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_CTYPE
336 # ifdef USE_LOCALE_TIME
337 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
338 # define _DUMMY_TIME LC_TIME_INDEX
340 # define _DUMMY_TIME _DUMMY_COLLATE
342 # ifdef USE_LOCALE_MESSAGES
343 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
346 # define _DUMMY_MESSAGES _DUMMY_TIME
348 # ifdef USE_LOCALE_MONETARY
349 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350 # define _DUMMY_MONETARY LC_MONETARY_INDEX
352 # define _DUMMY_MONETARY _DUMMY_MESSAGES
354 # ifdef USE_LOCALE_ADDRESS
355 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
358 # define _DUMMY_ADDRESS _DUMMY_MONETARY
360 # ifdef USE_LOCALE_IDENTIFICATION
361 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
364 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
366 # ifdef USE_LOCALE_MEASUREMENT
367 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
370 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
372 # ifdef USE_LOCALE_PAPER
373 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374 # define _DUMMY_PAPER LC_PAPER_INDEX
376 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
378 # ifdef USE_LOCALE_TELEPHONE
379 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
382 # define _DUMMY_TELEPHONE _DUMMY_PAPER
385 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
387 #endif /* ifdef USE_LOCALE */
389 /* Windows requres a customized base-level setlocale() */
391 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
393 # define my_setlocale(cat, locale) setlocale(cat, locale)
396 #ifndef USE_POSIX_2008_LOCALE
398 /* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
403 #else /* Below uses POSIX 2008 */
405 /* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411 # define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
415 /* A third array, parallel to the ones above to map from category to its
417 const int category_masks[] = {
418 # ifdef USE_LOCALE_NUMERIC
421 # ifdef USE_LOCALE_CTYPE
424 # ifdef USE_LOCALE_COLLATE
427 # ifdef USE_LOCALE_TIME
430 # ifdef USE_LOCALE_MESSAGES
433 # ifdef USE_LOCALE_MONETARY
436 # ifdef USE_LOCALE_ADDRESS
439 # ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
442 # ifdef USE_LOCALE_MEASUREMENT
445 # ifdef USE_LOCALE_PAPER
448 # ifdef USE_LOCALE_TELEPHONE
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
460 S_emulate_setlocale(const int category,
463 const bool is_index_valid
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
554 mask = category_masks[index];
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
580 # ifdef HAS_QUERYLOCALE
582 return (char *) querylocale(mask, cur_obj);
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
589 # if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
608 # ifdef USE_C_BACKTRACE
610 dump_c_backtrace(Perl_debug_log, 20, 1);
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
626 /* Without querylocale(), we have to use our record-keeping we've
629 if (category != LC_ALL) {
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
639 return PL_curlocales[index];
641 else { /* For LC_ALL */
643 Size_t names_len = 0;
645 bool are_all_categories_the_same_locale = TRUE;
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
658 return PL_curlocales[LC_ALL_INDEX];
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
666 for (i = 0; i < LC_ALL_INDEX; i++) {
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
676 names_len += strlen(category_names[i])
678 + strlen(PL_curlocales[i])
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
728 SETERRNO(EINVAL, LIB_INVARG);
738 /* Here, we are switching locales. */
740 # ifndef HAS_QUERYLOCALE
742 if (strEQ(locale, "")) {
744 /* For non-querylocale() systems, we do the setting of "" ourselves to
745 * be sure that we really know what's going on. We follow the Linux
746 * documented behavior (but if that differs from the actual behavior,
747 * this won't work exactly as the OS implements). We go out and
748 * examine the environment based on our understanding of how the system
749 * works, and use that to figure things out */
751 const char * const lc_all = PerlEnv_getenv("LC_ALL");
753 /* Use any "LC_ALL" environment variable, as it overrides everything
755 if (lc_all && strNE(lc_all, "")) {
760 /* Otherwise, we need to dig deeper. Unless overridden, the
761 * default is the LANG environment variable; if it doesn't exist,
764 const char * default_name;
766 /* To minimize other threads messing with the environment, we copy
767 * the variable, making it a temporary. But this doesn't work upon
768 * program initialization before any scopes are created, and at
769 * this time, there's nothing else going on that would interfere.
770 * So skip the copy in that case */
771 if (PL_scopestack_ix == 0) {
772 default_name = PerlEnv_getenv("LANG");
775 default_name = savepv(PerlEnv_getenv("LANG"));
778 if (! default_name || strEQ(default_name, "")) {
781 else if (PL_scopestack_ix != 0) {
782 SAVEFREEPV(default_name);
785 if (category != LC_ALL) {
786 const char * const name = PerlEnv_getenv(category_names[index]);
788 /* Here we are setting a single category. Assume will have the
790 locale = default_name;
792 /* But then look for an overriding environment variable */
793 if (name && strNE(name, "")) {
798 bool did_override = FALSE;
801 /* Here, we are getting LC_ALL. Any categories that don't have
802 * a corresponding environment variable set should be set to
803 * LANG, or to "C" if there is no LANG. If no individual
804 * categories differ from this, we can just set LC_ALL. This
805 * is buggy on systems that have extra categories that we don't
806 * know about. If there is an environment variable that sets
807 * that category, we won't know to look for it, and so our use
808 * of LANG or "C" improperly overrides it. On the other hand,
809 * if we don't do what is done here, and there is no
810 * environment variable, the category's locale should be set to
811 * LANG or "C". So there is no good solution. khw thinks the
812 * best is to look at systems to see what categories they have,
813 * and include them, and then to assume that we know the
816 for (i = 0; i < LC_ALL_INDEX; i++) {
817 const char * const env_override
818 = savepv(PerlEnv_getenv(category_names[i]));
819 const char * this_locale = ( env_override
820 && strNE(env_override, ""))
823 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
825 Safefree(env_override);
829 if (strNE(this_locale, default_name)) {
833 Safefree(env_override);
836 /* If all the categories are the same, we can set LC_ALL to
838 if (! did_override) {
839 locale = default_name;
843 /* Here, LC_ALL is no longer valid, as some individual
844 * categories don't match it. We call ourselves
845 * recursively, as that will execute the code that
846 * generates the proper locale string for this situation.
847 * We don't do the remainder of this function, as that is
848 * to update our records, and we've just done that for the
849 * individual categories in the loop above, and doing so
850 * would cause LC_ALL to be done as well */
851 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
856 else if (strchr(locale, ';')) {
858 /* LC_ALL may actually incude a conglomeration of various categories.
859 * Without querylocale, this code uses the glibc (as of this writing)
860 * syntax for representing that, but that is not a stable API, and
861 * other platforms do it differently, so we have to handle all cases
865 const char * s = locale;
866 const char * e = locale + strlen(locale);
868 const char * category_end;
869 const char * name_start;
870 const char * name_end;
872 /* If the string that gives what to set doesn't include all categories,
873 * the omitted ones get set to "C". To get this behavior, first set
874 * all the individual categories to "C", and override the furnished
876 for (i = 0; i < LC_ALL_INDEX; i++) {
877 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
884 /* Parse through the category */
885 while (isWORDCHAR(*p)) {
892 "panic: %s: %d: Unexpected character in locale name '%02X",
893 __FILE__, __LINE__, *(p-1));
896 /* Parse through the locale name */
898 while (p < e && *p != ';') {
901 "panic: %s: %d: Unexpected character in locale name '%02X",
902 __FILE__, __LINE__, *(p-1));
908 /* Space past the semi-colon */
913 /* Find the index of the category name in our lists */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
915 char * individ_locale;
917 /* Keep going if this isn't the index. The strnNE() avoids a
918 * Perl_form(), but would fail if ever a category name could be
919 * a substring of another one, like if there were a
921 if strnNE(s, category_names[i], category_end - s) {
925 /* If this index is for the single category we're changing, we
926 * have found the locale to set it to. */
927 if (category == categories[i]) {
928 locale = Perl_form(aTHX_ "%.*s",
929 (int) (name_end - name_start),
934 assert(category == LC_ALL);
935 individ_locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start), name_start);
937 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
946 /* Here we have set all the individual categories by recursive calls.
947 * These collectively should have fixed up LC_ALL, so can just query
948 * what that now is */
949 assert(category == LC_ALL);
951 return do_setlocale_c(LC_ALL, NULL);
956 /* Here at the end of having to deal with the absence of querylocale().
957 * Some cases have already been fully handled by recursive calls to this
958 * function. But at this point, we haven't dealt with those, but are now
959 * prepared to, knowing what the locale name to set this category to is.
960 * This would have come for free if this system had had querylocale() */
962 # endif /* end of ! querylocale */
964 assert(PL_C_locale_obj);
966 /* Switching locales generally entails freeing the current one's space (at
967 * the C library's discretion). We need to stop using that locale before
968 * the switch. So switch to a known locale object that we don't otherwise
969 * mess with. This returns the locale object in effect at the time of the
971 old_obj = uselocale(PL_C_locale_obj);
975 if (DEBUG_Lv_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
985 if (DEBUG_L_TEST || debug_initialization) {
987 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
998 if (DEBUG_Lv_TEST || debug_initialization) {
999 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1004 /* If we weren't in a thread safe locale, set so that newlocale() below
1005 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1006 The latter is defensive coding, so that, even if there is some bug, we
1007 will never end up trying to modify either of these, as if passed to
1008 newlocale(), they can be. */
1009 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1010 old_obj = (locale_t) 0;
1013 /* Ready to create a new locale by modification of the exising one */
1014 new_obj = newlocale(mask, locale, old_obj);
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1027 if (! uselocale(old_obj)) {
1031 if (DEBUG_L_TEST || debug_initialization) {
1032 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1044 if (DEBUG_Lv_TEST || debug_initialization) {
1045 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p; should have freed %p\n", __FILE__, __LINE__, new_obj, old_obj);
1050 /* And switch into it */
1051 if (! uselocale(new_obj)) {
1056 if (DEBUG_L_TEST || debug_initialization) {
1057 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1062 if (! uselocale(old_obj)) {
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1073 freelocale(new_obj);
1080 if (DEBUG_Lv_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1086 /* We are done, except for updating our records (if the system doesn't keep
1087 * them) and in the case of locale "", we don't actually know what the
1088 * locale that got switched to is, as it came from the environment. So
1089 * have to find it */
1091 # ifdef HAS_QUERYLOCALE
1093 if (strEQ(locale, "")) {
1094 locale = querylocale(mask, new_obj);
1099 /* Here, 'locale' is the return value */
1101 /* Without querylocale(), we have to update our records */
1103 if (category == LC_ALL) {
1106 /* For LC_ALL, we change all individual categories to correspond */
1107 /* PL_curlocales is a parallel array, so has same
1108 * length as 'categories' */
1109 for (i = 0; i <= LC_ALL_INDEX; i++) {
1110 Safefree(PL_curlocales[i]);
1111 PL_curlocales[i] = savepv(locale);
1116 /* For a single category, if it's not the same as the one in LC_ALL, we
1119 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1120 Safefree(PL_curlocales[LC_ALL_INDEX]);
1121 PL_curlocales[LC_ALL_INDEX] = NULL;
1124 /* Then update the category's record */
1125 Safefree(PL_curlocales[index]);
1126 PL_curlocales[index] = savepv(locale);
1134 #endif /* USE_POSIX_2008_LOCALE */
1136 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1137 didn't natively support them */
1139 /* The way this would work is that we would keep a per-thread list of the
1140 * correct locale for that thread. Any operation that was locale-sensitive
1141 * would have to be changed so that it would look like this:
1144 * setlocale to the correct locale for this operation
1148 * This leaves the global locale in the most recently used operation's, but it
1149 * was locked long enough to get the result. If that result is static, it
1150 * needs to be copied before the unlock.
1152 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1153 * the setup, but are no-ops when not needed, and similarly,
1154 * END_LOCALE_DEPENDENT_OP for the tear-down
1156 * But every call to a locale-sensitive function would have to be changed, and
1157 * if a module didn't cooperate by using the mutex, things would break.
1159 * This code was abandoned before being completed or tested, and is left as-is
1162 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1163 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1166 S_locking_setlocale(pTHX_
1168 const char * locale,
1170 const bool is_index_valid
1173 /* This function kind of performs a setlocale() on just the current thread;
1174 * thus it is kind of thread-safe. It does this by keeping a thread-level
1175 * array of the current locales for each category. Every time a locale is
1176 * switched to, it does the switch globally, but updates the thread's
1177 * array. A query as to what the current locale is just returns the
1178 * appropriate element from the array, and doesn't actually call the system
1179 * setlocale(). The saving into the array is done in an uninterruptible
1180 * section of code, so is unaffected by whatever any other threads might be
1183 * All locale-sensitive operations must work by first starting a critical
1184 * section, then switching to the thread's locale as kept by this function,
1185 * and then doing the operation, then ending the critical section. Thus,
1186 * each gets done in the appropriate locale. simulating thread-safety.
1188 * This function takes the same parameters, 'category' and 'locale', that
1189 * the regular setlocale() function does, but it also takes two additional
1190 * ones. This is because as described earlier. If we know on input the
1191 * index corresponding to the category into the array where we store the
1192 * current locales, we don't have to calculate it. If the caller knows at
1193 * compile time what the index is, it it can pass it, setting
1194 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1198 /* If the input index might be incorrect, calculate the correct one */
1199 if (! is_index_valid) {
1202 if (DEBUG_Lv_TEST || debug_initialization) {
1203 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1206 for (i = 0; i <= LC_ALL_INDEX; i++) {
1207 if (category == categories[i]) {
1213 /* Here, we don't know about this category, so can't handle it.
1214 * XXX best we can do is to unsafely set this
1217 return my_setlocale(category, locale);
1221 if (DEBUG_Lv_TEST || debug_initialization) {
1222 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1226 /* For a query, just return what's in our records */
1227 if (new_locale == NULL) {
1228 return curlocales[index];
1232 /* Otherwise, we need to do the switch, and save the result, all in a
1233 * critical section */
1235 Safefree(curlocales[[index]]);
1237 /* It might be that this is called from an already-locked section of code.
1238 * We would have to detect and skip the LOCK/UNLOCK if so */
1241 curlocales[index] = savepv(my_setlocale(category, new_locale));
1243 if (strEQ(new_locale, "")) {
1247 /* The locale values come from the environment, and may not all be the
1248 * same, so for LC_ALL, we have to update all the others, while the
1249 * mutex is still locked */
1251 if (category == LC_ALL) {
1253 for (i = 0; i < LC_ALL_INDEX) {
1254 curlocales[i] = my_setlocale(categories[i], NULL);
1263 return curlocales[index];
1270 S_set_numeric_radix(pTHX_ const bool use_locale)
1272 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1273 * TRUE, use the radix character derived from the current locale */
1275 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1276 || defined(HAS_NL_LANGINFO))
1278 const char * radix = (use_locale)
1279 ? my_nl_langinfo(RADIXCHAR, FALSE)
1280 /* FALSE => already in dest locale */
1283 sv_setpv(PL_numeric_radix_sv, radix);
1285 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1286 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1287 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1288 SvCUR(PL_numeric_radix_sv))
1289 && _is_cur_LC_category_utf8(LC_NUMERIC))
1291 SvUTF8_on(PL_numeric_radix_sv);
1296 if (DEBUG_L_TEST || debug_initialization) {
1297 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1298 SvPVX(PL_numeric_radix_sv),
1299 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1305 PERL_UNUSED_ARG(use_locale);
1307 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1312 S_new_numeric(pTHX_ const char *newnum)
1315 #ifndef USE_LOCALE_NUMERIC
1317 PERL_UNUSED_ARG(newnum);
1321 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1322 * core Perl this and that 'newnum' is the name of the new locale.
1323 * It installs this locale as the current underlying default.
1325 * The default locale and the C locale can be toggled between by use of the
1326 * set_numeric_underlying() and set_numeric_standard() functions, which
1327 * should probably not be called directly, but only via macros like
1328 * SET_NUMERIC_STANDARD() in perl.h.
1330 * The toggling is necessary mainly so that a non-dot radix decimal point
1331 * character can be output, while allowing internal calculations to use a
1334 * This sets several interpreter-level variables:
1335 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1336 * PL_numeric_underlying A boolean indicating if the toggled state is such
1337 * that the current locale is the program's underlying
1339 * PL_numeric_standard An int indicating if the toggled state is such
1340 * that the current locale is the C locale or
1341 * indistinguishable from the C locale. If non-zero, it
1342 * is in C; if > 1, it means it may not be toggled away
1344 * PL_numeric_underlying_is_standard A bool kept by this function
1345 * indicating that the underlying locale and the standard
1346 * C locale are indistinguishable for the purposes of
1347 * LC_NUMERIC. This happens when both of the above two
1348 * variables are true at the same time. (Toggling is a
1349 * no-op under these circumstances.) This variable is
1350 * used to avoid having to recalculate.
1356 Safefree(PL_numeric_name);
1357 PL_numeric_name = NULL;
1358 PL_numeric_standard = TRUE;
1359 PL_numeric_underlying = TRUE;
1360 PL_numeric_underlying_is_standard = TRUE;
1364 save_newnum = stdize_locale(savepv(newnum));
1365 PL_numeric_underlying = TRUE;
1366 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1368 #ifndef TS_W32_BROKEN_LOCALECONV
1370 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1371 * them. But on broken Windows systems calling my_nl_langinfo() for
1372 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1373 * and just always change the locale if not C nor POSIX on those systems */
1374 if (! PL_numeric_standard) {
1375 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1376 FALSE /* Don't toggle locale */ ))
1377 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1382 /* Save the new name if it isn't the same as the previous one, if any */
1383 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1384 Safefree(PL_numeric_name);
1385 PL_numeric_name = save_newnum;
1388 Safefree(save_newnum);
1391 PL_numeric_underlying_is_standard = PL_numeric_standard;
1393 # ifdef HAS_POSIX_2008_LOCALE
1395 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1397 PL_underlying_numeric_obj);
1401 if (DEBUG_L_TEST || debug_initialization) {
1402 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1405 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1406 * have to worry about the radix being a non-dot. (Core operations that
1407 * need the underlying locale change to it temporarily). */
1408 if (PL_numeric_standard) {
1409 set_numeric_radix(0);
1412 set_numeric_standard();
1415 #endif /* USE_LOCALE_NUMERIC */
1420 Perl_set_numeric_standard(pTHX)
1423 #ifdef USE_LOCALE_NUMERIC
1425 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1426 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1427 * macro avoids calling this routine if toggling isn't necessary according
1428 * to our records (which could be wrong if some XS code has changed the
1429 * locale behind our back) */
1433 if (DEBUG_L_TEST || debug_initialization) {
1434 PerlIO_printf(Perl_debug_log,
1435 "Setting LC_NUMERIC locale to standard C\n");
1440 do_setlocale_c(LC_NUMERIC, "C");
1441 PL_numeric_standard = TRUE;
1442 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1443 set_numeric_radix(0);
1445 #endif /* USE_LOCALE_NUMERIC */
1450 Perl_set_numeric_underlying(pTHX)
1453 #ifdef USE_LOCALE_NUMERIC
1455 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1456 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1457 * instead of calling this directly. The macro avoids calling this routine
1458 * if toggling isn't necessary according to our records (which could be
1459 * wrong if some XS code has changed the locale behind our back) */
1463 if (DEBUG_L_TEST || debug_initialization) {
1464 PerlIO_printf(Perl_debug_log,
1465 "Setting LC_NUMERIC locale to %s\n",
1471 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1472 PL_numeric_standard = PL_numeric_underlying_is_standard;
1473 PL_numeric_underlying = TRUE;
1474 set_numeric_radix(! PL_numeric_standard);
1476 #endif /* USE_LOCALE_NUMERIC */
1481 * Set up for a new ctype locale.
1484 S_new_ctype(pTHX_ const char *newctype)
1487 #ifndef USE_LOCALE_CTYPE
1489 PERL_UNUSED_ARG(newctype);
1490 PERL_UNUSED_CONTEXT;
1494 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1495 * core Perl this and that 'newctype' is the name of the new locale.
1497 * This function sets up the folding arrays for all 256 bytes, assuming
1498 * that tofold() is tolc() since fold case is not a concept in POSIX,
1500 * Any code changing the locale (outside this file) should use
1501 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1502 * this function should be called directly only from this file and from
1503 * POSIX::setlocale() */
1508 /* Don't check for problems if we are suppressing the warnings */
1509 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1511 PERL_ARGS_ASSERT_NEW_CTYPE;
1513 /* We will replace any bad locale warning with 1) nothing if the new one is
1514 * ok; or 2) a new warning for the bad new locale */
1515 if (PL_warn_locale) {
1516 SvREFCNT_dec_NN(PL_warn_locale);
1517 PL_warn_locale = NULL;
1520 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1522 /* A UTF-8 locale gets standard rules. But note that code still has to
1523 * handle this specially because of the three problematic code points */
1524 if (PL_in_utf8_CTYPE_locale) {
1525 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1528 /* We don't populate the other lists if a UTF-8 locale, but do check that
1529 * everything works as expected, unless checking turned off */
1530 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1531 /* Assume enough space for every character being bad. 4 spaces each
1532 * for the 94 printable characters that are output like "'x' "; and 5
1533 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1535 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1536 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1538 unsigned int bad_count = 0; /* Count of bad characters */
1540 for (i = 0; i < 256; i++) {
1541 if (! PL_in_utf8_CTYPE_locale) {
1543 PL_fold_locale[i] = (U8) tolower(i);
1544 else if (islower(i))
1545 PL_fold_locale[i] = (U8) toupper(i);
1547 PL_fold_locale[i] = (U8) i;
1550 /* If checking for locale problems, see if the native ASCII-range
1551 * printables plus \n and \t are in their expected categories in
1552 * the new locale. If not, this could mean big trouble, upending
1553 * Perl's and most programs' assumptions, like having a
1554 * metacharacter with special meaning become a \w. Fortunately,
1555 * it's very rare to find locales that aren't supersets of ASCII
1556 * nowadays. It isn't a problem for most controls to be changed
1557 * into something else; we check only \n and \t, though perhaps \r
1558 * could be an issue as well. */
1559 if ( check_for_problems
1560 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1562 bool is_bad = FALSE;
1563 char name[4] = { '\0' };
1565 /* Convert the name into a string */
1570 else if (i == '\n') {
1571 my_strlcpy(name, "\\n", sizeof(name));
1573 else if (i == '\t') {
1574 my_strlcpy(name, "\\t", sizeof(name));
1578 my_strlcpy(name, "' '", sizeof(name));
1581 /* Check each possibe class */
1582 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isalnum('%s') unexpectedly is %d\n",
1586 name, cBOOL(isalnum(i))));
1588 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isalpha('%s') unexpectedly is %d\n",
1592 name, cBOOL(isalpha(i))));
1594 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isdigit('%s') unexpectedly is %d\n",
1598 name, cBOOL(isdigit(i))));
1600 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "isgraph('%s') unexpectedly is %d\n",
1604 name, cBOOL(isgraph(i))));
1606 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "islower('%s') unexpectedly is %d\n",
1610 name, cBOOL(islower(i))));
1612 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "isprint('%s') unexpectedly is %d\n",
1616 name, cBOOL(isprint(i))));
1618 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "ispunct('%s') unexpectedly is %d\n",
1622 name, cBOOL(ispunct(i))));
1624 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1626 DEBUG_L(PerlIO_printf(Perl_debug_log,
1627 "isspace('%s') unexpectedly is %d\n",
1628 name, cBOOL(isspace(i))));
1630 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1632 DEBUG_L(PerlIO_printf(Perl_debug_log,
1633 "isupper('%s') unexpectedly is %d\n",
1634 name, cBOOL(isupper(i))));
1636 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1638 DEBUG_L(PerlIO_printf(Perl_debug_log,
1639 "isxdigit('%s') unexpectedly is %d\n",
1640 name, cBOOL(isxdigit(i))));
1642 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1644 DEBUG_L(PerlIO_printf(Perl_debug_log,
1645 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1646 name, tolower(i), (int) toLOWER_A(i)));
1648 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1650 DEBUG_L(PerlIO_printf(Perl_debug_log,
1651 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1652 name, toupper(i), (int) toUPPER_A(i)));
1654 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1656 DEBUG_L(PerlIO_printf(Perl_debug_log,
1657 "'\\n' (=%02X) is not a control\n", (int) i));
1660 /* Add to the list; Separate multiple entries with a blank */
1663 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1665 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1671 PL_in_utf8_turkic_locale = FALSE;
1675 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1676 * this locale requires more than one byte, there are going to be
1678 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1679 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1680 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1682 if ( check_for_problems && MB_CUR_MAX > 1
1683 && ! PL_in_utf8_CTYPE_locale
1685 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1686 * locale. Just assume that the implementation for them (plus
1687 * for POSIX) is correct and the > 1 value is spurious. (Since
1688 * these are specially handled to never be considered UTF-8
1689 * locales, as long as this is the only problem, everything
1690 * should work fine */
1691 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1693 multi_byte_locale = TRUE;
1698 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1699 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1700 PL_warn_locale = Perl_newSVpvf(aTHX_
1701 "Locale '%s' contains (at least) the following characters"
1702 " which have\nunexpected meanings: %s\nThe Perl program"
1703 " will use the expected meanings",
1704 newctype, bad_chars_list);
1707 PL_warn_locale = Perl_newSVpvf(aTHX_
1708 "Locale '%s' may not work well.%s%s%s\n",
1711 ? " Some characters in it are not recognized by"
1715 ? "\nThe following characters (and maybe others)"
1716 " may not have the same meaning as the Perl"
1717 " program expects:\n"
1725 # ifdef HAS_NL_LANGINFO
1727 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1728 /* parameter FALSE is a don't care here */
1729 my_nl_langinfo(CODESET, FALSE));
1733 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1735 /* If we are actually in the scope of the locale or are debugging,
1736 * output the message now. If not in that scope, we save the
1737 * message to be output at the first operation using this locale,
1738 * if that actually happens. Most programs don't use locales, so
1739 * they are immune to bad ones. */
1740 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1742 /* The '0' below suppresses a bogus gcc compiler warning */
1743 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1745 if (IN_LC(LC_CTYPE)) {
1746 SvREFCNT_dec_NN(PL_warn_locale);
1747 PL_warn_locale = NULL;
1753 #endif /* USE_LOCALE_CTYPE */
1758 Perl__warn_problematic_locale()
1761 #ifdef USE_LOCALE_CTYPE
1765 /* Internal-to-core function that outputs the message in PL_warn_locale,
1766 * and then NULLS it. Should be called only through the macro
1767 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1769 if (PL_warn_locale) {
1770 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1771 SvPVX(PL_warn_locale),
1772 0 /* dummy to avoid compiler warning */ );
1773 SvREFCNT_dec_NN(PL_warn_locale);
1774 PL_warn_locale = NULL;
1782 S_new_collate(pTHX_ const char *newcoll)
1785 #ifndef USE_LOCALE_COLLATE
1787 PERL_UNUSED_ARG(newcoll);
1788 PERL_UNUSED_CONTEXT;
1792 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1793 * core Perl this and that 'newcoll' is the name of the new locale.
1795 * The design of locale collation is that every locale change is given an
1796 * index 'PL_collation_ix'. The first time a string particpates in an
1797 * operation that requires collation while locale collation is active, it
1798 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1799 * magic includes the collation index, and the transformation of the string
1800 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1801 * instead of the string itself. If a string changes, the magic is
1802 * cleared. The next time the locale changes, the index is incremented,
1803 * and so we know during a comparison that the transformation is not
1804 * necessarily still valid, and so is recomputed. Note that if the locale
1805 * changes enough times, the index could wrap (a U32), and it is possible
1806 * that a transformation would improperly be considered valid, leading to
1807 * an unlikely bug */
1810 if (PL_collation_name) {
1812 Safefree(PL_collation_name);
1813 PL_collation_name = NULL;
1815 PL_collation_standard = TRUE;
1816 is_standard_collation:
1817 PL_collxfrm_base = 0;
1818 PL_collxfrm_mult = 2;
1819 PL_in_utf8_COLLATE_locale = FALSE;
1820 PL_strxfrm_NUL_replacement = '\0';
1821 PL_strxfrm_max_cp = 0;
1825 /* If this is not the same locale as currently, set the new one up */
1826 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1828 Safefree(PL_collation_name);
1829 PL_collation_name = stdize_locale(savepv(newcoll));
1830 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1831 if (PL_collation_standard) {
1832 goto is_standard_collation;
1835 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1836 PL_strxfrm_NUL_replacement = '\0';
1837 PL_strxfrm_max_cp = 0;
1839 /* A locale collation definition includes primary, secondary, tertiary,
1840 * etc. weights for each character. To sort, the primary weights are
1841 * used, and only if they compare equal, then the secondary weights are
1842 * used, and only if they compare equal, then the tertiary, etc.
1844 * strxfrm() works by taking the input string, say ABC, and creating an
1845 * output transformed string consisting of first the primary weights,
1846 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1847 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1848 * may not have weights at every level. In our example, let's say B
1849 * doesn't have a tertiary weight, and A doesn't have a secondary
1850 * weight. The constructed string is then going to be
1851 * A¹B¹C¹ B²C² A³C³ ....
1852 * This has the desired effect that strcmp() will look at the secondary
1853 * or tertiary weights only if the strings compare equal at all higher
1854 * priority weights. The spaces shown here, like in
1856 * are not just for readability. In the general case, these must
1857 * actually be bytes, which we will call here 'separator weights'; and
1858 * they must be smaller than any other weight value, but since these
1859 * are C strings, only the terminating one can be a NUL (some
1860 * implementations may include a non-NUL separator weight just before
1861 * the NUL). Implementations tend to reserve 01 for the separator
1862 * weights. They are needed so that a shorter string's secondary
1863 * weights won't be misconstrued as primary weights of a longer string,
1864 * etc. By making them smaller than any other weight, the shorter
1865 * string will sort first. (Actually, if all secondary weights are
1866 * smaller than all primary ones, there is no need for a separator
1867 * weight between those two levels, etc.)
1869 * The length of the transformed string is roughly a linear function of
1870 * the input string. It's not exactly linear because some characters
1871 * don't have weights at all levels. When we call strxfrm() we have to
1872 * allocate some memory to hold the transformed string. The
1873 * calculations below try to find coefficients 'm' and 'b' for this
1874 * locale so that m*x + b equals how much space we need, given the size
1875 * of the input string in 'x'. If we calculate too small, we increase
1876 * the size as needed, and call strxfrm() again, but it is better to
1877 * get it right the first time to avoid wasted expensive string
1878 * transformations. */
1881 /* We use the string below to find how long the tranformation of it
1882 * is. Almost all locales are supersets of ASCII, or at least the
1883 * ASCII letters. We use all of them, half upper half lower,
1884 * because if we used fewer, we might hit just the ones that are
1885 * outliers in a particular locale. Most of the strings being
1886 * collated will contain a preponderance of letters, and even if
1887 * they are above-ASCII, they are likely to have the same number of
1888 * weight levels as the ASCII ones. It turns out that digits tend
1889 * to have fewer levels, and some punctuation has more, but those
1890 * are relatively sparse in text, and khw believes this gives a
1891 * reasonable result, but it could be changed if experience so
1893 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1894 char * x_longer; /* Transformed 'longer' */
1895 Size_t x_len_longer; /* Length of 'x_longer' */
1897 char * x_shorter; /* We also transform a substring of 'longer' */
1898 Size_t x_len_shorter;
1900 /* _mem_collxfrm() is used get the transformation (though here we
1901 * are interested only in its length). It is used because it has
1902 * the intelligence to handle all cases, but to work, it needs some
1903 * values of 'm' and 'b' to get it started. For the purposes of
1904 * this calculation we use a very conservative estimate of 'm' and
1905 * 'b'. This assumes a weight can be multiple bytes, enough to
1906 * hold any UV on the platform, and there are 5 levels, 4 weight
1907 * bytes, and a trailing NUL. */
1908 PL_collxfrm_base = 5;
1909 PL_collxfrm_mult = 5 * sizeof(UV);
1911 /* Find out how long the transformation really is */
1912 x_longer = _mem_collxfrm(longer,
1916 /* We avoid converting to UTF-8 in the
1917 * called function by telling it the
1918 * string is in UTF-8 if the locale is a
1919 * UTF-8 one. Since the string passed
1920 * here is invariant under UTF-8, we can
1921 * claim it's UTF-8 even though it isn't.
1923 PL_in_utf8_COLLATE_locale);
1926 /* Find out how long the transformation of a substring of 'longer'
1927 * is. Together the lengths of these transformations are
1928 * sufficient to calculate 'm' and 'b'. The substring is all of
1929 * 'longer' except the first character. This minimizes the chances
1930 * of being swayed by outliers */
1931 x_shorter = _mem_collxfrm(longer + 1,
1934 PL_in_utf8_COLLATE_locale);
1935 Safefree(x_shorter);
1937 /* If the results are nonsensical for this simple test, the whole
1938 * locale definition is suspect. Mark it so that locale collation
1939 * is not active at all for it. XXX Should we warn? */
1940 if ( x_len_shorter == 0
1941 || x_len_longer == 0
1942 || x_len_shorter >= x_len_longer)
1944 PL_collxfrm_mult = 0;
1945 PL_collxfrm_base = 0;
1948 SSize_t base; /* Temporary */
1950 /* We have both: m * strlen(longer) + b = x_len_longer
1951 * m * strlen(shorter) + b = x_len_shorter;
1952 * subtracting yields:
1953 * m * (strlen(longer) - strlen(shorter))
1954 * = x_len_longer - x_len_shorter
1955 * But we have set things up so that 'shorter' is 1 byte smaller
1956 * than 'longer'. Hence:
1957 * m = x_len_longer - x_len_shorter
1959 * But if something went wrong, make sure the multiplier is at
1962 if (x_len_longer > x_len_shorter) {
1963 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1966 PL_collxfrm_mult = 1;
1971 * but in case something has gone wrong, make sure it is
1973 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1978 /* Add 1 for the trailing NUL */
1979 PL_collxfrm_base = base + 1;
1984 if (DEBUG_L_TEST || debug_initialization) {
1985 PerlIO_printf(Perl_debug_log,
1986 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1988 " collate multipler=%zu, collate base=%zu\n",
1990 PL_in_utf8_COLLATE_locale,
1991 x_len_shorter, x_len_longer,
1992 PL_collxfrm_mult, PL_collxfrm_base);
1999 #endif /* USE_LOCALE_COLLATE */
2008 S_win32_setlocale(pTHX_ int category, const char* locale)
2010 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2011 * difference between the two unless the input locale is "", which normally
2012 * means on Windows to get the machine default, which is set via the
2013 * computer's "Regional and Language Options" (or its current equivalent).
2014 * In POSIX, it instead means to find the locale from the user's
2015 * environment. This routine changes the Windows behavior to first look in
2016 * the environment, and, if anything is found, use that instead of going to
2017 * the machine default. If there is no environment override, the machine
2018 * default is used, by calling the real setlocale() with "".
2020 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2021 * use the particular category's variable if set; otherwise to use the LANG
2024 bool override_LC_ALL = FALSE;
2028 if (locale && strEQ(locale, "")) {
2032 locale = PerlEnv_getenv("LC_ALL");
2034 if (category == LC_ALL) {
2035 override_LC_ALL = TRUE;
2041 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2042 if (category == categories[i]) {
2043 locale = PerlEnv_getenv(category_names[i]);
2048 locale = PerlEnv_getenv("LANG");
2064 result = setlocale(category, locale);
2065 DEBUG_L(STMT_START {
2067 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2068 setlocale_debug_string(category, locale, result));
2072 if (! override_LC_ALL) {
2076 /* Here the input category was LC_ALL, and we have set it to what is in the
2077 * LANG variable or the system default if there is no LANG. But these have
2078 * lower priority than the other LC_foo variables, so override it for each
2079 * one that is set. (If they are set to "", it means to use the same thing
2080 * we just set LC_ALL to, so can skip) */
2082 for (i = 0; i < LC_ALL_INDEX; i++) {
2083 result = PerlEnv_getenv(category_names[i]);
2084 if (result && strNE(result, "")) {
2085 setlocale(categories[i], result);
2086 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2088 setlocale_debug_string(categories[i], result, "not captured")));
2092 result = setlocale(LC_ALL, NULL);
2093 DEBUG_L(STMT_START {
2095 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2097 setlocale_debug_string(LC_ALL, NULL, result));
2108 =head1 Locale-related functions and macros
2110 =for apidoc Perl_setlocale
2112 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2113 taking the same parameters, and returning the same information, except that it
2114 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2115 instead return C<C> if the underlying locale has a non-dot decimal point
2116 character, or a non-empty thousands separator for displaying floating point
2117 numbers. This is because perl keeps that locale category such that it has a
2118 dot and empty separator, changing the locale briefly during the operations
2119 where the underlying one is required. C<Perl_setlocale> knows about this, and
2120 compensates; regular C<setlocale> doesn't.
2122 Another reason it isn't completely a drop-in replacement is that it is
2123 declared to return S<C<const char *>>, whereas the system setlocale omits the
2124 C<const> (presumably because its API was specified long ago, and can't be
2125 updated; it is illegal to change the information C<setlocale> returns; doing
2126 so leads to segfaults.)
2128 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2129 C<setlocale> can be completely ineffective on some platforms under some
2132 C<Perl_setlocale> should not be used to change the locale except on systems
2133 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2134 the system C<setlocale()> is ineffective, returning the wrong information, and
2135 failing to actually change the locale. C<Perl_setlocale>, however works
2136 properly in all circumstances.
2138 The return points to a per-thread static buffer, which is overwritten the next
2139 time C<Perl_setlocale> is called from the same thread.
2146 Perl_setlocale(const int category, const char * locale)
2148 /* This wraps POSIX::setlocale() */
2152 PERL_UNUSED_ARG(category);
2153 PERL_UNUSED_ARG(locale);
2159 const char * retval;
2160 const char * newlocale;
2163 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2165 #ifdef USE_LOCALE_NUMERIC
2167 /* A NULL locale means only query what the current one is. We have the
2168 * LC_NUMERIC name saved, because we are normally switched into the C
2169 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2170 * the correct results. All other categories don't require special
2172 if (locale == NULL) {
2173 if (category == LC_NUMERIC) {
2175 /* We don't have to copy this return value, as it is a per-thread
2176 * variable, and won't change until a future setlocale */
2177 return PL_numeric_name;
2182 else if (category == LC_ALL) {
2183 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2192 retval = save_to_buffer(do_setlocale_r(category, locale),
2193 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2196 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2198 if (locale == NULL && category == LC_ALL) {
2199 RESTORE_LC_NUMERIC();
2204 DEBUG_L(PerlIO_printf(Perl_debug_log,
2205 "%s:%d: %s\n", __FILE__, __LINE__,
2206 setlocale_debug_string(category, locale, retval)));
2214 /* If locale == NULL, we are just querying the state */
2215 if (locale == NULL) {
2219 /* Now that have switched locales, we have to update our records to
2224 #ifdef USE_LOCALE_CTYPE
2231 #ifdef USE_LOCALE_COLLATE
2234 new_collate(retval);
2238 #ifdef USE_LOCALE_NUMERIC
2241 new_numeric(retval);
2249 /* LC_ALL updates all the things we care about. The values may not
2250 * be the same as 'retval', as the locale "" may have set things
2253 # ifdef USE_LOCALE_CTYPE
2255 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2256 new_ctype(newlocale);
2257 Safefree(newlocale);
2259 # endif /* USE_LOCALE_CTYPE */
2260 # ifdef USE_LOCALE_COLLATE
2262 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2263 new_collate(newlocale);
2264 Safefree(newlocale);
2267 # ifdef USE_LOCALE_NUMERIC
2269 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2270 new_numeric(newlocale);
2271 Safefree(newlocale);
2273 # endif /* USE_LOCALE_NUMERIC */
2286 PERL_STATIC_INLINE const char *
2287 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2289 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2290 * growing it if necessary */
2294 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2300 string_size = strlen(string) + offset + 1;
2302 if (*buf_size == 0) {
2303 Newx(*buf, string_size, char);
2304 *buf_size = string_size;
2306 else if (string_size > *buf_size) {
2307 Renew(*buf, string_size, char);
2308 *buf_size = string_size;
2311 Copy(string, *buf + offset, string_size - offset, char);
2317 =for apidoc Perl_langinfo
2319 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2320 taking the same C<item> parameter values, and returning the same information.
2321 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2322 of Perl's locale handling from your code, and can be used on systems that lack
2323 a native C<nl_langinfo>.
2331 The reason it isn't quite a drop-in replacement is actually an advantage. The
2332 only difference is that it returns S<C<const char *>>, whereas plain
2333 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2334 forbidden to write into the buffer. By declaring this C<const>, the compiler
2335 enforces this restriction, so if it is violated, you know at compilation time,
2336 rather than getting segfaults at runtime.
2340 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2341 without you having to write extra code. The reason for the extra code would be
2342 because these are from the C<LC_NUMERIC> locale category, which is normally
2343 kept set by Perl so that the radix is a dot, and the separator is the empty
2344 string, no matter what the underlying locale is supposed to be, and so to get
2345 the expected results, you have to temporarily toggle into the underlying
2346 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2347 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2348 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2349 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2350 (decimal point) character to be a dot.)
2354 The system function it replaces can have its static return buffer trashed,
2355 not only by a subesequent call to that function, but by a C<freelocale>,
2356 C<setlocale>, or other locale change. The returned buffer of this function is
2357 not changed until the next call to it, so the buffer is never in a trashed
2362 Its return buffer is per-thread, so it also is never overwritten by a call to
2363 this function from another thread; unlike the function it replaces.
2367 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2368 as Windows, hence makes your code more portable. Of the fifty-some possible
2369 items specified by the POSIX 2008 standard,
2370 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2371 only one is completely unimplemented, though on non-Windows platforms, another
2372 significant one is also not implemented). It uses various techniques to
2373 recover the other items, including calling C<L<localeconv(3)>>, and
2374 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2375 available. Later C<strftime()> versions have additional capabilities; C<""> is
2376 returned for those not available on your system.
2378 It is important to note that when called with an item that is recovered by
2379 using C<localeconv>, the buffer from any previous explicit call to
2380 C<localeconv> will be overwritten. This means you must save that buffer's
2381 contents if you need to access them after a call to this function. (But note
2382 that you might not want to be using C<localeconv()> directly anyway, because of
2383 issues like the ones listed in the second item of this list (above) for
2384 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2385 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2388 The details for those items which may deviate from what this emulation returns
2389 and what a native C<nl_langinfo()> would return are specified in
2394 When using C<Perl_langinfo> on systems that don't have a native
2395 C<nl_langinfo()>, you must
2397 #include "perl_langinfo.h"
2399 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2400 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2401 C<langinfo.h> would try to import into the namespace for code that doesn't need
2404 The original impetus for C<Perl_langinfo()> was so that code that needs to
2405 find out the current currency symbol, floating point radix character, or digit
2406 grouping separator can use, on all systems, the simpler and more
2407 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2408 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2409 is better to use the methods given in L<perlcall> to call
2410 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2417 #ifdef HAS_NL_LANGINFO
2418 Perl_langinfo(const nl_item item)
2420 Perl_langinfo(const int item)
2423 return my_nl_langinfo(item, TRUE);
2427 #ifdef HAS_NL_LANGINFO
2428 S_my_nl_langinfo(const nl_item item, bool toggle)
2430 S_my_nl_langinfo(const int item, bool toggle)
2434 const char * retval;
2436 #ifdef USE_LOCALE_NUMERIC
2438 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2439 * two items, and only if not already there */
2440 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2441 || PL_numeric_underlying))
2443 #endif /* No toggling needed if not using LC_NUMERIC */
2447 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2448 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2449 || ! defined(HAS_POSIX_2008_LOCALE) \
2450 || ! defined(DUPLOCALE)
2452 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2453 * for those items dependent on it. This must be copied to a buffer before
2454 * switching back, as some systems destroy the buffer when setlocale() is
2458 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2461 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2464 LOCALE_LOCK; /* Prevent interference from another thread executing
2465 this code section (the only call to nl_langinfo in
2469 /* Copy to a per-thread buffer, which is also one that won't be
2470 * destroyed by a subsequent setlocale(), such as the
2471 * RESTORE_LC_NUMERIC may do just below. */
2472 retval = save_to_buffer(nl_langinfo(item),
2473 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2478 RESTORE_LC_NUMERIC();
2482 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2485 bool do_free = FALSE;
2486 locale_t cur = uselocale((locale_t) 0);
2488 if (cur == LC_GLOBAL_LOCALE) {
2489 cur = duplocale(LC_GLOBAL_LOCALE);
2493 # ifdef USE_LOCALE_NUMERIC
2496 if (PL_underlying_numeric_obj) {
2497 cur = PL_underlying_numeric_obj;
2500 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2507 /* We have to save it to a buffer, because the freelocale() just below
2508 * can invalidate the internal one */
2509 retval = save_to_buffer(nl_langinfo_l(item, cur),
2510 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2519 if (strEQ(retval, "")) {
2520 if (item == YESSTR) {
2523 if (item == NOSTR) {
2530 #else /* Below, emulate nl_langinfo as best we can */
2534 # ifdef HAS_LOCALECONV
2536 const struct lconv* lc;
2538 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2540 # ifdef TS_W32_BROKEN_LOCALECONV
2542 const char * save_global;
2543 const char * save_thread;
2551 # ifdef HAS_STRFTIME
2554 bool return_format = FALSE; /* Return the %format, not the value */
2555 const char * format;
2559 /* We copy the results to a per-thread buffer, even if not
2560 * multi-threaded. This is in part to simplify this code, and partly
2561 * because we need a buffer anyway for strftime(), and partly because a
2562 * call of localeconv() could otherwise wipe out the buffer, and the
2563 * programmer would not be expecting this, as this is a nl_langinfo()
2564 * substitute after all, so s/he might be thinking their localeconv()
2565 * is safe until another localeconv() call. */
2570 /* This is unimplemented */
2571 case ERA: /* For use with strftime() %E modifier */
2576 /* We use only an English set, since we don't know any more */
2577 case YESEXPR: return "^[+1yY]";
2578 case YESSTR: return "yes";
2579 case NOEXPR: return "^[-0nN]";
2580 case NOSTR: return "no";
2586 /* On non-windows, this is unimplemented, in part because of
2587 * inconsistencies between vendors. The Darwin native
2588 * nl_langinfo() implementation simply looks at everything past
2589 * any dot in the name, but that doesn't work for other
2590 * vendors. Many Linux locales that don't have UTF-8 in their
2591 * names really are UTF-8, for example; z/OS locales that do
2592 * have UTF-8 in their names, aren't really UTF-8 */
2597 { /* But on Windows, the name does seem to be consistent, so
2602 const char * name = my_setlocale(LC_CTYPE, NULL);
2604 if (isNAME_C_OR_POSIX(name)) {
2605 return "ANSI_X3.4-1968";
2608 /* Find the dot in the locale name */
2609 first = (const char *) strchr(name, '.');
2615 /* Look at everything past the dot */
2620 if (! isDIGIT(*p)) {
2627 /* Here everything past the dot is a digit. Treat it as a
2629 retval = save_to_buffer("CP", &PL_langinfo_buf,
2630 &PL_langinfo_bufsize, 0);
2631 offset = STRLENs("CP");
2635 retval = save_to_buffer(first, &PL_langinfo_buf,
2636 &PL_langinfo_bufsize, offset);
2642 # ifdef HAS_LOCALECONV
2646 /* We don't bother with localeconv_l() because any system that
2647 * has it is likely to also have nl_langinfo() */
2649 LOCALE_LOCK_V; /* Prevent interference with other threads
2650 using localeconv() */
2652 # ifdef TS_W32_BROKEN_LOCALECONV
2654 /* This is a workaround for a Windows bug prior to VS 15.
2655 * What we do here is, while locked, switch to the global
2656 * locale so localeconv() works; then switch back just before
2657 * the unlock. This can screw things up if some thread is
2658 * already using the global locale while assuming no other is.
2659 * A different workaround would be to call GetCurrencyFormat on
2660 * a known value, and parse it; patches welcome
2662 * We have to use LC_ALL instead of LC_MONETARY because of
2663 * another bug in Windows */
2665 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2666 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2667 save_global= savepv(my_setlocale(LC_ALL, NULL));
2668 my_setlocale(LC_ALL, save_thread);
2674 || ! lc->currency_symbol
2675 || strEQ("", lc->currency_symbol))
2681 /* Leave the first spot empty to be filled in below */
2682 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2683 &PL_langinfo_bufsize, 1);
2684 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2685 { /* khw couldn't figure out how the localedef specifications
2686 would show that the $ should replace the radix; this is
2687 just a guess as to how it might work.*/
2688 PL_langinfo_buf[0] = '.';
2690 else if (lc->p_cs_precedes) {
2691 PL_langinfo_buf[0] = '-';
2694 PL_langinfo_buf[0] = '+';
2697 # ifdef TS_W32_BROKEN_LOCALECONV
2699 my_setlocale(LC_ALL, save_global);
2700 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2701 my_setlocale(LC_ALL, save_thread);
2702 Safefree(save_global);
2703 Safefree(save_thread);
2710 # ifdef TS_W32_BROKEN_LOCALECONV
2714 /* For this, we output a known simple floating point number to
2715 * a buffer, and parse it, looking for the radix */
2718 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2721 if (PL_langinfo_bufsize < 10) {
2722 PL_langinfo_bufsize = 10;
2723 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2726 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2728 if (needed_size >= (int) PL_langinfo_bufsize) {
2729 PL_langinfo_bufsize = needed_size + 1;
2730 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2731 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2733 assert(needed_size < (int) PL_langinfo_bufsize);
2736 ptr = PL_langinfo_buf;
2737 e = PL_langinfo_buf + PL_langinfo_bufsize;
2740 while (ptr < e && *ptr != '1') {
2747 while (ptr < e && *ptr != '5') {
2751 /* Everything in between is the radix string */
2753 PL_langinfo_buf[0] = '?';
2754 PL_langinfo_buf[1] = '\0';
2758 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2762 RESTORE_LC_NUMERIC();
2765 retval = PL_langinfo_buf;
2770 case RADIXCHAR: /* No special handling needed */
2777 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2780 LOCALE_LOCK_V; /* Prevent interference with other threads
2781 using localeconv() */
2783 # ifdef TS_W32_BROKEN_LOCALECONV
2785 /* This should only be for the thousands separator. A
2786 * different work around would be to use GetNumberFormat on a
2787 * known value and parse the result to find the separator */
2788 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2789 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2790 save_global = savepv(my_setlocale(LC_ALL, NULL));
2791 my_setlocale(LC_ALL, save_thread);
2793 /* This is the start of code that for broken Windows replaces
2794 * the above and below code, and instead calls
2795 * GetNumberFormat() and then would parse that to find the
2796 * thousands separator. It needs to handle UTF-16 vs -8
2799 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2800 DEBUG_L(PerlIO_printf(Perl_debug_log,
2801 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2802 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2812 temp = (item == RADIXCHAR)
2814 : lc->thousands_sep;
2820 retval = save_to_buffer(temp, &PL_langinfo_buf,
2821 &PL_langinfo_bufsize, 0);
2823 # ifdef TS_W32_BROKEN_LOCALECONV
2825 my_setlocale(LC_ALL, save_global);
2826 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2827 my_setlocale(LC_ALL, save_thread);
2828 Safefree(save_global);
2829 Safefree(save_thread);
2836 RESTORE_LC_NUMERIC();
2842 # ifdef HAS_STRFTIME
2844 /* These are defined by C89, so we assume that strftime supports
2845 * them, and so are returned unconditionally; they may not be what
2846 * the locale actually says, but should give good enough results
2847 * for someone using them as formats (as opposed to trying to parse
2848 * them to figure out what the locale says). The other format
2849 * items are actually tested to verify they work on the platform */
2850 case D_FMT: return "%x";
2851 case T_FMT: return "%X";
2852 case D_T_FMT: return "%c";
2854 /* These formats are only available in later strfmtime's */
2855 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2857 /* The rest can be gotten from most versions of strftime(). */
2858 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2859 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2861 case AM_STR: case PM_STR:
2862 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2863 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2864 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2865 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2866 case DAY_5: case DAY_6: case DAY_7:
2867 case MON_1: case MON_2: case MON_3: case MON_4:
2868 case MON_5: case MON_6: case MON_7: case MON_8:
2869 case MON_9: case MON_10: case MON_11: case MON_12:
2873 init_tm(&tm); /* Precaution against core dumps */
2877 tm.tm_year = 2017 - 1900;
2884 "panic: %s: %d: switch case: %d problem",
2885 __FILE__, __LINE__, item);
2886 NOT_REACHED; /* NOTREACHED */
2888 case PM_STR: tm.tm_hour = 18;
2893 case ABDAY_7: tm.tm_wday++;
2894 case ABDAY_6: tm.tm_wday++;
2895 case ABDAY_5: tm.tm_wday++;
2896 case ABDAY_4: tm.tm_wday++;
2897 case ABDAY_3: tm.tm_wday++;
2898 case ABDAY_2: tm.tm_wday++;
2903 case DAY_7: tm.tm_wday++;
2904 case DAY_6: tm.tm_wday++;
2905 case DAY_5: tm.tm_wday++;
2906 case DAY_4: tm.tm_wday++;
2907 case DAY_3: tm.tm_wday++;
2908 case DAY_2: tm.tm_wday++;
2913 case ABMON_12: tm.tm_mon++;
2914 case ABMON_11: tm.tm_mon++;
2915 case ABMON_10: tm.tm_mon++;
2916 case ABMON_9: tm.tm_mon++;
2917 case ABMON_8: tm.tm_mon++;
2918 case ABMON_7: tm.tm_mon++;
2919 case ABMON_6: tm.tm_mon++;
2920 case ABMON_5: tm.tm_mon++;
2921 case ABMON_4: tm.tm_mon++;
2922 case ABMON_3: tm.tm_mon++;
2923 case ABMON_2: tm.tm_mon++;
2928 case MON_12: tm.tm_mon++;
2929 case MON_11: tm.tm_mon++;
2930 case MON_10: tm.tm_mon++;
2931 case MON_9: tm.tm_mon++;
2932 case MON_8: tm.tm_mon++;
2933 case MON_7: tm.tm_mon++;
2934 case MON_6: tm.tm_mon++;
2935 case MON_5: tm.tm_mon++;
2936 case MON_4: tm.tm_mon++;
2937 case MON_3: tm.tm_mon++;
2938 case MON_2: tm.tm_mon++;
2945 return_format = TRUE;
2950 return_format = TRUE;
2955 return_format = TRUE;
2960 return_format = TRUE;
2965 format = "%Ow"; /* Find the alternate digit for 0 */
2969 /* We can't use my_strftime() because it doesn't look at
2971 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2974 /* A zero return means one of:
2975 * a) there wasn't enough space in PL_langinfo_buf
2976 * b) the format, like a plain %p, returns empty
2977 * c) it was an illegal format, though some
2978 * implementations of strftime will just return the
2979 * illegal format as a plain character sequence.
2981 * To quickly test for case 'b)', try again but precede
2982 * the format with a plain character. If that result is
2983 * still empty, the problem is either 'a)' or 'c)' */
2985 Size_t format_size = strlen(format) + 1;
2986 Size_t mod_size = format_size + 1;
2990 Newx(mod_format, mod_size, char);
2991 Newx(temp_result, PL_langinfo_bufsize, char);
2993 my_strlcpy(mod_format + 1, format, mod_size);
2994 len = strftime(temp_result,
2995 PL_langinfo_bufsize,
2997 Safefree(mod_format);
2998 Safefree(temp_result);
3000 /* If 'len' is non-zero, it means that we had a case like
3001 * %p which means the current locale doesn't use a.m. or
3002 * p.m., and that is valid */
3005 /* Here, still didn't work. If we get well beyond a
3006 * reasonable size, bail out to prevent an infinite
3009 if (PL_langinfo_bufsize > 100 * format_size) {
3010 *PL_langinfo_buf = '\0';
3013 /* Double the buffer size to retry; Add 1 in case
3014 * original was 0, so we aren't stuck at 0. */
3015 PL_langinfo_bufsize *= 2;
3016 PL_langinfo_bufsize++;
3017 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3025 /* Here, we got a result.
3027 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3028 * alternate format for wday 0. If the value is the same as
3029 * the normal 0, there isn't an alternate, so clear the buffer.
3031 if ( item == ALT_DIGITS
3032 && strEQ(PL_langinfo_buf, "0"))
3034 *PL_langinfo_buf = '\0';
3037 /* ALT_DIGITS is problematic. Experiments on it showed that
3038 * strftime() did not always work properly when going from
3039 * alt-9 to alt-10. Only a few locales have this item defined,
3040 * and in all of them on Linux that khw was able to find,
3041 * nl_langinfo() merely returned the alt-0 character, possibly
3042 * doubled. Most Unicode digits are in blocks of 10
3043 * consecutive code points, so that is sufficient information
3044 * for those scripts, as we can infer alt-1, alt-2, .... But
3045 * for a Japanese locale, a CJK ideographic 0 is returned, and
3046 * the CJK digits are not in code point order, so you can't
3047 * really infer anything. The localedef for this locale did
3048 * specify the succeeding digits, so that strftime() works
3049 * properly on them, without needing to infer anything. But
3050 * the nl_langinfo() return did not give sufficient information
3051 * for the caller to understand what's going on. So until
3052 * there is evidence that it should work differently, this
3053 * returns the alt-0 string for ALT_DIGITS.
3055 * wday was chosen because its range is all a single digit.
3056 * Things like tm_sec have two digits as the minimum: '00' */
3060 retval = PL_langinfo_buf;
3062 /* If to return the format, not the value, overwrite the buffer
3063 * with it. But some strftime()s will keep the original format
3064 * if illegal, so change those to "" */
3065 if (return_format) {
3066 if (strEQ(PL_langinfo_buf, format)) {
3067 *PL_langinfo_buf = '\0';
3070 retval = save_to_buffer(format, &PL_langinfo_buf,
3071 &PL_langinfo_bufsize, 0);
3089 * Initialize locale awareness.
3092 Perl_init_i18nl10n(pTHX_ int printwarn)
3096 * 0 if not to output warning when setup locale is bad
3097 * 1 if to output warning based on value of PERL_BADLANG
3098 * >1 if to output regardless of PERL_BADLANG
3101 * 1 = set ok or not applicable,
3102 * 0 = fallback to a locale of lower priority
3103 * -1 = fallback to all locales failed, not even to the C locale
3105 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3106 * set, debugging information is output.
3108 * This looks more complicated than it is, mainly due to the #ifdefs.
3110 * We try to set LC_ALL to the value determined by the environment. If
3111 * there is no LC_ALL on this platform, we try the individual categories we
3112 * know about. If this works, we are done.
3114 * But if it doesn't work, we have to do something else. We search the
3115 * environment variables ourselves instead of relying on the system to do
3116 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3117 * think there is one), and the ultimate fallback "C". This is all done in
3118 * the same loop as above to avoid duplicating code, but it makes things
3119 * more complex. The 'trial_locales' array is initialized with just one
3120 * element; it causes the behavior described in the paragraph above this to
3121 * happen. If that fails, we add elements to 'trial_locales', and do extra
3122 * loop iterations to cause the behavior described in this paragraph.
3124 * On Ultrix, the locale MUST come from the environment, so there is
3125 * preliminary code to set it. I (khw) am not sure that it is necessary,
3126 * and that this couldn't be folded into the loop, but barring any real
3127 * platforms to test on, it's staying as-is
3129 * A slight complication is that in embedded Perls, the locale may already
3130 * be set-up, and we don't want to get it from the normal environment
3131 * variables. This is handled by having a special environment variable
3132 * indicate we're in this situation. We simply set setlocale's 2nd
3133 * parameter to be a NULL instead of "". That indicates to setlocale that
3134 * it is not to change anything, but to return the current value,
3135 * effectively initializing perl's db to what the locale already is.
3137 * We play the same trick with NULL if a LC_ALL succeeds. We call
3138 * setlocale() on the individual categores with NULL to get their existing
3139 * values for our db, instead of trying to change them.
3146 PERL_UNUSED_ARG(printwarn);
3148 #else /* USE_LOCALE */
3151 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3155 /* NULL uses the existing already set up locale */
3156 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3159 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3160 unsigned int trial_locales_count;
3161 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3162 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3163 bool setlocale_failure = FALSE;
3166 /* A later getenv() could zap this, so only use here */
3167 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3169 const bool locwarn = (printwarn > 1
3171 && ( ! bad_lang_use_once
3173 /* disallow with "" or "0" */
3175 && strNE("0", bad_lang_use_once)))));
3177 /* setlocale() return vals; not copied so must be looked at immediately */
3178 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3180 /* current locale for given category; should have been copied so aren't
3182 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3186 /* In some systems you can find out the system default locale
3187 * and use that as the fallback locale. */
3188 # define SYSTEM_DEFAULT_LOCALE
3190 # ifdef SYSTEM_DEFAULT_LOCALE
3192 const char *system_default_locale = NULL;
3197 # define DEBUG_LOCALE_INIT(a,b,c)
3200 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3202 # define DEBUG_LOCALE_INIT(category, locale, result) \
3204 if (debug_initialization) { \
3205 PerlIO_printf(Perl_debug_log, \
3207 __FILE__, __LINE__, \
3208 setlocale_debug_string(category, \
3214 /* Make sure the parallel arrays are properly set up */
3215 # ifdef USE_LOCALE_NUMERIC
3216 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3217 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3218 # ifdef USE_POSIX_2008_LOCALE
3219 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3222 # ifdef USE_LOCALE_CTYPE
3223 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3224 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3225 # ifdef USE_POSIX_2008_LOCALE
3226 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3229 # ifdef USE_LOCALE_COLLATE
3230 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3231 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3232 # ifdef USE_POSIX_2008_LOCALE
3233 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3236 # ifdef USE_LOCALE_TIME
3237 assert(categories[LC_TIME_INDEX] == LC_TIME);
3238 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3239 # ifdef USE_POSIX_2008_LOCALE
3240 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3243 # ifdef USE_LOCALE_MESSAGES
3244 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3245 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3246 # ifdef USE_POSIX_2008_LOCALE
3247 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3250 # ifdef USE_LOCALE_MONETARY
3251 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3252 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3253 # ifdef USE_POSIX_2008_LOCALE
3254 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3257 # ifdef USE_LOCALE_ADDRESS
3258 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3259 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3260 # ifdef USE_POSIX_2008_LOCALE
3261 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3264 # ifdef USE_LOCALE_IDENTIFICATION
3265 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3266 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3267 # ifdef USE_POSIX_2008_LOCALE
3268 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3271 # ifdef USE_LOCALE_MEASUREMENT
3272 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3273 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3274 # ifdef USE_POSIX_2008_LOCALE
3275 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3278 # ifdef USE_LOCALE_PAPER
3279 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3280 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3281 # ifdef USE_POSIX_2008_LOCALE
3282 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3285 # ifdef USE_LOCALE_TELEPHONE
3286 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3287 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3288 # ifdef USE_POSIX_2008_LOCALE
3289 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3293 assert(categories[LC_ALL_INDEX] == LC_ALL);
3294 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3295 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3296 # ifdef USE_POSIX_2008_LOCALE
3297 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3300 # endif /* DEBUGGING */
3302 /* Initialize the cache of the program's UTF-8ness for the always known
3303 * locales C and POSIX */
3304 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3305 sizeof(PL_locale_utf8ness));
3307 # ifdef USE_THREAD_SAFE_LOCALE
3310 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3314 # ifdef USE_POSIX_2008_LOCALE
3316 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3317 if (! PL_C_locale_obj) {
3318 Perl_croak_nocontext(
3319 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3321 if (DEBUG_Lv_TEST || debug_initialization) {
3322 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3327 # ifdef USE_LOCALE_NUMERIC
3329 PL_numeric_radix_sv = newSVpvs(".");
3333 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3335 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3336 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3339 # ifdef LOCALE_ENVIRON_REQUIRED
3342 * Ultrix setlocale(..., "") fails if there are no environment
3343 * variables from which to get a locale name.
3347 # error Ultrix without LC_ALL not implemented
3353 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3354 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3355 if (sl_result[LC_ALL_INDEX])
3358 setlocale_failure = TRUE;
3360 if (! setlocale_failure) {
3361 const char * locale_param;
3362 for (i = 0; i < LC_ALL_INDEX; i++) {
3363 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3366 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3367 if (! sl_result[i]) {
3368 setlocale_failure = TRUE;
3370 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3375 # endif /* LC_ALL */
3376 # endif /* LOCALE_ENVIRON_REQUIRED */
3378 /* We try each locale in the list until we get one that works, or exhaust
3379 * the list. Normally the loop is executed just once. But if setting the
3380 * locale fails, inside the loop we add fallback trials to the array and so
3381 * will execute the loop multiple times */
3382 trial_locales[0] = setlocale_init;
3383 trial_locales_count = 1;
3385 for (i= 0; i < trial_locales_count; i++) {
3386 const char * trial_locale = trial_locales[i];
3390 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3391 * when i==0, but I (khw) don't think that behavior makes much
3393 setlocale_failure = FALSE;
3395 # ifdef SYSTEM_DEFAULT_LOCALE
3396 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3398 /* On Windows machines, an entry of "" after the 0th means to use
3399 * the system default locale, which we now proceed to get. */
3400 if (strEQ(trial_locale, "")) {
3403 /* Note that this may change the locale, but we are going to do
3404 * that anyway just below */
3405 system_default_locale = do_setlocale_c(LC_ALL, "");
3406 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3408 /* Skip if invalid or if it's already on the list of locales to
3410 if (! system_default_locale) {
3411 goto next_iteration;
3413 for (j = 0; j < trial_locales_count; j++) {
3414 if (strEQ(system_default_locale, trial_locales[j])) {
3415 goto next_iteration;
3419 trial_locale = system_default_locale;
3422 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3424 # endif /* SYSTEM_DEFAULT_LOCALE */
3430 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3431 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3432 if (! sl_result[LC_ALL_INDEX]) {
3433 setlocale_failure = TRUE;
3436 /* Since LC_ALL succeeded, it should have changed all the other
3437 * categories it can to its value; so we massage things so that the
3438 * setlocales below just return their category's current values.
3439 * This adequately handles the case in NetBSD where LC_COLLATE may
3440 * not be defined for a locale, and setting it individually will
3441 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3442 * the POSIX locale. */
3443 trial_locale = NULL;
3446 # endif /* LC_ALL */
3448 if (! setlocale_failure) {
3450 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3452 = savepv(do_setlocale_r(categories[j], trial_locale));
3453 if (! curlocales[j]) {
3454 setlocale_failure = TRUE;
3456 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3459 if (! setlocale_failure) { /* All succeeded */
3460 break; /* Exit trial_locales loop */
3464 /* Here, something failed; will need to try a fallback. */
3470 if (locwarn) { /* Output failure info only on the first one */
3474 PerlIO_printf(Perl_error_log,
3475 "perl: warning: Setting locale failed.\n");
3477 # else /* !LC_ALL */
3479 PerlIO_printf(Perl_error_log,
3480 "perl: warning: Setting locale failed for the categories:\n\t");
3482 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3483 if (! curlocales[j]) {
3484 PerlIO_printf(Perl_error_log, category_names[j]);
3487 Safefree(curlocales[j]);
3491 # endif /* LC_ALL */
3493 PerlIO_printf(Perl_error_log,
3494 "perl: warning: Please check that your locale settings:\n");
3498 PerlIO_printf(Perl_error_log,
3499 "\tLANGUAGE = %c%s%c,\n",
3500 language ? '"' : '(',
3501 language ? language : "unset",
3502 language ? '"' : ')');
3505 PerlIO_printf(Perl_error_log,
3506 "\tLC_ALL = %c%s%c,\n",
3508 lc_all ? lc_all : "unset",
3509 lc_all ? '"' : ')');
3511 # if defined(USE_ENVIRON_ARRAY)
3516 /* Look through the environment for any variables of the
3517 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3518 * already handled above. These are assumed to be locale
3519 * settings. Output them and their values. */
3520 for (e = environ; *e; e++) {
3521 const STRLEN prefix_len = sizeof("LC_") - 1;
3524 if ( strBEGINs(*e, "LC_")
3525 && ! strBEGINs(*e, "LC_ALL=")
3526 && (uppers_len = strspn(*e + prefix_len,
3527 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3528 && ((*e)[prefix_len + uppers_len] == '='))
3530 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3531 (int) (prefix_len + uppers_len), *e,
3532 *e + prefix_len + uppers_len + 1);
3539 PerlIO_printf(Perl_error_log,
3540 "\t(possibly more locale environment variables)\n");
3544 PerlIO_printf(Perl_error_log,
3545 "\tLANG = %c%s%c\n",
3547 lang ? lang : "unset",
3550 PerlIO_printf(Perl_error_log,
3551 " are supported and installed on your system.\n");
3554 /* Calculate what fallback locales to try. We have avoided this
3555 * until we have to, because failure is quite unlikely. This will
3556 * usually change the upper bound of the loop we are in.
3558 * Since the system's default way of setting the locale has not
3559 * found one that works, We use Perl's defined ordering: LC_ALL,
3560 * LANG, and the C locale. We don't try the same locale twice, so
3561 * don't add to the list if already there. (On POSIX systems, the
3562 * LC_ALL element will likely be a repeat of the 0th element "",
3563 * but there's no harm done by doing it explicitly.
3565 * Note that this tries the LC_ALL environment variable even on
3566 * systems which have no LC_ALL locale setting. This may or may
3567 * not have been originally intentional, but there's no real need
3568 * to change the behavior. */
3570 for (j = 0; j < trial_locales_count; j++) {
3571 if (strEQ(lc_all, trial_locales[j])) {
3575 trial_locales[trial_locales_count++] = lc_all;
3580 for (j = 0; j < trial_locales_count; j++) {
3581 if (strEQ(lang, trial_locales[j])) {
3585 trial_locales[trial_locales_count++] = lang;
3589 # if defined(WIN32) && defined(LC_ALL)
3591 /* For Windows, we also try the system default locale before "C".
3592 * (If there exists a Windows without LC_ALL we skip this because
3593 * it gets too complicated. For those, the "C" is the next
3594 * fallback possibility). The "" is the same as the 0th element of
3595 * the array, but the code at the loop above knows to treat it
3596 * differently when not the 0th */
3597 trial_locales[trial_locales_count++] = "";
3601 for (j = 0; j < trial_locales_count; j++) {
3602 if (strEQ("C", trial_locales[j])) {
3606 trial_locales[trial_locales_count++] = "C";
3609 } /* end of first time through the loop */
3617 } /* end of looping through the trial locales */
3619 if (ok < 1) { /* If we tried to fallback */
3621 if (! setlocale_failure) { /* fallback succeeded */
3622 msg = "Falling back to";
3624 else { /* fallback failed */
3627 /* We dropped off the end of the loop, so have to decrement i to
3628 * get back to the value the last time through */
3632 msg = "Failed to fall back to";
3634 /* To continue, we should use whatever values we've got */
3636 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3637 Safefree(curlocales[j]);
3638 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3639 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3644 const char * description;
3645 const char * name = "";
3646 if (strEQ(trial_locales[i], "C")) {
3647 description = "the standard locale";
3651 # ifdef SYSTEM_DEFAULT_LOCALE
3653 else if (strEQ(trial_locales[i], "")) {
3654 description = "the system default locale";
3655 if (system_default_locale) {
3656 name = system_default_locale;
3660 # endif /* SYSTEM_DEFAULT_LOCALE */
3663 description = "a fallback locale";
3664 name = trial_locales[i];
3666 if (name && strNE(name, "")) {
3667 PerlIO_printf(Perl_error_log,
3668 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3671 PerlIO_printf(Perl_error_log,
3672 "perl: warning: %s %s.\n", msg, description);
3675 } /* End of tried to fallback */
3677 /* Done with finding the locales; update our records */
3679 # ifdef USE_LOCALE_CTYPE
3681 new_ctype(curlocales[LC_CTYPE_INDEX]);
3684 # ifdef USE_LOCALE_COLLATE
3686 new_collate(curlocales[LC_COLLATE_INDEX]);
3689 # ifdef USE_LOCALE_NUMERIC
3691 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3695 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3697 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3699 /* This caches whether each category's locale is UTF-8 or not. This
3700 * may involve changing the locale. It is ok to do this at
3701 * initialization time before any threads have started, but not later
3702 * unless thread-safe operations are used.
3703 * Caching means that if the program heeds our dictate not to change
3704 * locales in threaded applications, this data will remain valid, and
3705 * it may get queried without having to change locales. If the
3706 * environment is such that all categories have the same locale, this
3707 * isn't needed, as the code will not change the locale; but this
3708 * handles the uncommon case where the environment has disparate
3709 * locales for the categories */
3710 (void) _is_cur_LC_category_utf8(categories[i]);
3714 Safefree(curlocales[i]);
3717 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3719 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3720 * locale is UTF-8. The call to new_ctype() just above has already
3721 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3722 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3723 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3724 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3725 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3727 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3728 This is an alternative to using the -C command line switch
3729 (the -C if present will override this). */
3731 const char *p = PerlEnv_getenv("PERL_UNICODE");
3732 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3733 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3747 #endif /* USE_LOCALE */
3750 /* So won't continue to output stuff */
3751 DEBUG_INITIALIZATION_set(FALSE);
3758 #ifdef USE_LOCALE_COLLATE
3761 Perl__mem_collxfrm(pTHX_ const char *input_string,
3762 STRLEN len, /* Length of 'input_string' */
3763 STRLEN *xlen, /* Set to length of returned string
3764 (not including the collation index
3766 bool utf8 /* Is the input in UTF-8? */
3770 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3771 * differences. First, it handles embedded NULs. Second, it allocates a bit
3772 * more memory than needed for the transformed data itself. The real
3773 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3774 * the length of that, and doesn't include the collation index size.
3775 * Please see sv_collxfrm() to see how this is used. */
3777 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3779 char * s = (char *) input_string;
3780 STRLEN s_strlen = strlen(input_string);
3782 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3783 STRLEN length_in_chars;
3784 bool first_time = TRUE; /* Cleared after first loop iteration */
3786 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3788 /* Must be NUL-terminated */
3789 assert(*(input_string + len) == '\0');
3791 /* If this locale has defective collation, skip */
3792 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3793 DEBUG_L(PerlIO_printf(Perl_debug_log,
3794 "_mem_collxfrm: locale's collation is defective\n"));
3798 /* Replace any embedded NULs with the control that sorts before any others.
3799 * This will give as good as possible results on strings that don't
3800 * otherwise contain that character, but otherwise there may be
3801 * less-than-perfect results with that character and NUL. This is
3802 * unavoidable unless we replace strxfrm with our own implementation. */
3803 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3807 STRLEN sans_nuls_len;
3808 int try_non_controls;
3809 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3810 making sure 2nd byte is NUL.
3812 STRLEN this_replacement_len;
3814 /* If we don't know what non-NUL control character sorts lowest for
3815 * this locale, find it */
3816 if (PL_strxfrm_NUL_replacement == '\0') {
3818 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3819 includes the collation index
3822 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3824 /* Unlikely, but it may be that no control will work to replace
3825 * NUL, in which case we instead look for any character. Controls
3826 * are preferred because collation order is, in general, context
3827 * sensitive, with adjoining characters affecting the order, and
3828 * controls are less likely to have such interactions, allowing the
3829 * NUL-replacement to stand on its own. (Another way to look at it
3830 * is to imagine what would happen if the NUL were replaced by a
3831 * combining character; it wouldn't work out all that well.) */
3832 for (try_non_controls = 0;
3833 try_non_controls < 2;
3836 /* Look through all legal code points (NUL isn't) */
3837 for (j = 1; j < 256; j++) {
3838 char * x; /* j's xfrm plus collation index */
3839 STRLEN x_len; /* length of 'x' */
3840 STRLEN trial_len = 1;
3841 char cur_source[] = { '\0', '\0' };
3843 /* Skip non-controls the first time through the loop. The
3844 * controls in a UTF-8 locale are the L1 ones */
3845 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3852 /* Create a 1-char string of the current code point */
3853 cur_source[0] = (char) j;
3855 /* Then transform it */
3856 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3857 0 /* The string is not in UTF-8 */);
3859 /* Ignore any character that didn't successfully transform.
3865 /* If this character's transformation is lower than
3866 * the current lowest, this one becomes the lowest */
3867 if ( cur_min_x == NULL
3868 || strLT(x + COLLXFRM_HDR_LEN,
3869 cur_min_x + COLLXFRM_HDR_LEN))
3871 PL_strxfrm_NUL_replacement = j;
3877 } /* end of loop through all 255 characters */
3879 /* Stop looking if found */
3884 /* Unlikely, but possible, if there aren't any controls that
3885 * work in the locale, repeat the loop, looking for any
3886 * character that works */
3887 DEBUG_L(PerlIO_printf(Perl_debug_log,
3888 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3889 } /* End of loop to try first the controls, then any char */
3892 DEBUG_L(PerlIO_printf(Perl_debug_log,
3893 "_mem_collxfrm: Couldn't find any character to replace"
3894 " embedded NULs in locale %s with", PL_collation_name));
3898 DEBUG_L(PerlIO_printf(Perl_debug_log,
3899 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3900 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3902 Safefree(cur_min_x);
3903 } /* End of determining the character that is to replace NULs */
3905 /* If the replacement is variant under UTF-8, it must match the
3906 * UTF8-ness of the original */
3907 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3908 this_replacement_char[0] =
3909 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3910 this_replacement_char[1] =
3911 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3912 this_replacement_len = 2;
3915 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3916 /* this_replacement_char[1] = '\0' was done at initialization */
3917 this_replacement_len = 1;
3920 /* The worst case length for the replaced string would be if every
3921 * character in it is NUL. Multiply that by the length of each
3922 * replacement, and allow for a trailing NUL */
3923 sans_nuls_len = (len * this_replacement_len) + 1;
3924 Newx(sans_nuls, sans_nuls_len, char);
3927 /* Replace each NUL with the lowest collating control. Loop until have
3928 * exhausted all the NULs */
3929 while (s + s_strlen < e) {
3930 my_strlcat(sans_nuls, s, sans_nuls_len);
3932 /* Do the actual replacement */
3933 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3935 /* Move past the input NUL */
3937 s_strlen = strlen(s);
3940 /* And add anything that trails the final NUL */
3941 my_strlcat(sans_nuls, s, sans_nuls_len);
3943 /* Switch so below we transform this modified string */
3946 } /* End of replacing NULs */
3948 /* Make sure the UTF8ness of the string and locale match */
3949 if (utf8 != PL_in_utf8_COLLATE_locale) {
3950 /* XXX convert above Unicode to 10FFFF? */
3951 const char * const t = s; /* Temporary so we can later find where the
3954 /* Here they don't match. Change the string's to be what the locale is
3957 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3958 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3961 else { /* locale is not UTF-8; but input is; downgrade the input */
3963 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3965 /* If the downgrade was successful we are done, but if the input
3966 * contains things that require UTF-8 to represent, have to do
3967 * damage control ... */
3968 if (UNLIKELY(utf8)) {
3970 /* What we do is construct a non-UTF-8 string with
3971 * 1) the characters representable by a single byte converted
3972 * to be so (if necessary);
3973 * 2) and the rest converted to collate the same as the
3974 * highest collating representable character. That makes
3975 * them collate at the end. This is similar to how we
3976 * handle embedded NULs, but we use the highest collating
3977 * code point instead of the smallest. Like the NUL case,
3978 * this isn't perfect, but is the best we can reasonably
3979 * do. Every above-255 code point will sort the same as
3980 * the highest-sorting 0-255 code point. If that code
3981 * point can combine in a sequence with some other code
3982 * points for weight calculations, us changing something to
3983 * be it can adversely affect the results. But in most
3984 * cases, it should work reasonably. And note that this is
3985 * really an illegal situation: using code points above 255
3986 * on a locale where only 0-255 are valid. If two strings
3987 * sort entirely equal, then the sort order for the
3988 * above-255 code points will be in code point order. */
3992 /* If we haven't calculated the code point with the maximum
3993 * collating order for this locale, do so now */
3994 if (! PL_strxfrm_max_cp) {
3997 /* The current transformed string that collates the
3998 * highest (except it also includes the prefixed collation
4000 char * cur_max_x = NULL;
4002 /* Look through all legal code points (NUL isn't) */
4003 for (j = 1; j < 256; j++) {
4006 char cur_source[] = { '\0', '\0' };
4008 /* Create a 1-char string of the current code point */
4009 cur_source[0] = (char) j;
4011 /* Then transform it */
4012 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4014 /* If something went wrong (which it shouldn't), just
4015 * ignore this code point */
4020 /* If this character's transformation is higher than
4021 * the current highest, this one becomes the highest */
4022 if ( cur_max_x == NULL
4023 || strGT(x + COLLXFRM_HDR_LEN,
4024 cur_max_x + COLLXFRM_HDR_LEN))
4026 PL_strxfrm_max_cp = j;
4035 DEBUG_L(PerlIO_printf(Perl_debug_log,
4036 "_mem_collxfrm: Couldn't find any character to"
4037 " replace above-Latin1 chars in locale %s with",
4038 PL_collation_name));
4042 DEBUG_L(PerlIO_printf(Perl_debug_log,
4043 "_mem_collxfrm: highest 1-byte collating character"
4044 " in locale %s is 0x%02X\n",
4046 PL_strxfrm_max_cp));
4048 Safefree(cur_max_x);
4051 /* Here we know which legal code point collates the highest.
4052 * We are ready to construct the non-UTF-8 string. The length
4053 * will be at least 1 byte smaller than the input string
4054 * (because we changed at least one 2-byte character into a
4055 * single byte), but that is eaten up by the trailing NUL */
4061 char * e = (char *) t + len;
4063 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4065 if (UTF8_IS_INVARIANT(cur_char)) {
4068 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4069 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4071 else { /* Replace illegal cp with highest collating
4073 s[d++] = PL_strxfrm_max_cp;
4077 Renew(s, d, char); /* Free up unused space */
4082 /* Here, we have constructed a modified version of the input. It could
4083 * be that we already had a modified copy before we did this version.
4084 * If so, that copy is no longer needed */
4085 if (t != input_string) {
4090 length_in_chars = (utf8)
4091 ? utf8_length((U8 *) s, (U8 *) s + len)
4094 /* The first element in the output is the collation id, used by
4095 * sv_collxfrm(); then comes the space for the transformed string. The
4096 * equation should give us a good estimate as to how much is needed */
4097 xAlloc = COLLXFRM_HDR_LEN
4099 + (PL_collxfrm_mult * length_in_chars);
4100 Newx(xbuf, xAlloc, char);
4101 if (UNLIKELY(! xbuf)) {
4102 DEBUG_L(PerlIO_printf(Perl_debug_log,
4103 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4107 /* Store the collation id */
4108 *(U32*)xbuf = PL_collation_ix;
4110 /* Then the transformation of the input. We loop until successful, or we
4114 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4116 /* If the transformed string occupies less space than we told strxfrm()
4117 * was available, it means it successfully transformed the whole
4119 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4121 /* Some systems include a trailing NUL in the returned length.
4122 * Ignore it, using a loop in case multiple trailing NULs are
4125 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4130 /* If the first try didn't get it, it means our prediction was low.
4131 * Modify the coefficients so that we predict a larger value in any
4132 * future transformations */
4134 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4135 STRLEN computed_guess = PL_collxfrm_base
4136 + (PL_collxfrm_mult * length_in_chars);
4138 /* On zero-length input, just keep current slope instead of
4140 const STRLEN new_m = (length_in_chars != 0)
4141 ? needed / length_in_chars
4144 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4145 "%s: %d: initial size of %zu bytes for a length "
4146 "%zu string was insufficient, %zu needed\n",
4148 computed_guess, length_in_chars, needed));
4150 /* If slope increased, use it, but discard this result for
4151 * length 1 strings, as we can't be sure that it's a real slope
4153 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4157 STRLEN old_m = PL_collxfrm_mult;
4158 STRLEN old_b = PL_collxfrm_base;
4162 PL_collxfrm_mult = new_m;
4163 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4164 computed_guess = PL_collxfrm_base
4165 + (PL_collxfrm_mult * length_in_chars);
4166 if (computed_guess < needed) {
4167 PL_collxfrm_base += needed - computed_guess;
4170 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4171 "%s: %d: slope is now %zu; was %zu, base "
4172 "is now %zu; was %zu\n",
4174 PL_collxfrm_mult, old_m,
4175 PL_collxfrm_base, old_b));
4177 else { /* Slope didn't change, but 'b' did */
4178 const STRLEN new_b = needed
4181 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4182 "%s: %d: base is now %zu; was %zu\n",
4184 new_b, PL_collxfrm_base));
4185 PL_collxfrm_base = new_b;
4192 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4193 DEBUG_L(PerlIO_printf(Perl_debug_log,
4194 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4195 *xlen, PERL_INT_MAX));
4199 /* A well-behaved strxfrm() returns exactly how much space it needs
4200 * (usually not including the trailing NUL) when it fails due to not
4201 * enough space being provided. Assume that this is the case unless
4202 * it's been proven otherwise */
4203 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4204 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4206 else { /* Here, either:
4207 * 1) The strxfrm() has previously shown bad behavior; or
4208 * 2) It isn't the first time through the loop, which means
4209 * that the strxfrm() is now showing bad behavior, because
4210 * we gave it what it said was needed in the previous
4211 * iteration, and it came back saying it needed still more.
4212 * (Many versions of cygwin fit this. When the buffer size
4213 * isn't sufficient, they return the input size instead of
4214 * how much is needed.)
4215 * Increase the buffer size by a fixed percentage and try again.
4217 xAlloc += (xAlloc / 4) + 1;
4218 PL_strxfrm_is_behaved = FALSE;
4222 if (DEBUG_Lv_TEST || debug_initialization) {
4223 PerlIO_printf(Perl_debug_log,
4224 "_mem_collxfrm required more space than previously calculated"
4225 " for locale %s, trying again with new guess=%d+%zu\n",
4226 PL_collation_name, (int) COLLXFRM_HDR_LEN,
4227 xAlloc - COLLXFRM_HDR_LEN);
4234 Renew(xbuf, xAlloc, char);
4235 if (UNLIKELY(! xbuf)) {
4236 DEBUG_L(PerlIO_printf(Perl_debug_log,
4237 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4247 if (DEBUG_Lv_TEST || debug_initialization) {
4249 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4250 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4251 PerlIO_printf(Perl_debug_log, "%s\n",
4252 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4258 /* Free up unneeded space; retain ehough for trailing NUL */
4259 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4261 if (s != input_string) {
4269 if (s != input_string) {
4276 if (DEBUG_Lv_TEST || debug_initialization) {
4277 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4288 S_print_collxfrm_input_and_return(pTHX_
4289 const char * const s,
4290 const char * const e,
4291 const STRLEN * const xlen,
4295 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4297 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
https://perl5.git.perl.org / perl5.git / blob