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
914 const char * s = locale;
915 const char * e = locale + strlen(locale);
917 const char * category_end;
918 const char * name_start;
919 const char * name_end;
921 /* If the string that gives what to set doesn't include all categories,
922 * the omitted ones get set to "C". To get this behavior, first set
923 * all the individual categories to "C", and override the furnished
925 for (i = 0; i < LC_ALL_INDEX; i++) {
926 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
933 /* Parse through the category */
934 while (isWORDCHAR(*p)) {
941 "panic: %s: %d: Unexpected character in locale name '%02X",
942 __FILE__, __LINE__, *(p-1));
945 /* Parse through the locale name */
947 while (p < e && *p != ';') {
950 "panic: %s: %d: Unexpected character in locale name '%02X",
951 __FILE__, __LINE__, *(p-1));
957 /* Space past the semi-colon */
962 /* Find the index of the category name in our lists */
963 for (i = 0; i < LC_ALL_INDEX; i++) {
964 char * individ_locale;
966 /* Keep going if this isn't the index. The strnNE() avoids a
967 * Perl_form(), but would fail if ever a category name could be
968 * a substring of another one, like if there were a
970 if strnNE(s, category_names[i], category_end - s) {
974 /* If this index is for the single category we're changing, we
975 * have found the locale to set it to. */
976 if (category == categories[i]) {
977 locale = Perl_form(aTHX_ "%.*s",
978 (int) (name_end - name_start),
983 assert(category == LC_ALL);
984 individ_locale = Perl_form(aTHX_ "%.*s",
985 (int) (name_end - name_start), name_start);
986 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
995 /* Here we have set all the individual categories by recursive calls.
996 * These collectively should have fixed up LC_ALL, so can just query
997 * what that now is */
998 assert(category == LC_ALL);
1000 return do_setlocale_c(LC_ALL, NULL);
1005 # endif /* end of ! querylocale */
1007 /* Ready to create a new locale by modification of the exising one */
1008 new_obj = newlocale(mask, locale, old_obj);
1015 if (DEBUG_L_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1021 if (! uselocale(old_obj)) {
1025 if (DEBUG_L_TEST || debug_initialization) {
1026 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1038 if (DEBUG_Lv_TEST || debug_initialization) {
1039 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1044 /* And switch into it */
1045 if (! uselocale(new_obj)) {
1050 if (DEBUG_L_TEST || debug_initialization) {
1051 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1056 if (! uselocale(old_obj)) {
1060 if (DEBUG_L_TEST || debug_initialization) {
1061 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1067 freelocale(new_obj);
1074 if (DEBUG_Lv_TEST || debug_initialization) {
1075 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1080 /* We are done, except for updating our records (if the system doesn't keep
1081 * them) and in the case of locale "", we don't actually know what the
1082 * locale that got switched to is, as it came from the environment. So
1083 * have to find it */
1085 # ifdef HAS_QUERYLOCALE
1087 if (strEQ(locale, "")) {
1088 locale = querylocale(mask, new_obj);
1093 /* Here, 'locale' is the return value */
1095 /* Without querylocale(), we have to update our records */
1097 if (category == LC_ALL) {
1100 /* For LC_ALL, we change all individual categories to correspond */
1101 /* PL_curlocales is a parallel array, so has same
1102 * length as 'categories' */
1103 for (i = 0; i <= LC_ALL_INDEX; i++) {
1104 Safefree(PL_curlocales[i]);
1105 PL_curlocales[i] = savepv(locale);
1110 /* For a single category, if it's not the same as the one in LC_ALL, we
1113 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1114 Safefree(PL_curlocales[LC_ALL_INDEX]);
1115 PL_curlocales[LC_ALL_INDEX] = NULL;
1118 /* Then update the category's record */
1119 Safefree(PL_curlocales[index]);
1120 PL_curlocales[index] = savepv(locale);
1128 #endif /* USE_POSIX_2008_LOCALE */
1130 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1131 didn't natively support them */
1133 /* The way this would work is that we would keep a per-thread list of the
1134 * correct locale for that thread. Any operation that was locale-sensitive
1135 * would have to be changed so that it would look like this:
1138 * setlocale to the correct locale for this operation
1142 * This leaves the global locale in the most recently used operation's, but it
1143 * was locked long enough to get the result. If that result is static, it
1144 * needs to be copied before the unlock.
1146 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1147 * the setup, but are no-ops when not needed, and similarly,
1148 * END_LOCALE_DEPENDENT_OP for the tear-down
1150 * But every call to a locale-sensitive function would have to be changed, and
1151 * if a module didn't cooperate by using the mutex, things would break.
1153 * This code was abandoned before being completed or tested, and is left as-is
1156 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1157 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1160 S_locking_setlocale(pTHX_
1162 const char * locale,
1164 const bool is_index_valid
1167 /* This function kind of performs a setlocale() on just the current thread;
1168 * thus it is kind of thread-safe. It does this by keeping a thread-level
1169 * array of the current locales for each category. Every time a locale is
1170 * switched to, it does the switch globally, but updates the thread's
1171 * array. A query as to what the current locale is just returns the
1172 * appropriate element from the array, and doesn't actually call the system
1173 * setlocale(). The saving into the array is done in an uninterruptible
1174 * section of code, so is unaffected by whatever any other threads might be
1177 * All locale-sensitive operations must work by first starting a critical
1178 * section, then switching to the thread's locale as kept by this function,
1179 * and then doing the operation, then ending the critical section. Thus,
1180 * each gets done in the appropriate locale. simulating thread-safety.
1182 * This function takes the same parameters, 'category' and 'locale', that
1183 * the regular setlocale() function does, but it also takes two additional
1184 * ones. This is because as described earlier. If we know on input the
1185 * index corresponding to the category into the array where we store the
1186 * current locales, we don't have to calculate it. If the caller knows at
1187 * compile time what the index is, it it can pass it, setting
1188 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1192 /* If the input index might be incorrect, calculate the correct one */
1193 if (! is_index_valid) {
1196 if (DEBUG_Lv_TEST || debug_initialization) {
1197 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1200 for (i = 0; i <= LC_ALL_INDEX; i++) {
1201 if (category == categories[i]) {
1207 /* Here, we don't know about this category, so can't handle it.
1208 * XXX best we can do is to unsafely set this
1211 return my_setlocale(category, locale);
1215 if (DEBUG_Lv_TEST || debug_initialization) {
1216 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1220 /* For a query, just return what's in our records */
1221 if (new_locale == NULL) {
1222 return curlocales[index];
1226 /* Otherwise, we need to do the switch, and save the result, all in a
1227 * critical section */
1229 Safefree(curlocales[[index]]);
1231 /* It might be that this is called from an already-locked section of code.
1232 * We would have to detect and skip the LOCK/UNLOCK if so */
1235 curlocales[index] = savepv(my_setlocale(category, new_locale));
1237 if (strEQ(new_locale, "")) {
1241 /* The locale values come from the environment, and may not all be the
1242 * same, so for LC_ALL, we have to update all the others, while the
1243 * mutex is still locked */
1245 if (category == LC_ALL) {
1247 for (i = 0; i < LC_ALL_INDEX) {
1248 curlocales[i] = my_setlocale(categories[i], NULL);
1257 return curlocales[index];
1263 S_set_numeric_radix(pTHX_ const bool use_locale)
1265 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1266 * TRUE, use the radix character derived from the current locale */
1268 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1269 || defined(HAS_NL_LANGINFO))
1271 const char * radix = (use_locale)
1272 ? my_nl_langinfo(RADIXCHAR, FALSE)
1273 /* FALSE => already in dest locale */
1276 sv_setpv(PL_numeric_radix_sv, radix);
1278 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1279 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1280 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1281 SvCUR(PL_numeric_radix_sv))
1282 && _is_cur_LC_category_utf8(LC_NUMERIC))
1284 SvUTF8_on(PL_numeric_radix_sv);
1289 if (DEBUG_L_TEST || debug_initialization) {
1290 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1291 SvPVX(PL_numeric_radix_sv),
1292 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1296 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1301 S_new_numeric(pTHX_ const char *newnum)
1304 #ifndef USE_LOCALE_NUMERIC
1306 PERL_UNUSED_ARG(newnum);
1310 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1311 * core Perl this and that 'newnum' is the name of the new locale.
1312 * It installs this locale as the current underlying default.
1314 * The default locale and the C locale can be toggled between by use of the
1315 * set_numeric_underlying() and set_numeric_standard() functions, which
1316 * should probably not be called directly, but only via macros like
1317 * SET_NUMERIC_STANDARD() in perl.h.
1319 * The toggling is necessary mainly so that a non-dot radix decimal point
1320 * character can be output, while allowing internal calculations to use a
1323 * This sets several interpreter-level variables:
1324 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1325 * PL_numeric_underlying A boolean indicating if the toggled state is such
1326 * that the current locale is the program's underlying
1328 * PL_numeric_standard An int indicating if the toggled state is such
1329 * that the current locale is the C locale or
1330 * indistinguishable from the C locale. If non-zero, it
1331 * is in C; if > 1, it means it may not be toggled away
1333 * PL_numeric_underlying_is_standard A bool kept by this function
1334 * indicating that the underlying locale and the standard
1335 * C locale are indistinguishable for the purposes of
1336 * LC_NUMERIC. This happens when both of the above two
1337 * variables are true at the same time. (Toggling is a
1338 * no-op under these circumstances.) This variable is
1339 * used to avoid having to recalculate.
1345 Safefree(PL_numeric_name);
1346 PL_numeric_name = NULL;
1347 PL_numeric_standard = TRUE;
1348 PL_numeric_underlying = TRUE;
1349 PL_numeric_underlying_is_standard = TRUE;
1353 save_newnum = stdize_locale(savepv(newnum));
1354 PL_numeric_underlying = TRUE;
1355 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1357 #ifndef TS_W32_BROKEN_LOCALECONV
1359 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1360 * them. But on broken Windows systems calling my_nl_langinfo() for
1361 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1362 * and just always change the locale if not C nor POSIX on those systems */
1363 if (! PL_numeric_standard) {
1364 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1365 FALSE /* Don't toggle locale */ ))
1366 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1371 /* Save the new name if it isn't the same as the previous one, if any */
1372 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1373 Safefree(PL_numeric_name);
1374 PL_numeric_name = save_newnum;
1377 Safefree(save_newnum);
1380 PL_numeric_underlying_is_standard = PL_numeric_standard;
1382 # ifdef HAS_POSIX_2008_LOCALE
1384 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1386 PL_underlying_numeric_obj);
1390 if (DEBUG_L_TEST || debug_initialization) {
1391 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1394 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1395 * have to worry about the radix being a non-dot. (Core operations that
1396 * need the underlying locale change to it temporarily). */
1397 if (PL_numeric_standard) {
1398 set_numeric_radix(0);
1401 set_numeric_standard();
1404 #endif /* USE_LOCALE_NUMERIC */
1409 Perl_set_numeric_standard(pTHX)
1412 #ifdef USE_LOCALE_NUMERIC
1414 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1415 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1416 * macro avoids calling this routine if toggling isn't necessary according
1417 * to our records (which could be wrong if some XS code has changed the
1418 * locale behind our back) */
1420 do_setlocale_c(LC_NUMERIC, "C");
1421 PL_numeric_standard = TRUE;
1422 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1423 set_numeric_radix(0);
1427 if (DEBUG_L_TEST || debug_initialization) {
1428 PerlIO_printf(Perl_debug_log,
1429 "LC_NUMERIC locale now is standard C\n");
1433 #endif /* USE_LOCALE_NUMERIC */
1438 Perl_set_numeric_underlying(pTHX)
1441 #ifdef USE_LOCALE_NUMERIC
1443 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1444 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1445 * instead of calling this directly. The macro avoids calling this routine
1446 * if toggling isn't necessary according to our records (which could be
1447 * wrong if some XS code has changed the locale behind our back) */
1449 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1450 PL_numeric_standard = PL_numeric_underlying_is_standard;
1451 PL_numeric_underlying = TRUE;
1452 set_numeric_radix(! PL_numeric_standard);
1456 if (DEBUG_L_TEST || debug_initialization) {
1457 PerlIO_printf(Perl_debug_log,
1458 "LC_NUMERIC locale now is %s\n",
1463 #endif /* USE_LOCALE_NUMERIC */
1468 * Set up for a new ctype locale.
1471 S_new_ctype(pTHX_ const char *newctype)
1474 #ifndef USE_LOCALE_CTYPE
1476 PERL_ARGS_ASSERT_NEW_CTYPE;
1477 PERL_UNUSED_ARG(newctype);
1478 PERL_UNUSED_CONTEXT;
1482 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1483 * core Perl this and that 'newctype' is the name of the new locale.
1485 * This function sets up the folding arrays for all 256 bytes, assuming
1486 * that tofold() is tolc() since fold case is not a concept in POSIX,
1488 * Any code changing the locale (outside this file) should use
1489 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1490 * this function should be called directly only from this file and from
1491 * POSIX::setlocale() */
1496 /* Don't check for problems if we are suppressing the warnings */
1497 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1499 PERL_ARGS_ASSERT_NEW_CTYPE;
1501 /* We will replace any bad locale warning with 1) nothing if the new one is
1502 * ok; or 2) a new warning for the bad new locale */
1503 if (PL_warn_locale) {
1504 SvREFCNT_dec_NN(PL_warn_locale);
1505 PL_warn_locale = NULL;
1508 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1510 /* A UTF-8 locale gets standard rules. But note that code still has to
1511 * handle this specially because of the three problematic code points */
1512 if (PL_in_utf8_CTYPE_locale) {
1513 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1516 /* We don't populate the other lists if a UTF-8 locale, but do check that
1517 * everything works as expected, unless checking turned off */
1518 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1519 /* Assume enough space for every character being bad. 4 spaces each
1520 * for the 94 printable characters that are output like "'x' "; and 5
1521 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1523 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1524 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1526 unsigned int bad_count = 0; /* Count of bad characters */
1528 for (i = 0; i < 256; i++) {
1529 if (! PL_in_utf8_CTYPE_locale) {
1531 PL_fold_locale[i] = (U8) tolower(i);
1532 else if (islower(i))
1533 PL_fold_locale[i] = (U8) toupper(i);
1535 PL_fold_locale[i] = (U8) i;
1538 /* If checking for locale problems, see if the native ASCII-range
1539 * printables plus \n and \t are in their expected categories in
1540 * the new locale. If not, this could mean big trouble, upending
1541 * Perl's and most programs' assumptions, like having a
1542 * metacharacter with special meaning become a \w. Fortunately,
1543 * it's very rare to find locales that aren't supersets of ASCII
1544 * nowadays. It isn't a problem for most controls to be changed
1545 * into something else; we check only \n and \t, though perhaps \r
1546 * could be an issue as well. */
1547 if ( check_for_problems
1548 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1550 bool is_bad = FALSE;
1551 char name[3] = { '\0' };
1553 /* Convert the name into a string */
1558 else if (i == '\n') {
1559 my_strlcpy(name, "\n", sizeof(name));
1562 my_strlcpy(name, "\t", sizeof(name));
1565 /* Check each possibe class */
1566 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1568 DEBUG_L(PerlIO_printf(Perl_debug_log,
1569 "isalnum('%s') unexpectedly is %d\n",
1570 name, cBOOL(isalnum(i))));
1572 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1574 DEBUG_L(PerlIO_printf(Perl_debug_log,
1575 "isalpha('%s') unexpectedly is %d\n",
1576 name, cBOOL(isalpha(i))));
1578 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1580 DEBUG_L(PerlIO_printf(Perl_debug_log,
1581 "isdigit('%s') unexpectedly is %d\n",
1582 name, cBOOL(isdigit(i))));
1584 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1586 DEBUG_L(PerlIO_printf(Perl_debug_log,
1587 "isgraph('%s') unexpectedly is %d\n",
1588 name, cBOOL(isgraph(i))));
1590 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1592 DEBUG_L(PerlIO_printf(Perl_debug_log,
1593 "islower('%s') unexpectedly is %d\n",
1594 name, cBOOL(islower(i))));
1596 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1598 DEBUG_L(PerlIO_printf(Perl_debug_log,
1599 "isprint('%s') unexpectedly is %d\n",
1600 name, cBOOL(isprint(i))));
1602 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1604 DEBUG_L(PerlIO_printf(Perl_debug_log,
1605 "ispunct('%s') unexpectedly is %d\n",
1606 name, cBOOL(ispunct(i))));
1608 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1610 DEBUG_L(PerlIO_printf(Perl_debug_log,
1611 "isspace('%s') unexpectedly is %d\n",
1612 name, cBOOL(isspace(i))));
1614 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1616 DEBUG_L(PerlIO_printf(Perl_debug_log,
1617 "isupper('%s') unexpectedly is %d\n",
1618 name, cBOOL(isupper(i))));
1620 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1622 DEBUG_L(PerlIO_printf(Perl_debug_log,
1623 "isxdigit('%s') unexpectedly is %d\n",
1624 name, cBOOL(isxdigit(i))));
1626 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1628 DEBUG_L(PerlIO_printf(Perl_debug_log,
1629 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1630 name, tolower(i), (int) toLOWER_A(i)));
1632 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1634 DEBUG_L(PerlIO_printf(Perl_debug_log,
1635 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1636 name, toupper(i), (int) toUPPER_A(i)));
1638 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1640 DEBUG_L(PerlIO_printf(Perl_debug_log,
1641 "'\\n' (=%02X) is not a control\n", (int) i));
1644 /* Add to the list; Separate multiple entries with a blank */
1647 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1649 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1657 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1658 * this locale requires more than one byte, there are going to be
1660 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1661 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1662 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1664 if ( check_for_problems && MB_CUR_MAX > 1
1665 && ! PL_in_utf8_CTYPE_locale
1667 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1668 * locale. Just assume that the implementation for them (plus
1669 * for POSIX) is correct and the > 1 value is spurious. (Since
1670 * these are specially handled to never be considered UTF-8
1671 * locales, as long as this is the only problem, everything
1672 * should work fine */
1673 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1675 multi_byte_locale = TRUE;
1680 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1681 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1682 PL_warn_locale = Perl_newSVpvf(aTHX_
1683 "Locale '%s' contains (at least) the following characters"
1684 " which have\nunexpected meanings: %s\nThe Perl program"
1685 " will use the expected meanings",
1686 newctype, bad_chars_list);
1689 PL_warn_locale = Perl_newSVpvf(aTHX_
1690 "Locale '%s' may not work well.%s%s%s\n",
1693 ? " Some characters in it are not recognized by"
1697 ? "\nThe following characters (and maybe others)"
1698 " may not have the same meaning as the Perl"
1699 " program expects:\n"
1707 # ifdef HAS_NL_LANGINFO
1709 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1710 /* parameter FALSE is a don't care here */
1711 my_nl_langinfo(CODESET, FALSE));
1715 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1717 /* If we are actually in the scope of the locale or are debugging,
1718 * output the message now. If not in that scope, we save the
1719 * message to be output at the first operation using this locale,
1720 * if that actually happens. Most programs don't use locales, so
1721 * they are immune to bad ones. */
1722 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1724 /* The '0' below suppresses a bogus gcc compiler warning */
1725 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1727 if (IN_LC(LC_CTYPE)) {
1728 SvREFCNT_dec_NN(PL_warn_locale);
1729 PL_warn_locale = NULL;
1735 #endif /* USE_LOCALE_CTYPE */
1740 Perl__warn_problematic_locale()
1743 #ifdef USE_LOCALE_CTYPE
1747 /* Internal-to-core function that outputs the message in PL_warn_locale,
1748 * and then NULLS it. Should be called only through the macro
1749 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1751 if (PL_warn_locale) {
1752 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1753 SvPVX(PL_warn_locale),
1754 0 /* dummy to avoid compiler warning */ );
1755 SvREFCNT_dec_NN(PL_warn_locale);
1756 PL_warn_locale = NULL;
1764 S_new_collate(pTHX_ const char *newcoll)
1767 #ifndef USE_LOCALE_COLLATE
1769 PERL_UNUSED_ARG(newcoll);
1770 PERL_UNUSED_CONTEXT;
1774 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1775 * core Perl this and that 'newcoll' is the name of the new locale.
1777 * The design of locale collation is that every locale change is given an
1778 * index 'PL_collation_ix'. The first time a string particpates in an
1779 * operation that requires collation while locale collation is active, it
1780 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1781 * magic includes the collation index, and the transformation of the string
1782 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1783 * instead of the string itself. If a string changes, the magic is
1784 * cleared. The next time the locale changes, the index is incremented,
1785 * and so we know during a comparison that the transformation is not
1786 * necessarily still valid, and so is recomputed. Note that if the locale
1787 * changes enough times, the index could wrap (a U32), and it is possible
1788 * that a transformation would improperly be considered valid, leading to
1789 * an unlikely bug */
1792 if (PL_collation_name) {
1794 Safefree(PL_collation_name);
1795 PL_collation_name = NULL;
1797 PL_collation_standard = TRUE;
1798 is_standard_collation:
1799 PL_collxfrm_base = 0;
1800 PL_collxfrm_mult = 2;
1801 PL_in_utf8_COLLATE_locale = FALSE;
1802 PL_strxfrm_NUL_replacement = '\0';
1803 PL_strxfrm_max_cp = 0;
1807 /* If this is not the same locale as currently, set the new one up */
1808 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1810 Safefree(PL_collation_name);
1811 PL_collation_name = stdize_locale(savepv(newcoll));
1812 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1813 if (PL_collation_standard) {
1814 goto is_standard_collation;
1817 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1818 PL_strxfrm_NUL_replacement = '\0';
1819 PL_strxfrm_max_cp = 0;
1821 /* A locale collation definition includes primary, secondary, tertiary,
1822 * etc. weights for each character. To sort, the primary weights are
1823 * used, and only if they compare equal, then the secondary weights are
1824 * used, and only if they compare equal, then the tertiary, etc.
1826 * strxfrm() works by taking the input string, say ABC, and creating an
1827 * output transformed string consisting of first the primary weights,
1828 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1829 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1830 * may not have weights at every level. In our example, let's say B
1831 * doesn't have a tertiary weight, and A doesn't have a secondary
1832 * weight. The constructed string is then going to be
1833 * A¹B¹C¹ B²C² A³C³ ....
1834 * This has the desired effect that strcmp() will look at the secondary
1835 * or tertiary weights only if the strings compare equal at all higher
1836 * priority weights. The spaces shown here, like in
1838 * are not just for readability. In the general case, these must
1839 * actually be bytes, which we will call here 'separator weights'; and
1840 * they must be smaller than any other weight value, but since these
1841 * are C strings, only the terminating one can be a NUL (some
1842 * implementations may include a non-NUL separator weight just before
1843 * the NUL). Implementations tend to reserve 01 for the separator
1844 * weights. They are needed so that a shorter string's secondary
1845 * weights won't be misconstrued as primary weights of a longer string,
1846 * etc. By making them smaller than any other weight, the shorter
1847 * string will sort first. (Actually, if all secondary weights are
1848 * smaller than all primary ones, there is no need for a separator
1849 * weight between those two levels, etc.)
1851 * The length of the transformed string is roughly a linear function of
1852 * the input string. It's not exactly linear because some characters
1853 * don't have weights at all levels. When we call strxfrm() we have to
1854 * allocate some memory to hold the transformed string. The
1855 * calculations below try to find coefficients 'm' and 'b' for this
1856 * locale so that m*x + b equals how much space we need, given the size
1857 * of the input string in 'x'. If we calculate too small, we increase
1858 * the size as needed, and call strxfrm() again, but it is better to
1859 * get it right the first time to avoid wasted expensive string
1860 * transformations. */
1863 /* We use the string below to find how long the tranformation of it
1864 * is. Almost all locales are supersets of ASCII, or at least the
1865 * ASCII letters. We use all of them, half upper half lower,
1866 * because if we used fewer, we might hit just the ones that are
1867 * outliers in a particular locale. Most of the strings being
1868 * collated will contain a preponderance of letters, and even if
1869 * they are above-ASCII, they are likely to have the same number of
1870 * weight levels as the ASCII ones. It turns out that digits tend
1871 * to have fewer levels, and some punctuation has more, but those
1872 * are relatively sparse in text, and khw believes this gives a
1873 * reasonable result, but it could be changed if experience so
1875 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1876 char * x_longer; /* Transformed 'longer' */
1877 Size_t x_len_longer; /* Length of 'x_longer' */
1879 char * x_shorter; /* We also transform a substring of 'longer' */
1880 Size_t x_len_shorter;
1882 /* _mem_collxfrm() is used get the transformation (though here we
1883 * are interested only in its length). It is used because it has
1884 * the intelligence to handle all cases, but to work, it needs some
1885 * values of 'm' and 'b' to get it started. For the purposes of
1886 * this calculation we use a very conservative estimate of 'm' and
1887 * 'b'. This assumes a weight can be multiple bytes, enough to
1888 * hold any UV on the platform, and there are 5 levels, 4 weight
1889 * bytes, and a trailing NUL. */
1890 PL_collxfrm_base = 5;
1891 PL_collxfrm_mult = 5 * sizeof(UV);
1893 /* Find out how long the transformation really is */
1894 x_longer = _mem_collxfrm(longer,
1898 /* We avoid converting to UTF-8 in the
1899 * called function by telling it the
1900 * string is in UTF-8 if the locale is a
1901 * UTF-8 one. Since the string passed
1902 * here is invariant under UTF-8, we can
1903 * claim it's UTF-8 even though it isn't.
1905 PL_in_utf8_COLLATE_locale);
1908 /* Find out how long the transformation of a substring of 'longer'
1909 * is. Together the lengths of these transformations are
1910 * sufficient to calculate 'm' and 'b'. The substring is all of
1911 * 'longer' except the first character. This minimizes the chances
1912 * of being swayed by outliers */
1913 x_shorter = _mem_collxfrm(longer + 1,
1916 PL_in_utf8_COLLATE_locale);
1917 Safefree(x_shorter);
1919 /* If the results are nonsensical for this simple test, the whole
1920 * locale definition is suspect. Mark it so that locale collation
1921 * is not active at all for it. XXX Should we warn? */
1922 if ( x_len_shorter == 0
1923 || x_len_longer == 0
1924 || x_len_shorter >= x_len_longer)
1926 PL_collxfrm_mult = 0;
1927 PL_collxfrm_base = 0;
1930 SSize_t base; /* Temporary */
1932 /* We have both: m * strlen(longer) + b = x_len_longer
1933 * m * strlen(shorter) + b = x_len_shorter;
1934 * subtracting yields:
1935 * m * (strlen(longer) - strlen(shorter))
1936 * = x_len_longer - x_len_shorter
1937 * But we have set things up so that 'shorter' is 1 byte smaller
1938 * than 'longer'. Hence:
1939 * m = x_len_longer - x_len_shorter
1941 * But if something went wrong, make sure the multiplier is at
1944 if (x_len_longer > x_len_shorter) {
1945 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1948 PL_collxfrm_mult = 1;
1953 * but in case something has gone wrong, make sure it is
1955 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1960 /* Add 1 for the trailing NUL */
1961 PL_collxfrm_base = base + 1;
1966 if (DEBUG_L_TEST || debug_initialization) {
1967 PerlIO_printf(Perl_debug_log,
1968 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1970 " collate multipler=%zu, collate base=%zu\n",
1972 PL_in_utf8_COLLATE_locale,
1973 x_len_shorter, x_len_longer,
1974 PL_collxfrm_mult, PL_collxfrm_base);
1981 #endif /* USE_LOCALE_COLLATE */
1988 S_win32_setlocale(pTHX_ int category, const char* locale)
1990 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1991 * difference between the two unless the input locale is "", which normally
1992 * means on Windows to get the machine default, which is set via the
1993 * computer's "Regional and Language Options" (or its current equivalent).
1994 * In POSIX, it instead means to find the locale from the user's
1995 * environment. This routine changes the Windows behavior to first look in
1996 * the environment, and, if anything is found, use that instead of going to
1997 * the machine default. If there is no environment override, the machine
1998 * default is used, by calling the real setlocale() with "".
2000 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2001 * use the particular category's variable if set; otherwise to use the LANG
2004 bool override_LC_ALL = FALSE;
2008 if (locale && strEQ(locale, "")) {
2012 locale = PerlEnv_getenv("LC_ALL");
2014 if (category == LC_ALL) {
2015 override_LC_ALL = TRUE;
2021 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2022 if (category == categories[i]) {
2023 locale = PerlEnv_getenv(category_names[i]);
2028 locale = PerlEnv_getenv("LANG");
2044 result = setlocale(category, locale);
2045 DEBUG_L(STMT_START {
2047 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2048 setlocale_debug_string(category, locale, result));
2052 if (! override_LC_ALL) {
2056 /* Here the input category was LC_ALL, and we have set it to what is in the
2057 * LANG variable or the system default if there is no LANG. But these have
2058 * lower priority than the other LC_foo variables, so override it for each
2059 * one that is set. (If they are set to "", it means to use the same thing
2060 * we just set LC_ALL to, so can skip) */
2062 for (i = 0; i < LC_ALL_INDEX; i++) {
2063 result = PerlEnv_getenv(category_names[i]);
2064 if (result && strNE(result, "")) {
2065 setlocale(categories[i], result);
2066 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2068 setlocale_debug_string(categories[i], result, "not captured")));
2072 result = setlocale(LC_ALL, NULL);
2073 DEBUG_L(STMT_START {
2075 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2077 setlocale_debug_string(LC_ALL, NULL, result));
2088 =head1 Locale-related functions and macros
2090 =for apidoc Perl_setlocale
2092 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2093 taking the same parameters, and returning the same information, except that it
2094 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2095 instead return C<C> if the underlying locale has a non-dot decimal point
2096 character, or a non-empty thousands separator for displaying floating point
2097 numbers. This is because perl keeps that locale category such that it has a
2098 dot and empty separator, changing the locale briefly during the operations
2099 where the underlying one is required. C<Perl_setlocale> knows about this, and
2100 compensates; regular C<setlocale> doesn't.
2102 Another reason it isn't completely a drop-in replacement is that it is
2103 declared to return S<C<const char *>>, whereas the system setlocale omits the
2104 C<const> (presumably because its API was specified long ago, and can't be
2105 updated; it is illegal to change the information C<setlocale> returns; doing
2106 so leads to segfaults.)
2108 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2109 C<setlocale> can be completely ineffective on some platforms under some
2112 C<Perl_setlocale> should not be used to change the locale except on systems
2113 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2114 the system C<setlocale()> is ineffective, returning the wrong information, and
2115 failing to actually change the locale. C<Perl_setlocale>, however works
2116 properly in all circumstances.
2118 The return points to a per-thread static buffer, which is overwritten the next
2119 time C<Perl_setlocale> is called from the same thread.
2126 Perl_setlocale(const int category, const char * locale)
2128 /* This wraps POSIX::setlocale() */
2130 const char * retval;
2131 const char * newlocale;
2133 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2136 #ifdef USE_LOCALE_NUMERIC
2138 /* A NULL locale means only query what the current one is. We have the
2139 * LC_NUMERIC name saved, because we are normally switched into the C
2140 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2141 * the correct results. All other categories don't require special
2143 if (locale == NULL) {
2144 if (category == LC_NUMERIC) {
2146 /* We don't have to copy this return value, as it is a per-thread
2147 * variable, and won't change until a future setlocale */
2148 return PL_numeric_name;
2153 else if (category == LC_ALL) {
2154 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2163 retval = save_to_buffer(do_setlocale_r(category, locale),
2164 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2167 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2169 if (locale == NULL && category == LC_ALL) {
2170 RESTORE_LC_NUMERIC();
2175 DEBUG_L(PerlIO_printf(Perl_debug_log,
2176 "%s:%d: %s\n", __FILE__, __LINE__,
2177 setlocale_debug_string(category, locale, retval)));
2185 /* If locale == NULL, we are just querying the state */
2186 if (locale == NULL) {
2190 /* Now that have switched locales, we have to update our records to
2195 #ifdef USE_LOCALE_CTYPE
2202 #ifdef USE_LOCALE_COLLATE
2205 new_collate(retval);
2209 #ifdef USE_LOCALE_NUMERIC
2212 new_numeric(retval);
2220 /* LC_ALL updates all the things we care about. The values may not
2221 * be the same as 'retval', as the locale "" may have set things
2224 # ifdef USE_LOCALE_CTYPE
2226 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2227 new_ctype(newlocale);
2228 Safefree(newlocale);
2230 # endif /* USE_LOCALE_CTYPE */
2231 # ifdef USE_LOCALE_COLLATE
2233 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2234 new_collate(newlocale);
2235 Safefree(newlocale);
2238 # ifdef USE_LOCALE_NUMERIC
2240 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2241 new_numeric(newlocale);
2242 Safefree(newlocale);
2244 # endif /* USE_LOCALE_NUMERIC */
2255 PERL_STATIC_INLINE const char *
2256 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2258 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2259 * growing it if necessary */
2263 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2269 string_size = strlen(string) + offset + 1;
2271 if (*buf_size == 0) {
2272 Newx(*buf, string_size, char);
2273 *buf_size = string_size;
2275 else if (string_size > *buf_size) {
2276 Renew(*buf, string_size, char);
2277 *buf_size = string_size;
2280 Copy(string, *buf + offset, string_size - offset, char);
2286 =for apidoc Perl_langinfo
2288 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2289 taking the same C<item> parameter values, and returning the same information.
2290 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2291 of Perl's locale handling from your code, and can be used on systems that lack
2292 a native C<nl_langinfo>.
2300 The reason it isn't quite a drop-in replacement is actually an advantage. The
2301 only difference is that it returns S<C<const char *>>, whereas plain
2302 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2303 forbidden to write into the buffer. By declaring this C<const>, the compiler
2304 enforces this restriction, so if it is violated, you know at compilation time,
2305 rather than getting segfaults at runtime.
2309 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2310 without you having to write extra code. The reason for the extra code would be
2311 because these are from the C<LC_NUMERIC> locale category, which is normally
2312 kept set by Perl so that the radix is a dot, and the separator is the empty
2313 string, no matter what the underlying locale is supposed to be, and so to get
2314 the expected results, you have to temporarily toggle into the underlying
2315 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2316 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2317 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2318 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2319 (decimal point) character to be a dot.)
2323 The system function it replaces can have its static return buffer trashed,
2324 not only by a subesequent call to that function, but by a C<freelocale>,
2325 C<setlocale>, or other locale change. The returned buffer of this function is
2326 not changed until the next call to it, so the buffer is never in a trashed
2331 Its return buffer is per-thread, so it also is never overwritten by a call to
2332 this function from another thread; unlike the function it replaces.
2336 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2337 as Windows, hence makes your code more portable. Of the fifty-some possible
2338 items specified by the POSIX 2008 standard,
2339 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2340 only one is completely unimplemented, though on non-Windows platforms, another
2341 significant one is also not implemented). It uses various techniques to
2342 recover the other items, including calling C<L<localeconv(3)>>, and
2343 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2344 available. Later C<strftime()> versions have additional capabilities; C<""> is
2345 returned for those not available on your system.
2347 It is important to note that when called with an item that is recovered by
2348 using C<localeconv>, the buffer from any previous explicit call to
2349 C<localeconv> will be overwritten. This means you must save that buffer's
2350 contents if you need to access them after a call to this function. (But note
2351 that you might not want to be using C<localeconv()> directly anyway, because of
2352 issues like the ones listed in the second item of this list (above) for
2353 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2354 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2357 The details for those items which may deviate from what this emulation returns
2358 and what a native C<nl_langinfo()> would return are specified in
2363 When using C<Perl_langinfo> on systems that don't have a native
2364 C<nl_langinfo()>, you must
2366 #include "perl_langinfo.h"
2368 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2369 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2370 C<langinfo.h> would try to import into the namespace for code that doesn't need
2373 The original impetus for C<Perl_langinfo()> was so that code that needs to
2374 find out the current currency symbol, floating point radix character, or digit
2375 grouping separator can use, on all systems, the simpler and more
2376 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2377 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2378 is better to use the methods given in L<perlcall> to call
2379 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2386 #ifdef HAS_NL_LANGINFO
2387 Perl_langinfo(const nl_item item)
2389 Perl_langinfo(const int item)
2392 return my_nl_langinfo(item, TRUE);
2396 #ifdef HAS_NL_LANGINFO
2397 S_my_nl_langinfo(const nl_item item, bool toggle)
2399 S_my_nl_langinfo(const int item, bool toggle)
2403 const char * retval;
2405 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2406 * two items, and only if not already there */
2407 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2408 || PL_numeric_underlying))
2413 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2414 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2415 || ! defined(HAS_POSIX_2008_LOCALE) \
2416 || ! defined(DUPLOCALE)
2418 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2419 * for those items dependent on it. This must be copied to a buffer before
2420 * switching back, as some systems destroy the buffer when setlocale() is
2424 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2427 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2430 LOCALE_LOCK; /* Prevent interference from another thread executing
2431 this code section (the only call to nl_langinfo in
2435 /* Copy to a per-thread buffer, which is also one that won't be
2436 * destroyed by a subsequent setlocale(), such as the
2437 * RESTORE_LC_NUMERIC may do just below. */
2438 retval = save_to_buffer(nl_langinfo(item),
2439 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2444 RESTORE_LC_NUMERIC();
2448 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2451 bool do_free = FALSE;
2452 locale_t cur = uselocale((locale_t) 0);
2454 if (cur == LC_GLOBAL_LOCALE) {
2455 cur = duplocale(LC_GLOBAL_LOCALE);
2460 if (PL_underlying_numeric_obj) {
2461 cur = PL_underlying_numeric_obj;
2464 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2469 /* We have to save it to a buffer, because the freelocale() just below
2470 * can invalidate the internal one */
2471 retval = save_to_buffer(nl_langinfo_l(item, cur),
2472 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2481 if (strEQ(retval, "")) {
2482 if (item == YESSTR) {
2485 if (item == NOSTR) {
2492 #else /* Below, emulate nl_langinfo as best we can */
2496 # ifdef HAS_LOCALECONV
2498 const struct lconv* lc;
2500 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2502 # ifdef TS_W32_BROKEN_LOCALECONV
2504 const char * save_global;
2505 const char * save_thread;
2513 # ifdef HAS_STRFTIME
2516 bool return_format = FALSE; /* Return the %format, not the value */
2517 const char * format;
2521 /* We copy the results to a per-thread buffer, even if not
2522 * multi-threaded. This is in part to simplify this code, and partly
2523 * because we need a buffer anyway for strftime(), and partly because a
2524 * call of localeconv() could otherwise wipe out the buffer, and the
2525 * programmer would not be expecting this, as this is a nl_langinfo()
2526 * substitute after all, so s/he might be thinking their localeconv()
2527 * is safe until another localeconv() call. */
2532 /* This is unimplemented */
2533 case ERA: /* For use with strftime() %E modifier */
2538 /* We use only an English set, since we don't know any more */
2539 case YESEXPR: return "^[+1yY]";
2540 case YESSTR: return "yes";
2541 case NOEXPR: return "^[-0nN]";
2542 case NOSTR: return "no";
2548 /* On non-windows, this is unimplemented, in part because of
2549 * inconsistencies between vendors. The Darwin native
2550 * nl_langinfo() implementation simply looks at everything past
2551 * any dot in the name, but that doesn't work for other
2552 * vendors. Many Linux locales that don't have UTF-8 in their
2553 * names really are UTF-8, for example; z/OS locales that do
2554 * have UTF-8 in their names, aren't really UTF-8 */
2559 { /* But on Windows, the name does seem to be consistent, so
2564 const char * name = my_setlocale(LC_CTYPE, NULL);
2566 if (isNAME_C_OR_POSIX(name)) {
2567 return "ANSI_X3.4-1968";
2570 /* Find the dot in the locale name */
2571 first = (const char *) strchr(name, '.');
2577 /* Look at everything past the dot */
2582 if (! isDIGIT(*p)) {
2589 /* Here everything past the dot is a digit. Treat it as a
2591 (void) save_to_buffer("CP", &PL_langinfo_buf,
2592 &PL_langinfo_bufsize, 0);
2593 offset = STRLENs("CP");
2597 retval = save_to_buffer(first, &PL_langinfo_buf,
2598 &PL_langinfo_bufsize, offset);
2604 # ifdef HAS_LOCALECONV
2608 /* We don't bother with localeconv_l() because any system that
2609 * has it is likely to also have nl_langinfo() */
2611 LOCALE_LOCK_V; /* Prevent interference with other threads
2612 using localeconv() */
2614 # ifdef TS_W32_BROKEN_LOCALECONV
2616 /* This is a workaround for a Windows bug prior to VS 15.
2617 * What we do here is, while locked, switch to the global
2618 * locale so localeconv() works; then switch back just before
2619 * the unlock. This can screw things up if some thread is
2620 * already using the global locale while assuming no other is.
2621 * A different workaround would be to call GetCurrencyFormat on
2622 * a known value, and parse it; patches welcome
2624 * We have to use LC_ALL instead of LC_MONETARY because of
2625 * another bug in Windows */
2627 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2628 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2629 save_global= savepv(my_setlocale(LC_ALL, NULL));
2630 my_setlocale(LC_ALL, save_thread);
2636 || ! lc->currency_symbol
2637 || strEQ("", lc->currency_symbol))
2643 /* Leave the first spot empty to be filled in below */
2644 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2645 &PL_langinfo_bufsize, 1);
2646 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2647 { /* khw couldn't figure out how the localedef specifications
2648 would show that the $ should replace the radix; this is
2649 just a guess as to how it might work.*/
2650 PL_langinfo_buf[0] = '.';
2652 else if (lc->p_cs_precedes) {
2653 PL_langinfo_buf[0] = '-';
2656 PL_langinfo_buf[0] = '+';
2659 # ifdef TS_W32_BROKEN_LOCALECONV
2661 my_setlocale(LC_ALL, save_global);
2662 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2663 my_setlocale(LC_ALL, save_thread);
2664 Safefree(save_global);
2665 Safefree(save_thread);
2672 # ifdef TS_W32_BROKEN_LOCALECONV
2676 /* For this, we output a known simple floating point number to
2677 * a buffer, and parse it, looking for the radix */
2680 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2683 if (PL_langinfo_bufsize < 10) {
2684 PL_langinfo_bufsize = 10;
2685 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2688 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2690 if (needed_size >= (int) PL_langinfo_bufsize) {
2691 PL_langinfo_bufsize = needed_size + 1;
2692 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2693 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2695 assert(needed_size < (int) PL_langinfo_bufsize);
2698 ptr = PL_langinfo_buf;
2699 e = PL_langinfo_buf + PL_langinfo_bufsize;
2702 while (ptr < e && *ptr != '1') {
2709 while (ptr < e && *ptr != '5') {
2713 /* Everything in between is the radix string */
2715 PL_langinfo_buf[0] = '?';
2716 PL_langinfo_buf[1] = '\0';
2720 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2724 RESTORE_LC_NUMERIC();
2727 retval = PL_langinfo_buf;
2732 case RADIXCHAR: /* No special handling needed */
2739 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2742 LOCALE_LOCK_V; /* Prevent interference with other threads
2743 using localeconv() */
2745 # ifdef TS_W32_BROKEN_LOCALECONV
2747 /* This should only be for the thousands separator. A
2748 * different work around would be to use GetNumberFormat on a
2749 * known value and parse the result to find the separator */
2750 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2751 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2752 save_global = savepv(my_setlocale(LC_ALL, NULL));
2753 my_setlocale(LC_ALL, save_thread);
2755 /* This is the start of code that for broken Windows replaces
2756 * the above and below code, and instead calls
2757 * GetNumberFormat() and then would parse that to find the
2758 * thousands separator. It needs to handle UTF-16 vs -8
2761 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2762 DEBUG_L(PerlIO_printf(Perl_debug_log,
2763 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2764 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2774 temp = (item == RADIXCHAR)
2776 : lc->thousands_sep;
2782 retval = save_to_buffer(temp, &PL_langinfo_buf,
2783 &PL_langinfo_bufsize, 0);
2785 # ifdef TS_W32_BROKEN_LOCALECONV
2787 my_setlocale(LC_ALL, save_global);
2788 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2789 my_setlocale(LC_ALL, save_thread);
2790 Safefree(save_global);
2791 Safefree(save_thread);
2798 RESTORE_LC_NUMERIC();
2804 # ifdef HAS_STRFTIME
2806 /* These are defined by C89, so we assume that strftime supports
2807 * them, and so are returned unconditionally; they may not be what
2808 * the locale actually says, but should give good enough results
2809 * for someone using them as formats (as opposed to trying to parse
2810 * them to figure out what the locale says). The other format
2811 * items are actually tested to verify they work on the platform */
2812 case D_FMT: return "%x";
2813 case T_FMT: return "%X";
2814 case D_T_FMT: return "%c";
2816 /* These formats are only available in later strfmtime's */
2817 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2819 /* The rest can be gotten from most versions of strftime(). */
2820 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2821 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2823 case AM_STR: case PM_STR:
2824 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2825 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2826 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2827 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2828 case DAY_5: case DAY_6: case DAY_7:
2829 case MON_1: case MON_2: case MON_3: case MON_4:
2830 case MON_5: case MON_6: case MON_7: case MON_8:
2831 case MON_9: case MON_10: case MON_11: case MON_12:
2835 init_tm(&tm); /* Precaution against core dumps */
2839 tm.tm_year = 2017 - 1900;
2846 "panic: %s: %d: switch case: %d problem",
2847 __FILE__, __LINE__, item);
2848 NOT_REACHED; /* NOTREACHED */
2850 case PM_STR: tm.tm_hour = 18;
2855 case ABDAY_7: tm.tm_wday++;
2856 case ABDAY_6: tm.tm_wday++;
2857 case ABDAY_5: tm.tm_wday++;
2858 case ABDAY_4: tm.tm_wday++;
2859 case ABDAY_3: tm.tm_wday++;
2860 case ABDAY_2: tm.tm_wday++;
2865 case DAY_7: tm.tm_wday++;
2866 case DAY_6: tm.tm_wday++;
2867 case DAY_5: tm.tm_wday++;
2868 case DAY_4: tm.tm_wday++;
2869 case DAY_3: tm.tm_wday++;
2870 case DAY_2: tm.tm_wday++;
2875 case ABMON_12: tm.tm_mon++;
2876 case ABMON_11: tm.tm_mon++;
2877 case ABMON_10: tm.tm_mon++;
2878 case ABMON_9: tm.tm_mon++;
2879 case ABMON_8: tm.tm_mon++;
2880 case ABMON_7: tm.tm_mon++;
2881 case ABMON_6: tm.tm_mon++;
2882 case ABMON_5: tm.tm_mon++;
2883 case ABMON_4: tm.tm_mon++;
2884 case ABMON_3: tm.tm_mon++;
2885 case ABMON_2: tm.tm_mon++;
2890 case MON_12: tm.tm_mon++;
2891 case MON_11: tm.tm_mon++;
2892 case MON_10: tm.tm_mon++;
2893 case MON_9: tm.tm_mon++;
2894 case MON_8: tm.tm_mon++;
2895 case MON_7: tm.tm_mon++;
2896 case MON_6: tm.tm_mon++;
2897 case MON_5: tm.tm_mon++;
2898 case MON_4: tm.tm_mon++;
2899 case MON_3: tm.tm_mon++;
2900 case MON_2: tm.tm_mon++;
2907 return_format = TRUE;
2912 return_format = TRUE;
2917 return_format = TRUE;
2922 return_format = TRUE;
2927 format = "%Ow"; /* Find the alternate digit for 0 */
2931 /* We can't use my_strftime() because it doesn't look at
2933 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2936 /* A zero return means one of:
2937 * a) there wasn't enough space in PL_langinfo_buf
2938 * b) the format, like a plain %p, returns empty
2939 * c) it was an illegal format, though some
2940 * implementations of strftime will just return the
2941 * illegal format as a plain character sequence.
2943 * To quickly test for case 'b)', try again but precede
2944 * the format with a plain character. If that result is
2945 * still empty, the problem is either 'a)' or 'c)' */
2947 Size_t format_size = strlen(format) + 1;
2948 Size_t mod_size = format_size + 1;
2952 Newx(mod_format, mod_size, char);
2953 Newx(temp_result, PL_langinfo_bufsize, char);
2955 my_strlcpy(mod_format + 1, format, mod_size);
2956 len = strftime(temp_result,
2957 PL_langinfo_bufsize,
2959 Safefree(mod_format);
2960 Safefree(temp_result);
2962 /* If 'len' is non-zero, it means that we had a case like
2963 * %p which means the current locale doesn't use a.m. or
2964 * p.m., and that is valid */
2967 /* Here, still didn't work. If we get well beyond a
2968 * reasonable size, bail out to prevent an infinite
2971 if (PL_langinfo_bufsize > 100 * format_size) {
2972 *PL_langinfo_buf = '\0';
2975 /* Double the buffer size to retry; Add 1 in case
2976 * original was 0, so we aren't stuck at 0. */
2977 PL_langinfo_bufsize *= 2;
2978 PL_langinfo_bufsize++;
2979 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2987 /* Here, we got a result.
2989 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2990 * alternate format for wday 0. If the value is the same as
2991 * the normal 0, there isn't an alternate, so clear the buffer.
2993 if ( item == ALT_DIGITS
2994 && strEQ(PL_langinfo_buf, "0"))
2996 *PL_langinfo_buf = '\0';
2999 /* ALT_DIGITS is problematic. Experiments on it showed that
3000 * strftime() did not always work properly when going from
3001 * alt-9 to alt-10. Only a few locales have this item defined,
3002 * and in all of them on Linux that khw was able to find,
3003 * nl_langinfo() merely returned the alt-0 character, possibly
3004 * doubled. Most Unicode digits are in blocks of 10
3005 * consecutive code points, so that is sufficient information
3006 * for those scripts, as we can infer alt-1, alt-2, .... But
3007 * for a Japanese locale, a CJK ideographic 0 is returned, and
3008 * the CJK digits are not in code point order, so you can't
3009 * really infer anything. The localedef for this locale did
3010 * specify the succeeding digits, so that strftime() works
3011 * properly on them, without needing to infer anything. But
3012 * the nl_langinfo() return did not give sufficient information
3013 * for the caller to understand what's going on. So until
3014 * there is evidence that it should work differently, this
3015 * returns the alt-0 string for ALT_DIGITS.
3017 * wday was chosen because its range is all a single digit.
3018 * Things like tm_sec have two digits as the minimum: '00' */
3022 retval = PL_langinfo_buf;
3024 /* If to return the format, not the value, overwrite the buffer
3025 * with it. But some strftime()s will keep the original format
3026 * if illegal, so change those to "" */
3027 if (return_format) {
3028 if (strEQ(PL_langinfo_buf, format)) {
3029 *PL_langinfo_buf = '\0';
3032 retval = save_to_buffer(format, &PL_langinfo_buf,
3033 &PL_langinfo_bufsize, 0);
3051 * Initialize locale awareness.
3054 Perl_init_i18nl10n(pTHX_ int printwarn)
3058 * 0 if not to output warning when setup locale is bad
3059 * 1 if to output warning based on value of PERL_BADLANG
3060 * >1 if to output regardless of PERL_BADLANG
3063 * 1 = set ok or not applicable,
3064 * 0 = fallback to a locale of lower priority
3065 * -1 = fallback to all locales failed, not even to the C locale
3067 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3068 * set, debugging information is output.
3070 * This looks more complicated than it is, mainly due to the #ifdefs.
3072 * We try to set LC_ALL to the value determined by the environment. If
3073 * there is no LC_ALL on this platform, we try the individual categories we
3074 * know about. If this works, we are done.
3076 * But if it doesn't work, we have to do something else. We search the
3077 * environment variables ourselves instead of relying on the system to do
3078 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3079 * think there is one), and the ultimate fallback "C". This is all done in
3080 * the same loop as above to avoid duplicating code, but it makes things
3081 * more complex. The 'trial_locales' array is initialized with just one
3082 * element; it causes the behavior described in the paragraph above this to
3083 * happen. If that fails, we add elements to 'trial_locales', and do extra
3084 * loop iterations to cause the behavior described in this paragraph.
3086 * On Ultrix, the locale MUST come from the environment, so there is
3087 * preliminary code to set it. I (khw) am not sure that it is necessary,
3088 * and that this couldn't be folded into the loop, but barring any real
3089 * platforms to test on, it's staying as-is
3091 * A slight complication is that in embedded Perls, the locale may already
3092 * be set-up, and we don't want to get it from the normal environment
3093 * variables. This is handled by having a special environment variable
3094 * indicate we're in this situation. We simply set setlocale's 2nd
3095 * parameter to be a NULL instead of "". That indicates to setlocale that
3096 * it is not to change anything, but to return the current value,
3097 * effectively initializing perl's db to what the locale already is.
3099 * We play the same trick with NULL if a LC_ALL succeeds. We call
3100 * setlocale() on the individual categores with NULL to get their existing
3101 * values for our db, instead of trying to change them.
3108 PERL_UNUSED_ARG(printwarn);
3110 #else /* USE_LOCALE */
3113 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3117 /* NULL uses the existing already set up locale */
3118 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3121 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3122 unsigned int trial_locales_count;
3123 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3124 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3125 bool setlocale_failure = FALSE;
3128 /* A later getenv() could zap this, so only use here */
3129 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3131 const bool locwarn = (printwarn > 1
3133 && ( ! bad_lang_use_once
3135 /* disallow with "" or "0" */
3137 && strNE("0", bad_lang_use_once)))));
3139 /* setlocale() return vals; not copied so must be looked at immediately */
3140 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3142 /* current locale for given category; should have been copied so aren't
3144 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3148 /* In some systems you can find out the system default locale
3149 * and use that as the fallback locale. */
3150 # define SYSTEM_DEFAULT_LOCALE
3152 # ifdef SYSTEM_DEFAULT_LOCALE
3154 const char *system_default_locale = NULL;
3159 # define DEBUG_LOCALE_INIT(a,b,c)
3162 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3164 # define DEBUG_LOCALE_INIT(category, locale, result) \
3166 if (debug_initialization) { \
3167 PerlIO_printf(Perl_debug_log, \
3169 __FILE__, __LINE__, \
3170 setlocale_debug_string(category, \
3176 /* Make sure the parallel arrays are properly set up */
3177 # ifdef USE_LOCALE_NUMERIC
3178 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3179 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3180 # ifdef USE_POSIX_2008_LOCALE
3181 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3184 # ifdef USE_LOCALE_CTYPE
3185 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3186 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3187 # ifdef USE_POSIX_2008_LOCALE
3188 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3191 # ifdef USE_LOCALE_COLLATE
3192 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3193 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3194 # ifdef USE_POSIX_2008_LOCALE
3195 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3198 # ifdef USE_LOCALE_TIME
3199 assert(categories[LC_TIME_INDEX] == LC_TIME);
3200 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3201 # ifdef USE_POSIX_2008_LOCALE
3202 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3205 # ifdef USE_LOCALE_MESSAGES
3206 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3207 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3208 # ifdef USE_POSIX_2008_LOCALE
3209 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3212 # ifdef USE_LOCALE_MONETARY
3213 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3214 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3215 # ifdef USE_POSIX_2008_LOCALE
3216 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3219 # ifdef USE_LOCALE_ADDRESS
3220 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3221 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3222 # ifdef USE_POSIX_2008_LOCALE
3223 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3226 # ifdef USE_LOCALE_IDENTIFICATION
3227 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3228 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3229 # ifdef USE_POSIX_2008_LOCALE
3230 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3233 # ifdef USE_LOCALE_MEASUREMENT
3234 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3235 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3236 # ifdef USE_POSIX_2008_LOCALE
3237 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3240 # ifdef USE_LOCALE_PAPER
3241 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3242 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3243 # ifdef USE_POSIX_2008_LOCALE
3244 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3247 # ifdef USE_LOCALE_TELEPHONE
3248 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3249 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3250 # ifdef USE_POSIX_2008_LOCALE
3251 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3255 assert(categories[LC_ALL_INDEX] == LC_ALL);
3256 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3257 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3258 # ifdef USE_POSIX_2008_LOCALE
3259 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3262 # endif /* DEBUGGING */
3264 /* Initialize the cache of the program's UTF-8ness for the always known
3265 * locales C and POSIX */
3266 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3267 sizeof(PL_locale_utf8ness));
3269 # ifdef USE_THREAD_SAFE_LOCALE
3272 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3276 # ifdef USE_POSIX_2008_LOCALE
3278 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3279 if (! PL_C_locale_obj) {
3280 Perl_croak_nocontext(
3281 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3283 if (DEBUG_Lv_TEST || debug_initialization) {
3284 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3289 PL_numeric_radix_sv = newSVpvs(".");
3291 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3293 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3294 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3297 # ifdef LOCALE_ENVIRON_REQUIRED
3300 * Ultrix setlocale(..., "") fails if there are no environment
3301 * variables from which to get a locale name.
3305 # error Ultrix without LC_ALL not implemented
3311 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3312 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3313 if (sl_result[LC_ALL_INDEX])
3316 setlocale_failure = TRUE;
3318 if (! setlocale_failure) {
3319 const char * locale_param;
3320 for (i = 0; i < LC_ALL_INDEX; i++) {
3321 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3324 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3325 if (! sl_result[i]) {
3326 setlocale_failure = TRUE;
3328 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3333 # endif /* LC_ALL */
3334 # endif /* LOCALE_ENVIRON_REQUIRED */
3336 /* We try each locale in the list until we get one that works, or exhaust
3337 * the list. Normally the loop is executed just once. But if setting the
3338 * locale fails, inside the loop we add fallback trials to the array and so
3339 * will execute the loop multiple times */
3340 trial_locales[0] = setlocale_init;
3341 trial_locales_count = 1;
3343 for (i= 0; i < trial_locales_count; i++) {
3344 const char * trial_locale = trial_locales[i];
3348 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3349 * when i==0, but I (khw) don't think that behavior makes much
3351 setlocale_failure = FALSE;
3353 # ifdef SYSTEM_DEFAULT_LOCALE
3354 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3356 /* On Windows machines, an entry of "" after the 0th means to use
3357 * the system default locale, which we now proceed to get. */
3358 if (strEQ(trial_locale, "")) {
3361 /* Note that this may change the locale, but we are going to do
3362 * that anyway just below */
3363 system_default_locale = do_setlocale_c(LC_ALL, "");
3364 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3366 /* Skip if invalid or if it's already on the list of locales to
3368 if (! system_default_locale) {
3369 goto next_iteration;
3371 for (j = 0; j < trial_locales_count; j++) {
3372 if (strEQ(system_default_locale, trial_locales[j])) {
3373 goto next_iteration;
3377 trial_locale = system_default_locale;
3380 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3382 # endif /* SYSTEM_DEFAULT_LOCALE */
3388 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3389 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3390 if (! sl_result[LC_ALL_INDEX]) {
3391 setlocale_failure = TRUE;
3394 /* Since LC_ALL succeeded, it should have changed all the other
3395 * categories it can to its value; so we massage things so that the
3396 * setlocales below just return their category's current values.
3397 * This adequately handles the case in NetBSD where LC_COLLATE may
3398 * not be defined for a locale, and setting it individually will
3399 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3400 * the POSIX locale. */
3401 trial_locale = NULL;
3404 # endif /* LC_ALL */
3406 if (! setlocale_failure) {
3408 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3410 = savepv(do_setlocale_r(categories[j], trial_locale));
3411 if (! curlocales[j]) {
3412 setlocale_failure = TRUE;
3414 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3417 if (! setlocale_failure) { /* All succeeded */
3418 break; /* Exit trial_locales loop */
3422 /* Here, something failed; will need to try a fallback. */
3428 if (locwarn) { /* Output failure info only on the first one */
3432 PerlIO_printf(Perl_error_log,
3433 "perl: warning: Setting locale failed.\n");
3435 # else /* !LC_ALL */
3437 PerlIO_printf(Perl_error_log,
3438 "perl: warning: Setting locale failed for the categories:\n\t");
3440 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3441 if (! curlocales[j]) {
3442 PerlIO_printf(Perl_error_log, category_names[j]);
3445 Safefree(curlocales[j]);
3449 # endif /* LC_ALL */
3451 PerlIO_printf(Perl_error_log,
3452 "perl: warning: Please check that your locale settings:\n");
3456 PerlIO_printf(Perl_error_log,
3457 "\tLANGUAGE = %c%s%c,\n",
3458 language ? '"' : '(',
3459 language ? language : "unset",
3460 language ? '"' : ')');
3463 PerlIO_printf(Perl_error_log,
3464 "\tLC_ALL = %c%s%c,\n",
3466 lc_all ? lc_all : "unset",
3467 lc_all ? '"' : ')');
3469 # if defined(USE_ENVIRON_ARRAY)
3474 /* Look through the environment for any variables of the
3475 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3476 * already handled above. These are assumed to be locale
3477 * settings. Output them and their values. */
3478 for (e = environ; *e; e++) {
3479 const STRLEN prefix_len = sizeof("LC_") - 1;
3482 if ( strBEGINs(*e, "LC_")
3483 && ! strBEGINs(*e, "LC_ALL=")
3484 && (uppers_len = strspn(*e + prefix_len,
3485 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3486 && ((*e)[prefix_len + uppers_len] == '='))
3488 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3489 (int) (prefix_len + uppers_len), *e,
3490 *e + prefix_len + uppers_len + 1);
3497 PerlIO_printf(Perl_error_log,
3498 "\t(possibly more locale environment variables)\n");
3502 PerlIO_printf(Perl_error_log,
3503 "\tLANG = %c%s%c\n",
3505 lang ? lang : "unset",
3508 PerlIO_printf(Perl_error_log,
3509 " are supported and installed on your system.\n");
3512 /* Calculate what fallback locales to try. We have avoided this
3513 * until we have to, because failure is quite unlikely. This will
3514 * usually change the upper bound of the loop we are in.
3516 * Since the system's default way of setting the locale has not
3517 * found one that works, We use Perl's defined ordering: LC_ALL,
3518 * LANG, and the C locale. We don't try the same locale twice, so
3519 * don't add to the list if already there. (On POSIX systems, the
3520 * LC_ALL element will likely be a repeat of the 0th element "",
3521 * but there's no harm done by doing it explicitly.
3523 * Note that this tries the LC_ALL environment variable even on
3524 * systems which have no LC_ALL locale setting. This may or may
3525 * not have been originally intentional, but there's no real need
3526 * to change the behavior. */
3528 for (j = 0; j < trial_locales_count; j++) {
3529 if (strEQ(lc_all, trial_locales[j])) {
3533 trial_locales[trial_locales_count++] = lc_all;
3538 for (j = 0; j < trial_locales_count; j++) {
3539 if (strEQ(lang, trial_locales[j])) {
3543 trial_locales[trial_locales_count++] = lang;
3547 # if defined(WIN32) && defined(LC_ALL)
3549 /* For Windows, we also try the system default locale before "C".
3550 * (If there exists a Windows without LC_ALL we skip this because
3551 * it gets too complicated. For those, the "C" is the next
3552 * fallback possibility). The "" is the same as the 0th element of
3553 * the array, but the code at the loop above knows to treat it
3554 * differently when not the 0th */
3555 trial_locales[trial_locales_count++] = "";
3559 for (j = 0; j < trial_locales_count; j++) {
3560 if (strEQ("C", trial_locales[j])) {
3564 trial_locales[trial_locales_count++] = "C";
3567 } /* end of first time through the loop */
3575 } /* end of looping through the trial locales */
3577 if (ok < 1) { /* If we tried to fallback */
3579 if (! setlocale_failure) { /* fallback succeeded */
3580 msg = "Falling back to";
3582 else { /* fallback failed */
3585 /* We dropped off the end of the loop, so have to decrement i to
3586 * get back to the value the last time through */
3590 msg = "Failed to fall back to";
3592 /* To continue, we should use whatever values we've got */
3594 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3595 Safefree(curlocales[j]);
3596 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3597 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3602 const char * description;
3603 const char * name = "";
3604 if (strEQ(trial_locales[i], "C")) {
3605 description = "the standard locale";
3609 # ifdef SYSTEM_DEFAULT_LOCALE
3611 else if (strEQ(trial_locales[i], "")) {
3612 description = "the system default locale";
3613 if (system_default_locale) {
3614 name = system_default_locale;
3618 # endif /* SYSTEM_DEFAULT_LOCALE */
3621 description = "a fallback locale";
3622 name = trial_locales[i];
3624 if (name && strNE(name, "")) {
3625 PerlIO_printf(Perl_error_log,
3626 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3629 PerlIO_printf(Perl_error_log,
3630 "perl: warning: %s %s.\n", msg, description);
3633 } /* End of tried to fallback */
3635 /* Done with finding the locales; update our records */
3637 # ifdef USE_LOCALE_CTYPE
3639 new_ctype(curlocales[LC_CTYPE_INDEX]);
3642 # ifdef USE_LOCALE_COLLATE
3644 new_collate(curlocales[LC_COLLATE_INDEX]);
3647 # ifdef USE_LOCALE_NUMERIC
3649 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3653 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3655 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3657 /* This caches whether each category's locale is UTF-8 or not. This
3658 * may involve changing the locale. It is ok to do this at
3659 * initialization time before any threads have started, but not later
3660 * unless thread-safe operations are used.
3661 * Caching means that if the program heeds our dictate not to change
3662 * locales in threaded applications, this data will remain valid, and
3663 * it may get queried without having to change locales. If the
3664 * environment is such that all categories have the same locale, this
3665 * isn't needed, as the code will not change the locale; but this
3666 * handles the uncommon case where the environment has disparate
3667 * locales for the categories */
3668 (void) _is_cur_LC_category_utf8(categories[i]);
3672 Safefree(curlocales[i]);
3675 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3677 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3678 * locale is UTF-8. The call to new_ctype() just above has already
3679 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3680 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3681 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3682 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3683 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3685 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3686 This is an alternative to using the -C command line switch
3687 (the -C if present will override this). */
3689 const char *p = PerlEnv_getenv("PERL_UNICODE");
3690 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3691 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3705 #endif /* USE_LOCALE */
3708 /* So won't continue to output stuff */
3709 DEBUG_INITIALIZATION_set(FALSE);
3716 #ifdef USE_LOCALE_COLLATE
3719 Perl__mem_collxfrm(pTHX_ const char *input_string,
3720 STRLEN len, /* Length of 'input_string' */
3721 STRLEN *xlen, /* Set to length of returned string
3722 (not including the collation index
3724 bool utf8 /* Is the input in UTF-8? */
3728 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3729 * differences. First, it handles embedded NULs. Second, it allocates a bit
3730 * more memory than needed for the transformed data itself. The real
3731 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3732 * the length of that, and doesn't include the collation index size.
3733 * Please see sv_collxfrm() to see how this is used. */
3735 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3737 char * s = (char *) input_string;
3738 STRLEN s_strlen = strlen(input_string);
3740 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3741 STRLEN length_in_chars;
3742 bool first_time = TRUE; /* Cleared after first loop iteration */
3744 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3746 /* Must be NUL-terminated */
3747 assert(*(input_string + len) == '\0');
3749 /* If this locale has defective collation, skip */
3750 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3751 DEBUG_L(PerlIO_printf(Perl_debug_log,
3752 "_mem_collxfrm: locale's collation is defective\n"));
3756 /* Replace any embedded NULs with the control that sorts before any others.
3757 * This will give as good as possible results on strings that don't
3758 * otherwise contain that character, but otherwise there may be
3759 * less-than-perfect results with that character and NUL. This is
3760 * unavoidable unless we replace strxfrm with our own implementation. */
3761 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3765 STRLEN sans_nuls_len;
3766 int try_non_controls;
3767 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3768 making sure 2nd byte is NUL.
3770 STRLEN this_replacement_len;
3772 /* If we don't know what non-NUL control character sorts lowest for
3773 * this locale, find it */
3774 if (PL_strxfrm_NUL_replacement == '\0') {
3776 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3777 includes the collation index
3780 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3782 /* Unlikely, but it may be that no control will work to replace
3783 * NUL, in which case we instead look for any character. Controls
3784 * are preferred because collation order is, in general, context
3785 * sensitive, with adjoining characters affecting the order, and
3786 * controls are less likely to have such interactions, allowing the
3787 * NUL-replacement to stand on its own. (Another way to look at it
3788 * is to imagine what would happen if the NUL were replaced by a
3789 * combining character; it wouldn't work out all that well.) */
3790 for (try_non_controls = 0;
3791 try_non_controls < 2;
3794 /* Look through all legal code points (NUL isn't) */
3795 for (j = 1; j < 256; j++) {
3796 char * x; /* j's xfrm plus collation index */
3797 STRLEN x_len; /* length of 'x' */
3798 STRLEN trial_len = 1;
3799 char cur_source[] = { '\0', '\0' };
3801 /* Skip non-controls the first time through the loop. The
3802 * controls in a UTF-8 locale are the L1 ones */
3803 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3810 /* Create a 1-char string of the current code point */
3811 cur_source[0] = (char) j;
3813 /* Then transform it */
3814 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3815 0 /* The string is not in UTF-8 */);
3817 /* Ignore any character that didn't successfully transform.
3823 /* If this character's transformation is lower than
3824 * the current lowest, this one becomes the lowest */
3825 if ( cur_min_x == NULL
3826 || strLT(x + COLLXFRM_HDR_LEN,
3827 cur_min_x + COLLXFRM_HDR_LEN))
3829 PL_strxfrm_NUL_replacement = j;
3835 } /* end of loop through all 255 characters */
3837 /* Stop looking if found */
3842 /* Unlikely, but possible, if there aren't any controls that
3843 * work in the locale, repeat the loop, looking for any
3844 * character that works */
3845 DEBUG_L(PerlIO_printf(Perl_debug_log,
3846 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3847 } /* End of loop to try first the controls, then any char */
3850 DEBUG_L(PerlIO_printf(Perl_debug_log,
3851 "_mem_collxfrm: Couldn't find any character to replace"
3852 " embedded NULs in locale %s with", PL_collation_name));
3856 DEBUG_L(PerlIO_printf(Perl_debug_log,
3857 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3858 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3860 Safefree(cur_min_x);
3861 } /* End of determining the character that is to replace NULs */
3863 /* If the replacement is variant under UTF-8, it must match the
3864 * UTF8-ness of the original */
3865 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3866 this_replacement_char[0] =
3867 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3868 this_replacement_char[1] =
3869 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3870 this_replacement_len = 2;
3873 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3874 /* this_replacement_char[1] = '\0' was done at initialization */
3875 this_replacement_len = 1;
3878 /* The worst case length for the replaced string would be if every
3879 * character in it is NUL. Multiply that by the length of each
3880 * replacement, and allow for a trailing NUL */
3881 sans_nuls_len = (len * this_replacement_len) + 1;
3882 Newx(sans_nuls, sans_nuls_len, char);
3885 /* Replace each NUL with the lowest collating control. Loop until have
3886 * exhausted all the NULs */
3887 while (s + s_strlen < e) {
3888 my_strlcat(sans_nuls, s, sans_nuls_len);
3890 /* Do the actual replacement */
3891 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3893 /* Move past the input NUL */
3895 s_strlen = strlen(s);
3898 /* And add anything that trails the final NUL */
3899 my_strlcat(sans_nuls, s, sans_nuls_len);
3901 /* Switch so below we transform this modified string */
3904 } /* End of replacing NULs */
3906 /* Make sure the UTF8ness of the string and locale match */
3907 if (utf8 != PL_in_utf8_COLLATE_locale) {
3908 /* XXX convert above Unicode to 10FFFF? */
3909 const char * const t = s; /* Temporary so we can later find where the
3912 /* Here they don't match. Change the string's to be what the locale is
3915 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3916 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3919 else { /* locale is not UTF-8; but input is; downgrade the input */
3921 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3923 /* If the downgrade was successful we are done, but if the input
3924 * contains things that require UTF-8 to represent, have to do
3925 * damage control ... */
3926 if (UNLIKELY(utf8)) {
3928 /* What we do is construct a non-UTF-8 string with
3929 * 1) the characters representable by a single byte converted
3930 * to be so (if necessary);
3931 * 2) and the rest converted to collate the same as the
3932 * highest collating representable character. That makes
3933 * them collate at the end. This is similar to how we
3934 * handle embedded NULs, but we use the highest collating
3935 * code point instead of the smallest. Like the NUL case,
3936 * this isn't perfect, but is the best we can reasonably
3937 * do. Every above-255 code point will sort the same as
3938 * the highest-sorting 0-255 code point. If that code
3939 * point can combine in a sequence with some other code
3940 * points for weight calculations, us changing something to
3941 * be it can adversely affect the results. But in most
3942 * cases, it should work reasonably. And note that this is
3943 * really an illegal situation: using code points above 255
3944 * on a locale where only 0-255 are valid. If two strings
3945 * sort entirely equal, then the sort order for the
3946 * above-255 code points will be in code point order. */
3950 /* If we haven't calculated the code point with the maximum
3951 * collating order for this locale, do so now */
3952 if (! PL_strxfrm_max_cp) {
3955 /* The current transformed string that collates the
3956 * highest (except it also includes the prefixed collation
3958 char * cur_max_x = NULL;
3960 /* Look through all legal code points (NUL isn't) */
3961 for (j = 1; j < 256; j++) {
3964 char cur_source[] = { '\0', '\0' };
3966 /* Create a 1-char string of the current code point */
3967 cur_source[0] = (char) j;
3969 /* Then transform it */
3970 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3972 /* If something went wrong (which it shouldn't), just
3973 * ignore this code point */
3978 /* If this character's transformation is higher than
3979 * the current highest, this one becomes the highest */
3980 if ( cur_max_x == NULL
3981 || strGT(x + COLLXFRM_HDR_LEN,
3982 cur_max_x + COLLXFRM_HDR_LEN))
3984 PL_strxfrm_max_cp = j;
3993 DEBUG_L(PerlIO_printf(Perl_debug_log,
3994 "_mem_collxfrm: Couldn't find any character to"
3995 " replace above-Latin1 chars in locale %s with",
3996 PL_collation_name));
4000 DEBUG_L(PerlIO_printf(Perl_debug_log,
4001 "_mem_collxfrm: highest 1-byte collating character"
4002 " in locale %s is 0x%02X\n",
4004 PL_strxfrm_max_cp));
4006 Safefree(cur_max_x);
4009 /* Here we know which legal code point collates the highest.
4010 * We are ready to construct the non-UTF-8 string. The length
4011 * will be at least 1 byte smaller than the input string
4012 * (because we changed at least one 2-byte character into a
4013 * single byte), but that is eaten up by the trailing NUL */
4019 char * e = (char *) t + len;
4021 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4023 if (UTF8_IS_INVARIANT(cur_char)) {
4026 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4027 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4029 else { /* Replace illegal cp with highest collating
4031 s[d++] = PL_strxfrm_max_cp;
4035 Renew(s, d, char); /* Free up unused space */
4040 /* Here, we have constructed a modified version of the input. It could
4041 * be that we already had a modified copy before we did this version.
4042 * If so, that copy is no longer needed */
4043 if (t != input_string) {
4048 length_in_chars = (utf8)
4049 ? utf8_length((U8 *) s, (U8 *) s + len)
4052 /* The first element in the output is the collation id, used by
4053 * sv_collxfrm(); then comes the space for the transformed string. The
4054 * equation should give us a good estimate as to how much is needed */
4055 xAlloc = COLLXFRM_HDR_LEN
4057 + (PL_collxfrm_mult * length_in_chars);
4058 Newx(xbuf, xAlloc, char);
4059 if (UNLIKELY(! xbuf)) {
4060 DEBUG_L(PerlIO_printf(Perl_debug_log,
4061 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4065 /* Store the collation id */
4066 *(U32*)xbuf = PL_collation_ix;
4068 /* Then the transformation of the input. We loop until successful, or we
4072 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4074 /* If the transformed string occupies less space than we told strxfrm()
4075 * was available, it means it successfully transformed the whole
4077 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4079 /* Some systems include a trailing NUL in the returned length.
4080 * Ignore it, using a loop in case multiple trailing NULs are
4083 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4088 /* If the first try didn't get it, it means our prediction was low.
4089 * Modify the coefficients so that we predict a larger value in any
4090 * future transformations */
4092 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4093 STRLEN computed_guess = PL_collxfrm_base
4094 + (PL_collxfrm_mult * length_in_chars);
4096 /* On zero-length input, just keep current slope instead of
4098 const STRLEN new_m = (length_in_chars != 0)
4099 ? needed / length_in_chars
4102 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4103 "%s: %d: initial size of %zu bytes for a length "
4104 "%zu string was insufficient, %zu needed\n",
4106 computed_guess, length_in_chars, needed));
4108 /* If slope increased, use it, but discard this result for
4109 * length 1 strings, as we can't be sure that it's a real slope
4111 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4115 STRLEN old_m = PL_collxfrm_mult;
4116 STRLEN old_b = PL_collxfrm_base;
4120 PL_collxfrm_mult = new_m;
4121 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4122 computed_guess = PL_collxfrm_base
4123 + (PL_collxfrm_mult * length_in_chars);
4124 if (computed_guess < needed) {
4125 PL_collxfrm_base += needed - computed_guess;
4128 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4129 "%s: %d: slope is now %zu; was %zu, base "
4130 "is now %zu; was %zu\n",
4132 PL_collxfrm_mult, old_m,
4133 PL_collxfrm_base, old_b));
4135 else { /* Slope didn't change, but 'b' did */
4136 const STRLEN new_b = needed
4139 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4140 "%s: %d: base is now %zu; was %zu\n",
4142 new_b, PL_collxfrm_base));
4143 PL_collxfrm_base = new_b;
4150 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4151 DEBUG_L(PerlIO_printf(Perl_debug_log,
4152 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4153 *xlen, PERL_INT_MAX));
4157 /* A well-behaved strxfrm() returns exactly how much space it needs
4158 * (usually not including the trailing NUL) when it fails due to not
4159 * enough space being provided. Assume that this is the case unless
4160 * it's been proven otherwise */
4161 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4162 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4164 else { /* Here, either:
4165 * 1) The strxfrm() has previously shown bad behavior; or
4166 * 2) It isn't the first time through the loop, which means
4167 * that the strxfrm() is now showing bad behavior, because
4168 * we gave it what it said was needed in the previous
4169 * iteration, and it came back saying it needed still more.
4170 * (Many versions of cygwin fit this. When the buffer size
4171 * isn't sufficient, they return the input size instead of
4172 * how much is needed.)
4173 * Increase the buffer size by a fixed percentage and try again.
4175 xAlloc += (xAlloc / 4) + 1;
4176 PL_strxfrm_is_behaved = FALSE;
4180 if (DEBUG_Lv_TEST || debug_initialization) {
4181 PerlIO_printf(Perl_debug_log,
4182 "_mem_collxfrm required more space than previously calculated"
4183 " for locale %s, trying again with new guess=%d+%zu\n",
4184 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4185 xAlloc - COLLXFRM_HDR_LEN);
4192 Renew(xbuf, xAlloc, char);
4193 if (UNLIKELY(! xbuf)) {
4194 DEBUG_L(PerlIO_printf(Perl_debug_log,
4195 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4205 if (DEBUG_Lv_TEST || debug_initialization) {
4207 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4208 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4209 PerlIO_printf(Perl_debug_log, "%s\n",
4210 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4216 /* Free up unneeded space; retain ehough for trailing NUL */
4217 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4219 if (s != input_string) {
4227 if (s != input_string) {
4234 if (DEBUG_Lv_TEST || debug_initialization) {
4235 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4246 S_print_collxfrm_input_and_return(pTHX_
4247 const char * const s,
4248 const char * const e,
4249 const STRLEN * const xlen,
4253 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4255 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4256 (UV)PL_collation_ix);
4258 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4261 PerlIO_printf(Perl_debug_log, "NULL");
4263 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4265 print_bytes_for_locale(s, e, is_utf8);
4267 PerlIO_printf(Perl_debug_log, "'\n");
4271 S_print_bytes_for_locale(pTHX_
4272 const char * const s,
4273 const char * const e,
4277 bool prev_was_printable = TRUE;
4278 bool first_time = TRUE;
4280 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4284 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4287 if (! prev_was_printable) {
4288 PerlIO_printf(Perl_debug_log, " ");
4290 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
https://perl5.git.perl.org / perl5.git / blob