910 * other platforms do it differently, so we have to handle all cases
913 const char * s = locale;
914 const char * e = locale + strlen(locale);
916 const char * category_end;
917 const char * name_start;
918 const char * name_end;
923 /* Parse through the category */
924 while (isWORDCHAR(*p)) {
931 "panic: %s: %d: Unexpected character in locale name '%02X",
932 __FILE__, __LINE__, *(p-1));
935 /* Parse through the locale name */
937 while (isGRAPH(*p) && *p != ';') {
944 "panic: %s: %d: Unexpected character in locale name '%02X",
945 __FILE__, __LINE__, *(p-1));
948 /* Find the index of the category name in our lists */
949 for (i = 0; i < LC_ALL_INDEX; i++) {
950 char * individ_locale;
952 /* Keep going if this isn't the index. The strnNE() avoids a
953 * Perl_form(), but would fail if ever a category name could be
954 * a substring of another one, like if there were a
956 if strnNE(s, category_names[i], category_end - s) {
960 /* If this index is for the single category we're changing, we
961 * have found the locale to set it to. */
962 if (category == categories[i]) {
963 locale = Perl_form(aTHX_ "%.*s",
964 (int) (name_end - name_start),
969 assert(category == LC_ALL);
970 individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
971 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
980 /* Here we have set all the individual categories by recursive calls.
981 * These collectively should have fixed up LC_ALL, so can just query
982 * what that now is */
983 assert(category == LC_ALL);
985 return do_setlocale_c(LC_ALL, NULL);
990 # endif /* end of ! querylocale */
992 /* Ready to create a new locale by modification of the exising one */
993 new_obj = newlocale(mask, locale, old_obj);
1000 if (DEBUG_L_TEST || debug_initialization) {
1001 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1006 if (! uselocale(old_obj)) {
1010 if (DEBUG_L_TEST || debug_initialization) {
1011 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1023 if (DEBUG_Lv_TEST || debug_initialization) {
1024 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1029 /* And switch into it */
1030 if (! uselocale(new_obj)) {
1035 if (DEBUG_L_TEST || debug_initialization) {
1036 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1041 if (! uselocale(old_obj)) {
1045 if (DEBUG_L_TEST || debug_initialization) {
1046 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1052 freelocale(new_obj);
1059 if (DEBUG_Lv_TEST || debug_initialization) {
1060 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1065 /* We are done, except for updating our records (if the system doesn't keep
1066 * them) and in the case of locale "", we don't actually know what the
1067 * locale that got switched to is, as it came from the environment. So
1068 * have to find it */
1070 # ifdef HAS_QUERYLOCALE
1072 if (strEQ(locale, "")) {
1073 locale = querylocale(mask, new_obj);
1078 /* Here, 'locale' is the return value */
1080 /* Without querylocale(), we have to update our records */
1082 if (category == LC_ALL) {
1085 /* For LC_ALL, we change all individual categories to correspond */
1086 /* PL_curlocales is a parallel array, so has same
1087 * length as 'categories' */
1088 for (i = 0; i <= LC_ALL_INDEX; i++) {
1089 Safefree(PL_curlocales[i]);
1090 PL_curlocales[i] = savepv(locale);
1095 /* For a single category, if it's not the same as the one in LC_ALL, we
1098 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1099 Safefree(PL_curlocales[LC_ALL_INDEX]);
1100 PL_curlocales[LC_ALL_INDEX] = NULL;
1103 /* Then update the category's record */
1104 Safefree(PL_curlocales[index]);
1105 PL_curlocales[index] = savepv(locale);
1113 #endif /* USE_POSIX_2008_LOCALE */
1115 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1116 didn't natively support them */
1118 /* The way this would work is that we would keep a per-thread list of the
1119 * correct locale for that thread. Any operation that was locale-sensitive
1120 * would have to be changed so that it would look like this:
1123 * setlocale to the correct locale for this operation
1127 * This leaves the global locale in the most recently used operation's, but it
1128 * was locked long enough to get the result. If that result is static, it
1129 * needs to be copied before the unlock.
1131 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1132 * the setup, but are no-ops when not needed, and similarly,
1133 * END_LOCALE_DEPENDENT_OP for the tear-down
1135 * But every call to a locale-sensitive function would have to be changed, and
1136 * if a module didn't cooperate by using the mutex, things would break.
1138 * This code was abandoned before being completed or tested, and is left as-is
1141 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1142 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1145 S_locking_setlocale(pTHX_
1147 const char * locale,
1149 const bool is_index_valid
1152 /* This function kind of performs a setlocale() on just the current thread;
1153 * thus it is kind of thread-safe. It does this by keeping a thread-level
1154 * array of the current locales for each category. Every time a locale is
1155 * switched to, it does the switch globally, but updates the thread's
1156 * array. A query as to what the current locale is just returns the
1157 * appropriate element from the array, and doesn't actually call the system
1158 * setlocale(). The saving into the array is done in an uninterruptible
1159 * section of code, so is unaffected by whatever any other threads might be
1162 * All locale-sensitive operations must work by first starting a critical
1163 * section, then switching to the thread's locale as kept by this function,
1164 * and then doing the operation, then ending the critical section. Thus,
1165 * each gets done in the appropriate locale. simulating thread-safety.
1167 * This function takes the same parameters, 'category' and 'locale', that
1168 * the regular setlocale() function does, but it also takes two additional
1169 * ones. This is because as described earlier. If we know on input the
1170 * index corresponding to the category into the array where we store the
1171 * current locales, we don't have to calculate it. If the caller knows at
1172 * compile time what the index is, it it can pass it, setting
1173 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1177 /* If the input index might be incorrect, calculate the correct one */
1178 if (! is_index_valid) {
1181 if (DEBUG_Lv_TEST || debug_initialization) {
1182 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1185 for (i = 0; i <= LC_ALL_INDEX; i++) {
1186 if (category == categories[i]) {
1192 /* Here, we don't know about this category, so can't handle it.
1193 * XXX best we can do is to unsafely set this
1196 return my_setlocale(category, locale);
1200 if (DEBUG_Lv_TEST || debug_initialization) {
1201 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1205 /* For a query, just return what's in our records */
1206 if (new_locale == NULL) {
1207 return curlocales[index];
1211 /* Otherwise, we need to do the switch, and save the result, all in a
1212 * critical section */
1214 Safefree(curlocales[[index]]);
1216 /* It might be that this is called from an already-locked section of code.
1217 * We would have to detect and skip the LOCK/UNLOCK if so */
1220 curlocales[index] = savepv(my_setlocale(category, new_locale));
1222 if (strEQ(new_locale, "")) {
1226 /* The locale values come from the environment, and may not all be the
1227 * same, so for LC_ALL, we have to update all the others, while the
1228 * mutex is still locked */
1230 if (category == LC_ALL) {
1232 for (i = 0; i < LC_ALL_INDEX) {
1233 curlocales[i] = my_setlocale(categories[i], NULL);
1242 return curlocales[index];
1248 S_set_numeric_radix(pTHX_ const bool use_locale)
1250 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1251 * TRUE, use the radix character derived from the current locale */
1253 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1254 || defined(HAS_NL_LANGINFO))
1256 const char * radix = (use_locale)
1257 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1258 /* FALSE => already in dest locale */
1261 sv_setpv(PL_numeric_radix_sv, radix);
1263 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1264 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1265 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1266 SvCUR(PL_numeric_radix_sv))
1267 && _is_cur_LC_category_utf8(LC_NUMERIC))
1269 SvUTF8_on(PL_numeric_radix_sv);
1274 if (DEBUG_L_TEST || debug_initialization) {
1275 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1276 SvPVX(PL_numeric_radix_sv),
1277 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1281 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1286 S_new_numeric(pTHX_ const char *newnum)
1289 #ifndef USE_LOCALE_NUMERIC
1291 PERL_UNUSED_ARG(newnum);
1295 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1296 * core Perl this and that 'newnum' is the name of the new locale.
1297 * It installs this locale as the current underlying default.
1299 * The default locale and the C locale can be toggled between by use of the
1300 * set_numeric_underlying() and set_numeric_standard() functions, which
1301 * should probably not be called directly, but only via macros like
1302 * SET_NUMERIC_STANDARD() in perl.h.
1304 * The toggling is necessary mainly so that a non-dot radix decimal point
1305 * character can be output, while allowing internal calculations to use a
1308 * This sets several interpreter-level variables:
1309 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1310 * PL_numeric_underlying A boolean indicating if the toggled state is such
1311 * that the current locale is the program's underlying
1313 * PL_numeric_standard An int indicating if the toggled state is such
1314 * that the current locale is the C locale or
1315 * indistinguishable from the C locale. If non-zero, it
1316 * is in C; if > 1, it means it may not be toggled away
1318 * PL_numeric_underlying_is_standard A bool kept by this function
1319 * indicating that the underlying locale and the standard
1320 * C locale are indistinguishable for the purposes of
1321 * LC_NUMERIC. This happens when both of the above two
1322 * variables are true at the same time. (Toggling is a
1323 * no-op under these circumstances.) This variable is
1324 * used to avoid having to recalculate.
1330 Safefree(PL_numeric_name);
1331 PL_numeric_name = NULL;
1332 PL_numeric_standard = TRUE;
1333 PL_numeric_underlying = TRUE;
1334 PL_numeric_underlying_is_standard = TRUE;
1338 save_newnum = stdize_locale(savepv(newnum));
1339 PL_numeric_underlying = TRUE;
1340 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1342 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1344 if (! PL_numeric_standard) {
1345 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1346 FALSE /* Don't toggle locale */ ))
1347 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1351 /* Save the new name if it isn't the same as the previous one, if any */
1352 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1353 Safefree(PL_numeric_name);
1354 PL_numeric_name = save_newnum;
1357 Safefree(save_newnum);
1360 PL_numeric_underlying_is_standard = PL_numeric_standard;
1362 # ifdef HAS_POSIX_2008_LOCALE
1364 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1366 PL_underlying_numeric_obj);
1370 if (DEBUG_L_TEST || debug_initialization) {
1371 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1374 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1375 * have to worry about the radix being a non-dot. (Core operations that
1376 * need the underlying locale change to it temporarily). */
1377 if (PL_numeric_standard) {
1378 set_numeric_radix(0);
1381 set_numeric_standard();
1384 #endif /* USE_LOCALE_NUMERIC */
1389 Perl_set_numeric_standard(pTHX)
1392 #ifdef USE_LOCALE_NUMERIC
1394 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1395 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1396 * macro avoids calling this routine if toggling isn't necessary according
1397 * to our records (which could be wrong if some XS code has changed the
1398 * locale behind our back) */
1400 do_setlocale_c(LC_NUMERIC, "C");
1401 PL_numeric_standard = TRUE;
1402 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1403 set_numeric_radix(0);
1407 if (DEBUG_L_TEST || debug_initialization) {
1408 PerlIO_printf(Perl_debug_log,
1409 "LC_NUMERIC locale now is standard C\n");
1413 #endif /* USE_LOCALE_NUMERIC */
1418 Perl_set_numeric_underlying(pTHX)
1421 #ifdef USE_LOCALE_NUMERIC
1423 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1424 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1425 * instead of calling this directly. The macro avoids calling this routine
1426 * if toggling isn't necessary according to our records (which could be
1427 * wrong if some XS code has changed the locale behind our back) */
1429 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1430 PL_numeric_standard = PL_numeric_underlying_is_standard;
1431 PL_numeric_underlying = TRUE;
1432 set_numeric_radix(! PL_numeric_standard);
1436 if (DEBUG_L_TEST || debug_initialization) {
1437 PerlIO_printf(Perl_debug_log,
1438 "LC_NUMERIC locale now is %s\n",
1443 #endif /* USE_LOCALE_NUMERIC */
1448 * Set up for a new ctype locale.
1451 S_new_ctype(pTHX_ const char *newctype)
1454 #ifndef USE_LOCALE_CTYPE
1456 PERL_ARGS_ASSERT_NEW_CTYPE;
1457 PERL_UNUSED_ARG(newctype);
1458 PERL_UNUSED_CONTEXT;
1462 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1463 * core Perl this and that 'newctype' is the name of the new locale.
1465 * This function sets up the folding arrays for all 256 bytes, assuming
1466 * that tofold() is tolc() since fold case is not a concept in POSIX,
1468 * Any code changing the locale (outside this file) should use
1469 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1470 * this function should be called directly only from this file and from
1471 * POSIX::setlocale() */
1476 /* Don't check for problems if we are suppressing the warnings */
1477 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1479 PERL_ARGS_ASSERT_NEW_CTYPE;
1481 /* We will replace any bad locale warning with 1) nothing if the new one is
1482 * ok; or 2) a new warning for the bad new locale */
1483 if (PL_warn_locale) {
1484 SvREFCNT_dec_NN(PL_warn_locale);
1485 PL_warn_locale = NULL;
1488 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1490 /* A UTF-8 locale gets standard rules. But note that code still has to
1491 * handle this specially because of the three problematic code points */
1492 if (PL_in_utf8_CTYPE_locale) {
1493 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1496 /* We don't populate the other lists if a UTF-8 locale, but do check that
1497 * everything works as expected, unless checking turned off */
1498 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1499 /* Assume enough space for every character being bad. 4 spaces each
1500 * for the 94 printable characters that are output like "'x' "; and 5
1501 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1503 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1504 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1506 unsigned int bad_count = 0; /* Count of bad characters */
1508 for (i = 0; i < 256; i++) {
1509 if (! PL_in_utf8_CTYPE_locale) {
1511 PL_fold_locale[i] = (U8) tolower(i);
1512 else if (islower(i))
1513 PL_fold_locale[i] = (U8) toupper(i);
1515 PL_fold_locale[i] = (U8) i;
1518 /* If checking for locale problems, see if the native ASCII-range
1519 * printables plus \n and \t are in their expected categories in
1520 * the new locale. If not, this could mean big trouble, upending
1521 * Perl's and most programs' assumptions, like having a
1522 * metacharacter with special meaning become a \w. Fortunately,
1523 * it's very rare to find locales that aren't supersets of ASCII
1524 * nowadays. It isn't a problem for most controls to be changed
1525 * into something else; we check only \n and \t, though perhaps \r
1526 * could be an issue as well. */
1527 if ( check_for_problems
1528 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1530 bool is_bad = FALSE;
1531 char name[3] = { '\0' };
1533 /* Convert the name into a string */
1538 else if (i == '\n') {
1539 my_strlcpy(name, "\n", sizeof(name));
1542 my_strlcpy(name, "\t", sizeof(name));
1545 /* Check each possibe class */
1546 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1548 DEBUG_L(PerlIO_printf(Perl_debug_log,
1549 "isalnum('%s') unexpectedly is %d\n",
1550 name, cBOOL(isalnum(i))));
1552 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1554 DEBUG_L(PerlIO_printf(Perl_debug_log,
1555 "isalpha('%s') unexpectedly is %d\n",
1556 name, cBOOL(isalpha(i))));
1558 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1560 DEBUG_L(PerlIO_printf(Perl_debug_log,
1561 "isdigit('%s') unexpectedly is %d\n",
1562 name, cBOOL(isdigit(i))));
1564 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1566 DEBUG_L(PerlIO_printf(Perl_debug_log,
1567 "isgraph('%s') unexpectedly is %d\n",
1568 name, cBOOL(isgraph(i))));
1570 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1572 DEBUG_L(PerlIO_printf(Perl_debug_log,
1573 "islower('%s') unexpectedly is %d\n",
1574 name, cBOOL(islower(i))));
1576 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1578 DEBUG_L(PerlIO_printf(Perl_debug_log,
1579 "isprint('%s') unexpectedly is %d\n",
1580 name, cBOOL(isprint(i))));
1582 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "ispunct('%s') unexpectedly is %d\n",
1586 name, cBOOL(ispunct(i))));
1588 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isspace('%s') unexpectedly is %d\n",
1592 name, cBOOL(isspace(i))));
1594 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isupper('%s') unexpectedly is %d\n",
1598 name, cBOOL(isupper(i))));
1600 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "isxdigit('%s') unexpectedly is %d\n",
1604 name, cBOOL(isxdigit(i))));
1606 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1610 name, tolower(i), (int) toLOWER_A(i)));
1612 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1616 name, toupper(i), (int) toUPPER_A(i)));
1618 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "'\\n' (=%02X) is not a control\n", (int) i));
1624 /* Add to the list; Separate multiple entries with a blank */
1627 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1629 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1637 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1638 * this locale requires more than one byte, there are going to be
1640 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1641 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1642 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1644 if ( check_for_problems && MB_CUR_MAX > 1
1645 && ! PL_in_utf8_CTYPE_locale
1647 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1648 * locale. Just assume that the implementation for them (plus
1649 * for POSIX) is correct and the > 1 value is spurious. (Since
1650 * these are specially handled to never be considered UTF-8
1651 * locales, as long as this is the only problem, everything
1652 * should work fine */
1653 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1655 multi_byte_locale = TRUE;
1660 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1661 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1662 PL_warn_locale = Perl_newSVpvf(aTHX_
1663 "Locale '%s' contains (at least) the following characters"
1664 " which have\nunexpected meanings: %s\nThe Perl program"
1665 " will use the expected meanings",
1666 newctype, bad_chars_list);
1669 PL_warn_locale = Perl_newSVpvf(aTHX_
1670 "Locale '%s' may not work well.%s%s%s\n",
1673 ? " Some characters in it are not recognized by"
1677 ? "\nThe following characters (and maybe others)"
1678 " may not have the same meaning as the Perl"
1679 " program expects:\n"
1687 # ifdef HAS_NL_LANGINFO
1689 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1690 /* parameter FALSE is a don't care here */
1691 my_nl_langinfo(PERL_CODESET, FALSE));
1695 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1697 /* If we are actually in the scope of the locale or are debugging,
1698 * output the message now. If not in that scope, we save the
1699 * message to be output at the first operation using this locale,
1700 * if that actually happens. Most programs don't use locales, so
1701 * they are immune to bad ones. */
1702 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1704 /* The '0' below suppresses a bogus gcc compiler warning */
1705 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1707 if (IN_LC(LC_CTYPE)) {
1708 SvREFCNT_dec_NN(PL_warn_locale);
1709 PL_warn_locale = NULL;
1715 #endif /* USE_LOCALE_CTYPE */
1720 Perl__warn_problematic_locale()
1723 #ifdef USE_LOCALE_CTYPE
1727 /* Internal-to-core function that outputs the message in PL_warn_locale,
1728 * and then NULLS it. Should be called only through the macro
1729 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1731 if (PL_warn_locale) {
1732 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1733 SvPVX(PL_warn_locale),
1734 0 /* dummy to avoid compiler warning */ );
1735 SvREFCNT_dec_NN(PL_warn_locale);
1736 PL_warn_locale = NULL;
1744 S_new_collate(pTHX_ const char *newcoll)
1747 #ifndef USE_LOCALE_COLLATE
1749 PERL_UNUSED_ARG(newcoll);
1750 PERL_UNUSED_CONTEXT;
1754 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1755 * core Perl this and that 'newcoll' is the name of the new locale.
1757 * The design of locale collation is that every locale change is given an
1758 * index 'PL_collation_ix'. The first time a string particpates in an
1759 * operation that requires collation while locale collation is active, it
1760 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1761 * magic includes the collation index, and the transformation of the string
1762 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1763 * instead of the string itself. If a string changes, the magic is
1764 * cleared. The next time the locale changes, the index is incremented,
1765 * and so we know during a comparison that the transformation is not
1766 * necessarily still valid, and so is recomputed. Note that if the locale
1767 * changes enough times, the index could wrap (a U32), and it is possible
1768 * that a transformation would improperly be considered valid, leading to
1769 * an unlikely bug */
1772 if (PL_collation_name) {
1774 Safefree(PL_collation_name);
1775 PL_collation_name = NULL;
1777 PL_collation_standard = TRUE;
1778 is_standard_collation:
1779 PL_collxfrm_base = 0;
1780 PL_collxfrm_mult = 2;
1781 PL_in_utf8_COLLATE_locale = FALSE;
1782 PL_strxfrm_NUL_replacement = '\0';
1783 PL_strxfrm_max_cp = 0;
1787 /* If this is not the same locale as currently, set the new one up */
1788 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1790 Safefree(PL_collation_name);
1791 PL_collation_name = stdize_locale(savepv(newcoll));
1792 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1793 if (PL_collation_standard) {
1794 goto is_standard_collation;
1797 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1798 PL_strxfrm_NUL_replacement = '\0';
1799 PL_strxfrm_max_cp = 0;
1801 /* A locale collation definition includes primary, secondary, tertiary,
1802 * etc. weights for each character. To sort, the primary weights are
1803 * used, and only if they compare equal, then the secondary weights are
1804 * used, and only if they compare equal, then the tertiary, etc.
1806 * strxfrm() works by taking the input string, say ABC, and creating an
1807 * output transformed string consisting of first the primary weights,
1808 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1809 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1810 * may not have weights at every level. In our example, let's say B
1811 * doesn't have a tertiary weight, and A doesn't have a secondary
1812 * weight. The constructed string is then going to be
1813 * A¹B¹C¹ B²C² A³C³ ....
1814 * This has the desired effect that strcmp() will look at the secondary
1815 * or tertiary weights only if the strings compare equal at all higher
1816 * priority weights. The spaces shown here, like in
1818 * are not just for readability. In the general case, these must
1819 * actually be bytes, which we will call here 'separator weights'; and
1820 * they must be smaller than any other weight value, but since these
1821 * are C strings, only the terminating one can be a NUL (some
1822 * implementations may include a non-NUL separator weight just before
1823 * the NUL). Implementations tend to reserve 01 for the separator
1824 * weights. They are needed so that a shorter string's secondary
1825 * weights won't be misconstrued as primary weights of a longer string,
1826 * etc. By making them smaller than any other weight, the shorter
1827 * string will sort first. (Actually, if all secondary weights are
1828 * smaller than all primary ones, there is no need for a separator
1829 * weight between those two levels, etc.)
1831 * The length of the transformed string is roughly a linear function of
1832 * the input string. It's not exactly linear because some characters
1833 * don't have weights at all levels. When we call strxfrm() we have to
1834 * allocate some memory to hold the transformed string. The
1835 * calculations below try to find coefficients 'm' and 'b' for this
1836 * locale so that m*x + b equals how much space we need, given the size
1837 * of the input string in 'x'. If we calculate too small, we increase
1838 * the size as needed, and call strxfrm() again, but it is better to
1839 * get it right the first time to avoid wasted expensive string
1840 * transformations. */
1843 /* We use the string below to find how long the tranformation of it
1844 * is. Almost all locales are supersets of ASCII, or at least the
1845 * ASCII letters. We use all of them, half upper half lower,
1846 * because if we used fewer, we might hit just the ones that are
1847 * outliers in a particular locale. Most of the strings being
1848 * collated will contain a preponderance of letters, and even if
1849 * they are above-ASCII, they are likely to have the same number of
1850 * weight levels as the ASCII ones. It turns out that digits tend
1851 * to have fewer levels, and some punctuation has more, but those
1852 * are relatively sparse in text, and khw believes this gives a
1853 * reasonable result, but it could be changed if experience so
1855 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1856 char * x_longer; /* Transformed 'longer' */
1857 Size_t x_len_longer; /* Length of 'x_longer' */
1859 char * x_shorter; /* We also transform a substring of 'longer' */
1860 Size_t x_len_shorter;
1862 /* _mem_collxfrm() is used get the transformation (though here we
1863 * are interested only in its length). It is used because it has
1864 * the intelligence to handle all cases, but to work, it needs some
1865 * values of 'm' and 'b' to get it started. For the purposes of
1866 * this calculation we use a very conservative estimate of 'm' and
1867 * 'b'. This assumes a weight can be multiple bytes, enough to
1868 * hold any UV on the platform, and there are 5 levels, 4 weight
1869 * bytes, and a trailing NUL. */
1870 PL_collxfrm_base = 5;
1871 PL_collxfrm_mult = 5 * sizeof(UV);
1873 /* Find out how long the transformation really is */
1874 x_longer = _mem_collxfrm(longer,
1878 /* We avoid converting to UTF-8 in the
1879 * called function by telling it the
1880 * string is in UTF-8 if the locale is a
1881 * UTF-8 one. Since the string passed
1882 * here is invariant under UTF-8, we can
1883 * claim it's UTF-8 even though it isn't.
1885 PL_in_utf8_COLLATE_locale);
1888 /* Find out how long the transformation of a substring of 'longer'
1889 * is. Together the lengths of these transformations are
1890 * sufficient to calculate 'm' and 'b'. The substring is all of
1891 * 'longer' except the first character. This minimizes the chances
1892 * of being swayed by outliers */
1893 x_shorter = _mem_collxfrm(longer + 1,
1896 PL_in_utf8_COLLATE_locale);
1897 Safefree(x_shorter);
1899 /* If the results are nonsensical for this simple test, the whole
1900 * locale definition is suspect. Mark it so that locale collation
1901 * is not active at all for it. XXX Should we warn? */
1902 if ( x_len_shorter == 0
1903 || x_len_longer == 0
1904 || x_len_shorter >= x_len_longer)
1906 PL_collxfrm_mult = 0;
1907 PL_collxfrm_base = 0;
1910 SSize_t base; /* Temporary */
1912 /* We have both: m * strlen(longer) + b = x_len_longer
1913 * m * strlen(shorter) + b = x_len_shorter;
1914 * subtracting yields:
1915 * m * (strlen(longer) - strlen(shorter))
1916 * = x_len_longer - x_len_shorter
1917 * But we have set things up so that 'shorter' is 1 byte smaller
1918 * than 'longer'. Hence:
1919 * m = x_len_longer - x_len_shorter
1921 * But if something went wrong, make sure the multiplier is at
1924 if (x_len_longer > x_len_shorter) {
1925 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1928 PL_collxfrm_mult = 1;
1933 * but in case something has gone wrong, make sure it is
1935 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1940 /* Add 1 for the trailing NUL */
1941 PL_collxfrm_base = base + 1;
1946 if (DEBUG_L_TEST || debug_initialization) {
1947 PerlIO_printf(Perl_debug_log,
1948 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1950 " collate multipler=%zu, collate base=%zu\n",
1952 PL_in_utf8_COLLATE_locale,
1953 x_len_shorter, x_len_longer,
1954 PL_collxfrm_mult, PL_collxfrm_base);
1961 #endif /* USE_LOCALE_COLLATE */
1968 S_win32_setlocale(pTHX_ int category, const char* locale)
1970 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1971 * difference between the two unless the input locale is "", which normally
1972 * means on Windows to get the machine default, which is set via the
1973 * computer's "Regional and Language Options" (or its current equivalent).
1974 * In POSIX, it instead means to find the locale from the user's
1975 * environment. This routine changes the Windows behavior to first look in
1976 * the environment, and, if anything is found, use that instead of going to
1977 * the machine default. If there is no environment override, the machine
1978 * default is used, by calling the real setlocale() with "".
1980 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1981 * use the particular category's variable if set; otherwise to use the LANG
1984 bool override_LC_ALL = FALSE;
1988 if (locale && strEQ(locale, "")) {
1992 locale = PerlEnv_getenv("LC_ALL");
1994 if (category == LC_ALL) {
1995 override_LC_ALL = TRUE;
2001 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2002 if (category == categories[i]) {
2003 locale = PerlEnv_getenv(category_names[i]);
2008 locale = PerlEnv_getenv("LANG");
2024 result = setlocale(category, locale);
2025 DEBUG_L(STMT_START {
2027 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2028 setlocale_debug_string(category, locale, result));
2032 if (! override_LC_ALL) {
2036 /* Here the input category was LC_ALL, and we have set it to what is in the
2037 * LANG variable or the system default if there is no LANG. But these have
2038 * lower priority than the other LC_foo variables, so override it for each
2039 * one that is set. (If they are set to "", it means to use the same thing
2040 * we just set LC_ALL to, so can skip) */
2042 for (i = 0; i < LC_ALL_INDEX; i++) {
2043 result = PerlEnv_getenv(category_names[i]);
2044 if (result && strNE(result, "")) {
2045 setlocale(categories[i], result);
2046 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2048 setlocale_debug_string(categories[i], result, "not captured")));
2052 result = setlocale(LC_ALL, NULL);
2053 DEBUG_L(STMT_START {
2055 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2057 setlocale_debug_string(LC_ALL, NULL, result));
2068 =head1 Locale-related functions and macros
2070 =for apidoc Perl_setlocale
2072 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2073 taking the same parameters, and returning the same information, except that it
2074 returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2075 perl keeps that locale category as C<C>, changing it briefly during the
2076 operations where the underlying one is required.
2078 Another reason it isn't completely a drop-in replacement is that it is
2079 declared to return S<C<const char *>>, whereas the system setlocale omits the
2080 C<const>. (If it were being written today, plain setlocale would be declared
2081 const, since it is illegal to change the information it returns; doing so leads
2084 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2085 C<setlocale> can be completely ineffective on some platforms under some
2088 C<Perl_setlocale> should not be used to change the locale except on systems
2089 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2090 the system C<setlocale()> is ineffective, returning the wrong information, and
2091 failing to actually change the locale. C<Perl_setlocale>, however works
2092 properly in all circumstances.
2094 The return points to a per-thread static buffer, which is overwritten the next
2095 time C<Perl_setlocale> is called from the same thread.
2102 Perl_setlocale(const int category, const char * locale)
2104 /* This wraps POSIX::setlocale() */
2106 const char * retval;
2107 const char * newlocale;
2109 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2112 #ifdef USE_LOCALE_NUMERIC
2114 /* A NULL locale means only query what the current one is. We have the
2115 * LC_NUMERIC name saved, because we are normally switched into the C
2116 * locale for it. For an LC_ALL query, switch back to get the correct
2117 * results. All other categories don't require special handling */
2118 if (locale == NULL) {
2119 if (category == LC_NUMERIC) {
2121 /* We don't have to copy this return value, as it is a per-thread
2122 * variable, and won't change until a future setlocale */
2123 return PL_numeric_name;
2128 else if (category == LC_ALL) {
2129 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2138 retval = save_to_buffer(do_setlocale_r(category, locale),
2139 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2142 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2144 if (locale == NULL && category == LC_ALL) {
2145 RESTORE_LC_NUMERIC();
2150 DEBUG_L(PerlIO_printf(Perl_debug_log,
2151 "%s:%d: %s\n", __FILE__, __LINE__,
2152 setlocale_debug_string(category, locale, retval)));
2160 /* If locale == NULL, we are just querying the state */
2161 if (locale == NULL) {
2165 /* Now that have switched locales, we have to update our records to
2170 #ifdef USE_LOCALE_CTYPE
2177 #ifdef USE_LOCALE_COLLATE
2180 new_collate(retval);
2184 #ifdef USE_LOCALE_NUMERIC
2187 new_numeric(retval);
2195 /* LC_ALL updates all the things we care about. The values may not
2196 * be the same as 'retval', as the locale "" may have set things
2199 # ifdef USE_LOCALE_CTYPE
2201 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2202 new_ctype(newlocale);
2204 # endif /* USE_LOCALE_CTYPE */
2205 # ifdef USE_LOCALE_COLLATE
2207 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2208 new_collate(newlocale);
2211 # ifdef USE_LOCALE_NUMERIC
2213 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2214 new_numeric(newlocale);
2216 # endif /* USE_LOCALE_NUMERIC */
2227 PERL_STATIC_INLINE const char *
2228 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2230 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2231 * growing it if necessary */
2235 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2241 string_size = strlen(string) + offset + 1;
2243 if (*buf_size == 0) {
2244 Newx(*buf, string_size, char);
2245 *buf_size = string_size;
2247 else if (string_size > *buf_size) {
2248 Renew(*buf, string_size, char);
2249 *buf_size = string_size;
2252 Copy(string, *buf + offset, string_size - offset, char);
2258 =for apidoc Perl_langinfo
2260 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2261 taking the same C<item> parameter values, and returning the same information.
2262 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2263 of Perl's locale handling from your code, and can be used on systems that lack
2264 a native C<nl_langinfo>.
2272 The reason it isn't quite a drop-in replacement is actually an advantage. The
2273 only difference is that it returns S<C<const char *>>, whereas plain
2274 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2275 forbidden to write into the buffer. By declaring this C<const>, the compiler
2276 enforces this restriction, so if it is violated, you know at compilation time,
2277 rather than getting segfaults at runtime.
2281 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2282 without you having to write extra code. The reason for the extra code would be
2283 because these are from the C<LC_NUMERIC> locale category, which is normally
2284 kept set to the C locale by Perl, no matter what the underlying locale is
2285 supposed to be, and so to get the expected results, you have to temporarily
2286 toggle into the underlying locale, and later toggle back. (You could use plain
2287 C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2288 then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2289 C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2290 radix (decimal point) character to be a dot.)
2294 The system function it replaces can have its static return buffer trashed,
2295 not only by a subesequent call to that function, but by a C<freelocale>,
2296 C<setlocale>, or other locale change. The returned buffer of this function is
2297 not changed until the next call to it, so the buffer is never in a trashed
2302 Its return buffer is per-thread, so it also is never overwritten by a call to
2303 this function from another thread; unlike the function it replaces.
2307 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2308 as Windows, hence makes your code more portable. Of the fifty-some possible
2309 items specified by the POSIX 2008 standard,
2310 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2311 only two are completely unimplemented (though the loss of one of these is
2312 significant). It uses various techniques to recover the other items, including
2313 calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2314 in C89, so should be always be available. Later C<strftime()> versions have
2315 additional capabilities; C<""> is returned for those not available on your
2318 It is important to note that when called with an item that is recovered by
2319 using C<localeconv>, the buffer from any previous explicit call to
2320 C<localeconv> will be overwritten. This means you must save that buffer's
2321 contents if you need to access them after a call to this function.
2323 The details for those items which may differ from what this emulation returns
2324 and what a native C<nl_langinfo()> would return are:
2332 Unimplemented, so returns C<"">.
2342 Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
2343 removed from POSIX 2008, and are retained here for backwards compatibility.
2344 Your platform's C<nl_langinfo> may not support them.
2348 Always evaluates to C<%x>, the locale's appropriate date representation.
2352 Always evaluates to C<%X>, the locale's appropriate time representation.
2356 Always evaluates to C<%c>, the locale's appropriate date and time
2361 The return may be incorrect for those rare locales where the currency symbol
2362 replaces the radix character.
2363 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2364 to work differently.
2368 Currently this gives the same results as Linux does.
2369 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2370 to work differently.
2376 =item C<ERA_D_T_FMT>
2380 These are derived by using C<strftime()>, and not all versions of that function
2381 know about them. C<""> is returned for these on such systems.
2387 When using C<Perl_langinfo> on systems that don't have a native
2388 C<nl_langinfo()>, you must
2390 #include "perl_langinfo.h"
2392 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2393 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2394 C<langinfo.h> imports into the namespace for code that doesn't need it.)
2396 You also should not use the bare C<langinfo.h> item names, but should preface
2397 them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2398 The C<PERL_I<foo>> versions will also work for this function on systems that do
2399 have a native C<nl_langinfo>.
2401 The original impetus for C<Perl_langinfo()> was so that code that needs to
2402 find out the current currency symbol, floating point radix character, or digit
2403 grouping separator can use, on all systems, the simpler and more
2404 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2405 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2406 is better to use the methods given in L<perlcall> to call
2407 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2414 #ifdef HAS_NL_LANGINFO
2415 Perl_langinfo(const nl_item item)
2417 Perl_langinfo(const int item)
2420 return my_nl_langinfo(item, TRUE);
2424 #ifdef HAS_NL_LANGINFO
2425 S_my_nl_langinfo(const nl_item item, bool toggle)
2427 S_my_nl_langinfo(const int item, bool toggle)
2431 const char * retval;
2433 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2434 * two items, and only if not already there */
2435 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2436 || PL_numeric_underlying))
2441 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2442 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2443 || ! defined(HAS_POSIX_2008_LOCALE)
2445 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2446 * for those items dependent on it. This must be copied to a buffer before
2447 * switching back, as some systems destroy the buffer when setlocale() is
2451 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2454 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2457 LOCALE_LOCK; /* Prevent interference from another thread executing
2458 this code section (the only call to nl_langinfo in
2462 /* Copy to a per-thread buffer, which is also one that won't be
2463 * destroyed by a subsequent setlocale(), such as the
2464 * RESTORE_LC_NUMERIC may do just below. */
2465 retval = save_to_buffer(nl_langinfo(item),
2466 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2471 RESTORE_LC_NUMERIC();
2475 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2478 bool do_free = FALSE;
2479 locale_t cur = uselocale((locale_t) 0);
2481 if (cur == LC_GLOBAL_LOCALE) {
2482 cur = duplocale(LC_GLOBAL_LOCALE);
2487 if (PL_underlying_numeric_obj) {
2488 cur = PL_underlying_numeric_obj;
2491 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2496 /* We have to save it to a buffer, because the freelocale() just below
2497 * can invalidate the internal one */
2498 retval = save_to_buffer(nl_langinfo_l(item, cur),
2499 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2508 if (strEQ(retval, "")) {
2509 if (item == PERL_YESSTR) {
2512 if (item == PERL_NOSTR) {
2519 #else /* Below, emulate nl_langinfo as best we can */
2523 # ifdef HAS_LOCALECONV
2525 const struct lconv* lc;
2527 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2530 # ifdef HAS_STRFTIME
2533 bool return_format = FALSE; /* Return the %format, not the value */
2534 const char * format;
2538 /* We copy the results to a per-thread buffer, even if not
2539 * multi-threaded. This is in part to simplify this code, and partly
2540 * because we need a buffer anyway for strftime(), and partly because a
2541 * call of localeconv() could otherwise wipe out the buffer, and the
2542 * programmer would not be expecting this, as this is a nl_langinfo()
2543 * substitute after all, so s/he might be thinking their localeconv()
2544 * is safe until another localeconv() call. */
2549 /* These 2 are unimplemented */
2551 case PERL_ERA: /* For use with strftime() %E modifier */
2556 /* We use only an English set, since we don't know any more */
2557 case PERL_YESEXPR: return "^[+1yY]";
2558 case PERL_YESSTR: return "yes";
2559 case PERL_NOEXPR: return "^[-0nN]";
2560 case PERL_NOSTR: return "no";
2562 # ifdef HAS_LOCALECONV
2566 /* We don't bother with localeconv_l() because any system that
2567 * has it is likely to also have nl_langinfo() */
2569 LOCALE_LOCK; /* Prevent interference with other threads
2570 using localeconv() */
2574 || ! lc->currency_symbol
2575 || strEQ("", lc->currency_symbol))
2581 /* Leave the first spot empty to be filled in below */
2582 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2583 &PL_langinfo_bufsize, 1);
2584 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2585 { /* khw couldn't figure out how the localedef specifications
2586 would show that the $ should replace the radix; this is
2587 just a guess as to how it might work.*/
2588 PL_langinfo_buf[0] = '.';
2590 else if (lc->p_cs_precedes) {
2591 PL_langinfo_buf[0] = '-';
2594 PL_langinfo_buf[0] = '+';
2600 case PERL_RADIXCHAR:
2604 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2607 LOCALE_LOCK; /* Prevent interference with other threads
2608 using localeconv() */
2615 temp = (item == PERL_RADIXCHAR)
2617 : lc->thousands_sep;
2623 retval = save_to_buffer(temp, &PL_langinfo_buf,
2624 &PL_langinfo_bufsize, 0);
2629 RESTORE_LC_NUMERIC();
2635 # ifdef HAS_STRFTIME
2637 /* These are defined by C89, so we assume that strftime supports
2638 * them, and so are returned unconditionally; they may not be what
2639 * the locale actually says, but should give good enough results
2640 * for someone using them as formats (as opposed to trying to parse
2641 * them to figure out what the locale says). The other format
2642 * items are actually tested to verify they work on the platform */
2643 case PERL_D_FMT: return "%x";
2644 case PERL_T_FMT: return "%X";
2645 case PERL_D_T_FMT: return "%c";
2647 /* These formats are only available in later strfmtime's */
2648 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2649 case PERL_T_FMT_AMPM:
2651 /* The rest can be gotten from most versions of strftime(). */
2652 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2653 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2655 case PERL_ALT_DIGITS:
2656 case PERL_AM_STR: case PERL_PM_STR:
2657 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2658 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2659 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2660 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2661 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2662 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2663 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2664 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2665 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2670 init_tm(&tm); /* Precaution against core dumps */
2674 tm.tm_year = 2017 - 1900;
2681 "panic: %s: %d: switch case: %d problem",
2682 __FILE__, __LINE__, item);
2683 NOT_REACHED; /* NOTREACHED */
2685 case PERL_PM_STR: tm.tm_hour = 18;
2690 case PERL_ABDAY_7: tm.tm_wday++;
2691 case PERL_ABDAY_6: tm.tm_wday++;
2692 case PERL_ABDAY_5: tm.tm_wday++;
2693 case PERL_ABDAY_4: tm.tm_wday++;
2694 case PERL_ABDAY_3: tm.tm_wday++;
2695 case PERL_ABDAY_2: tm.tm_wday++;
2700 case PERL_DAY_7: tm.tm_wday++;
2701 case PERL_DAY_6: tm.tm_wday++;
2702 case PERL_DAY_5: tm.tm_wday++;
2703 case PERL_DAY_4: tm.tm_wday++;
2704 case PERL_DAY_3: tm.tm_wday++;
2705 case PERL_DAY_2: tm.tm_wday++;
2710 case PERL_ABMON_12: tm.tm_mon++;
2711 case PERL_ABMON_11: tm.tm_mon++;
2712 case PERL_ABMON_10: tm.tm_mon++;
2713 case PERL_ABMON_9: tm.tm_mon++;
2714 case PERL_ABMON_8: tm.tm_mon++;
2715 case PERL_ABMON_7: tm.tm_mon++;
2716 case PERL_ABMON_6: tm.tm_mon++;
2717 case PERL_ABMON_5: tm.tm_mon++;
2718 case PERL_ABMON_4: tm.tm_mon++;
2719 case PERL_ABMON_3: tm.tm_mon++;
2720 case PERL_ABMON_2: tm.tm_mon++;
2725 case PERL_MON_12: tm.tm_mon++;
2726 case PERL_MON_11: tm.tm_mon++;
2727 case PERL_MON_10: tm.tm_mon++;
2728 case PERL_MON_9: tm.tm_mon++;
2729 case PERL_MON_8: tm.tm_mon++;
2730 case PERL_MON_7: tm.tm_mon++;
2731 case PERL_MON_6: tm.tm_mon++;
2732 case PERL_MON_5: tm.tm_mon++;
2733 case PERL_MON_4: tm.tm_mon++;
2734 case PERL_MON_3: tm.tm_mon++;
2735 case PERL_MON_2: tm.tm_mon++;
2740 case PERL_T_FMT_AMPM:
2742 return_format = TRUE;
2745 case PERL_ERA_D_FMT:
2747 return_format = TRUE;
2750 case PERL_ERA_T_FMT:
2752 return_format = TRUE;
2755 case PERL_ERA_D_T_FMT:
2757 return_format = TRUE;
2760 case PERL_ALT_DIGITS:
2762 format = "%Ow"; /* Find the alternate digit for 0 */
2766 /* We can't use my_strftime() because it doesn't look at
2768 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2771 /* A zero return means one of:
2772 * a) there wasn't enough space in PL_langinfo_buf
2773 * b) the format, like a plain %p, returns empty
2774 * c) it was an illegal format, though some
2775 * implementations of strftime will just return the
2776 * illegal format as a plain character sequence.
2778 * To quickly test for case 'b)', try again but precede
2779 * the format with a plain character. If that result is
2780 * still empty, the problem is either 'a)' or 'c)' */
2782 Size_t format_size = strlen(format) + 1;
2783 Size_t mod_size = format_size + 1;
2787 Newx(mod_format, mod_size, char);
2788 Newx(temp_result, PL_langinfo_bufsize, char);
2790 my_strlcpy(mod_format + 1, format, mod_size);
2791 len = strftime(temp_result,
2792 PL_langinfo_bufsize,
2794 Safefree(mod_format);
2795 Safefree(temp_result);
2797 /* If 'len' is non-zero, it means that we had a case like
2798 * %p which means the current locale doesn't use a.m. or
2799 * p.m., and that is valid */
2802 /* Here, still didn't work. If we get well beyond a
2803 * reasonable size, bail out to prevent an infinite
2806 if (PL_langinfo_bufsize > 100 * format_size) {
2807 *PL_langinfo_buf = '\0';
2810 /* Double the buffer size to retry; Add 1 in case
2811 * original was 0, so we aren't stuck at 0. */
2812 PL_langinfo_bufsize *= 2;
2813 PL_langinfo_bufsize++;
2814 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2822 /* Here, we got a result.
2824 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2825 * alternate format for wday 0. If the value is the same as
2826 * the normal 0, there isn't an alternate, so clear the buffer.
2828 if ( item == PERL_ALT_DIGITS
2829 && strEQ(PL_langinfo_buf, "0"))
2831 *PL_langinfo_buf = '\0';
2834 /* ALT_DIGITS is problematic. Experiments on it showed that
2835 * strftime() did not always work properly when going from
2836 * alt-9 to alt-10. Only a few locales have this item defined,
2837 * and in all of them on Linux that khw was able to find,
2838 * nl_langinfo() merely returned the alt-0 character, possibly
2839 * doubled. Most Unicode digits are in blocks of 10
2840 * consecutive code points, so that is sufficient information
2841 * for those scripts, as we can infer alt-1, alt-2, .... But
2842 * for a Japanese locale, a CJK ideographic 0 is returned, and
2843 * the CJK digits are not in code point order, so you can't
2844 * really infer anything. The localedef for this locale did
2845 * specify the succeeding digits, so that strftime() works
2846 * properly on them, without needing to infer anything. But
2847 * the nl_langinfo() return did not give sufficient information
2848 * for the caller to understand what's going on. So until
2849 * there is evidence that it should work differently, this
2850 * returns the alt-0 string for ALT_DIGITS.
2852 * wday was chosen because its range is all a single digit.
2853 * Things like tm_sec have two digits as the minimum: '00' */
2857 retval = PL_langinfo_buf;
2859 /* If to return the format, not the value, overwrite the buffer
2860 * with it. But some strftime()s will keep the original format
2861 * if illegal, so change those to "" */
2862 if (return_format) {
2863 if (strEQ(PL_langinfo_buf, format)) {
2864 *PL_langinfo_buf = '\0';
2867 retval = save_to_buffer(format, &PL_langinfo_buf,
2868 &PL_langinfo_bufsize, 0);
2886 * Initialize locale awareness.
2889 Perl_init_i18nl10n(pTHX_ int printwarn)
2893 * 0 if not to output warning when setup locale is bad
2894 * 1 if to output warning based on value of PERL_BADLANG
2895 * >1 if to output regardless of PERL_BADLANG
2898 * 1 = set ok or not applicable,
2899 * 0 = fallback to a locale of lower priority
2900 * -1 = fallback to all locales failed, not even to the C locale
2902 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2903 * set, debugging information is output.
2905 * This looks more complicated than it is, mainly due to the #ifdefs.
2907 * We try to set LC_ALL to the value determined by the environment. If
2908 * there is no LC_ALL on this platform, we try the individual categories we
2909 * know about. If this works, we are done.
2911 * But if it doesn't work, we have to do something else. We search the
2912 * environment variables ourselves instead of relying on the system to do
2913 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2914 * think there is one), and the ultimate fallback "C". This is all done in
2915 * the same loop as above to avoid duplicating code, but it makes things
2916 * more complex. The 'trial_locales' array is initialized with just one
2917 * element; it causes the behavior described in the paragraph above this to
2918 * happen. If that fails, we add elements to 'trial_locales', and do extra
2919 * loop iterations to cause the behavior described in this paragraph.
2921 * On Ultrix, the locale MUST come from the environment, so there is
2922 * preliminary code to set it. I (khw) am not sure that it is necessary,
2923 * and that this couldn't be folded into the loop, but barring any real
2924 * platforms to test on, it's staying as-is
2926 * A slight complication is that in embedded Perls, the locale may already
2927 * be set-up, and we don't want to get it from the normal environment
2928 * variables. This is handled by having a special environment variable
2929 * indicate we're in this situation. We simply set setlocale's 2nd
2930 * parameter to be a NULL instead of "". That indicates to setlocale that
2931 * it is not to change anything, but to return the current value,
2932 * effectively initializing perl's db to what the locale already is.
2934 * We play the same trick with NULL if a LC_ALL succeeds. We call
2935 * setlocale() on the individual categores with NULL to get their existing
2936 * values for our db, instead of trying to change them.
2943 PERL_UNUSED_ARG(printwarn);
2945 #else /* USE_LOCALE */
2948 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
2952 /* NULL uses the existing already set up locale */
2953 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2956 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2957 unsigned int trial_locales_count;
2958 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2959 const char * const lang = savepv(PerlEnv_getenv("LANG"));
2960 bool setlocale_failure = FALSE;
2963 /* A later getenv() could zap this, so only use here */
2964 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2966 const bool locwarn = (printwarn > 1
2968 && ( ! bad_lang_use_once
2970 /* disallow with "" or "0" */
2972 && strNE("0", bad_lang_use_once)))));
2974 /* setlocale() return vals; not copied so must be looked at immediately */
2975 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
2977 /* current locale for given category; should have been copied so aren't
2979 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
2983 /* In some systems you can find out the system default locale
2984 * and use that as the fallback locale. */
2985 # define SYSTEM_DEFAULT_LOCALE
2987 # ifdef SYSTEM_DEFAULT_LOCALE
2989 const char *system_default_locale = NULL;
2994 # define DEBUG_LOCALE_INIT(a,b,c)
2997 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
2999 # define DEBUG_LOCALE_INIT(category, locale, result) \
3001 if (debug_initialization) { \
3002 PerlIO_printf(Perl_debug_log, \
3004 __FILE__, __LINE__, \
3005 setlocale_debug_string(category, \
3011 /* Make sure the parallel arrays are properly set up */
3012 # ifdef USE_LOCALE_NUMERIC
3013 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3014 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3015 # ifdef USE_POSIX_2008_LOCALE
3016 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3019 # ifdef USE_LOCALE_CTYPE
3020 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3021 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3022 # ifdef USE_POSIX_2008_LOCALE
3023 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3026 # ifdef USE_LOCALE_COLLATE
3027 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3028 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3029 # ifdef USE_POSIX_2008_LOCALE
3030 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3033 # ifdef USE_LOCALE_TIME
3034 assert(categories[LC_TIME_INDEX] == LC_TIME);
3035 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3036 # ifdef USE_POSIX_2008_LOCALE
3037 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3040 # ifdef USE_LOCALE_MESSAGES
3041 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3042 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3043 # ifdef USE_POSIX_2008_LOCALE
3044 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3047 # ifdef USE_LOCALE_MONETARY
3048 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3049 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3050 # ifdef USE_POSIX_2008_LOCALE
3051 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3054 # ifdef USE_LOCALE_ADDRESS
3055 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3056 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3057 # ifdef USE_POSIX_2008_LOCALE
3058 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3061 # ifdef USE_LOCALE_IDENTIFICATION
3062 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3063 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3064 # ifdef USE_POSIX_2008_LOCALE
3065 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3068 # ifdef USE_LOCALE_MEASUREMENT
3069 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3070 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3071 # ifdef USE_POSIX_2008_LOCALE
3072 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3075 # ifdef USE_LOCALE_PAPER
3076 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3077 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3078 # ifdef USE_POSIX_2008_LOCALE
3079 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3082 # ifdef USE_LOCALE_TELEPHONE
3083 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3084 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3085 # ifdef USE_POSIX_2008_LOCALE
3086 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3090 assert(categories[LC_ALL_INDEX] == LC_ALL);
3091 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3092 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3093 # ifdef USE_POSIX_2008_LOCALE
3094 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3097 # endif /* DEBUGGING */
3099 /* Initialize the cache of the program's UTF-8ness for the always known
3100 * locales C and POSIX */
3101 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3102 sizeof(PL_locale_utf8ness));
3104 # ifdef USE_THREAD_SAFE_LOCALE
3107 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3111 # if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3113 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3114 if (! PL_C_locale_obj) {
3115 Perl_croak_nocontext(
3116 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3118 if (DEBUG_Lv_TEST || debug_initialization) {
3119 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3124 PL_numeric_radix_sv = newSVpvs(".");
3126 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3128 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3129 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3132 # ifdef LOCALE_ENVIRON_REQUIRED
3135 * Ultrix setlocale(..., "") fails if there are no environment
3136 * variables from which to get a locale name.
3140 # error Ultrix without LC_ALL not implemented
3146 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3147 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3148 if (sl_result[LC_ALL_INDEX])
3151 setlocale_failure = TRUE;
3153 if (! setlocale_failure) {
3154 const char * locale_param;
3155 for (i = 0; i < LC_ALL_INDEX; i++) {
3156 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3159 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3160 if (! sl_result[i]) {
3161 setlocale_failure = TRUE;
3163 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3168 # endif /* LC_ALL */
3169 # endif /* LOCALE_ENVIRON_REQUIRED */
3171 /* We try each locale in the list until we get one that works, or exhaust
3172 * the list. Normally the loop is executed just once. But if setting the
3173 * locale fails, inside the loop we add fallback trials to the array and so
3174 * will execute the loop multiple times */
3175 trial_locales[0] = setlocale_init;
3176 trial_locales_count = 1;
3178 for (i= 0; i < trial_locales_count; i++) {
3179 const char * trial_locale = trial_locales[i];
3183 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3184 * when i==0, but I (khw) don't think that behavior makes much
3186 setlocale_failure = FALSE;
3188 # ifdef SYSTEM_DEFAULT_LOCALE
3189 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3191 /* On Windows machines, an entry of "" after the 0th means to use
3192 * the system default locale, which we now proceed to get. */
3193 if (strEQ(trial_locale, "")) {
3196 /* Note that this may change the locale, but we are going to do
3197 * that anyway just below */
3198 system_default_locale = do_setlocale_c(LC_ALL, "");
3199 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3201 /* Skip if invalid or if it's already on the list of locales to
3203 if (! system_default_locale) {
3204 goto next_iteration;
3206 for (j = 0; j < trial_locales_count; j++) {
3207 if (strEQ(system_default_locale, trial_locales[j])) {
3208 goto next_iteration;
3212 trial_locale = system_default_locale;
3215 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3217 # endif /* SYSTEM_DEFAULT_LOCALE */
3223 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3224 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3225 if (! sl_result[LC_ALL_INDEX]) {
3226 setlocale_failure = TRUE;
3229 /* Since LC_ALL succeeded, it should have changed all the other
3230 * categories it can to its value; so we massage things so that the
3231 * setlocales below just return their category's current values.
3232 * This adequately handles the case in NetBSD where LC_COLLATE may
3233 * not be defined for a locale, and setting it individually will
3234 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3235 * the POSIX locale. */
3236 trial_locale = NULL;
3239 # endif /* LC_ALL */
3241 if (! setlocale_failure) {
3243 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3245 = savepv(do_setlocale_r(categories[j], trial_locale));
3246 if (! curlocales[j]) {
3247 setlocale_failure = TRUE;
3249 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3252 if (! setlocale_failure) { /* All succeeded */
3253 break; /* Exit trial_locales loop */
3257 /* Here, something failed; will need to try a fallback. */
3263 if (locwarn) { /* Output failure info only on the first one */
3267 PerlIO_printf(Perl_error_log,
3268 "perl: warning: Setting locale failed.\n");
3270 # else /* !LC_ALL */
3272 PerlIO_printf(Perl_error_log,
3273 "perl: warning: Setting locale failed for the categories:\n\t");
3275 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3276 if (! curlocales[j]) {
3277 PerlIO_printf(Perl_error_log, category_names[j]);
3280 Safefree(curlocales[j]);
3284 # endif /* LC_ALL */
3286 PerlIO_printf(Perl_error_log,
3287 "perl: warning: Please check that your locale settings:\n");
3291 PerlIO_printf(Perl_error_log,
3292 "\tLANGUAGE = %c%s%c,\n",
3293 language ? '"' : '(',
3294 language ? language : "unset",
3295 language ? '"' : ')');
3298 PerlIO_printf(Perl_error_log,
3299 "\tLC_ALL = %c%s%c,\n",
3301 lc_all ? lc_all : "unset",
3302 lc_all ? '"' : ')');
3304 # if defined(USE_ENVIRON_ARRAY)
3309 /* Look through the environment for any variables of the
3310 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3311 * already handled above. These are assumed to be locale
3312 * settings. Output them and their values. */
3313 for (e = environ; *e; e++) {
3314 const STRLEN prefix_len = sizeof("LC_") - 1;
3317 if ( strBEGINs(*e, "LC_")
3318 && ! strBEGINs(*e, "LC_ALL=")
3319 && (uppers_len = strspn(*e + prefix_len,
3320 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3321 && ((*e)[prefix_len + uppers_len] == '='))
3323 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3324 (int) (prefix_len + uppers_len), *e,
3325 *e + prefix_len + uppers_len + 1);
3332 PerlIO_printf(Perl_error_log,
3333 "\t(possibly more locale environment variables)\n");
3337 PerlIO_printf(Perl_error_log,
3338 "\tLANG = %c%s%c\n",
3340 lang ? lang : "unset",
3343 PerlIO_printf(Perl_error_log,
3344 " are supported and installed on your system.\n");
3347 /* Calculate what fallback locales to try. We have avoided this
3348 * until we have to, because failure is quite unlikely. This will
3349 * usually change the upper bound of the loop we are in.
3351 * Since the system's default way of setting the locale has not
3352 * found one that works, We use Perl's defined ordering: LC_ALL,
3353 * LANG, and the C locale. We don't try the same locale twice, so
3354 * don't add to the list if already there. (On POSIX systems, the
3355 * LC_ALL element will likely be a repeat of the 0th element "",
3356 * but there's no harm done by doing it explicitly.
3358 * Note that this tries the LC_ALL environment variable even on
3359 * systems which have no LC_ALL locale setting. This may or may
3360 * not have been originally intentional, but there's no real need
3361 * to change the behavior. */
3363 for (j = 0; j < trial_locales_count; j++) {
3364 if (strEQ(lc_all, trial_locales[j])) {
3368 trial_locales[trial_locales_count++] = lc_all;
3373 for (j = 0; j < trial_locales_count; j++) {
3374 if (strEQ(lang, trial_locales[j])) {
3378 trial_locales[trial_locales_count++] = lang;
3382 # if defined(WIN32) && defined(LC_ALL)
3384 /* For Windows, we also try the system default locale before "C".
3385 * (If there exists a Windows without LC_ALL we skip this because
3386 * it gets too complicated. For those, the "C" is the next
3387 * fallback possibility). The "" is the same as the 0th element of
3388 * the array, but the code at the loop above knows to treat it
3389 * differently when not the 0th */
3390 trial_locales[trial_locales_count++] = "";
3394 for (j = 0; j < trial_locales_count; j++) {
3395 if (strEQ("C", trial_locales[j])) {
3399 trial_locales[trial_locales_count++] = "C";
3402 } /* end of first time through the loop */
3410 } /* end of looping through the trial locales */
3412 if (ok < 1) { /* If we tried to fallback */
3414 if (! setlocale_failure) { /* fallback succeeded */
3415 msg = "Falling back to";
3417 else { /* fallback failed */
3420 /* We dropped off the end of the loop, so have to decrement i to
3421 * get back to the value the last time through */
3425 msg = "Failed to fall back to";
3427 /* To continue, we should use whatever values we've got */
3429 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3430 Safefree(curlocales[j]);
3431 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3432 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3437 const char * description;
3438 const char * name = "";
3439 if (strEQ(trial_locales[i], "C")) {
3440 description = "the standard locale";
3444 # ifdef SYSTEM_DEFAULT_LOCALE
3446 else if (strEQ(trial_locales[i], "")) {
3447 description = "the system default locale";
3448 if (system_default_locale) {
3449 name = system_default_locale;
3453 # endif /* SYSTEM_DEFAULT_LOCALE */
3456 description = "a fallback locale";
3457 name = trial_locales[i];
3459 if (name && strNE(name, "")) {
3460 PerlIO_printf(Perl_error_log,
3461 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3464 PerlIO_printf(Perl_error_log,
3465 "perl: warning: %s %s.\n", msg, description);
3468 } /* End of tried to fallback */
3470 /* Done with finding the locales; update our records */
3472 # ifdef USE_LOCALE_CTYPE
3474 new_ctype(curlocales[LC_CTYPE_INDEX]);
3477 # ifdef USE_LOCALE_COLLATE
3479 new_collate(curlocales[LC_COLLATE_INDEX]);
3482 # ifdef USE_LOCALE_NUMERIC
3484 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3488 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3490 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3492 /* This caches whether each category's locale is UTF-8 or not. This
3493 * may involve changing the locale. It is ok to do this at
3494 * initialization time before any threads have started, but not later
3495 * unless thread-safe operations are used.
3496 * Caching means that if the program heeds our dictate not to change
3497 * locales in threaded applications, this data will remain valid, and
3498 * it may get queried without having to change locales. If the
3499 * environment is such that all categories have the same locale, this
3500 * isn't needed, as the code will not change the locale; but this
3501 * handles the uncommon case where the environment has disparate
3502 * locales for the categories */
3503 (void) _is_cur_LC_category_utf8(categories[i]);
3507 Safefree(curlocales[i]);
3510 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3512 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3513 * locale is UTF-8. The call to new_ctype() just above has already
3514 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3515 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3516 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3517 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3518 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3520 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3521 This is an alternative to using the -C command line switch
3522 (the -C if present will override this). */
3524 const char *p = PerlEnv_getenv("PERL_UNICODE");
3525 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3526 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3540 #endif /* USE_LOCALE */
3543 /* So won't continue to output stuff */
3544 DEBUG_INITIALIZATION_set(FALSE);
3551 #ifdef USE_LOCALE_COLLATE
3554 Perl__mem_collxfrm(pTHX_ const char *input_string,
3555 STRLEN len, /* Length of 'input_string' */
3556 STRLEN *xlen, /* Set to length of returned string
3557 (not including the collation index
3559 bool utf8 /* Is the input in UTF-8? */
3563 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3564 * differences. First, it handles embedded NULs. Second, it allocates a bit
3565 * more memory than needed for the transformed data itself. The real
3566 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3567 * the length of that, and doesn't include the collation index size.
3568 * Please see sv_collxfrm() to see how this is used. */
3570 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3572 char * s = (char *) input_string;
3573 STRLEN s_strlen = strlen(input_string);
3575 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3576 STRLEN length_in_chars;
3577 bool first_time = TRUE; /* Cleared after first loop iteration */
3579 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3581 /* Must be NUL-terminated */
3582 assert(*(input_string + len) == '\0');
3584 /* If this locale has defective collation, skip */
3585 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3586 DEBUG_L(PerlIO_printf(Perl_debug_log,
3587 "_mem_collxfrm: locale's collation is defective\n"));
3591 /* Replace any embedded NULs with the control that sorts before any others.
3592 * This will give as good as possible results on strings that don't
3593 * otherwise contain that character, but otherwise there may be
3594 * less-than-perfect results with that character and NUL. This is
3595 * unavoidable unless we replace strxfrm with our own implementation. */
3596 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3600 STRLEN sans_nuls_len;
3601 int try_non_controls;
3602 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3603 making sure 2nd byte is NUL.
3605 STRLEN this_replacement_len;
3607 /* If we don't know what non-NUL control character sorts lowest for
3608 * this locale, find it */
3609 if (PL_strxfrm_NUL_replacement == '\0') {
3611 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3612 includes the collation index
3615 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3617 /* Unlikely, but it may be that no control will work to replace
3618 * NUL, in which case we instead look for any character. Controls
3619 * are preferred because collation order is, in general, context
3620 * sensitive, with adjoining characters affecting the order, and
3621 * controls are less likely to have such interactions, allowing the
3622 * NUL-replacement to stand on its own. (Another way to look at it
3623 * is to imagine what would happen if the NUL were replaced by a
3624 * combining character; it wouldn't work out all that well.) */
3625 for (try_non_controls = 0;
3626 try_non_controls < 2;
3629 /* Look through all legal code points (NUL isn't) */
3630 for (j = 1; j < 256; j++) {
3631 char * x; /* j's xfrm plus collation index */
3632 STRLEN x_len; /* length of 'x' */
3633 STRLEN trial_len = 1;
3634 char cur_source[] = { '\0', '\0' };
3636 /* Skip non-controls the first time through the loop. The
3637 * controls in a UTF-8 locale are the L1 ones */
3638 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3645 /* Create a 1-char string of the current code point */
3646 cur_source[0] = (char) j;
3648 /* Then transform it */
3649 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3650 0 /* The string is not in UTF-8 */);
3652 /* Ignore any character that didn't successfully transform.
3658 /* If this character's transformation is lower than
3659 * the current lowest, this one becomes the lowest */
3660 if ( cur_min_x == NULL
3661 || strLT(x + COLLXFRM_HDR_LEN,
3662 cur_min_x + COLLXFRM_HDR_LEN))
3664 PL_strxfrm_NUL_replacement = j;
3670 } /* end of loop through all 255 characters */
3672 /* Stop looking if found */
3677 /* Unlikely, but possible, if there aren't any controls that
3678 * work in the locale, repeat the loop, looking for any
3679 * character that works */
3680 DEBUG_L(PerlIO_printf(Perl_debug_log,
3681 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3682 } /* End of loop to try first the controls, then any char */
3685 DEBUG_L(PerlIO_printf(Perl_debug_log,
3686 "_mem_collxfrm: Couldn't find any character to replace"
3687 " embedded NULs in locale %s with", PL_collation_name));
3691 DEBUG_L(PerlIO_printf(Perl_debug_log,
3692 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3693 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3695 Safefree(cur_min_x);
3696 } /* End of determining the character that is to replace NULs */
3698 /* If the replacement is variant under UTF-8, it must match the
3699 * UTF8-ness of the original */
3700 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3701 this_replacement_char[0] =
3702 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3703 this_replacement_char[1] =
3704 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3705 this_replacement_len = 2;
3708 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3709 /* this_replacement_char[1] = '\0' was done at initialization */
3710 this_replacement_len = 1;
3713 /* The worst case length for the replaced string would be if every
3714 * character in it is NUL. Multiply that by the length of each
3715 * replacement, and allow for a trailing NUL */
3716 sans_nuls_len = (len * this_replacement_len) + 1;
3717 Newx(sans_nuls, sans_nuls_len, char);
3720 /* Replace each NUL with the lowest collating control. Loop until have
3721 * exhausted all the NULs */
3722 while (s + s_strlen < e) {
3723 my_strlcat(sans_nuls, s, sans_nuls_len);
3725 /* Do the actual replacement */
3726 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3728 /* Move past the input NUL */
3730 s_strlen = strlen(s);
3733 /* And add anything that trails the final NUL */
3734 my_strlcat(sans_nuls, s, sans_nuls_len);
3736 /* Switch so below we transform this modified string */
3739 } /* End of replacing NULs */
3741 /* Make sure the UTF8ness of the string and locale match */
3742 if (utf8 != PL_in_utf8_COLLATE_locale) {
3743 /* XXX convert above Unicode to 10FFFF? */
3744 const char * const t = s; /* Temporary so we can later find where the
3747 /* Here they don't match. Change the string's to be what the locale is
3750 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3751 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3754 else { /* locale is not UTF-8; but input is; downgrade the input */
3756 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3758 /* If the downgrade was successful we are done, but if the input
3759 * contains things that require UTF-8 to represent, have to do
3760 * damage control ... */
3761 if (UNLIKELY(utf8)) {
3763 /* What we do is construct a non-UTF-8 string with
3764 * 1) the characters representable by a single byte converted
3765 * to be so (if necessary);
3766 * 2) and the rest converted to collate the same as the
3767 * highest collating representable character. That makes
3768 * them collate at the end. This is similar to how we
3769 * handle embedded NULs, but we use the highest collating
3770 * code point instead of the smallest. Like the NUL case,
3771 * this isn't perfect, but is the best we can reasonably
3772 * do. Every above-255 code point will sort the same as
3773 * the highest-sorting 0-255 code point. If that code
3774 * point can combine in a sequence with some other code
3775 * points for weight calculations, us changing something to
3776 * be it can adversely affect the results. But in most
3777 * cases, it should work reasonably. And note that this is
3778 * really an illegal situation: using code points above 255
3779 * on a locale where only 0-255 are valid. If two strings
3780 * sort entirely equal, then the sort order for the
3781 * above-255 code points will be in code point order. */
3785 /* If we haven't calculated the code point with the maximum
3786 * collating order for this locale, do so now */
3787 if (! PL_strxfrm_max_cp) {
3790 /* The current transformed string that collates the
3791 * highest (except it also includes the prefixed collation
3793 char * cur_max_x = NULL;
3795 /* Look through all legal code points (NUL isn't) */
3796 for (j = 1; j < 256; j++) {
3799 char cur_source[] = { '\0', '\0' };
3801 /* Create a 1-char string of the current code point */
3802 cur_source[0] = (char) j;
3804 /* Then transform it */
3805 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3807 /* If something went wrong (which it shouldn't), just
3808 * ignore this code point */
3813 /* If this character's transformation is higher than
3814 * the current highest, this one becomes the highest */
3815 if ( cur_max_x == NULL
3816 || strGT(x + COLLXFRM_HDR_LEN,
3817 cur_max_x + COLLXFRM_HDR_LEN))
3819 PL_strxfrm_max_cp = j;
3828 DEBUG_L(PerlIO_printf(Perl_debug_log,
3829 "_mem_collxfrm: Couldn't find any character to"
3830 " replace above-Latin1 chars in locale %s with",
3831 PL_collation_name));
3835 DEBUG_L(PerlIO_printf(Perl_debug_log,
3836 "_mem_collxfrm: highest 1-byte collating character"
3837 " in locale %s is 0x%02X\n",
3839 PL_strxfrm_max_cp));
3841 Safefree(cur_max_x);
3844 /* Here we know which legal code point collates the highest.
3845 * We are ready to construct the non-UTF-8 string. The length
3846 * will be at least 1 byte smaller than the input string
3847 * (because we changed at least one 2-byte character into a
3848 * single byte), but that is eaten up by the trailing NUL */
3854 char * e = (char *) t + len;
3856 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3858 if (UTF8_IS_INVARIANT(cur_char)) {
3861 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
3862 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3864 else { /* Replace illegal cp with highest collating
3866 s[d++] = PL_strxfrm_max_cp;
3870 Renew(s, d, char); /* Free up unused space */
3875 /* Here, we have constructed a modified version of the input. It could
3876 * be that we already had a modified copy before we did this version.
3877 * If so, that copy is no longer needed */
3878 if (t != input_string) {
3883 length_in_chars = (utf8)
3884 ? utf8_length((U8 *) s, (U8 *) s + len)
3887 /* The first element in the output is the collation id, used by
3888 * sv_collxfrm(); then comes the space for the transformed string. The
3889 * equation should give us a good estimate as to how much is needed */
3890 xAlloc = COLLXFRM_HDR_LEN
3892 + (PL_collxfrm_mult * length_in_chars);
3893 Newx(xbuf, xAlloc, char);
3894 if (UNLIKELY(! xbuf)) {
3895 DEBUG_L(PerlIO_printf(Perl_debug_log,
3896 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
3900 /* Store the collation id */
3901 *(U32*)xbuf = PL_collation_ix;
3903 /* Then the transformation of the input. We loop until successful, or we
3907 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
3909 /* If the transformed string occupies less space than we told strxfrm()
3910 * was available, it means it successfully transformed the whole
3912 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
3914 /* Some systems include a trailing NUL in the returned length.
3915 * Ignore it, using a loop in case multiple trailing NULs are
3918 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3923 /* If the first try didn't get it, it means our prediction was low.
3924 * Modify the coefficients so that we predict a larger value in any
3925 * future transformations */
3927 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3928 STRLEN computed_guess = PL_collxfrm_base
3929 + (PL_collxfrm_mult * length_in_chars);
3931 /* On zero-length input, just keep current slope instead of
3933 const STRLEN new_m = (length_in_chars != 0)
3934 ? needed / length_in_chars
3937 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3938 "%s: %d: initial size of %zu bytes for a length "
3939 "%zu string was insufficient, %zu needed\n",
3941 computed_guess, length_in_chars, needed));
3943 /* If slope increased, use it, but discard this result for
3944 * length 1 strings, as we can't be sure that it's a real slope
3946 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
3950 STRLEN old_m = PL_collxfrm_mult;
3951 STRLEN old_b = PL_collxfrm_base;
3955 PL_collxfrm_mult = new_m;
3956 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3957 computed_guess = PL_collxfrm_base
3958 + (PL_collxfrm_mult * length_in_chars);
3959 if (computed_guess < needed) {
3960 PL_collxfrm_base += needed - computed_guess;
3963 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3964 "%s: %d: slope is now %zu; was %zu, base "
3965 "is now %zu; was %zu\n",
3967 PL_collxfrm_mult, old_m,
3968 PL_collxfrm_base, old_b));
3970 else { /* Slope didn't change, but 'b' did */
3971 const STRLEN new_b = needed
3974 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3975 "%s: %d: base is now %zu; was %zu\n",
3977 new_b, PL_collxfrm_base));
3978 PL_collxfrm_base = new_b;
3985 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3986 DEBUG_L(PerlIO_printf(Perl_debug_log,
3987 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3988 *xlen, PERL_INT_MAX));
3992 /* A well-behaved strxfrm() returns exactly how much space it needs
3993 * (usually not including the trailing NUL) when it fails due to not
3994 * enough space being provided. Assume that this is the case unless
3995 * it's been proven otherwise */
3996 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
3997 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
3999 else { /* Here, either:
4000 * 1) The strxfrm() has previously shown bad behavior; or
4001 * 2) It isn't the first time through the loop, which means
4002 * that the strxfrm() is now showing bad behavior, because
4003 * we gave it what it said was needed in the previous
4004 * iteration, and it came back saying it needed still more.
4005 * (Many versions of cygwin fit this. When the buffer size
4006 * isn't sufficient, they return the input size instead of
4007 * how much is needed.)
4008 * Increase the buffer size by a fixed percentage and try again.
4010 xAlloc += (xAlloc / 4) + 1;
4011 PL_strxfrm_is_behaved = FALSE;
4015 if (DEBUG_Lv_TEST || debug_initialization) {
4016 PerlIO_printf(Perl_debug_log,
4017 "_mem_collxfrm required more space than previously calculated"
4018 " for locale %s, trying again with new guess=%d+%zu\n",
4019 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4020 xAlloc - COLLXFRM_HDR_LEN);
4027 Renew(xbuf, xAlloc, char);
4028 if (UNLIKELY(! xbuf)) {
4029 DEBUG_L(PerlIO_printf(Perl_debug_log,
4030 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4040 if (DEBUG_Lv_TEST || debug_initialization) {
4042 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4043 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4044 PerlIO_printf(Perl_debug_log, "%s\n",
4045 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4051 /* Free up unneeded space; retain ehough for trailing NUL */
4052 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4054 if (s != input_string) {
4062 if (s != input_string) {
4069 if (DEBUG_Lv_TEST || debug_initialization) {
4070 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4081 S_print_collxfrm_input_and_return(pTHX_
4082 const char * const s,
4083 const char * const e,
4084 const STRLEN * const xlen,
4088 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4090 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4091 (UV)PL_collation_ix);
4093 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4096 PerlIO_printf(Perl_debug_log, "NULL");
4098 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4100 print_bytes_for_locale(s, e, is_utf8);
4102 PerlIO_printf(Perl_debug_log, "'\n");
4106 S_print_bytes_for_locale(pTHX_
4107 const char * const s,
4108 const char * const e,
4112 bool prev_was_printable = TRUE;
4113 bool first_time = TRUE;
4115 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4119 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4122 if (! prev_was_printable) {
4123 PerlIO_printf(Perl_debug_log, " ");
4125 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4126 prev_was_printable = TRUE;
4130 PerlIO_printf(Perl_debug_log, " ");
4132 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4133 prev_was_printable = FALSE;
4135 t += (is_utf8) ? UTF8SKIP(t) : 1;
4140 # endif /* #ifdef DEBUGGING */
4141 #endif /* USE_LOCALE_COLLATE */
4146 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4148 /* Changes the locale for LC_'switch_category" to that of
4149 * LC_'template_category', if they aren't already the same. If not NULL,
4150 * 'template_locale' is the locale that 'template_category' is in.
4152 * Returns a copy of the name of the original locale for 'switch_category'
4153 * so can be switched back to with the companion function
4154 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4156 char * restore_to_locale = NULL;
4158 if (switch_category == template_category) { /* No changes needed */
4162 /* Find the original locale of the category we may need to change, so that
4163 * it can be restored to later */
4164 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4166 if (! restore_to_locale) {
4168 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4169 __FILE__, __LINE__, category_name(switch_category), errno);
4172 /* If the locale of the template category wasn't passed in, find it now */
4173 if (template_locale == NULL) {
4174 template_locale = do_setlocale_r(template_category, NULL);
4175 if (! template_locale) {
4177 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4178 __FILE__, __LINE__, category_name(template_category), errno);
4182 /* It the locales are the same, there's nothing to do */
4183 if (strEQ(restore_to_locale, template_locale)) {
4184 Safefree(restore_to_locale);
4186 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4187 category_name(switch_category), restore_to_locale));
4192 /* Finally, change the locale to the template one */
4193 if (! do_setlocale_r(switch_category, template_locale)) {
4195 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4196 __FILE__, __LINE__, category_name(switch_category),
4197 template_locale, errno);
4200 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4201 category_name(switch_category), template_locale));
4203 return restore_to_locale;
4207 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4209 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4210 * copy that will be freed by this function), or do nothing if the latter
4211 * parameter is NULL */
4213 if (original_locale == NULL) {
4217 if (! do_setlocale_r(category, original_locale)) {
4219 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4221 category_name(category), original_locale, errno);
4224 Safefree(original_locale);
4228 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4230 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4231 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4232 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4233 * could give the wrong result. The result will very likely be correct for
4234 * languages that have commonly used non-ASCII characters, but for notably
4235 * English, it comes down to if the locale's name ends in something like
4236 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4238 * If the platform is early C89, not containing mbtowc(), or we are
4239 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4240 * These work very well for non-Latin locales or those whose currency
4241 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4242 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4243 * on the name containing UTF-8 or not. */
4245 /* Name of current locale corresponding to the input category */
4246 const char *save_input_locale = NULL;
4248 bool is_utf8 = FALSE; /* The return value */
4250 /* The variables below are for the cache of previous lookups using this
4251 * function. The cache is a C string, described at the definition for
4252 * 'C_and_POSIX_utf8ness'.
4254 * The first part of the cache is fixed, for the C and POSIX locales. The
4255 * varying part starts just after them. */
4256 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4258 Size_t utf8ness_cache_size; /* Size of the varying portion */
4259 Size_t input_name_len; /* Length in bytes of save_input_locale */
4260 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4261 the name in the cache */
4262 char * delimited; /* The name plus the delimiters used to store
4264 char * name_pos; /* position of 'delimited' in the cache, or 0
4270 assert(category != LC_ALL);
4274 /* Get the desired category's locale */
4275 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4276 if (! save_input_locale) {
4278 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4279 __FILE__, __LINE__, category_name(category), errno);
4282 DEBUG_L(PerlIO_printf(Perl_debug_log,
4283 "Current locale for %s is %s\n",
4284 category_name(category), save_input_locale));
4286 input_name_len = strlen(save_input_locale);
4288 /* In our cache, each name is accompanied by two delimiters and a single
4290 input_name_len_with_overhead = input_name_len + 3;
4292 /* Allocate and populate space for a copy of the name surrounded by the
4294 Newx(delimited, input_name_len_with_overhead, char);
4295 delimited[0] = UTF8NESS_SEP[0];
4296 Copy(save_input_locale, delimited + 1, input_name_len, char);
4297 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4298 delimited[input_name_len+2] = '\0';
4300 /* And see if that is in the cache */
4301 name_pos = instr(PL_locale_utf8ness, delimited);
4303 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4307 if (DEBUG_Lv_TEST || debug_initialization) {
4308 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4309 save_input_locale, is_utf8);
4314 /* And, if not already in that position, move it to the beginning of
4315 * the non-constant portion of the list, since it is the most recently
4316 * used. (We don't have to worry about overflow, since just moving
4317 * existing names around) */
4318 if (name_pos > utf8ness_cache) {
4319 Move(utf8ness_cache,
4320 utf8ness_cache + input_name_len_with_overhead,
4321 name_pos - utf8ness_cache, char);
https://perl5.git.perl.org / perl5.git / blob