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;
646 /* If we have a valid LC_ALL value, just return it */
647 if (PL_curlocales[LC_ALL_INDEX]) {
651 if (DEBUG_Lv_TEST || debug_initialization) {
652 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
657 return PL_curlocales[LC_ALL_INDEX];
660 /* Otherwise, we need to construct a string of name=value pairs.
661 * We use the glibc syntax, like
662 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
663 * First calculate the needed size. */
664 for (i = 0; i < LC_ALL_INDEX; i++) {
668 if (DEBUG_Lv_TEST || debug_initialization) {
669 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]);
674 names_len += strlen(category_names[i])
676 + strlen(PL_curlocales[i])
679 names_len++; /* Trailing '\0' */
680 SAVEFREEPV(Newx(all_string, names_len, char));
683 /* Then fill in the string */
684 for (i = 0; i < LC_ALL_INDEX; i++) {
688 if (DEBUG_Lv_TEST || debug_initialization) {
689 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]);
694 my_strlcat(all_string, category_names[i], names_len);
695 my_strlcat(all_string, "=", names_len);
696 my_strlcat(all_string, PL_curlocales[i], names_len);
697 my_strlcat(all_string, ";", names_len);
702 if (DEBUG_L_TEST || debug_initialization) {
703 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
713 SETERRNO(EINVAL, LIB_INVARG);
723 assert(PL_C_locale_obj);
725 /* Otherwise, we are switching locales. This will generally entail freeing
726 * the current one's space (at the C library's discretion). We need to
727 * stop using that locale before the switch. So switch to a known locale
728 * object that we don't otherwise mess with. This returns the locale
729 * object in effect at the time of the switch. */
730 old_obj = uselocale(PL_C_locale_obj);
734 if (DEBUG_Lv_TEST || debug_initialization) {
735 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
744 if (DEBUG_L_TEST || debug_initialization) {
746 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
757 if (DEBUG_Lv_TEST || debug_initialization) {
758 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
763 /* If we weren't in a thread safe locale, set so that newlocale() below
764 which uses 'old_obj', uses an empty one. Same for our reserved C object.
765 The latter is defensive coding, so that, even if there is some bug, we
766 will never end up trying to modify either of these, as if passed to
767 newlocale(), they can be. */
768 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
769 old_obj = (locale_t) 0;
772 /* Create the new locale (it may actually modify the current one). */
774 # ifndef HAS_QUERYLOCALE
776 if (strEQ(locale, "")) {
778 /* For non-querylocale() systems, we do the setting of "" ourselves to
779 * be sure that we really know what's going on. We follow the Linux
780 * documented behavior (but if that differs from the actual behavior,
781 * this won't work exactly as the OS implements). We go out and
782 * examine the environment based on our understanding of how the system
783 * works, and use that to figure things out */
785 const char * const lc_all = PerlEnv_getenv("LC_ALL");
787 /* Use any "LC_ALL" environment variable, as it overrides everything
789 if (lc_all && strNE(lc_all, "")) {
794 /* Otherwise, we need to dig deeper. Unless overridden, the
795 * default is the LANG environment variable; if it doesn't exist,
798 const char * default_name;
800 /* To minimize other threads messing with the environment, we copy
801 * the variable, making it a temporary. But this doesn't work upon
802 * program initialization before any scopes are created, and at
803 * this time, there's nothing else going on that would interfere.
804 * So skip the copy in that case */
805 if (PL_scopestack_ix == 0) {
806 default_name = PerlEnv_getenv("LANG");
809 default_name = savepv(PerlEnv_getenv("LANG"));
812 if (! default_name || strEQ(default_name, "")) {
815 else if (PL_scopestack_ix != 0) {
816 SAVEFREEPV(default_name);
819 if (category != LC_ALL) {
820 const char * const name = PerlEnv_getenv(category_names[index]);
822 /* Here we are setting a single category. Assume will have the
824 locale = default_name;
826 /* But then look for an overriding environment variable */
827 if (name && strNE(name, "")) {
832 bool did_override = FALSE;
835 /* Here, we are getting LC_ALL. Any categories that don't have
836 * a corresponding environment variable set should be set to
837 * LANG, or to "C" if there is no LANG. If no individual
838 * categories differ from this, we can just set LC_ALL. This
839 * is buggy on systems that have extra categories that we don't
840 * know about. If there is an environment variable that sets
841 * that category, we won't know to look for it, and so our use
842 * of LANG or "C" improperly overrides it. On the other hand,
843 * if we don't do what is done here, and there is no
844 * environment variable, the category's locale should be set to
845 * LANG or "C". So there is no good solution. khw thinks the
846 * best is to look at systems to see what categories they have,
847 * and include them, and then to assume that we know the
850 for (i = 0; i < LC_ALL_INDEX; i++) {
851 const char * const env_override
852 = savepv(PerlEnv_getenv(category_names[i]));
853 const char * this_locale = ( env_override
854 && strNE(env_override, ""))
857 emulate_setlocale(categories[i], this_locale, i, TRUE);
859 if (strNE(this_locale, default_name)) {
863 Safefree(env_override);
866 /* If all the categories are the same, we can set LC_ALL to
868 if (! did_override) {
869 locale = default_name;
873 /* Here, LC_ALL is no longer valid, as some individual
874 * categories don't match it. We call ourselves
875 * recursively, as that will execute the code that
876 * generates the proper locale string for this situation.
877 * We don't do the remainder of this function, as that is
878 * to update our records, and we've just done that for the
879 * individual categories in the loop above, and doing so
880 * would cause LC_ALL to be done as well */
881 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
886 else if (strchr(locale, ';')) {
888 /* LC_ALL may actually incude a conglomeration of various categories.
889 * Without querylocale, this code uses the glibc (as of this writing)
890 * syntax for representing that, but that is not a stable API, and
891 * other platforms do it differently, so we have to handle all cases
894 const char * s = locale;
895 const char * e = locale + strlen(locale);
897 const char * category_end;
898 const char * name_start;
899 const char * name_end;
904 /* Parse through the category */
905 while (isWORDCHAR(*p)) {
912 "panic: %s: %d: Unexpected character in locale name '%02X",
913 __FILE__, __LINE__, *(p-1));
916 /* Parse through the locale name */
918 while (isGRAPH(*p) && *p != ';') {
925 "panic: %s: %d: Unexpected character in locale name '%02X",
926 __FILE__, __LINE__, *(p-1));
929 /* Find the index of the category name in our lists */
930 for (i = 0; i < LC_ALL_INDEX; i++) {
932 /* Keep going if this isn't the index. The strnNE() avoids a
933 * Perl_form(), but would fail if ever a category name could be
934 * a substring of another one, like if there were a
936 if strnNE(s, category_names[i], category_end - s) {
940 /* If this index is for the single category we're changing, we
941 * have found the locale to set it to. */
942 if (category == categories[i]) {
943 locale = Perl_form(aTHX_ "%.*s",
944 (int) (name_end - name_start),
949 if (category == LC_ALL) {
950 char * individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
951 emulate_setlocale(categories[i], individ_locale, i, TRUE);
952 Safefree(individ_locale);
959 /* Here we have set all the individual categories by recursive calls.
960 * These collectively should have fixed up LC_ALL, so can just query
961 * what that now is */
962 assert(category == LC_ALL);
964 return do_setlocale_c(LC_ALL, NULL);
969 # endif /* end of ! querylocale */
971 /* Ready to create a new locale by modification of the exising one */
972 new_obj = newlocale(mask, locale, old_obj);
979 if (DEBUG_L_TEST || debug_initialization) {
980 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
985 if (! uselocale(old_obj)) {
990 if (DEBUG_L_TEST || debug_initialization) {
991 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1003 if (DEBUG_Lv_TEST || debug_initialization) {
1004 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1009 /* And switch into it */
1010 if (! uselocale(new_obj)) {
1015 if (DEBUG_L_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
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);
1032 freelocale(new_obj);
1039 if (DEBUG_Lv_TEST || debug_initialization) {
1040 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1045 /* We are done, except for updating our records (if the system doesn't keep
1046 * them) and in the case of locale "", we don't actually know what the
1047 * locale that got switched to is, as it came from the environment. So
1048 * have to find it */
1050 # ifdef HAS_QUERYLOCALE
1052 if (strEQ(locale, "")) {
1053 locale = querylocale(mask, new_obj);
1058 /* Here, 'locale' is the return value */
1060 /* Without querylocale(), we have to update our records */
1062 if (category == LC_ALL) {
1065 /* For LC_ALL, we change all individual categories to correspond */
1066 /* PL_curlocales is a parallel array, so has same
1067 * length as 'categories' */
1068 for (i = 0; i <= LC_ALL_INDEX; i++) {
1069 Safefree(PL_curlocales[i]);
1070 PL_curlocales[i] = savepv(locale);
1075 /* For a single category, if it's not the same as the one in LC_ALL, we
1078 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1079 Safefree(PL_curlocales[LC_ALL_INDEX]);
1080 PL_curlocales[LC_ALL_INDEX] = NULL;
1083 /* Then update the category's record */
1084 Safefree(PL_curlocales[index]);
1085 PL_curlocales[index] = savepv(locale);
1093 #endif /* USE_POSIX_2008_LOCALE */
1095 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1096 didn't natively support them */
1098 /* The way this would work is that we would keep a per-thread list of the
1099 * correct locale for that thread. Any operation that was locale-sensitive
1100 * would have to be changed so that it would look like this:
1103 * setlocale to the correct locale for this operation
1107 * This leaves the global locale in the most recently used operation's, but it
1108 * was locked long enough to get the result. If that result is static, it
1109 * needs to be copied before the unlock.
1111 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1112 * the setup, but are no-ops when not needed, and similarly,
1113 * END_LOCALE_DEPENDENT_OP for the tear-down
1115 * But every call to a locale-sensitive function would have to be changed, and
1116 * if a module didn't cooperate by using the mutex, things would break.
1118 * This code was abandoned before being completed or tested, and is left as-is
1121 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1122 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1125 S_locking_setlocale(pTHX_
1127 const char * locale,
1129 const bool is_index_valid
1132 /* This function kind of performs a setlocale() on just the current thread;
1133 * thus it is kind of thread-safe. It does this by keeping a thread-level
1134 * array of the current locales for each category. Every time a locale is
1135 * switched to, it does the switch globally, but updates the thread's
1136 * array. A query as to what the current locale is just returns the
1137 * appropriate element from the array, and doesn't actually call the system
1138 * setlocale(). The saving into the array is done in an uninterruptible
1139 * section of code, so is unaffected by whatever any other threads might be
1142 * All locale-sensitive operations must work by first starting a critical
1143 * section, then switching to the thread's locale as kept by this function,
1144 * and then doing the operation, then ending the critical section. Thus,
1145 * each gets done in the appropriate locale. simulating thread-safety.
1147 * This function takes the same parameters, 'category' and 'locale', that
1148 * the regular setlocale() function does, but it also takes two additional
1149 * ones. This is because as described earlier. If we know on input the
1150 * index corresponding to the category into the array where we store the
1151 * current locales, we don't have to calculate it. If the caller knows at
1152 * compile time what the index is, it it can pass it, setting
1153 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1157 /* If the input index might be incorrect, calculate the correct one */
1158 if (! is_index_valid) {
1161 if (DEBUG_Lv_TEST || debug_initialization) {
1162 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1165 for (i = 0; i <= LC_ALL_INDEX; i++) {
1166 if (category == categories[i]) {
1172 /* Here, we don't know about this category, so can't handle it.
1173 * XXX best we can do is to unsafely set this
1176 return my_setlocale(category, locale);
1180 if (DEBUG_Lv_TEST || debug_initialization) {
1181 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1185 /* For a query, just return what's in our records */
1186 if (new_locale == NULL) {
1187 return curlocales[index];
1191 /* Otherwise, we need to do the switch, and save the result, all in a
1192 * critical section */
1194 Safefree(curlocales[[index]]);
1196 /* It might be that this is called from an already-locked section of code.
1197 * We would have to detect and skip the LOCK/UNLOCK if so */
1200 curlocales[index] = savepv(my_setlocale(category, new_locale));
1202 if (strEQ(new_locale, "")) {
1206 /* The locale values come from the environment, and may not all be the
1207 * same, so for LC_ALL, we have to update all the others, while the
1208 * mutex is still locked */
1210 if (category == LC_ALL) {
1212 for (i = 0; i < LC_ALL_INDEX) {
1213 curlocales[i] = my_setlocale(categories[i], NULL);
1222 return curlocales[index];
1228 S_set_numeric_radix(pTHX_ const bool use_locale)
1230 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1231 * TRUE, use the radix character derived from the current locale */
1233 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1234 || defined(HAS_NL_LANGINFO))
1236 const char * radix = (use_locale)
1237 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1238 /* FALSE => already in dest locale */
1241 sv_setpv(PL_numeric_radix_sv, radix);
1243 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1244 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1245 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1246 SvCUR(PL_numeric_radix_sv))
1247 && _is_cur_LC_category_utf8(LC_NUMERIC))
1249 SvUTF8_on(PL_numeric_radix_sv);
1254 if (DEBUG_L_TEST || debug_initialization) {
1255 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1256 SvPVX(PL_numeric_radix_sv),
1257 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1261 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1266 S_new_numeric(pTHX_ const char *newnum)
1269 #ifndef USE_LOCALE_NUMERIC
1271 PERL_UNUSED_ARG(newnum);
1275 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1276 * core Perl this and that 'newnum' is the name of the new locale.
1277 * It installs this locale as the current underlying default.
1279 * The default locale and the C locale can be toggled between by use of the
1280 * set_numeric_underlying() and set_numeric_standard() functions, which
1281 * should probably not be called directly, but only via macros like
1282 * SET_NUMERIC_STANDARD() in perl.h.
1284 * The toggling is necessary mainly so that a non-dot radix decimal point
1285 * character can be output, while allowing internal calculations to use a
1288 * This sets several interpreter-level variables:
1289 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1290 * PL_numeric_underlying A boolean indicating if the toggled state is such
1291 * that the current locale is the program's underlying
1293 * PL_numeric_standard An int indicating if the toggled state is such
1294 * that the current locale is the C locale or
1295 * indistinguishable from the C locale. If non-zero, it
1296 * is in C; if > 1, it means it may not be toggled away
1298 * PL_numeric_underlying_is_standard A bool kept by this function
1299 * indicating that the underlying locale and the standard
1300 * C locale are indistinguishable for the purposes of
1301 * LC_NUMERIC. This happens when both of the above two
1302 * variables are true at the same time. (Toggling is a
1303 * no-op under these circumstances.) This variable is
1304 * used to avoid having to recalculate.
1310 Safefree(PL_numeric_name);
1311 PL_numeric_name = NULL;
1312 PL_numeric_standard = TRUE;
1313 PL_numeric_underlying = TRUE;
1314 PL_numeric_underlying_is_standard = TRUE;
1318 save_newnum = stdize_locale(savepv(newnum));
1319 PL_numeric_underlying = TRUE;
1320 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1322 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1324 if (! PL_numeric_standard) {
1325 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1326 FALSE /* Don't toggle locale */ ))
1327 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1331 /* Save the new name if it isn't the same as the previous one, if any */
1332 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1333 Safefree(PL_numeric_name);
1334 PL_numeric_name = save_newnum;
1337 Safefree(save_newnum);
1340 PL_numeric_underlying_is_standard = PL_numeric_standard;
1342 # ifdef HAS_POSIX_2008_LOCALE
1344 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1346 PL_underlying_numeric_obj);
1350 if (DEBUG_L_TEST || debug_initialization) {
1351 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1354 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1355 * have to worry about the radix being a non-dot. (Core operations that
1356 * need the underlying locale change to it temporarily). */
1357 if (PL_numeric_standard) {
1358 set_numeric_radix(0);
1361 set_numeric_standard();
1364 #endif /* USE_LOCALE_NUMERIC */
1369 Perl_set_numeric_standard(pTHX)
1372 #ifdef USE_LOCALE_NUMERIC
1374 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1375 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1376 * macro avoids calling this routine if toggling isn't necessary according
1377 * to our records (which could be wrong if some XS code has changed the
1378 * locale behind our back) */
1380 do_setlocale_c(LC_NUMERIC, "C");
1381 PL_numeric_standard = TRUE;
1382 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1383 set_numeric_radix(0);
1387 if (DEBUG_L_TEST || debug_initialization) {
1388 PerlIO_printf(Perl_debug_log,
1389 "LC_NUMERIC locale now is standard C\n");
1393 #endif /* USE_LOCALE_NUMERIC */
1398 Perl_set_numeric_underlying(pTHX)
1401 #ifdef USE_LOCALE_NUMERIC
1403 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1404 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1405 * instead of calling this directly. The macro avoids calling this routine
1406 * if toggling isn't necessary according to our records (which could be
1407 * wrong if some XS code has changed the locale behind our back) */
1409 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1410 PL_numeric_standard = PL_numeric_underlying_is_standard;
1411 PL_numeric_underlying = TRUE;
1412 set_numeric_radix(! PL_numeric_standard);
1416 if (DEBUG_L_TEST || debug_initialization) {
1417 PerlIO_printf(Perl_debug_log,
1418 "LC_NUMERIC locale now is %s\n",
1423 #endif /* USE_LOCALE_NUMERIC */
1428 * Set up for a new ctype locale.
1431 S_new_ctype(pTHX_ const char *newctype)
1434 #ifndef USE_LOCALE_CTYPE
1436 PERL_ARGS_ASSERT_NEW_CTYPE;
1437 PERL_UNUSED_ARG(newctype);
1438 PERL_UNUSED_CONTEXT;
1442 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1443 * core Perl this and that 'newctype' is the name of the new locale.
1445 * This function sets up the folding arrays for all 256 bytes, assuming
1446 * that tofold() is tolc() since fold case is not a concept in POSIX,
1448 * Any code changing the locale (outside this file) should use
1449 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1450 * this function should be called directly only from this file and from
1451 * POSIX::setlocale() */
1456 /* Don't check for problems if we are suppressing the warnings */
1457 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1459 PERL_ARGS_ASSERT_NEW_CTYPE;
1461 /* We will replace any bad locale warning with 1) nothing if the new one is
1462 * ok; or 2) a new warning for the bad new locale */
1463 if (PL_warn_locale) {
1464 SvREFCNT_dec_NN(PL_warn_locale);
1465 PL_warn_locale = NULL;
1468 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1470 /* A UTF-8 locale gets standard rules. But note that code still has to
1471 * handle this specially because of the three problematic code points */
1472 if (PL_in_utf8_CTYPE_locale) {
1473 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1476 /* We don't populate the other lists if a UTF-8 locale, but do check that
1477 * everything works as expected, unless checking turned off */
1478 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1479 /* Assume enough space for every character being bad. 4 spaces each
1480 * for the 94 printable characters that are output like "'x' "; and 5
1481 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1483 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1484 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1486 unsigned int bad_count = 0; /* Count of bad characters */
1488 for (i = 0; i < 256; i++) {
1489 if (! PL_in_utf8_CTYPE_locale) {
1491 PL_fold_locale[i] = (U8) tolower(i);
1492 else if (islower(i))
1493 PL_fold_locale[i] = (U8) toupper(i);
1495 PL_fold_locale[i] = (U8) i;
1498 /* If checking for locale problems, see if the native ASCII-range
1499 * printables plus \n and \t are in their expected categories in
1500 * the new locale. If not, this could mean big trouble, upending
1501 * Perl's and most programs' assumptions, like having a
1502 * metacharacter with special meaning become a \w. Fortunately,
1503 * it's very rare to find locales that aren't supersets of ASCII
1504 * nowadays. It isn't a problem for most controls to be changed
1505 * into something else; we check only \n and \t, though perhaps \r
1506 * could be an issue as well. */
1507 if ( check_for_problems
1508 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1510 bool is_bad = FALSE;
1511 char name[3] = { '\0' };
1513 /* Convert the name into a string */
1518 else if (i == '\n') {
1519 my_strlcpy(name, "\n", sizeof(name));
1522 my_strlcpy(name, "\t", sizeof(name));
1525 /* Check each possibe class */
1526 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1528 DEBUG_L(PerlIO_printf(Perl_debug_log,
1529 "isalnum('%s') unexpectedly is %d\n",
1530 name, cBOOL(isalnum(i))));
1532 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1534 DEBUG_L(PerlIO_printf(Perl_debug_log,
1535 "isalpha('%s') unexpectedly is %d\n",
1536 name, cBOOL(isalpha(i))));
1538 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1540 DEBUG_L(PerlIO_printf(Perl_debug_log,
1541 "isdigit('%s') unexpectedly is %d\n",
1542 name, cBOOL(isdigit(i))));
1544 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1546 DEBUG_L(PerlIO_printf(Perl_debug_log,
1547 "isgraph('%s') unexpectedly is %d\n",
1548 name, cBOOL(isgraph(i))));
1550 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1552 DEBUG_L(PerlIO_printf(Perl_debug_log,
1553 "islower('%s') unexpectedly is %d\n",
1554 name, cBOOL(islower(i))));
1556 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1558 DEBUG_L(PerlIO_printf(Perl_debug_log,
1559 "isprint('%s') unexpectedly is %d\n",
1560 name, cBOOL(isprint(i))));
1562 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1564 DEBUG_L(PerlIO_printf(Perl_debug_log,
1565 "ispunct('%s') unexpectedly is %d\n",
1566 name, cBOOL(ispunct(i))));
1568 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1570 DEBUG_L(PerlIO_printf(Perl_debug_log,
1571 "isspace('%s') unexpectedly is %d\n",
1572 name, cBOOL(isspace(i))));
1574 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1576 DEBUG_L(PerlIO_printf(Perl_debug_log,
1577 "isupper('%s') unexpectedly is %d\n",
1578 name, cBOOL(isupper(i))));
1580 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1582 DEBUG_L(PerlIO_printf(Perl_debug_log,
1583 "isxdigit('%s') unexpectedly is %d\n",
1584 name, cBOOL(isxdigit(i))));
1586 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1588 DEBUG_L(PerlIO_printf(Perl_debug_log,
1589 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1590 name, tolower(i), (int) toLOWER_A(i)));
1592 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1594 DEBUG_L(PerlIO_printf(Perl_debug_log,
1595 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1596 name, toupper(i), (int) toUPPER_A(i)));
1598 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1600 DEBUG_L(PerlIO_printf(Perl_debug_log,
1601 "'\\n' (=%02X) is not a control\n", (int) i));
1604 /* Add to the list; Separate multiple entries with a blank */
1607 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1609 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1617 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1618 * this locale requires more than one byte, there are going to be
1620 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1621 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1622 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1624 if ( check_for_problems && MB_CUR_MAX > 1
1625 && ! PL_in_utf8_CTYPE_locale
1627 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1628 * locale. Just assume that the implementation for them (plus
1629 * for POSIX) is correct and the > 1 value is spurious. (Since
1630 * these are specially handled to never be considered UTF-8
1631 * locales, as long as this is the only problem, everything
1632 * should work fine */
1633 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1635 multi_byte_locale = TRUE;
1640 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1641 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1642 PL_warn_locale = Perl_newSVpvf(aTHX_
1643 "Locale '%s' contains (at least) the following characters"
1644 " which have\nnon-standard meanings: %s\nThe Perl program"
1645 " will use the standard meanings",
1646 newctype, bad_chars_list);
1650 PL_warn_locale = Perl_newSVpvf(aTHX_
1651 "Locale '%s' may not work well.%s%s%s\n",
1654 ? " Some characters in it are not recognized by"
1658 ? "\nThe following characters (and maybe others)"
1659 " may not have the same meaning as the Perl"
1660 " program expects:\n"
1668 # ifdef HAS_NL_LANGINFO
1670 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1671 /* parameter FALSE is a don't care here */
1672 my_nl_langinfo(PERL_CODESET, FALSE));
1676 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1678 /* If we are actually in the scope of the locale or are debugging,
1679 * output the message now. If not in that scope, we save the
1680 * message to be output at the first operation using this locale,
1681 * if that actually happens. Most programs don't use locales, so
1682 * they are immune to bad ones. */
1683 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1685 /* The '0' below suppresses a bogus gcc compiler warning */
1686 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1688 if (IN_LC(LC_CTYPE)) {
1689 SvREFCNT_dec_NN(PL_warn_locale);
1690 PL_warn_locale = NULL;
1696 #endif /* USE_LOCALE_CTYPE */
1701 Perl__warn_problematic_locale()
1704 #ifdef USE_LOCALE_CTYPE
1708 /* Internal-to-core function that outputs the message in PL_warn_locale,
1709 * and then NULLS it. Should be called only through the macro
1710 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1712 if (PL_warn_locale) {
1713 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1714 SvPVX(PL_warn_locale),
1715 0 /* dummy to avoid compiler warning */ );
1716 SvREFCNT_dec_NN(PL_warn_locale);
1717 PL_warn_locale = NULL;
1725 S_new_collate(pTHX_ const char *newcoll)
1728 #ifndef USE_LOCALE_COLLATE
1730 PERL_UNUSED_ARG(newcoll);
1731 PERL_UNUSED_CONTEXT;
1735 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1736 * core Perl this and that 'newcoll' is the name of the new locale.
1738 * The design of locale collation is that every locale change is given an
1739 * index 'PL_collation_ix'. The first time a string particpates in an
1740 * operation that requires collation while locale collation is active, it
1741 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1742 * magic includes the collation index, and the transformation of the string
1743 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1744 * instead of the string itself. If a string changes, the magic is
1745 * cleared. The next time the locale changes, the index is incremented,
1746 * and so we know during a comparison that the transformation is not
1747 * necessarily still valid, and so is recomputed. Note that if the locale
1748 * changes enough times, the index could wrap (a U32), and it is possible
1749 * that a transformation would improperly be considered valid, leading to
1750 * an unlikely bug */
1753 if (PL_collation_name) {
1755 Safefree(PL_collation_name);
1756 PL_collation_name = NULL;
1758 PL_collation_standard = TRUE;
1759 is_standard_collation:
1760 PL_collxfrm_base = 0;
1761 PL_collxfrm_mult = 2;
1762 PL_in_utf8_COLLATE_locale = FALSE;
1763 PL_strxfrm_NUL_replacement = '\0';
1764 PL_strxfrm_max_cp = 0;
1768 /* If this is not the same locale as currently, set the new one up */
1769 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1771 Safefree(PL_collation_name);
1772 PL_collation_name = stdize_locale(savepv(newcoll));
1773 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1774 if (PL_collation_standard) {
1775 goto is_standard_collation;
1778 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1779 PL_strxfrm_NUL_replacement = '\0';
1780 PL_strxfrm_max_cp = 0;
1782 /* A locale collation definition includes primary, secondary, tertiary,
1783 * etc. weights for each character. To sort, the primary weights are
1784 * used, and only if they compare equal, then the secondary weights are
1785 * used, and only if they compare equal, then the tertiary, etc.
1787 * strxfrm() works by taking the input string, say ABC, and creating an
1788 * output transformed string consisting of first the primary weights,
1789 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1790 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1791 * may not have weights at every level. In our example, let's say B
1792 * doesn't have a tertiary weight, and A doesn't have a secondary
1793 * weight. The constructed string is then going to be
1794 * A¹B¹C¹ B²C² A³C³ ....
1795 * This has the desired effect that strcmp() will look at the secondary
1796 * or tertiary weights only if the strings compare equal at all higher
1797 * priority weights. The spaces shown here, like in
1799 * are not just for readability. In the general case, these must
1800 * actually be bytes, which we will call here 'separator weights'; and
1801 * they must be smaller than any other weight value, but since these
1802 * are C strings, only the terminating one can be a NUL (some
1803 * implementations may include a non-NUL separator weight just before
1804 * the NUL). Implementations tend to reserve 01 for the separator
1805 * weights. They are needed so that a shorter string's secondary
1806 * weights won't be misconstrued as primary weights of a longer string,
1807 * etc. By making them smaller than any other weight, the shorter
1808 * string will sort first. (Actually, if all secondary weights are
1809 * smaller than all primary ones, there is no need for a separator
1810 * weight between those two levels, etc.)
1812 * The length of the transformed string is roughly a linear function of
1813 * the input string. It's not exactly linear because some characters
1814 * don't have weights at all levels. When we call strxfrm() we have to
1815 * allocate some memory to hold the transformed string. The
1816 * calculations below try to find coefficients 'm' and 'b' for this
1817 * locale so that m*x + b equals how much space we need, given the size
1818 * of the input string in 'x'. If we calculate too small, we increase
1819 * the size as needed, and call strxfrm() again, but it is better to
1820 * get it right the first time to avoid wasted expensive string
1821 * transformations. */
1824 /* We use the string below to find how long the tranformation of it
1825 * is. Almost all locales are supersets of ASCII, or at least the
1826 * ASCII letters. We use all of them, half upper half lower,
1827 * because if we used fewer, we might hit just the ones that are
1828 * outliers in a particular locale. Most of the strings being
1829 * collated will contain a preponderance of letters, and even if
1830 * they are above-ASCII, they are likely to have the same number of
1831 * weight levels as the ASCII ones. It turns out that digits tend
1832 * to have fewer levels, and some punctuation has more, but those
1833 * are relatively sparse in text, and khw believes this gives a
1834 * reasonable result, but it could be changed if experience so
1836 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1837 char * x_longer; /* Transformed 'longer' */
1838 Size_t x_len_longer; /* Length of 'x_longer' */
1840 char * x_shorter; /* We also transform a substring of 'longer' */
1841 Size_t x_len_shorter;
1843 /* _mem_collxfrm() is used get the transformation (though here we
1844 * are interested only in its length). It is used because it has
1845 * the intelligence to handle all cases, but to work, it needs some
1846 * values of 'm' and 'b' to get it started. For the purposes of
1847 * this calculation we use a very conservative estimate of 'm' and
1848 * 'b'. This assumes a weight can be multiple bytes, enough to
1849 * hold any UV on the platform, and there are 5 levels, 4 weight
1850 * bytes, and a trailing NUL. */
1851 PL_collxfrm_base = 5;
1852 PL_collxfrm_mult = 5 * sizeof(UV);
1854 /* Find out how long the transformation really is */
1855 x_longer = _mem_collxfrm(longer,
1859 /* We avoid converting to UTF-8 in the
1860 * called function by telling it the
1861 * string is in UTF-8 if the locale is a
1862 * UTF-8 one. Since the string passed
1863 * here is invariant under UTF-8, we can
1864 * claim it's UTF-8 even though it isn't.
1866 PL_in_utf8_COLLATE_locale);
1869 /* Find out how long the transformation of a substring of 'longer'
1870 * is. Together the lengths of these transformations are
1871 * sufficient to calculate 'm' and 'b'. The substring is all of
1872 * 'longer' except the first character. This minimizes the chances
1873 * of being swayed by outliers */
1874 x_shorter = _mem_collxfrm(longer + 1,
1877 PL_in_utf8_COLLATE_locale);
1878 Safefree(x_shorter);
1880 /* If the results are nonsensical for this simple test, the whole
1881 * locale definition is suspect. Mark it so that locale collation
1882 * is not active at all for it. XXX Should we warn? */
1883 if ( x_len_shorter == 0
1884 || x_len_longer == 0
1885 || x_len_shorter >= x_len_longer)
1887 PL_collxfrm_mult = 0;
1888 PL_collxfrm_base = 0;
1891 SSize_t base; /* Temporary */
1893 /* We have both: m * strlen(longer) + b = x_len_longer
1894 * m * strlen(shorter) + b = x_len_shorter;
1895 * subtracting yields:
1896 * m * (strlen(longer) - strlen(shorter))
1897 * = x_len_longer - x_len_shorter
1898 * But we have set things up so that 'shorter' is 1 byte smaller
1899 * than 'longer'. Hence:
1900 * m = x_len_longer - x_len_shorter
1902 * But if something went wrong, make sure the multiplier is at
1905 if (x_len_longer > x_len_shorter) {
1906 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1909 PL_collxfrm_mult = 1;
1914 * but in case something has gone wrong, make sure it is
1916 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1921 /* Add 1 for the trailing NUL */
1922 PL_collxfrm_base = base + 1;
1927 if (DEBUG_L_TEST || debug_initialization) {
1928 PerlIO_printf(Perl_debug_log,
1929 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1931 " collate multipler=%zu, collate base=%zu\n",
1933 PL_in_utf8_COLLATE_locale,
1934 x_len_shorter, x_len_longer,
1935 PL_collxfrm_mult, PL_collxfrm_base);
1942 #endif /* USE_LOCALE_COLLATE */
1949 S_win32_setlocale(pTHX_ int category, const char* locale)
1951 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1952 * difference between the two unless the input locale is "", which normally
1953 * means on Windows to get the machine default, which is set via the
1954 * computer's "Regional and Language Options" (or its current equivalent).
1955 * In POSIX, it instead means to find the locale from the user's
1956 * environment. This routine changes the Windows behavior to first look in
1957 * the environment, and, if anything is found, use that instead of going to
1958 * the machine default. If there is no environment override, the machine
1959 * default is used, by calling the real setlocale() with "".
1961 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1962 * use the particular category's variable if set; otherwise to use the LANG
1965 bool override_LC_ALL = FALSE;
1969 if (locale && strEQ(locale, "")) {
1973 locale = PerlEnv_getenv("LC_ALL");
1975 if (category == LC_ALL) {
1976 override_LC_ALL = TRUE;
1982 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1983 if (category == categories[i]) {
1984 locale = PerlEnv_getenv(category_names[i]);
1989 locale = PerlEnv_getenv("LANG");
2005 result = setlocale(category, locale);
2006 DEBUG_L(STMT_START {
2008 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2009 setlocale_debug_string(category, locale, result));
2013 if (! override_LC_ALL) {
2017 /* Here the input category was LC_ALL, and we have set it to what is in the
2018 * LANG variable or the system default if there is no LANG. But these have
2019 * lower priority than the other LC_foo variables, so override it for each
2020 * one that is set. (If they are set to "", it means to use the same thing
2021 * we just set LC_ALL to, so can skip) */
2023 for (i = 0; i < LC_ALL_INDEX; i++) {
2024 result = PerlEnv_getenv(category_names[i]);
2025 if (result && strNE(result, "")) {
2026 setlocale(categories[i], result);
2027 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2029 setlocale_debug_string(categories[i], result, "not captured")));
2033 result = setlocale(LC_ALL, NULL);
2034 DEBUG_L(STMT_START {
2036 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2038 setlocale_debug_string(LC_ALL, NULL, result));
2049 =head1 Locale-related functions and macros
2051 =for apidoc Perl_setlocale
2053 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2054 taking the same parameters, and returning the same information, except that it
2055 returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2056 perl keeps that locale category as C<C>, changing it briefly during the
2057 operations where the underlying one is required.
2059 Another reason it isn't completely a drop-in replacement is that it is
2060 declared to return S<C<const char *>>, whereas the system setlocale omits the
2061 C<const>. (If it were being written today, plain setlocale would be declared
2062 const, since it is illegal to change the information it returns; doing so leads
2065 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2066 C<setlocale> can be completely ineffective on some platforms under some
2069 C<Perl_setlocale> should not be used to change the locale except on systems
2070 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2071 the system C<setlocale()> is ineffective, returning the wrong information, and
2072 failing to actually change the locale. C<Perl_setlocale>, however works
2073 properly in all circumstances.
2075 The return points to a per-thread static buffer, which is overwritten the next
2076 time C<Perl_setlocale> is called from the same thread.
2083 Perl_setlocale(const int category, const char * locale)
2085 /* This wraps POSIX::setlocale() */
2087 const char * retval;
2088 const char * newlocale;
2090 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2093 #ifdef USE_LOCALE_NUMERIC
2095 /* A NULL locale means only query what the current one is. We have the
2096 * LC_NUMERIC name saved, because we are normally switched into the C
2097 * locale for it. For an LC_ALL query, switch back to get the correct
2098 * results. All other categories don't require special handling */
2099 if (locale == NULL) {
2100 if (category == LC_NUMERIC) {
2102 /* We don't have to copy this return value, as it is a per-thread
2103 * variable, and won't change until a future setlocale */
2104 return PL_numeric_name;
2109 else if (category == LC_ALL) {
2110 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2119 retval = do_setlocale_r(category, locale);
2122 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2124 if (locale == NULL && category == LC_ALL) {
2125 RESTORE_LC_NUMERIC();
2130 DEBUG_L(PerlIO_printf(Perl_debug_log,
2131 "%s:%d: %s\n", __FILE__, __LINE__,
2132 setlocale_debug_string(category, locale, retval)));
2140 save_to_buffer(retval, &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2141 retval = PL_setlocale_buf;
2143 /* If locale == NULL, we are just querying the state */
2144 if (locale == NULL) {
2148 /* Now that have switched locales, we have to update our records to
2153 #ifdef USE_LOCALE_CTYPE
2160 #ifdef USE_LOCALE_COLLATE
2163 new_collate(retval);
2167 #ifdef USE_LOCALE_NUMERIC
2170 new_numeric(retval);
2178 /* LC_ALL updates all the things we care about. The values may not
2179 * be the same as 'retval', as the locale "" may have set things
2182 # ifdef USE_LOCALE_CTYPE
2184 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2185 new_ctype(newlocale);
2187 # endif /* USE_LOCALE_CTYPE */
2188 # ifdef USE_LOCALE_COLLATE
2190 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2191 new_collate(newlocale);
2194 # ifdef USE_LOCALE_NUMERIC
2196 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2197 new_numeric(newlocale);
2199 # endif /* USE_LOCALE_NUMERIC */
2210 PERL_STATIC_INLINE const char *
2211 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2213 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2214 * growing it if necessary */
2216 const Size_t string_size = strlen(string) + offset + 1;
2218 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2220 if (*buf_size == 0) {
2221 Newx(*buf, string_size, char);
2222 *buf_size = string_size;
2224 else if (string_size > *buf_size) {
2225 Renew(*buf, string_size, char);
2226 *buf_size = string_size;
2229 Copy(string, *buf + offset, string_size - offset, char);
2235 =for apidoc Perl_langinfo
2237 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2238 taking the same C<item> parameter values, and returning the same information.
2239 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2240 of Perl's locale handling from your code, and can be used on systems that lack
2241 a native C<nl_langinfo>.
2249 The reason it isn't quite a drop-in replacement is actually an advantage. The
2250 only difference is that it returns S<C<const char *>>, whereas plain
2251 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2252 forbidden to write into the buffer. By declaring this C<const>, the compiler
2253 enforces this restriction, so if it is violated, you know at compilation time,
2254 rather than getting segfaults at runtime.
2258 It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
2259 without you having to write extra code. The reason for the extra code would be
2260 because these are from the C<LC_NUMERIC> locale category, which is normally
2261 kept set to the C locale by Perl, no matter what the underlying locale is
2262 supposed to be, and so to get the expected results, you have to temporarily
2263 toggle into the underlying locale, and later toggle back. (You could use plain
2264 C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2265 then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2266 C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2267 radix (decimal point) character to be a dot.)
2271 The system function it replaces can have its static return buffer trashed,
2272 not only by a subesequent call to that function, but by a C<freelocale>,
2273 C<setlocale>, or other locale change. The returned buffer of this function is
2274 not changed until the next call to it, so the buffer is never in a trashed
2279 Its return buffer is per-thread, so it also is never overwritten by a call to
2280 this function from another thread; unlike the function it replaces.
2284 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2285 as Windows, hence makes your code more portable. Of the fifty-some possible
2286 items specified by the POSIX 2008 standard,
2287 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2288 only two are completely unimplemented (though the loss of one of these is
2289 significant). It uses various techniques to recover the other items, including
2290 calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2291 in C89, so should be always be available. Later C<strftime()> versions have
2292 additional capabilities; C<""> is returned for those not available on your
2295 It is important to note that when called with an item that is recovered by
2296 using C<localeconv>, the buffer from any previous explicit call to
2297 C<localeconv> will be overwritten. This means you must save that buffer's
2298 contents if you need to access them after a call to this function.
2300 The details for those items which may differ from what this emulation returns
2301 and what a native C<nl_langinfo()> would return are:
2309 Unimplemented, so returns C<"">.
2319 Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
2320 removed from POSIX 2008, and are retained here for backwards compatibility.
2321 Your platform's C<nl_langinfo> may not support them.
2325 Always evaluates to C<%x>, the locale's appropriate date representation.
2329 Always evaluates to C<%X>, the locale's appropriate time representation.
2333 Always evaluates to C<%c>, the locale's appropriate date and time
2338 The return may be incorrect for those rare locales where the currency symbol
2339 replaces the radix character.
2340 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2341 to work differently.
2345 Currently this gives the same results as Linux does.
2346 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2347 to work differently.
2353 =item C<ERA_D_T_FMT>
2357 These are derived by using C<strftime()>, and not all versions of that function
2358 know about them. C<""> is returned for these on such systems.
2364 When using C<Perl_langinfo> on systems that don't have a native
2365 C<nl_langinfo()>, you must
2367 #include "perl_langinfo.h"
2369 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2370 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2371 C<langinfo.h> imports into the namespace for code that doesn't need it.)
2373 You also should not use the bare C<langinfo.h> item names, but should preface
2374 them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2375 The C<PERL_I<foo>> versions will also work for this function on systems that do
2376 have a native C<nl_langinfo>.
2378 The original impetus for C<Perl_langinfo()> was so that code that needs to
2379 find out the current currency symbol, floating point radix character, or digit
2380 grouping separator can use, on all systems, the simpler and more
2381 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2382 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2383 is better to use the methods given in L<perlcall> to call
2384 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2391 #ifdef HAS_NL_LANGINFO
2392 Perl_langinfo(const nl_item item)
2394 Perl_langinfo(const int item)
2397 return my_nl_langinfo(item, TRUE);
2401 #ifdef HAS_NL_LANGINFO
2402 S_my_nl_langinfo(const nl_item item, bool toggle)
2404 S_my_nl_langinfo(const int item, bool toggle)
2409 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2410 * two items, and only if not already there */
2411 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2412 || PL_numeric_underlying))
2417 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2418 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2419 || ! defined(HAS_POSIX_2008_LOCALE)
2421 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2422 * for those items dependent on it. This must be copied to a buffer before
2423 * switching back, as some systems destroy the buffer when setlocale() is
2427 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2430 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2433 LOCALE_LOCK; /* Prevent interference from another thread executing
2434 this code section (the only call to nl_langinfo in
2438 /* Copy to a per-thread buffer, which is also one that won't be
2439 * destroyed by a subsequent setlocale(), such as the
2440 * RESTORE_LC_NUMERIC may do just below. */
2441 save_to_buffer(nl_langinfo(item),
2442 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2447 RESTORE_LC_NUMERIC();
2451 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2454 bool do_free = FALSE;
2455 locale_t cur = uselocale((locale_t) 0);
2457 if (cur == LC_GLOBAL_LOCALE) {
2458 cur = duplocale(LC_GLOBAL_LOCALE);
2463 if (PL_underlying_numeric_obj) {
2464 cur = PL_underlying_numeric_obj;
2467 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2472 /* We have to save it to a buffer, because the freelocale() just below
2473 * can invalidate the internal one */
2474 save_to_buffer(nl_langinfo_l(item, cur),
2475 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2484 if (strEQ(PL_langinfo_buf, "")) {
2485 if (item == PERL_YESSTR) {
2488 if (item == PERL_NOSTR) {
2493 return PL_langinfo_buf;
2495 #else /* Below, emulate nl_langinfo as best we can */
2499 # ifdef HAS_LOCALECONV
2501 const struct lconv* lc;
2503 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2506 # ifdef HAS_STRFTIME
2509 bool return_format = FALSE; /* Return the %format, not the value */
2510 const char * format;
2514 /* We copy the results to a per-thread buffer, even if not
2515 * multi-threaded. This is in part to simplify this code, and partly
2516 * because we need a buffer anyway for strftime(), and partly because a
2517 * call of localeconv() could otherwise wipe out the buffer, and the
2518 * programmer would not be expecting this, as this is a nl_langinfo()
2519 * substitute after all, so s/he might be thinking their localeconv()
2520 * is safe until another localeconv() call. */
2525 /* These 2 are unimplemented */
2527 case PERL_ERA: /* For use with strftime() %E modifier */
2532 /* We use only an English set, since we don't know any more */
2533 case PERL_YESEXPR: return "^[+1yY]";
2534 case PERL_YESSTR: return "yes";
2535 case PERL_NOEXPR: return "^[-0nN]";
2536 case PERL_NOSTR: return "no";
2538 # ifdef HAS_LOCALECONV
2542 /* We don't bother with localeconv_l() because any system that
2543 * has it is likely to also have nl_langinfo() */
2545 LOCALE_LOCK; /* Prevent interference with other threads
2546 using localeconv() */
2550 || ! lc->currency_symbol
2551 || strEQ("", lc->currency_symbol))
2557 /* Leave the first spot empty to be filled in below */
2558 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2559 &PL_langinfo_bufsize, 1);
2560 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2561 { /* khw couldn't figure out how the localedef specifications
2562 would show that the $ should replace the radix; this is
2563 just a guess as to how it might work.*/
2564 *PL_langinfo_buf = '.';
2566 else if (lc->p_cs_precedes) {
2567 *PL_langinfo_buf = '-';
2570 *PL_langinfo_buf = '+';
2576 case PERL_RADIXCHAR:
2580 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2583 LOCALE_LOCK; /* Prevent interference with other threads
2584 using localeconv() */
2591 temp = (item == PERL_RADIXCHAR)
2593 : lc->thousands_sep;
2599 save_to_buffer(temp, &PL_langinfo_buf,
2600 &PL_langinfo_bufsize, 0);
2605 RESTORE_LC_NUMERIC();
2611 # ifdef HAS_STRFTIME
2613 /* These are defined by C89, so we assume that strftime supports
2614 * them, and so are returned unconditionally; they may not be what
2615 * the locale actually says, but should give good enough results
2616 * for someone using them as formats (as opposed to trying to parse
2617 * them to figure out what the locale says). The other format
2618 * items are actually tested to verify they work on the platform */
2619 case PERL_D_FMT: return "%x";
2620 case PERL_T_FMT: return "%X";
2621 case PERL_D_T_FMT: return "%c";
2623 /* These formats are only available in later strfmtime's */
2624 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2625 case PERL_T_FMT_AMPM:
2627 /* The rest can be gotten from most versions of strftime(). */
2628 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2629 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2631 case PERL_ALT_DIGITS:
2632 case PERL_AM_STR: case PERL_PM_STR:
2633 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2634 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2635 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2636 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2637 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2638 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2639 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2640 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2641 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2646 init_tm(&tm); /* Precaution against core dumps */
2650 tm.tm_year = 2017 - 1900;
2657 "panic: %s: %d: switch case: %d problem",
2658 __FILE__, __LINE__, item);
2659 NOT_REACHED; /* NOTREACHED */
2661 case PERL_PM_STR: tm.tm_hour = 18;
2666 case PERL_ABDAY_7: tm.tm_wday++;
2667 case PERL_ABDAY_6: tm.tm_wday++;
2668 case PERL_ABDAY_5: tm.tm_wday++;
2669 case PERL_ABDAY_4: tm.tm_wday++;
2670 case PERL_ABDAY_3: tm.tm_wday++;
2671 case PERL_ABDAY_2: tm.tm_wday++;
2676 case PERL_DAY_7: tm.tm_wday++;
2677 case PERL_DAY_6: tm.tm_wday++;
2678 case PERL_DAY_5: tm.tm_wday++;
2679 case PERL_DAY_4: tm.tm_wday++;
2680 case PERL_DAY_3: tm.tm_wday++;
2681 case PERL_DAY_2: tm.tm_wday++;
2686 case PERL_ABMON_12: tm.tm_mon++;
2687 case PERL_ABMON_11: tm.tm_mon++;
2688 case PERL_ABMON_10: tm.tm_mon++;
2689 case PERL_ABMON_9: tm.tm_mon++;
2690 case PERL_ABMON_8: tm.tm_mon++;
2691 case PERL_ABMON_7: tm.tm_mon++;
2692 case PERL_ABMON_6: tm.tm_mon++;
2693 case PERL_ABMON_5: tm.tm_mon++;
2694 case PERL_ABMON_4: tm.tm_mon++;
2695 case PERL_ABMON_3: tm.tm_mon++;
2696 case PERL_ABMON_2: tm.tm_mon++;
2701 case PERL_MON_12: tm.tm_mon++;
2702 case PERL_MON_11: tm.tm_mon++;
2703 case PERL_MON_10: tm.tm_mon++;
2704 case PERL_MON_9: tm.tm_mon++;
2705 case PERL_MON_8: tm.tm_mon++;
2706 case PERL_MON_7: tm.tm_mon++;
2707 case PERL_MON_6: tm.tm_mon++;
2708 case PERL_MON_5: tm.tm_mon++;
2709 case PERL_MON_4: tm.tm_mon++;
2710 case PERL_MON_3: tm.tm_mon++;
2711 case PERL_MON_2: tm.tm_mon++;
2716 case PERL_T_FMT_AMPM:
2718 return_format = TRUE;
2721 case PERL_ERA_D_FMT:
2723 return_format = TRUE;
2726 case PERL_ERA_T_FMT:
2728 return_format = TRUE;
2731 case PERL_ERA_D_T_FMT:
2733 return_format = TRUE;
2736 case PERL_ALT_DIGITS:
2738 format = "%Ow"; /* Find the alternate digit for 0 */
2742 /* We can't use my_strftime() because it doesn't look at
2744 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2747 /* A zero return means one of:
2748 * a) there wasn't enough space in PL_langinfo_buf
2749 * b) the format, like a plain %p, returns empty
2750 * c) it was an illegal format, though some
2751 * implementations of strftime will just return the
2752 * illegal format as a plain character sequence.
2754 * To quickly test for case 'b)', try again but precede
2755 * the format with a plain character. If that result is
2756 * still empty, the problem is either 'a)' or 'c)' */
2758 Size_t format_size = strlen(format) + 1;
2759 Size_t mod_size = format_size + 1;
2763 Newx(mod_format, mod_size, char);
2764 Newx(temp_result, PL_langinfo_bufsize, char);
2766 my_strlcpy(mod_format + 1, format, mod_size);
2767 len = strftime(temp_result,
2768 PL_langinfo_bufsize,
2770 Safefree(mod_format);
2771 Safefree(temp_result);
2773 /* If 'len' is non-zero, it means that we had a case like
2774 * %p which means the current locale doesn't use a.m. or
2775 * p.m., and that is valid */
2778 /* Here, still didn't work. If we get well beyond a
2779 * reasonable size, bail out to prevent an infinite
2782 if (PL_langinfo_bufsize > 100 * format_size) {
2783 *PL_langinfo_buf = '\0';
2786 /* Double the buffer size to retry; Add 1 in case
2787 * original was 0, so we aren't stuck at 0. */
2788 PL_langinfo_bufsize *= 2;
2789 PL_langinfo_bufsize++;
2790 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2798 /* Here, we got a result.
2800 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2801 * alternate format for wday 0. If the value is the same as
2802 * the normal 0, there isn't an alternate, so clear the buffer.
2804 if ( item == PERL_ALT_DIGITS
2805 && strEQ(PL_langinfo_buf, "0"))
2807 *PL_langinfo_buf = '\0';
2810 /* ALT_DIGITS is problematic. Experiments on it showed that
2811 * strftime() did not always work properly when going from
2812 * alt-9 to alt-10. Only a few locales have this item defined,
2813 * and in all of them on Linux that khw was able to find,
2814 * nl_langinfo() merely returned the alt-0 character, possibly
2815 * doubled. Most Unicode digits are in blocks of 10
2816 * consecutive code points, so that is sufficient information
2817 * for those scripts, as we can infer alt-1, alt-2, .... But
2818 * for a Japanese locale, a CJK ideographic 0 is returned, and
2819 * the CJK digits are not in code point order, so you can't
2820 * really infer anything. The localedef for this locale did
2821 * specify the succeeding digits, so that strftime() works
2822 * properly on them, without needing to infer anything. But
2823 * the nl_langinfo() return did not give sufficient information
2824 * for the caller to understand what's going on. So until
2825 * there is evidence that it should work differently, this
2826 * returns the alt-0 string for ALT_DIGITS.
2828 * wday was chosen because its range is all a single digit.
2829 * Things like tm_sec have two digits as the minimum: '00' */
2833 /* If to return the format, not the value, overwrite the buffer
2834 * with it. But some strftime()s will keep the original format
2835 * if illegal, so change those to "" */
2836 if (return_format) {
2837 if (strEQ(PL_langinfo_buf, format)) {
2838 *PL_langinfo_buf = '\0';
2841 save_to_buffer(format, &PL_langinfo_buf,
2842 &PL_langinfo_bufsize, 0);
2853 return PL_langinfo_buf;
2860 * Initialize locale awareness.
2863 Perl_init_i18nl10n(pTHX_ int printwarn)
2867 * 0 if not to output warning when setup locale is bad
2868 * 1 if to output warning based on value of PERL_BADLANG
2869 * >1 if to output regardless of PERL_BADLANG
2872 * 1 = set ok or not applicable,
2873 * 0 = fallback to a locale of lower priority
2874 * -1 = fallback to all locales failed, not even to the C locale
2876 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2877 * set, debugging information is output.
2879 * This looks more complicated than it is, mainly due to the #ifdefs.
2881 * We try to set LC_ALL to the value determined by the environment. If
2882 * there is no LC_ALL on this platform, we try the individual categories we
2883 * know about. If this works, we are done.
2885 * But if it doesn't work, we have to do something else. We search the
2886 * environment variables ourselves instead of relying on the system to do
2887 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2888 * think there is one), and the ultimate fallback "C". This is all done in
2889 * the same loop as above to avoid duplicating code, but it makes things
2890 * more complex. The 'trial_locales' array is initialized with just one
2891 * element; it causes the behavior described in the paragraph above this to
2892 * happen. If that fails, we add elements to 'trial_locales', and do extra
2893 * loop iterations to cause the behavior described in this paragraph.
2895 * On Ultrix, the locale MUST come from the environment, so there is
2896 * preliminary code to set it. I (khw) am not sure that it is necessary,
2897 * and that this couldn't be folded into the loop, but barring any real
2898 * platforms to test on, it's staying as-is
2900 * A slight complication is that in embedded Perls, the locale may already
2901 * be set-up, and we don't want to get it from the normal environment
2902 * variables. This is handled by having a special environment variable
2903 * indicate we're in this situation. We simply set setlocale's 2nd
2904 * parameter to be a NULL instead of "". That indicates to setlocale that
2905 * it is not to change anything, but to return the current value,
2906 * effectively initializing perl's db to what the locale already is.
2908 * We play the same trick with NULL if a LC_ALL succeeds. We call
2909 * setlocale() on the individual categores with NULL to get their existing
2910 * values for our db, instead of trying to change them.
2917 PERL_UNUSED_ARG(printwarn);
2919 #else /* USE_LOCALE */
2922 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
2926 /* NULL uses the existing already set up locale */
2927 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2930 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2931 unsigned int trial_locales_count;
2932 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2933 const char * const lang = savepv(PerlEnv_getenv("LANG"));
2934 bool setlocale_failure = FALSE;
2937 /* A later getenv() could zap this, so only use here */
2938 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2940 const bool locwarn = (printwarn > 1
2942 && ( ! bad_lang_use_once
2944 /* disallow with "" or "0" */
2946 && strNE("0", bad_lang_use_once)))));
2948 /* setlocale() return vals; not copied so must be looked at immediately */
2949 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
2951 /* current locale for given category; should have been copied so aren't
2953 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
2957 /* In some systems you can find out the system default locale
2958 * and use that as the fallback locale. */
2959 # define SYSTEM_DEFAULT_LOCALE
2961 # ifdef SYSTEM_DEFAULT_LOCALE
2963 const char *system_default_locale = NULL;
2968 # define DEBUG_LOCALE_INIT(a,b,c)
2971 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
2973 # define DEBUG_LOCALE_INIT(category, locale, result) \
2975 if (debug_initialization) { \
2976 PerlIO_printf(Perl_debug_log, \
2978 __FILE__, __LINE__, \
2979 setlocale_debug_string(category, \
2985 /* Make sure the parallel arrays are properly set up */
2986 # ifdef USE_LOCALE_NUMERIC
2987 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2988 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
2989 # ifdef USE_POSIX_2008_LOCALE
2990 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
2993 # ifdef USE_LOCALE_CTYPE
2994 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2995 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
2996 # ifdef USE_POSIX_2008_LOCALE
2997 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3000 # ifdef USE_LOCALE_COLLATE
3001 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3002 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3003 # ifdef USE_POSIX_2008_LOCALE
3004 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3007 # ifdef USE_LOCALE_TIME
3008 assert(categories[LC_TIME_INDEX] == LC_TIME);
3009 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3010 # ifdef USE_POSIX_2008_LOCALE
3011 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3014 # ifdef USE_LOCALE_MESSAGES
3015 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3016 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3017 # ifdef USE_POSIX_2008_LOCALE
3018 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3021 # ifdef USE_LOCALE_MONETARY
3022 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3023 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3024 # ifdef USE_POSIX_2008_LOCALE
3025 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3028 # ifdef USE_LOCALE_ADDRESS
3029 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3030 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3031 # ifdef USE_POSIX_2008_LOCALE
3032 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3035 # ifdef USE_LOCALE_IDENTIFICATION
3036 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3037 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3038 # ifdef USE_POSIX_2008_LOCALE
3039 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3042 # ifdef USE_LOCALE_MEASUREMENT
3043 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3044 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3045 # ifdef USE_POSIX_2008_LOCALE
3046 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3049 # ifdef USE_LOCALE_PAPER
3050 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3051 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3052 # ifdef USE_POSIX_2008_LOCALE
3053 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3056 # ifdef USE_LOCALE_TELEPHONE
3057 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3058 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3059 # ifdef USE_POSIX_2008_LOCALE
3060 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3064 assert(categories[LC_ALL_INDEX] == LC_ALL);
3065 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3066 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3067 # ifdef USE_POSIX_2008_LOCALE
3068 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3071 # endif /* DEBUGGING */
3073 /* Initialize the cache of the program's UTF-8ness for the always known
3074 * locales C and POSIX */
3075 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3076 sizeof(PL_locale_utf8ness));
3078 # ifdef USE_THREAD_SAFE_LOCALE
3081 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3085 # if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3087 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3088 if (! PL_C_locale_obj) {
3089 Perl_croak_nocontext(
3090 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3092 if (DEBUG_Lv_TEST || debug_initialization) {
3093 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3098 PL_numeric_radix_sv = newSVpvs(".");
3100 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3102 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3103 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3106 # ifdef LOCALE_ENVIRON_REQUIRED
3109 * Ultrix setlocale(..., "") fails if there are no environment
3110 * variables from which to get a locale name.
3114 # error Ultrix without LC_ALL not implemented
3120 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3121 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3122 if (sl_result[LC_ALL_INDEX])
3125 setlocale_failure = TRUE;
3127 if (! setlocale_failure) {
3128 const char * locale_param;
3129 for (i = 0; i < LC_ALL_INDEX; i++) {
3130 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3133 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3134 if (! sl_result[i]) {
3135 setlocale_failure = TRUE;
3137 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3142 # endif /* LC_ALL */
3143 # endif /* LOCALE_ENVIRON_REQUIRED */
3145 /* We try each locale in the list until we get one that works, or exhaust
3146 * the list. Normally the loop is executed just once. But if setting the
3147 * locale fails, inside the loop we add fallback trials to the array and so
3148 * will execute the loop multiple times */
3149 trial_locales[0] = setlocale_init;
3150 trial_locales_count = 1;
3152 for (i= 0; i < trial_locales_count; i++) {
3153 const char * trial_locale = trial_locales[i];
3157 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3158 * when i==0, but I (khw) don't think that behavior makes much
3160 setlocale_failure = FALSE;
3162 # ifdef SYSTEM_DEFAULT_LOCALE
3163 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3165 /* On Windows machines, an entry of "" after the 0th means to use
3166 * the system default locale, which we now proceed to get. */
3167 if (strEQ(trial_locale, "")) {
3170 /* Note that this may change the locale, but we are going to do
3171 * that anyway just below */
3172 system_default_locale = do_setlocale_c(LC_ALL, "");
3173 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3175 /* Skip if invalid or if it's already on the list of locales to
3177 if (! system_default_locale) {
3178 goto next_iteration;
3180 for (j = 0; j < trial_locales_count; j++) {
3181 if (strEQ(system_default_locale, trial_locales[j])) {
3182 goto next_iteration;
3186 trial_locale = system_default_locale;
3189 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3191 # endif /* SYSTEM_DEFAULT_LOCALE */
3197 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3198 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3199 if (! sl_result[LC_ALL_INDEX]) {
3200 setlocale_failure = TRUE;
3203 /* Since LC_ALL succeeded, it should have changed all the other
3204 * categories it can to its value; so we massage things so that the
3205 * setlocales below just return their category's current values.
3206 * This adequately handles the case in NetBSD where LC_COLLATE may
3207 * not be defined for a locale, and setting it individually will
3208 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3209 * the POSIX locale. */
3210 trial_locale = NULL;
3213 # endif /* LC_ALL */
3215 if (! setlocale_failure) {
3217 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3219 = savepv(do_setlocale_r(categories[j], trial_locale));
3220 if (! curlocales[j]) {
3221 setlocale_failure = TRUE;
3223 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3226 if (! setlocale_failure) { /* All succeeded */
3227 break; /* Exit trial_locales loop */
3231 /* Here, something failed; will need to try a fallback. */
3237 if (locwarn) { /* Output failure info only on the first one */
3241 PerlIO_printf(Perl_error_log,
3242 "perl: warning: Setting locale failed.\n");
3244 # else /* !LC_ALL */
3246 PerlIO_printf(Perl_error_log,
3247 "perl: warning: Setting locale failed for the categories:\n\t");
3249 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3250 if (! curlocales[j]) {
3251 PerlIO_printf(Perl_error_log, category_names[j]);
3254 Safefree(curlocales[j]);
3258 # endif /* LC_ALL */
3260 PerlIO_printf(Perl_error_log,
3261 "perl: warning: Please check that your locale settings:\n");
3265 PerlIO_printf(Perl_error_log,
3266 "\tLANGUAGE = %c%s%c,\n",
3267 language ? '"' : '(',
3268 language ? language : "unset",
3269 language ? '"' : ')');
3272 PerlIO_printf(Perl_error_log,
3273 "\tLC_ALL = %c%s%c,\n",
3275 lc_all ? lc_all : "unset",
3276 lc_all ? '"' : ')');
3278 # if defined(USE_ENVIRON_ARRAY)
3283 /* Look through the environment for any variables of the
3284 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3285 * already handled above. These are assumed to be locale
3286 * settings. Output them and their values. */
3287 for (e = environ; *e; e++) {
3288 const STRLEN prefix_len = sizeof("LC_") - 1;
3291 if ( strBEGINs(*e, "LC_")
3292 && ! strBEGINs(*e, "LC_ALL=")
3293 && (uppers_len = strspn(*e + prefix_len,
3294 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3295 && ((*e)[prefix_len + uppers_len] == '='))
3297 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3298 (int) (prefix_len + uppers_len), *e,
3299 *e + prefix_len + uppers_len + 1);
3306 PerlIO_printf(Perl_error_log,
3307 "\t(possibly more locale environment variables)\n");
3311 PerlIO_printf(Perl_error_log,
3312 "\tLANG = %c%s%c\n",
3314 lang ? lang : "unset",
3317 PerlIO_printf(Perl_error_log,
3318 " are supported and installed on your system.\n");
3321 /* Calculate what fallback locales to try. We have avoided this
3322 * until we have to, because failure is quite unlikely. This will
3323 * usually change the upper bound of the loop we are in.
3325 * Since the system's default way of setting the locale has not
3326 * found one that works, We use Perl's defined ordering: LC_ALL,
3327 * LANG, and the C locale. We don't try the same locale twice, so
3328 * don't add to the list if already there. (On POSIX systems, the
3329 * LC_ALL element will likely be a repeat of the 0th element "",
3330 * but there's no harm done by doing it explicitly.
3332 * Note that this tries the LC_ALL environment variable even on
3333 * systems which have no LC_ALL locale setting. This may or may
3334 * not have been originally intentional, but there's no real need
3335 * to change the behavior. */
3337 for (j = 0; j < trial_locales_count; j++) {
3338 if (strEQ(lc_all, trial_locales[j])) {
3342 trial_locales[trial_locales_count++] = lc_all;
3347 for (j = 0; j < trial_locales_count; j++) {
3348 if (strEQ(lang, trial_locales[j])) {
3352 trial_locales[trial_locales_count++] = lang;
3356 # if defined(WIN32) && defined(LC_ALL)
3358 /* For Windows, we also try the system default locale before "C".
3359 * (If there exists a Windows without LC_ALL we skip this because
3360 * it gets too complicated. For those, the "C" is the next
3361 * fallback possibility). The "" is the same as the 0th element of
3362 * the array, but the code at the loop above knows to treat it
3363 * differently when not the 0th */
3364 trial_locales[trial_locales_count++] = "";
3368 for (j = 0; j < trial_locales_count; j++) {
3369 if (strEQ("C", trial_locales[j])) {
3373 trial_locales[trial_locales_count++] = "C";
3376 } /* end of first time through the loop */
3384 } /* end of looping through the trial locales */
3386 if (ok < 1) { /* If we tried to fallback */
3388 if (! setlocale_failure) { /* fallback succeeded */
3389 msg = "Falling back to";
3391 else { /* fallback failed */
3394 /* We dropped off the end of the loop, so have to decrement i to
3395 * get back to the value the last time through */
3399 msg = "Failed to fall back to";
3401 /* To continue, we should use whatever values we've got */
3403 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3404 Safefree(curlocales[j]);
3405 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3406 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3411 const char * description;
3412 const char * name = "";
3413 if (strEQ(trial_locales[i], "C")) {
3414 description = "the standard locale";
3418 # ifdef SYSTEM_DEFAULT_LOCALE
3420 else if (strEQ(trial_locales[i], "")) {
3421 description = "the system default locale";
3422 if (system_default_locale) {
3423 name = system_default_locale;
3427 # endif /* SYSTEM_DEFAULT_LOCALE */
3430 description = "a fallback locale";
3431 name = trial_locales[i];
3433 if (name && strNE(name, "")) {
3434 PerlIO_printf(Perl_error_log,
3435 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3438 PerlIO_printf(Perl_error_log,
3439 "perl: warning: %s %s.\n", msg, description);
3442 } /* End of tried to fallback */
3444 /* Done with finding the locales; update our records */
3446 # ifdef USE_LOCALE_CTYPE
3448 new_ctype(curlocales[LC_CTYPE_INDEX]);
3451 # ifdef USE_LOCALE_COLLATE
3453 new_collate(curlocales[LC_COLLATE_INDEX]);
3456 # ifdef USE_LOCALE_NUMERIC
3458 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3462 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3464 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3466 /* This caches whether each category's locale is UTF-8 or not. This
3467 * may involve changing the locale. It is ok to do this at
3468 * initialization time before any threads have started, but not later
3469 * unless thread-safe operations are used.
3470 * Caching means that if the program heeds our dictate not to change
3471 * locales in threaded applications, this data will remain valid, and
3472 * it may get queried without having to change locales. If the
3473 * environment is such that all categories have the same locale, this
3474 * isn't needed, as the code will not change the locale; but this
3475 * handles the uncommon case where the environment has disparate
3476 * locales for the categories */
3477 (void) _is_cur_LC_category_utf8(categories[i]);
3481 Safefree(curlocales[i]);
3484 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3486 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3487 * locale is UTF-8. The call to new_ctype() just above has already
3488 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3489 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3490 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3491 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3492 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3494 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3495 This is an alternative to using the -C command line switch
3496 (the -C if present will override this). */
3498 const char *p = PerlEnv_getenv("PERL_UNICODE");
3499 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3500 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3514 #endif /* USE_LOCALE */
3517 /* So won't continue to output stuff */
3518 DEBUG_INITIALIZATION_set(FALSE);
3525 #ifdef USE_LOCALE_COLLATE
3528 Perl__mem_collxfrm(pTHX_ const char *input_string,
3529 STRLEN len, /* Length of 'input_string' */
3530 STRLEN *xlen, /* Set to length of returned string
3531 (not including the collation index
3533 bool utf8 /* Is the input in UTF-8? */
3537 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3538 * differences. First, it handles embedded NULs. Second, it allocates a bit
3539 * more memory than needed for the transformed data itself. The real
3540 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3541 * the length of that, and doesn't include the collation index size.
3542 * Please see sv_collxfrm() to see how this is used. */
3544 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3546 char * s = (char *) input_string;
3547 STRLEN s_strlen = strlen(input_string);
3549 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3550 STRLEN length_in_chars;
3551 bool first_time = TRUE; /* Cleared after first loop iteration */
3553 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3555 /* Must be NUL-terminated */
3556 assert(*(input_string + len) == '\0');
3558 /* If this locale has defective collation, skip */
3559 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3560 DEBUG_L(PerlIO_printf(Perl_debug_log,
3561 "_mem_collxfrm: locale's collation is defective\n"));
3565 /* Replace any embedded NULs with the control that sorts before any others.
3566 * This will give as good as possible results on strings that don't
3567 * otherwise contain that character, but otherwise there may be
3568 * less-than-perfect results with that character and NUL. This is
3569 * unavoidable unless we replace strxfrm with our own implementation. */
3570 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3574 STRLEN sans_nuls_len;
3575 int try_non_controls;
3576 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3577 making sure 2nd byte is NUL.
3579 STRLEN this_replacement_len;
3581 /* If we don't know what non-NUL control character sorts lowest for
3582 * this locale, find it */
3583 if (PL_strxfrm_NUL_replacement == '\0') {
3585 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3586 includes the collation index
3589 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3591 /* Unlikely, but it may be that no control will work to replace
3592 * NUL, in which case we instead look for any character. Controls
3593 * are preferred because collation order is, in general, context
3594 * sensitive, with adjoining characters affecting the order, and
3595 * controls are less likely to have such interactions, allowing the
3596 * NUL-replacement to stand on its own. (Another way to look at it
3597 * is to imagine what would happen if the NUL were replaced by a
3598 * combining character; it wouldn't work out all that well.) */
3599 for (try_non_controls = 0;
3600 try_non_controls < 2;
3603 /* Look through all legal code points (NUL isn't) */
3604 for (j = 1; j < 256; j++) {
3605 char * x; /* j's xfrm plus collation index */
3606 STRLEN x_len; /* length of 'x' */
3607 STRLEN trial_len = 1;
3608 char cur_source[] = { '\0', '\0' };
3610 /* Skip non-controls the first time through the loop. The
3611 * controls in a UTF-8 locale are the L1 ones */
3612 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3619 /* Create a 1-char string of the current code point */
3620 cur_source[0] = (char) j;
3622 /* Then transform it */
3623 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3624 0 /* The string is not in UTF-8 */);
3626 /* Ignore any character that didn't successfully transform.
3632 /* If this character's transformation is lower than
3633 * the current lowest, this one becomes the lowest */
3634 if ( cur_min_x == NULL
3635 || strLT(x + COLLXFRM_HDR_LEN,
3636 cur_min_x + COLLXFRM_HDR_LEN))
3638 PL_strxfrm_NUL_replacement = j;
3644 } /* end of loop through all 255 characters */
3646 /* Stop looking if found */
3651 /* Unlikely, but possible, if there aren't any controls that
3652 * work in the locale, repeat the loop, looking for any
3653 * character that works */
3654 DEBUG_L(PerlIO_printf(Perl_debug_log,
3655 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3656 } /* End of loop to try first the controls, then any char */
3659 DEBUG_L(PerlIO_printf(Perl_debug_log,
3660 "_mem_collxfrm: Couldn't find any character to replace"
3661 " embedded NULs in locale %s with", PL_collation_name));
3665 DEBUG_L(PerlIO_printf(Perl_debug_log,
3666 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3667 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3669 Safefree(cur_min_x);
3670 } /* End of determining the character that is to replace NULs */
3672 /* If the replacement is variant under UTF-8, it must match the
3673 * UTF8-ness of the original */
3674 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3675 this_replacement_char[0] =
3676 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3677 this_replacement_char[1] =
3678 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3679 this_replacement_len = 2;
3682 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3683 /* this_replacement_char[1] = '\0' was done at initialization */
3684 this_replacement_len = 1;
3687 /* The worst case length for the replaced string would be if every
3688 * character in it is NUL. Multiply that by the length of each
3689 * replacement, and allow for a trailing NUL */
3690 sans_nuls_len = (len * this_replacement_len) + 1;
3691 Newx(sans_nuls, sans_nuls_len, char);
3694 /* Replace each NUL with the lowest collating control. Loop until have
3695 * exhausted all the NULs */
3696 while (s + s_strlen < e) {
3697 my_strlcat(sans_nuls, s, sans_nuls_len);
3699 /* Do the actual replacement */
3700 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3702 /* Move past the input NUL */
3704 s_strlen = strlen(s);
3707 /* And add anything that trails the final NUL */
3708 my_strlcat(sans_nuls, s, sans_nuls_len);
3710 /* Switch so below we transform this modified string */
3713 } /* End of replacing NULs */
3715 /* Make sure the UTF8ness of the string and locale match */
3716 if (utf8 != PL_in_utf8_COLLATE_locale) {
3717 /* XXX convert above Unicode to 10FFFF? */
3718 const char * const t = s; /* Temporary so we can later find where the
3721 /* Here they don't match. Change the string's to be what the locale is
3724 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3725 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3728 else { /* locale is not UTF-8; but input is; downgrade the input */
3730 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3732 /* If the downgrade was successful we are done, but if the input
3733 * contains things that require UTF-8 to represent, have to do
3734 * damage control ... */
3735 if (UNLIKELY(utf8)) {
3737 /* What we do is construct a non-UTF-8 string with
3738 * 1) the characters representable by a single byte converted
3739 * to be so (if necessary);
3740 * 2) and the rest converted to collate the same as the
3741 * highest collating representable character. That makes
3742 * them collate at the end. This is similar to how we
3743 * handle embedded NULs, but we use the highest collating
3744 * code point instead of the smallest. Like the NUL case,
3745 * this isn't perfect, but is the best we can reasonably
3746 * do. Every above-255 code point will sort the same as
3747 * the highest-sorting 0-255 code point. If that code
3748 * point can combine in a sequence with some other code
3749 * points for weight calculations, us changing something to
3750 * be it can adversely affect the results. But in most
3751 * cases, it should work reasonably. And note that this is
3752 * really an illegal situation: using code points above 255
3753 * on a locale where only 0-255 are valid. If two strings
3754 * sort entirely equal, then the sort order for the
3755 * above-255 code points will be in code point order. */
3759 /* If we haven't calculated the code point with the maximum
3760 * collating order for this locale, do so now */
3761 if (! PL_strxfrm_max_cp) {
3764 /* The current transformed string that collates the
3765 * highest (except it also includes the prefixed collation
3767 char * cur_max_x = NULL;
3769 /* Look through all legal code points (NUL isn't) */
3770 for (j = 1; j < 256; j++) {
3773 char cur_source[] = { '\0', '\0' };
3775 /* Create a 1-char string of the current code point */
3776 cur_source[0] = (char) j;
3778 /* Then transform it */
3779 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3781 /* If something went wrong (which it shouldn't), just
3782 * ignore this code point */
3787 /* If this character's transformation is higher than
3788 * the current highest, this one becomes the highest */
3789 if ( cur_max_x == NULL
3790 || strGT(x + COLLXFRM_HDR_LEN,
3791 cur_max_x + COLLXFRM_HDR_LEN))
3793 PL_strxfrm_max_cp = j;
3802 DEBUG_L(PerlIO_printf(Perl_debug_log,
3803 "_mem_collxfrm: Couldn't find any character to"
3804 " replace above-Latin1 chars in locale %s with",
3805 PL_collation_name));
3809 DEBUG_L(PerlIO_printf(Perl_debug_log,
3810 "_mem_collxfrm: highest 1-byte collating character"
3811 " in locale %s is 0x%02X\n",
3813 PL_strxfrm_max_cp));
3815 Safefree(cur_max_x);
3818 /* Here we know which legal code point collates the highest.
3819 * We are ready to construct the non-UTF-8 string. The length
3820 * will be at least 1 byte smaller than the input string
3821 * (because we changed at least one 2-byte character into a
3822 * single byte), but that is eaten up by the trailing NUL */
3828 char * e = (char *) t + len;
3830 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3832 if (UTF8_IS_INVARIANT(cur_char)) {
3835 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
3836 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3838 else { /* Replace illegal cp with highest collating
3840 s[d++] = PL_strxfrm_max_cp;
3844 Renew(s, d, char); /* Free up unused space */
3849 /* Here, we have constructed a modified version of the input. It could
3850 * be that we already had a modified copy before we did this version.
3851 * If so, that copy is no longer needed */
3852 if (t != input_string) {
3857 length_in_chars = (utf8)
3858 ? utf8_length((U8 *) s, (U8 *) s + len)
3861 /* The first element in the output is the collation id, used by
3862 * sv_collxfrm(); then comes the space for the transformed string. The
3863 * equation should give us a good estimate as to how much is needed */
3864 xAlloc = COLLXFRM_HDR_LEN
3866 + (PL_collxfrm_mult * length_in_chars);
3867 Newx(xbuf, xAlloc, char);
3868 if (UNLIKELY(! xbuf)) {
3869 DEBUG_L(PerlIO_printf(Perl_debug_log,
3870 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
3874 /* Store the collation id */
3875 *(U32*)xbuf = PL_collation_ix;
3877 /* Then the transformation of the input. We loop until successful, or we
3881 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
3883 /* If the transformed string occupies less space than we told strxfrm()
3884 * was available, it means it successfully transformed the whole
3886 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
3888 /* Some systems include a trailing NUL in the returned length.
3889 * Ignore it, using a loop in case multiple trailing NULs are
3892 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3897 /* If the first try didn't get it, it means our prediction was low.
3898 * Modify the coefficients so that we predict a larger value in any
3899 * future transformations */
3901 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3902 STRLEN computed_guess = PL_collxfrm_base
3903 + (PL_collxfrm_mult * length_in_chars);
3905 /* On zero-length input, just keep current slope instead of
3907 const STRLEN new_m = (length_in_chars != 0)
3908 ? needed / length_in_chars
3911 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3912 "%s: %d: initial size of %zu bytes for a length "
3913 "%zu string was insufficient, %zu needed\n",
3915 computed_guess, length_in_chars, needed));
3917 /* If slope increased, use it, but discard this result for
3918 * length 1 strings, as we can't be sure that it's a real slope
3920 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
3924 STRLEN old_m = PL_collxfrm_mult;
3925 STRLEN old_b = PL_collxfrm_base;
3929 PL_collxfrm_mult = new_m;
3930 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3931 computed_guess = PL_collxfrm_base
3932 + (PL_collxfrm_mult * length_in_chars);
3933 if (computed_guess < needed) {
3934 PL_collxfrm_base += needed - computed_guess;
3937 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3938 "%s: %d: slope is now %zu; was %zu, base "
3939 "is now %zu; was %zu\n",
3941 PL_collxfrm_mult, old_m,
3942 PL_collxfrm_base, old_b));
3944 else { /* Slope didn't change, but 'b' did */
3945 const STRLEN new_b = needed
3948 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3949 "%s: %d: base is now %zu; was %zu\n",
3951 new_b, PL_collxfrm_base));
3952 PL_collxfrm_base = new_b;
3959 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3960 DEBUG_L(PerlIO_printf(Perl_debug_log,
3961 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3962 *xlen, PERL_INT_MAX));
3966 /* A well-behaved strxfrm() returns exactly how much space it needs
3967 * (usually not including the trailing NUL) when it fails due to not
3968 * enough space being provided. Assume that this is the case unless
3969 * it's been proven otherwise */
3970 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
3971 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
3973 else { /* Here, either:
3974 * 1) The strxfrm() has previously shown bad behavior; or
3975 * 2) It isn't the first time through the loop, which means
3976 * that the strxfrm() is now showing bad behavior, because
3977 * we gave it what it said was needed in the previous
3978 * iteration, and it came back saying it needed still more.
3979 * (Many versions of cygwin fit this. When the buffer size
3980 * isn't sufficient, they return the input size instead of
3981 * how much is needed.)
3982 * Increase the buffer size by a fixed percentage and try again.
3984 xAlloc += (xAlloc / 4) + 1;
3985 PL_strxfrm_is_behaved = FALSE;
3989 if (DEBUG_Lv_TEST || debug_initialization) {
3990 PerlIO_printf(Perl_debug_log,
3991 "_mem_collxfrm required more space than previously calculated"
3992 " for locale %s, trying again with new guess=%d+%zu\n",
3993 PL_collation_name, (int) COLLXFRM_HDR_LEN,
3994 xAlloc - COLLXFRM_HDR_LEN);
4001 Renew(xbuf, xAlloc, char);
4002 if (UNLIKELY(! xbuf)) {
4003 DEBUG_L(PerlIO_printf(Perl_debug_log,
4004 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4014 if (DEBUG_Lv_TEST || debug_initialization) {
4016 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4017 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4018 PerlIO_printf(Perl_debug_log, "%s\n",
4019 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4025 /* Free up unneeded space; retain ehough for trailing NUL */
4026 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4028 if (s != input_string) {
4036 if (s != input_string) {
4043 if (DEBUG_Lv_TEST || debug_initialization) {
4044 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4055 S_print_collxfrm_input_and_return(pTHX_
4056 const char * const s,
4057 const char * const e,
4058 const STRLEN * const xlen,
4062 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4064 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4065 (UV)PL_collation_ix);
4067 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4070 PerlIO_printf(Perl_debug_log, "NULL");
4072 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4074 print_bytes_for_locale(s, e, is_utf8);
4076 PerlIO_printf(Perl_debug_log, "'\n");
4080 S_print_bytes_for_locale(pTHX_
4081 const char * const s,
4082 const char * const e,
4086 bool prev_was_printable = TRUE;
4087 bool first_time = TRUE;
4089 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4093 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4096 if (! prev_was_printable) {
4097 PerlIO_printf(Perl_debug_log, " ");
4099 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4100 prev_was_printable = TRUE;
4104 PerlIO_printf(Perl_debug_log, " ");
4106 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4107 prev_was_printable = FALSE;
4109 t += (is_utf8) ? UTF8SKIP(t) : 1;
4114 # endif /* #ifdef DEBUGGING */
4115 #endif /* USE_LOCALE_COLLATE */
4120 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4122 /* Changes the locale for LC_'switch_category" to that of
4123 * LC_'template_category', if they aren't already the same. If not NULL,
4124 * 'template_locale' is the locale that 'template_category' is in.
4126 * Returns a copy of the name of the original locale for 'switch_category'
4127 * so can be switched back to with the companion function
4128 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4130 char * restore_to_locale = NULL;
4132 if (switch_category == template_category) { /* No changes needed */
4136 /* Find the original locale of the category we may need to change, so that
4137 * it can be restored to later */
4138 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4140 if (! restore_to_locale) {
4142 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4143 __FILE__, __LINE__, category_name(switch_category), errno);
4146 /* If the locale of the template category wasn't passed in, find it now */
4147 if (template_locale == NULL) {
4148 template_locale = do_setlocale_r(template_category, NULL);
4149 if (! template_locale) {
4151 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4152 __FILE__, __LINE__, category_name(template_category), errno);
4156 /* It the locales are the same, there's nothing to do */
4157 if (strEQ(restore_to_locale, template_locale)) {
4158 Safefree(restore_to_locale);
4160 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4161 category_name(switch_category), restore_to_locale));
4166 /* Finally, change the locale to the template one */
4167 if (! do_setlocale_r(switch_category, template_locale)) {
4169 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4170 __FILE__, __LINE__, category_name(switch_category),
4171 template_locale, errno);
4174 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4175 category_name(switch_category), template_locale));
4177 return restore_to_locale;
4181 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4183 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4184 * copy that will be freed by this function), or do nothing if the latter
4185 * parameter is NULL */
4187 if (original_locale == NULL) {
4191 if (! do_setlocale_r(category, original_locale)) {
4193 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4195 category_name(category), original_locale, errno);
4198 Safefree(original_locale);
4202 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4204 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4205 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4206 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4207 * could give the wrong result. The result will very likely be correct for
4208 * languages that have commonly used non-ASCII characters, but for notably
4209 * English, it comes down to if the locale's name ends in something like
4210 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4212 * If the platform is early C89, not containing mbtowc(), or we are
4213 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4214 * These work very well for non-Latin locales or those whose currency
4215 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4216 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4217 * on the name containing UTF-8 or not. */
4219 /* Name of current locale corresponding to the input category */
4220 const char *save_input_locale = NULL;
4222 bool is_utf8 = FALSE; /* The return value */
4224 /* The variables below are for the cache of previous lookups using this
4225 * function. The cache is a C string, described at the definition for
4226 * 'C_and_POSIX_utf8ness'.
4228 * The first part of the cache is fixed, for the C and POSIX locales. The
4229 * varying part starts just after them. */
4230 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4232 Size_t utf8ness_cache_size; /* Size of the varying portion */
4233 Size_t input_name_len; /* Length in bytes of save_input_locale */
4234 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4235 the name in the cache */
4236 char * delimited; /* The name plus the delimiters used to store
4238 char * name_pos; /* position of 'delimited' in the cache, or 0
4244 assert(category != LC_ALL);
4248 /* Get the desired category's locale */
4249 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4250 if (! save_input_locale) {
4252 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4253 __FILE__, __LINE__, category_name(category), errno);
4256 DEBUG_L(PerlIO_printf(Perl_debug_log,
4257 "Current locale for %s is %s\n",
4258 category_name(category), save_input_locale));
4260 input_name_len = strlen(save_input_locale);
4262 /* In our cache, each name is accompanied by two delimiters and a single
4264 input_name_len_with_overhead = input_name_len + 3;
4266 /* Allocate and populate space for a copy of the name surrounded by the
4268 Newx(delimited, input_name_len_with_overhead, char);
4269 delimited[0] = UTF8NESS_SEP[0];
4270 Copy(save_input_locale, delimited + 1, input_name_len, char);
4271 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4272 delimited[input_name_len+2] = '\0';
4274 /* And see if that is in the cache */
4275 name_pos = instr(PL_locale_utf8ness, delimited);
4277 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4281 if (DEBUG_Lv_TEST || debug_initialization) {
4282 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4283 save_input_locale, is_utf8);
4288 /* And, if not already in that position, move it to the beginning of
4289 * the non-constant portion of the list, since it is the most recently
4290 * used. (We don't have to worry about overflow, since just moving
4291 * existing names around) */
4292 if (name_pos > utf8ness_cache) {
4293 Move(utf8ness_cache,
4294 utf8ness_cache + input_name_len_with_overhead,
4295 name_pos - utf8ness_cache, char);
4298 input_name_len_with_overhead - 1, char);
4299 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4302 Safefree(delimited);
4303 Safefree(save_input_locale);
4307 /* Here we don't have stored the utf8ness for the input locale. We have to
4310 # if defined(USE_LOCALE_CTYPE) \
4311 && ( (defined(HAS_NL_LANGINFO) && defined(CODESET)) \
4312 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4315 const char *original_ctype_locale
4316 = switch_category_locale_to_template(LC_CTYPE,
4320 /* Here the current LC_CTYPE is set to the locale of the category whose
4321 * information is desired. This means that nl_langinfo() and mbtowc()
https://perl5.git.perl.org / perl5.git / blob