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"
60 /* If the environment says to, we can output debugging information during
61 * initialization. This is done before option parsing, and before any thread
62 * creation, so can be a file-level static */
63 #if ! defined(DEBUGGING)
64 # define debug_initialization 0
65 # define DEBUG_INITIALIZATION_set(v)
67 static bool debug_initialization = FALSE;
68 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
72 /* Returns the Unix errno portion; ignoring any others. This is a macro here
73 * instead of putting it into perl.h, because unclear to khw what should be
75 #define GET_ERRNO saved_errno
77 /* strlen() of a literal string constant. We might want this more general,
78 * but using it in just this file for now. A problem with more generality is
79 * the compiler warnings about comparing unlike signs */
80 #define STRLENs(s) (sizeof("" s "") - 1)
82 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
83 * return of setlocale(), then this is extremely likely to be the C or POSIX
84 * locale. However, the output of setlocale() is documented to be opaque, but
85 * the odds are extremely small that it would return these two strings for some
86 * other locale. Note that VMS in these two locales includes many non-ASCII
87 * characters as controls and punctuation (below are hex bytes):
89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
90 * Oddly, none there are listed as alphas, though some represent alphabetics
91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
92 #define isNAME_C_OR_POSIX(name) \
94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
95 || strEQ((name), "POSIX")))
99 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
100 * looked up. This is in the form of a C string: */
102 #define UTF8NESS_SEP "\v"
103 #define UTF8NESS_PREFIX "\f"
105 /* So, the string looks like:
107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
109 * where the digit 0 after the \a indicates that the locale starting just
110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
112 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
113 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
115 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
118 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
119 * kept there always. The remining portion of the cache is LRU, with the
120 * oldest looked-up locale at the tail end */
123 S_stdize_locale(pTHX_ char *locs)
125 /* Standardize the locale name from a string returned by 'setlocale',
126 * possibly modifying that string.
128 * The typical return value of setlocale() is either
129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
131 * (the space-separated values represent the various sublocales,
132 * in some unspecified order). This is not handled by this function.
134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
135 * which is harmful for further use of the string in setlocale(). This
136 * function removes the trailing new line and everything up through the '='
139 const char * const s = strchr(locs, '=');
142 PERL_ARGS_ASSERT_STDIZE_LOCALE;
145 const char * const t = strchr(s, '.');
148 const char * const u = strchr(t, '\n');
149 if (u && (u[1] == 0)) {
150 const STRLEN len = u - s;
151 Move(s + 1, locs, len, char);
159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
164 /* Two parallel arrays; first the locale categories Perl uses on this system;
165 * the second array is their names. These arrays are in mostly arbitrary
168 const int categories[] = {
170 # ifdef USE_LOCALE_NUMERIC
173 # ifdef USE_LOCALE_CTYPE
176 # ifdef USE_LOCALE_COLLATE
179 # ifdef USE_LOCALE_TIME
182 # ifdef USE_LOCALE_MESSAGES
185 # ifdef USE_LOCALE_MONETARY
188 # ifdef USE_LOCALE_ADDRESS
191 # ifdef USE_LOCALE_IDENTIFICATION
194 # ifdef USE_LOCALE_MEASUREMENT
197 # ifdef USE_LOCALE_PAPER
200 # ifdef USE_LOCALE_TELEPHONE
203 # ifdef USE_LOCALE_SYNTAX
206 # ifdef USE_LOCALE_TOD
212 -1 /* Placeholder because C doesn't allow a
213 trailing comma, and it would get complicated
214 with all the #ifdef's */
217 /* The top-most real element is LC_ALL */
219 const char * const category_names[] = {
221 # ifdef USE_LOCALE_NUMERIC
224 # ifdef USE_LOCALE_CTYPE
227 # ifdef USE_LOCALE_COLLATE
230 # ifdef USE_LOCALE_TIME
233 # ifdef USE_LOCALE_MESSAGES
236 # ifdef USE_LOCALE_MONETARY
239 # ifdef USE_LOCALE_ADDRESS
242 # ifdef USE_LOCALE_IDENTIFICATION
245 # ifdef USE_LOCALE_MEASUREMENT
248 # ifdef USE_LOCALE_PAPER
251 # ifdef USE_LOCALE_TELEPHONE
254 # ifdef USE_LOCALE_SYNTAX
257 # ifdef USE_LOCALE_TOD
263 NULL /* Placeholder */
268 /* On systems with LC_ALL, it is kept in the highest index position. (-2
269 * to account for the final unused placeholder element.) */
270 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
274 /* On systems without LC_ALL, we pretend it is there, one beyond the real
275 * top element, hence in the unused placeholder element. */
276 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
280 /* Pretending there is an LC_ALL element just above allows us to avoid most
281 * special cases. Most loops through these arrays in the code below are
282 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
283 * on either type of system. But the code must be written to not access the
284 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
285 * checked for at compile time by using the #define LC_ALL_INDEX which is only
286 * defined if we do have LC_ALL. */
289 S_category_name(const int category)
295 if (category == LC_ALL) {
301 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
302 if (category == categories[i]) {
303 return category_names[i];
308 const char suffix[] = " (unknown)";
310 Size_t length = sizeof(suffix) + 1;
319 /* Calculate the number of digits */
325 Newx(unknown, length, char);
326 my_snprintf(unknown, length, "%d%s", category, suffix);
332 /* Now create LC_foo_INDEX #defines for just those categories on this system */
333 # ifdef USE_LOCALE_NUMERIC
334 # define LC_NUMERIC_INDEX 0
335 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
337 # define _DUMMY_NUMERIC -1
339 # ifdef USE_LOCALE_CTYPE
340 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
341 # define _DUMMY_CTYPE LC_CTYPE_INDEX
343 # define _DUMMY_CTYPE _DUMMY_NUMERIC
345 # ifdef USE_LOCALE_COLLATE
346 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
347 # define _DUMMY_COLLATE LC_COLLATE_INDEX
349 # define _DUMMY_COLLATE _DUMMY_CTYPE
351 # ifdef USE_LOCALE_TIME
352 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
353 # define _DUMMY_TIME LC_TIME_INDEX
355 # define _DUMMY_TIME _DUMMY_COLLATE
357 # ifdef USE_LOCALE_MESSAGES
358 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
359 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
361 # define _DUMMY_MESSAGES _DUMMY_TIME
363 # ifdef USE_LOCALE_MONETARY
364 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
365 # define _DUMMY_MONETARY LC_MONETARY_INDEX
367 # define _DUMMY_MONETARY _DUMMY_MESSAGES
369 # ifdef USE_LOCALE_ADDRESS
370 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
371 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
373 # define _DUMMY_ADDRESS _DUMMY_MONETARY
375 # ifdef USE_LOCALE_IDENTIFICATION
376 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
377 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
379 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
381 # ifdef USE_LOCALE_MEASUREMENT
382 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
383 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
385 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
387 # ifdef USE_LOCALE_PAPER
388 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
389 # define _DUMMY_PAPER LC_PAPER_INDEX
391 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
393 # ifdef USE_LOCALE_TELEPHONE
394 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
395 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
397 # define _DUMMY_TELEPHONE _DUMMY_PAPER
399 # ifdef USE_LOCALE_SYNTAX
400 # define LC_SYNTAX_INDEX _DUMMY_TELEPHONE + 1
401 # define _DUMMY_SYNTAX LC_SYNTAX_INDEX
403 # define _DUMMY_SYNTAX _DUMMY_TELEPHONE
405 # ifdef USE_LOCALE_TOD
406 # define LC_TOD_INDEX _DUMMY_SYNTAX + 1
407 # define _DUMMY_TOD LC_TOD_INDEX
409 # define _DUMMY_TOD _DUMMY_SYNTAX
412 # define LC_ALL_INDEX _DUMMY_TOD + 1
414 #endif /* ifdef USE_LOCALE */
416 /* Windows requres a customized base-level setlocale() */
418 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
420 # define my_setlocale(cat, locale) setlocale(cat, locale)
423 #ifndef USE_POSIX_2008_LOCALE
425 /* "do_setlocale_c" is intended to be called when the category is a constant
426 * known at compile time; "do_setlocale_r", not known until run time */
427 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
428 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
429 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
431 #else /* Below uses POSIX 2008 */
433 /* We emulate setlocale with our own function. LC_foo is not valid for the
434 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
435 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
436 * proper offset into the array 'category_masks[]'. At runtime, we have to
437 * search through the array (as the actual numbers may not be small contiguous
438 * positive integers which would lend themselves to array lookup). */
439 # define do_setlocale_c(cat, locale) \
440 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
441 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
443 # if ! defined(__GLIBC__) || ! defined(USE_LOCALE_MESSAGES)
445 # define FIX_GLIBC_LC_MESSAGES_BUG(i)
447 # else /* Invalidate glibc cache of loaded translations, see [perl #134264] */
449 # include <libintl.h>
450 # define FIX_GLIBC_LC_MESSAGES_BUG(i) \
452 if ((i) == LC_MESSAGES_INDEX) { \
453 textdomain(textdomain(NULL)); \
459 /* A third array, parallel to the ones above to map from category to its
461 const int category_masks[] = {
462 # ifdef USE_LOCALE_NUMERIC
465 # ifdef USE_LOCALE_CTYPE
468 # ifdef USE_LOCALE_COLLATE
471 # ifdef USE_LOCALE_TIME
474 # ifdef USE_LOCALE_MESSAGES
477 # ifdef USE_LOCALE_MONETARY
480 # ifdef USE_LOCALE_ADDRESS
483 # ifdef USE_LOCALE_IDENTIFICATION
484 LC_IDENTIFICATION_MASK,
486 # ifdef USE_LOCALE_MEASUREMENT
489 # ifdef USE_LOCALE_PAPER
492 # ifdef USE_LOCALE_TELEPHONE
495 # ifdef USE_LOCALE_SYNTAX
498 # ifdef USE_LOCALE_TOD
501 /* LC_ALL can't be turned off by a Configure
502 * option, and in Posix 2008, should always be
503 * here, so compile it in unconditionally.
504 * This could catch some glitches at compile
510 S_emulate_setlocale(const int category,
513 const bool is_index_valid
516 /* This function effectively performs a setlocale() on just the current
517 * thread; thus it is thread-safe. It does this by using the POSIX 2008
518 * locale functions to emulate the behavior of setlocale(). Similar to
519 * regular setlocale(), the return from this function points to memory that
520 * can be overwritten by other system calls, so needs to be copied
521 * immediately if you need to retain it. The difference here is that
522 * system calls besides another setlocale() can overwrite it.
524 * By doing this, most locale-sensitive functions become thread-safe. The
525 * exceptions are mostly those that return a pointer to static memory.
527 * This function takes the same parameters, 'category' and 'locale', that
528 * the regular setlocale() function does, but it also takes two additional
529 * ones. This is because the 2008 functions don't use a category; instead
530 * they use a corresponding mask. Because this function operates in both
531 * worlds, it may need one or the other or both. This function can
532 * calculate the mask from the input category, but to avoid this
533 * calculation, if the caller knows at compile time what the mask is, it
534 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
535 * parameter is ignored.
537 * POSIX 2008, for some sick reason, chose not to provide a method to find
538 * the category name of a locale. Some vendors have created a
539 * querylocale() function to do just that. This function is a lot simpler
540 * to implement on systems that have this. Otherwise, we have to keep
541 * track of what the locale has been set to, so that we can return its
542 * name to emulate setlocale(). It's also possible for C code in some
543 * library to change the locale without us knowing it, though as of
544 * September 2017, there are no occurrences in CPAN of uselocale(). Some
545 * libraries do use setlocale(), but that changes the global locale, and
546 * threads using per-thread locales will just ignore those changes.
547 * Another problem is that without querylocale(), we have to guess at what
548 * was meant by setting a locale of "". We handle this by not actually
549 * ever setting to "" (unless querylocale exists), but to emulate what we
550 * think should happen for "".
560 if (DEBUG_Lv_TEST || debug_initialization) {
561 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);
566 /* If the input mask might be incorrect, calculate the correct one */
567 if (! is_index_valid) {
572 if (DEBUG_Lv_TEST || debug_initialization) {
573 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
578 for (i = 0; i <= LC_ALL_INDEX; i++) {
579 if (category == categories[i]) {
585 /* Here, we don't know about this category, so can't handle it.
586 * Fallback to the early POSIX usages */
587 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
588 "Unknown locale category %d; can't set it to %s\n",
596 if (DEBUG_Lv_TEST || debug_initialization) {
597 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
604 mask = category_masks[index];
608 if (DEBUG_Lv_TEST || debug_initialization) {
609 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
614 /* If just querying what the existing locale is ... */
615 if (locale == NULL) {
616 locale_t cur_obj = uselocale((locale_t) 0);
620 if (DEBUG_Lv_TEST || debug_initialization) {
621 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
626 if (cur_obj == LC_GLOBAL_LOCALE) {
627 return my_setlocale(category, NULL);
630 # ifdef HAS_QUERYLOCALE
632 return (char *) querylocale(mask, cur_obj);
636 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
637 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
639 # if defined(_NL_LOCALE_NAME) \
640 && defined(DEBUGGING) \
641 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
642 /* On systems that accept any locale name, the real underlying locale
643 * is often returned by this internal function, so we can't use it */
645 /* Internal glibc for querylocale(), but doesn't handle
646 * empty-string ("") locale properly; who knows what other
647 * glitches. Check for it now, under debug. */
649 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
650 uselocale((locale_t) 0));
652 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
653 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
654 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
656 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
657 if ( strNE(PL_curlocales[index], temp_name)
658 && ! ( isNAME_C_OR_POSIX(temp_name)
659 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
661 # ifdef USE_C_BACKTRACE
663 dump_c_backtrace(Perl_debug_log, 20, 1);
667 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
668 " (%s) and what internal glibc thinks"
669 " (%s)\n", category_names[index],
670 PL_curlocales[index], temp_name);
679 /* Without querylocale(), we have to use our record-keeping we've
682 if (category != LC_ALL) {
686 if (DEBUG_Lv_TEST || debug_initialization) {
687 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
692 return PL_curlocales[index];
694 else { /* For LC_ALL */
696 Size_t names_len = 0;
698 bool are_all_categories_the_same_locale = TRUE;
700 /* If we have a valid LC_ALL value, just return it */
701 if (PL_curlocales[LC_ALL_INDEX]) {
705 if (DEBUG_Lv_TEST || debug_initialization) {
706 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
711 return PL_curlocales[LC_ALL_INDEX];
714 /* Otherwise, we need to construct a string of name=value pairs.
715 * We use the glibc syntax, like
716 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
717 * First calculate the needed size. Along the way, check if all
718 * the locale names are the same */
719 for (i = 0; i < LC_ALL_INDEX; i++) {
723 if (DEBUG_Lv_TEST || debug_initialization) {
724 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]);
729 names_len += strlen(category_names[i])
731 + strlen(PL_curlocales[i])
734 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
735 are_all_categories_the_same_locale = FALSE;
739 /* If they are the same, we don't actually have to construct the
740 * string; we just make the entry in LC_ALL_INDEX valid, and be
741 * that single name */
742 if (are_all_categories_the_same_locale) {
743 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
744 return PL_curlocales[LC_ALL_INDEX];
747 names_len++; /* Trailing '\0' */
748 SAVEFREEPV(Newx(all_string, names_len, char));
751 /* Then fill in the string */
752 for (i = 0; i < LC_ALL_INDEX; i++) {
756 if (DEBUG_Lv_TEST || debug_initialization) {
757 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]);
762 my_strlcat(all_string, category_names[i], names_len);
763 my_strlcat(all_string, "=", names_len);
764 my_strlcat(all_string, PL_curlocales[i], names_len);
765 my_strlcat(all_string, ";", names_len);
770 if (DEBUG_L_TEST || debug_initialization) {
771 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
781 SETERRNO(EINVAL, LIB_INVARG);
789 } /* End of this being setlocale(LC_foo, NULL) */
791 /* Here, we are switching locales. */
793 # ifndef HAS_QUERYLOCALE
795 if (strEQ(locale, "")) {
797 /* For non-querylocale() systems, we do the setting of "" ourselves to
798 * be sure that we really know what's going on. We follow the Linux
799 * documented behavior (but if that differs from the actual behavior,
800 * this won't work exactly as the OS implements). We go out and
801 * examine the environment based on our understanding of how the system
802 * works, and use that to figure things out */
804 const char * const lc_all = PerlEnv_getenv("LC_ALL");
806 /* Use any "LC_ALL" environment variable, as it overrides everything
808 if (lc_all && strNE(lc_all, "")) {
813 /* Otherwise, we need to dig deeper. Unless overridden, the
814 * default is the LANG environment variable; if it doesn't exist,
817 const char * default_name;
819 default_name = PerlEnv_getenv("LANG");
821 if (! default_name || strEQ(default_name, "")) {
825 if (category != LC_ALL) {
826 const char * const name = PerlEnv_getenv(category_names[index]);
828 /* Here we are setting a single category. Assume will have the
830 locale = default_name;
832 /* But then look for an overriding environment variable */
833 if (name && strNE(name, "")) {
838 bool did_override = FALSE;
841 /* Here, we are getting LC_ALL. Any categories that don't have
842 * a corresponding environment variable set should be set to
843 * LANG, or to "C" if there is no LANG. If no individual
844 * categories differ from this, we can just set LC_ALL. This
845 * is buggy on systems that have extra categories that we don't
846 * know about. If there is an environment variable that sets
847 * that category, we won't know to look for it, and so our use
848 * of LANG or "C" improperly overrides it. On the other hand,
849 * if we don't do what is done here, and there is no
850 * environment variable, the category's locale should be set to
851 * LANG or "C". So there is no good solution. khw thinks the
852 * best is to look at systems to see what categories they have,
853 * and include them, and then to assume that we know the
856 for (i = 0; i < LC_ALL_INDEX; i++) {
857 const char * const env_override
858 = PerlEnv_getenv(category_names[i]);
859 const char * this_locale = ( env_override
860 && strNE(env_override, ""))
863 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
868 if (strNE(this_locale, default_name)) {
873 /* If all the categories are the same, we can set LC_ALL to
875 if (! did_override) {
876 locale = default_name;
880 /* Here, LC_ALL is no longer valid, as some individual
881 * categories don't match it. We call ourselves
882 * recursively, as that will execute the code that
883 * generates the proper locale string for this situation.
884 * We don't do the remainder of this function, as that is
885 * to update our records, and we've just done that for the
886 * individual categories in the loop above, and doing so
887 * would cause LC_ALL to be done as well */
888 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
892 } /* End of this being setlocale(LC_foo, "") */
893 else if (strchr(locale, ';')) {
895 /* LC_ALL may actually incude a conglomeration of various categories.
896 * Without querylocale, this code uses the glibc (as of this writing)
897 * syntax for representing that, but that is not a stable API, and
898 * other platforms do it differently, so we have to handle all cases
902 const char * s = locale;
903 const char * e = locale + strlen(locale);
905 const char * category_end;
906 const char * name_start;
907 const char * name_end;
909 /* If the string that gives what to set doesn't include all categories,
910 * the omitted ones get set to "C". To get this behavior, first set
911 * all the individual categories to "C", and override the furnished
913 for (i = 0; i < LC_ALL_INDEX; i++) {
914 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
921 /* Parse through the category */
922 while (isWORDCHAR(*p)) {
929 "panic: %s: %d: Unexpected character in locale name '%02X",
930 __FILE__, __LINE__, *(p-1));
933 /* Parse through the locale name */
935 while (p < e && *p != ';') {
938 "panic: %s: %d: Unexpected character in locale name '%02X",
939 __FILE__, __LINE__, *(p-1));
945 /* Space past the semi-colon */
950 /* Find the index of the category name in our lists */
951 for (i = 0; i < LC_ALL_INDEX; i++) {
952 char * individ_locale;
954 /* Keep going if this isn't the index. The strnNE() avoids a
955 * Perl_form(), but would fail if ever a category name could be
956 * a substring of another one, like if there were a
958 if strnNE(s, category_names[i], category_end - s) {
962 /* If this index is for the single category we're changing, we
963 * have found the locale to set it to. */
964 if (category == categories[i]) {
965 locale = Perl_form(aTHX_ "%.*s",
966 (int) (name_end - name_start),
971 assert(category == LC_ALL);
972 individ_locale = Perl_form(aTHX_ "%.*s",
973 (int) (name_end - name_start), name_start);
974 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
983 /* Here we have set all the individual categories by recursive calls.
984 * These collectively should have fixed up LC_ALL, so can just query
985 * what that now is */
986 assert(category == LC_ALL);
988 return do_setlocale_c(LC_ALL, NULL);
989 } /* End of this being setlocale(LC_ALL,
990 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
994 /* Here at the end of having to deal with the absence of querylocale().
995 * Some cases have already been fully handled by recursive calls to this
996 * function. But at this point, we haven't dealt with those, but are now
997 * prepared to, knowing what the locale name to set this category to is.
998 * This would have come for free if this system had had querylocale() */
1000 # endif /* end of ! querylocale */
1002 assert(PL_C_locale_obj);
1004 /* Switching locales generally entails freeing the current one's space (at
1005 * the C library's discretion). We need to stop using that locale before
1006 * the switch. So switch to a known locale object that we don't otherwise
1007 * mess with. This returns the locale object in effect at the time of the
1009 old_obj = uselocale(PL_C_locale_obj);
1013 if (DEBUG_Lv_TEST || debug_initialization) {
1014 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
1023 if (DEBUG_L_TEST || debug_initialization) {
1025 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1036 if (DEBUG_Lv_TEST || debug_initialization) {
1037 PerlIO_printf(Perl_debug_log,
1038 "%s:%d: emulate_setlocale now using %p\n",
1039 __FILE__, __LINE__, PL_C_locale_obj);
1044 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1045 * in fact, we already have switched to it (in preparation for what
1046 * normally is to come). But since we're already there, continue to use
1047 * it instead of trying to create a new locale */
1048 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1052 if (DEBUG_Lv_TEST || debug_initialization) {
1053 PerlIO_printf(Perl_debug_log,
1054 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1059 new_obj = PL_C_locale_obj;
1061 /* We already had switched to the C locale in preparation for freeing
1063 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1064 freelocale(old_obj);
1068 /* If we weren't in a thread safe locale, set so that newlocale() below
1069 * which uses 'old_obj', uses an empty one. Same for our reserved C
1070 * object. The latter is defensive coding, so that, even if there is
1071 * some bug, we will never end up trying to modify either of these, as
1072 * if passed to newlocale(), they can be. */
1073 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1074 old_obj = (locale_t) 0;
1077 /* Ready to create a new locale by modification of the exising one */
1078 new_obj = newlocale(mask, locale, old_obj);
1085 if (DEBUG_L_TEST || debug_initialization) {
1086 PerlIO_printf(Perl_debug_log,
1087 "%s:%d: emulate_setlocale creating new object"
1088 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1093 if (! uselocale(old_obj)) {
1097 if (DEBUG_L_TEST || debug_initialization) {
1098 PerlIO_printf(Perl_debug_log,
1099 "%s:%d: switching back failed: %d\n",
1100 __FILE__, __LINE__, GET_ERRNO);
1112 if (DEBUG_Lv_TEST || debug_initialization) {
1113 PerlIO_printf(Perl_debug_log,
1114 "%s:%d: emulate_setlocale created %p",
1115 __FILE__, __LINE__, new_obj);
1117 PerlIO_printf(Perl_debug_log,
1118 "; should have freed %p", old_obj);
1120 PerlIO_printf(Perl_debug_log, "\n");
1125 /* And switch into it */
1126 if (! uselocale(new_obj)) {
1131 if (DEBUG_L_TEST || debug_initialization) {
1132 PerlIO_printf(Perl_debug_log,
1133 "%s:%d: emulate_setlocale switching to new object"
1134 " failed\n", __FILE__, __LINE__);
1139 if (! uselocale(old_obj)) {
1143 if (DEBUG_L_TEST || debug_initialization) {
1144 PerlIO_printf(Perl_debug_log,
1145 "%s:%d: switching back failed: %d\n",
1146 __FILE__, __LINE__, GET_ERRNO);
1152 freelocale(new_obj);
1160 if (DEBUG_Lv_TEST || debug_initialization) {
1161 PerlIO_printf(Perl_debug_log,
1162 "%s:%d: emulate_setlocale now using %p\n",
1163 __FILE__, __LINE__, new_obj);
1168 /* We are done, except for updating our records (if the system doesn't keep
1169 * them) and in the case of locale "", we don't actually know what the
1170 * locale that got switched to is, as it came from the environment. So
1171 * have to find it */
1173 # ifdef HAS_QUERYLOCALE
1175 if (strEQ(locale, "")) {
1176 locale = querylocale(mask, new_obj);
1181 /* Here, 'locale' is the return value */
1183 /* Without querylocale(), we have to update our records */
1185 if (category == LC_ALL) {
1188 /* For LC_ALL, we change all individual categories to correspond */
1189 /* PL_curlocales is a parallel array, so has same
1190 * length as 'categories' */
1191 for (i = 0; i <= LC_ALL_INDEX; i++) {
1192 Safefree(PL_curlocales[i]);
1193 PL_curlocales[i] = savepv(locale);
1196 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
1200 /* For a single category, if it's not the same as the one in LC_ALL, we
1203 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1204 Safefree(PL_curlocales[LC_ALL_INDEX]);
1205 PL_curlocales[LC_ALL_INDEX] = NULL;
1208 /* Then update the category's record */
1209 Safefree(PL_curlocales[index]);
1210 PL_curlocales[index] = savepv(locale);
1212 FIX_GLIBC_LC_MESSAGES_BUG(index);
1220 #endif /* USE_POSIX_2008_LOCALE */
1222 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1223 didn't natively support them */
1225 /* The way this would work is that we would keep a per-thread list of the
1226 * correct locale for that thread. Any operation that was locale-sensitive
1227 * would have to be changed so that it would look like this:
1230 * setlocale to the correct locale for this operation
1234 * This leaves the global locale in the most recently used operation's, but it
1235 * was locked long enough to get the result. If that result is static, it
1236 * needs to be copied before the unlock.
1238 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1239 * the setup, but are no-ops when not needed, and similarly,
1240 * END_LOCALE_DEPENDENT_OP for the tear-down
1242 * But every call to a locale-sensitive function would have to be changed, and
1243 * if a module didn't cooperate by using the mutex, things would break.
1245 * This code was abandoned before being completed or tested, and is left as-is
1248 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1249 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1252 S_locking_setlocale(pTHX_
1254 const char * locale,
1256 const bool is_index_valid
1259 /* This function kind of performs a setlocale() on just the current thread;
1260 * thus it is kind of thread-safe. It does this by keeping a thread-level
1261 * array of the current locales for each category. Every time a locale is
1262 * switched to, it does the switch globally, but updates the thread's
1263 * array. A query as to what the current locale is just returns the
1264 * appropriate element from the array, and doesn't actually call the system
1265 * setlocale(). The saving into the array is done in an uninterruptible
1266 * section of code, so is unaffected by whatever any other threads might be
1269 * All locale-sensitive operations must work by first starting a critical
1270 * section, then switching to the thread's locale as kept by this function,
1271 * and then doing the operation, then ending the critical section. Thus,
1272 * each gets done in the appropriate locale. simulating thread-safety.
1274 * This function takes the same parameters, 'category' and 'locale', that
1275 * the regular setlocale() function does, but it also takes two additional
1276 * ones. This is because as described earlier. If we know on input the
1277 * index corresponding to the category into the array where we store the
1278 * current locales, we don't have to calculate it. If the caller knows at
1279 * compile time what the index is, it can pass it, setting
1280 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1284 /* If the input index might be incorrect, calculate the correct one */
1285 if (! is_index_valid) {
1288 if (DEBUG_Lv_TEST || debug_initialization) {
1289 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1292 for (i = 0; i <= LC_ALL_INDEX; i++) {
1293 if (category == categories[i]) {
1299 /* Here, we don't know about this category, so can't handle it.
1300 * XXX best we can do is to unsafely set this
1303 return my_setlocale(category, locale);
1307 if (DEBUG_Lv_TEST || debug_initialization) {
1308 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1312 /* For a query, just return what's in our records */
1313 if (new_locale == NULL) {
1314 return curlocales[index];
1318 /* Otherwise, we need to do the switch, and save the result, all in a
1319 * critical section */
1321 Safefree(curlocales[[index]]);
1323 /* It might be that this is called from an already-locked section of code.
1324 * We would have to detect and skip the LOCK/UNLOCK if so */
1327 curlocales[index] = savepv(my_setlocale(category, new_locale));
1329 if (strEQ(new_locale, "")) {
1333 /* The locale values come from the environment, and may not all be the
1334 * same, so for LC_ALL, we have to update all the others, while the
1335 * mutex is still locked */
1337 if (category == LC_ALL) {
1339 for (i = 0; i < LC_ALL_INDEX) {
1340 curlocales[i] = my_setlocale(categories[i], NULL);
1349 return curlocales[index];
1356 S_set_numeric_radix(pTHX_ const bool use_locale)
1358 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1359 * TRUE, use the radix character derived from the current locale */
1361 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1362 || defined(HAS_NL_LANGINFO))
1364 const char * radix = (use_locale)
1365 ? my_nl_langinfo(RADIXCHAR, FALSE)
1366 /* FALSE => already in dest locale */
1369 sv_setpv(PL_numeric_radix_sv, radix);
1371 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1372 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1373 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1374 SvCUR(PL_numeric_radix_sv))
1375 && _is_cur_LC_category_utf8(LC_NUMERIC))
1377 SvUTF8_on(PL_numeric_radix_sv);
1382 if (DEBUG_L_TEST || debug_initialization) {
1383 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1384 SvPVX(PL_numeric_radix_sv),
1385 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1391 PERL_UNUSED_ARG(use_locale);
1393 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1398 S_new_numeric(pTHX_ const char *newnum)
1401 #ifndef USE_LOCALE_NUMERIC
1403 PERL_UNUSED_ARG(newnum);
1407 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1408 * core Perl this and that 'newnum' is the name of the new locale.
1409 * It installs this locale as the current underlying default.
1411 * The default locale and the C locale can be toggled between by use of the
1412 * set_numeric_underlying() and set_numeric_standard() functions, which
1413 * should probably not be called directly, but only via macros like
1414 * SET_NUMERIC_STANDARD() in perl.h.
1416 * The toggling is necessary mainly so that a non-dot radix decimal point
1417 * character can be output, while allowing internal calculations to use a
1420 * This sets several interpreter-level variables:
1421 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1422 * PL_numeric_underlying A boolean indicating if the toggled state is such
1423 * that the current locale is the program's underlying
1425 * PL_numeric_standard An int indicating if the toggled state is such
1426 * that the current locale is the C locale or
1427 * indistinguishable from the C locale. If non-zero, it
1428 * is in C; if > 1, it means it may not be toggled away
1430 * PL_numeric_underlying_is_standard A bool kept by this function
1431 * indicating that the underlying locale and the standard
1432 * C locale are indistinguishable for the purposes of
1433 * LC_NUMERIC. This happens when both of the above two
1434 * variables are true at the same time. (Toggling is a
1435 * no-op under these circumstances.) This variable is
1436 * used to avoid having to recalculate.
1442 Safefree(PL_numeric_name);
1443 PL_numeric_name = NULL;
1444 PL_numeric_standard = TRUE;
1445 PL_numeric_underlying = TRUE;
1446 PL_numeric_underlying_is_standard = TRUE;
1450 save_newnum = stdize_locale(savepv(newnum));
1451 PL_numeric_underlying = TRUE;
1452 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1454 #ifndef TS_W32_BROKEN_LOCALECONV
1456 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1457 * them. But on broken Windows systems calling my_nl_langinfo() for
1458 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1459 * and just always change the locale if not C nor POSIX on those systems */
1460 if (! PL_numeric_standard) {
1461 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1462 FALSE /* Don't toggle locale */ ))
1463 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1468 /* Save the new name if it isn't the same as the previous one, if any */
1469 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1470 Safefree(PL_numeric_name);
1471 PL_numeric_name = save_newnum;
1474 Safefree(save_newnum);
1477 PL_numeric_underlying_is_standard = PL_numeric_standard;
1479 # ifdef HAS_POSIX_2008_LOCALE
1481 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1483 PL_underlying_numeric_obj);
1487 if (DEBUG_L_TEST || debug_initialization) {
1488 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1491 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1492 * have to worry about the radix being a non-dot. (Core operations that
1493 * need the underlying locale change to it temporarily). */
1494 if (PL_numeric_standard) {
1495 set_numeric_radix(0);
1498 set_numeric_standard();
1501 #endif /* USE_LOCALE_NUMERIC */
1506 Perl_set_numeric_standard(pTHX)
1509 #ifdef USE_LOCALE_NUMERIC
1511 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1512 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1513 * macro avoids calling this routine if toggling isn't necessary according
1514 * to our records (which could be wrong if some XS code has changed the
1515 * locale behind our back) */
1519 if (DEBUG_L_TEST || debug_initialization) {
1520 PerlIO_printf(Perl_debug_log,
1521 "Setting LC_NUMERIC locale to standard C\n");
1526 do_setlocale_c(LC_NUMERIC, "C");
1527 PL_numeric_standard = TRUE;
1528 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1529 set_numeric_radix(0);
1531 #endif /* USE_LOCALE_NUMERIC */
1536 Perl_set_numeric_underlying(pTHX)
1539 #ifdef USE_LOCALE_NUMERIC
1541 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1542 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1543 * instead of calling this directly. The macro avoids calling this routine
1544 * if toggling isn't necessary according to our records (which could be
1545 * wrong if some XS code has changed the locale behind our back) */
1549 if (DEBUG_L_TEST || debug_initialization) {
1550 PerlIO_printf(Perl_debug_log,
1551 "Setting LC_NUMERIC locale to %s\n",
1557 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1558 PL_numeric_standard = PL_numeric_underlying_is_standard;
1559 PL_numeric_underlying = TRUE;
1560 set_numeric_radix(! PL_numeric_standard);
1562 #endif /* USE_LOCALE_NUMERIC */
1567 * Set up for a new ctype locale.
1570 S_new_ctype(pTHX_ const char *newctype)
1573 #ifndef USE_LOCALE_CTYPE
1575 PERL_UNUSED_ARG(newctype);
1576 PERL_UNUSED_CONTEXT;
1580 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1581 * core Perl this and that 'newctype' is the name of the new locale.
1583 * This function sets up the folding arrays for all 256 bytes, assuming
1584 * that tofold() is tolc() since fold case is not a concept in POSIX,
1586 * Any code changing the locale (outside this file) should use
1587 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1588 * this function should be called directly only from this file and from
1589 * POSIX::setlocale() */
1593 /* Don't check for problems if we are suppressing the warnings */
1594 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1595 bool maybe_utf8_turkic = FALSE;
1597 PERL_ARGS_ASSERT_NEW_CTYPE;
1599 /* We will replace any bad locale warning with 1) nothing if the new one is
1600 * ok; or 2) a new warning for the bad new locale */
1601 if (PL_warn_locale) {
1602 SvREFCNT_dec_NN(PL_warn_locale);
1603 PL_warn_locale = NULL;
1606 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1608 /* A UTF-8 locale gets standard rules. But note that code still has to
1609 * handle this specially because of the three problematic code points */
1610 if (PL_in_utf8_CTYPE_locale) {
1611 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1613 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1614 * Turkic. Make sure these two are the only anomalies. (We don't use
1615 * towupper and towlower because they aren't in C89.) */
1617 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1619 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1623 if (toupper('i') == 'i' && tolower('I') == 'I') {
1626 check_for_problems = TRUE;
1627 maybe_utf8_turkic = TRUE;
1631 /* We don't populate the other lists if a UTF-8 locale, but do check that
1632 * everything works as expected, unless checking turned off */
1633 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1634 /* Assume enough space for every character being bad. 4 spaces each
1635 * for the 94 printable characters that are output like "'x' "; and 5
1636 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1638 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1639 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1641 unsigned int bad_count = 0; /* Count of bad characters */
1643 for (i = 0; i < 256; i++) {
1644 if (! PL_in_utf8_CTYPE_locale) {
1646 PL_fold_locale[i] = (U8) tolower(i);
1647 else if (islower(i))
1648 PL_fold_locale[i] = (U8) toupper(i);
1650 PL_fold_locale[i] = (U8) i;
1653 /* If checking for locale problems, see if the native ASCII-range
1654 * printables plus \n and \t are in their expected categories in
1655 * the new locale. If not, this could mean big trouble, upending
1656 * Perl's and most programs' assumptions, like having a
1657 * metacharacter with special meaning become a \w. Fortunately,
1658 * it's very rare to find locales that aren't supersets of ASCII
1659 * nowadays. It isn't a problem for most controls to be changed
1660 * into something else; we check only \n and \t, though perhaps \r
1661 * could be an issue as well. */
1662 if ( check_for_problems
1663 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1665 bool is_bad = FALSE;
1666 char name[4] = { '\0' };
1668 /* Convert the name into a string */
1673 else if (i == '\n') {
1674 my_strlcpy(name, "\\n", sizeof(name));
1676 else if (i == '\t') {
1677 my_strlcpy(name, "\\t", sizeof(name));
1681 my_strlcpy(name, "' '", sizeof(name));
1684 /* Check each possibe class */
1685 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1687 DEBUG_L(PerlIO_printf(Perl_debug_log,
1688 "isalnum('%s') unexpectedly is %d\n",
1689 name, cBOOL(isalnum(i))));
1691 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1693 DEBUG_L(PerlIO_printf(Perl_debug_log,
1694 "isalpha('%s') unexpectedly is %d\n",
1695 name, cBOOL(isalpha(i))));
1697 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1699 DEBUG_L(PerlIO_printf(Perl_debug_log,
1700 "isdigit('%s') unexpectedly is %d\n",
1701 name, cBOOL(isdigit(i))));
1703 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1705 DEBUG_L(PerlIO_printf(Perl_debug_log,
1706 "isgraph('%s') unexpectedly is %d\n",
1707 name, cBOOL(isgraph(i))));
1709 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1711 DEBUG_L(PerlIO_printf(Perl_debug_log,
1712 "islower('%s') unexpectedly is %d\n",
1713 name, cBOOL(islower(i))));
1715 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1717 DEBUG_L(PerlIO_printf(Perl_debug_log,
1718 "isprint('%s') unexpectedly is %d\n",
1719 name, cBOOL(isprint(i))));
1721 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1723 DEBUG_L(PerlIO_printf(Perl_debug_log,
1724 "ispunct('%s') unexpectedly is %d\n",
1725 name, cBOOL(ispunct(i))));
1727 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1729 DEBUG_L(PerlIO_printf(Perl_debug_log,
1730 "isspace('%s') unexpectedly is %d\n",
1731 name, cBOOL(isspace(i))));
1733 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1735 DEBUG_L(PerlIO_printf(Perl_debug_log,
1736 "isupper('%s') unexpectedly is %d\n",
1737 name, cBOOL(isupper(i))));
1739 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1741 DEBUG_L(PerlIO_printf(Perl_debug_log,
1742 "isxdigit('%s') unexpectedly is %d\n",
1743 name, cBOOL(isxdigit(i))));
1745 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1747 DEBUG_L(PerlIO_printf(Perl_debug_log,
1748 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1749 name, tolower(i), (int) toLOWER_A(i)));
1751 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1753 DEBUG_L(PerlIO_printf(Perl_debug_log,
1754 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1755 name, toupper(i), (int) toUPPER_A(i)));
1757 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1759 DEBUG_L(PerlIO_printf(Perl_debug_log,
1760 "'\\n' (=%02X) is not a control\n", (int) i));
1763 /* Add to the list; Separate multiple entries with a blank */
1766 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1768 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1774 if (bad_count == 2 && maybe_utf8_turkic) {
1776 *bad_chars_list = '\0';
1777 PL_fold_locale['I'] = 'I';
1778 PL_fold_locale['i'] = 'i';
1779 PL_in_utf8_turkic_locale = TRUE;
1780 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1781 __FILE__, __LINE__, newctype));
1784 PL_in_utf8_turkic_locale = FALSE;
1789 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1790 * this locale requires more than one byte, there are going to be
1792 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1793 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1794 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1796 if ( check_for_problems && MB_CUR_MAX > 1
1797 && ! PL_in_utf8_CTYPE_locale
1799 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1800 * locale. Just assume that the implementation for them (plus
1801 * for POSIX) is correct and the > 1 value is spurious. (Since
1802 * these are specially handled to never be considered UTF-8
1803 * locales, as long as this is the only problem, everything
1804 * should work fine */
1805 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1807 multi_byte_locale = TRUE;
1812 /* If we found problems and we want them output, do so */
1813 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1814 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1816 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1817 PL_warn_locale = Perl_newSVpvf(aTHX_
1818 "Locale '%s' contains (at least) the following characters"
1819 " which have\nunexpected meanings: %s\nThe Perl program"
1820 " will use the expected meanings",
1821 newctype, bad_chars_list);
1824 PL_warn_locale = Perl_newSVpvf(aTHX_
1825 "Locale '%s' may not work well.%s%s%s\n",
1828 ? " Some characters in it are not recognized by"
1832 ? "\nThe following characters (and maybe others)"
1833 " may not have the same meaning as the Perl"
1834 " program expects:\n"
1842 # ifdef HAS_NL_LANGINFO
1844 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1845 /* parameter FALSE is a don't care here */
1846 my_nl_langinfo(CODESET, FALSE));
1850 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1852 /* If we are actually in the scope of the locale or are debugging,
1853 * output the message now. If not in that scope, we save the
1854 * message to be output at the first operation using this locale,
1855 * if that actually happens. Most programs don't use locales, so
1856 * they are immune to bad ones. */
1857 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1859 /* The '0' below suppresses a bogus gcc compiler warning */
1860 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1862 if (IN_LC(LC_CTYPE)) {
1863 SvREFCNT_dec_NN(PL_warn_locale);
1864 PL_warn_locale = NULL;
1870 #endif /* USE_LOCALE_CTYPE */
1875 Perl__warn_problematic_locale()
1878 #ifdef USE_LOCALE_CTYPE
1882 /* Internal-to-core function that outputs the message in PL_warn_locale,
1883 * and then NULLS it. Should be called only through the macro
1884 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1886 if (PL_warn_locale) {
1887 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1888 SvPVX(PL_warn_locale),
1889 0 /* dummy to avoid compiler warning */ );
1890 SvREFCNT_dec_NN(PL_warn_locale);
1891 PL_warn_locale = NULL;
1899 S_new_collate(pTHX_ const char *newcoll)
1902 #ifndef USE_LOCALE_COLLATE
1904 PERL_UNUSED_ARG(newcoll);
1905 PERL_UNUSED_CONTEXT;
1909 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1910 * core Perl this and that 'newcoll' is the name of the new locale.
1912 * The design of locale collation is that every locale change is given an
1913 * index 'PL_collation_ix'. The first time a string particpates in an
1914 * operation that requires collation while locale collation is active, it
1915 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1916 * magic includes the collation index, and the transformation of the string
1917 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1918 * instead of the string itself. If a string changes, the magic is
1919 * cleared. The next time the locale changes, the index is incremented,
1920 * and so we know during a comparison that the transformation is not
1921 * necessarily still valid, and so is recomputed. Note that if the locale
1922 * changes enough times, the index could wrap (a U32), and it is possible
1923 * that a transformation would improperly be considered valid, leading to
1924 * an unlikely bug */
1927 if (PL_collation_name) {
1929 Safefree(PL_collation_name);
1930 PL_collation_name = NULL;
1932 PL_collation_standard = TRUE;
1933 is_standard_collation:
1934 PL_collxfrm_base = 0;
1935 PL_collxfrm_mult = 2;
1936 PL_in_utf8_COLLATE_locale = FALSE;
1937 PL_strxfrm_NUL_replacement = '\0';
1938 PL_strxfrm_max_cp = 0;
1942 /* If this is not the same locale as currently, set the new one up */
1943 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1945 Safefree(PL_collation_name);
1946 PL_collation_name = stdize_locale(savepv(newcoll));
1947 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1948 if (PL_collation_standard) {
1949 goto is_standard_collation;
1952 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1953 PL_strxfrm_NUL_replacement = '\0';
1954 PL_strxfrm_max_cp = 0;
1956 /* A locale collation definition includes primary, secondary, tertiary,
1957 * etc. weights for each character. To sort, the primary weights are
1958 * used, and only if they compare equal, then the secondary weights are
1959 * used, and only if they compare equal, then the tertiary, etc.
1961 * strxfrm() works by taking the input string, say ABC, and creating an
1962 * output transformed string consisting of first the primary weights,
1963 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1964 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1965 * may not have weights at every level. In our example, let's say B
1966 * doesn't have a tertiary weight, and A doesn't have a secondary
1967 * weight. The constructed string is then going to be
1968 * A¹B¹C¹ B²C² A³C³ ....
1969 * This has the desired effect that strcmp() will look at the secondary
1970 * or tertiary weights only if the strings compare equal at all higher
1971 * priority weights. The spaces shown here, like in
1973 * are not just for readability. In the general case, these must
1974 * actually be bytes, which we will call here 'separator weights'; and
1975 * they must be smaller than any other weight value, but since these
1976 * are C strings, only the terminating one can be a NUL (some
1977 * implementations may include a non-NUL separator weight just before
1978 * the NUL). Implementations tend to reserve 01 for the separator
1979 * weights. They are needed so that a shorter string's secondary
1980 * weights won't be misconstrued as primary weights of a longer string,
1981 * etc. By making them smaller than any other weight, the shorter
1982 * string will sort first. (Actually, if all secondary weights are
1983 * smaller than all primary ones, there is no need for a separator
1984 * weight between those two levels, etc.)
1986 * The length of the transformed string is roughly a linear function of
1987 * the input string. It's not exactly linear because some characters
1988 * don't have weights at all levels. When we call strxfrm() we have to
1989 * allocate some memory to hold the transformed string. The
1990 * calculations below try to find coefficients 'm' and 'b' for this
1991 * locale so that m*x + b equals how much space we need, given the size
1992 * of the input string in 'x'. If we calculate too small, we increase
1993 * the size as needed, and call strxfrm() again, but it is better to
1994 * get it right the first time to avoid wasted expensive string
1995 * transformations. */
1998 /* We use the string below to find how long the tranformation of it
1999 * is. Almost all locales are supersets of ASCII, or at least the
2000 * ASCII letters. We use all of them, half upper half lower,
2001 * because if we used fewer, we might hit just the ones that are
2002 * outliers in a particular locale. Most of the strings being
2003 * collated will contain a preponderance of letters, and even if
2004 * they are above-ASCII, they are likely to have the same number of
2005 * weight levels as the ASCII ones. It turns out that digits tend
2006 * to have fewer levels, and some punctuation has more, but those
2007 * are relatively sparse in text, and khw believes this gives a
2008 * reasonable result, but it could be changed if experience so
2010 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
2011 char * x_longer; /* Transformed 'longer' */
2012 Size_t x_len_longer; /* Length of 'x_longer' */
2014 char * x_shorter; /* We also transform a substring of 'longer' */
2015 Size_t x_len_shorter;
2017 /* _mem_collxfrm() is used get the transformation (though here we
2018 * are interested only in its length). It is used because it has
2019 * the intelligence to handle all cases, but to work, it needs some
2020 * values of 'm' and 'b' to get it started. For the purposes of
2021 * this calculation we use a very conservative estimate of 'm' and
2022 * 'b'. This assumes a weight can be multiple bytes, enough to
2023 * hold any UV on the platform, and there are 5 levels, 4 weight
2024 * bytes, and a trailing NUL. */
2025 PL_collxfrm_base = 5;
2026 PL_collxfrm_mult = 5 * sizeof(UV);
2028 /* Find out how long the transformation really is */
2029 x_longer = _mem_collxfrm(longer,
2033 /* We avoid converting to UTF-8 in the
2034 * called function by telling it the
2035 * string is in UTF-8 if the locale is a
2036 * UTF-8 one. Since the string passed
2037 * here is invariant under UTF-8, we can
2038 * claim it's UTF-8 even though it isn't.
2040 PL_in_utf8_COLLATE_locale);
2043 /* Find out how long the transformation of a substring of 'longer'
2044 * is. Together the lengths of these transformations are
2045 * sufficient to calculate 'm' and 'b'. The substring is all of
2046 * 'longer' except the first character. This minimizes the chances
2047 * of being swayed by outliers */
2048 x_shorter = _mem_collxfrm(longer + 1,
2051 PL_in_utf8_COLLATE_locale);
2052 Safefree(x_shorter);
2054 /* If the results are nonsensical for this simple test, the whole
2055 * locale definition is suspect. Mark it so that locale collation
2056 * is not active at all for it. XXX Should we warn? */
2057 if ( x_len_shorter == 0
2058 || x_len_longer == 0
2059 || x_len_shorter >= x_len_longer)
2061 PL_collxfrm_mult = 0;
2062 PL_collxfrm_base = 0;
2065 SSize_t base; /* Temporary */
2067 /* We have both: m * strlen(longer) + b = x_len_longer
2068 * m * strlen(shorter) + b = x_len_shorter;
2069 * subtracting yields:
2070 * m * (strlen(longer) - strlen(shorter))
2071 * = x_len_longer - x_len_shorter
2072 * But we have set things up so that 'shorter' is 1 byte smaller
2073 * than 'longer'. Hence:
2074 * m = x_len_longer - x_len_shorter
2076 * But if something went wrong, make sure the multiplier is at
2079 if (x_len_longer > x_len_shorter) {
2080 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2083 PL_collxfrm_mult = 1;
2088 * but in case something has gone wrong, make sure it is
2090 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2095 /* Add 1 for the trailing NUL */
2096 PL_collxfrm_base = base + 1;
2101 if (DEBUG_L_TEST || debug_initialization) {
2102 PerlIO_printf(Perl_debug_log,
2103 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2105 " collate multipler=%zu, collate base=%zu\n",
2107 PL_in_utf8_COLLATE_locale,
2108 x_len_shorter, x_len_longer,
2109 PL_collxfrm_mult, PL_collxfrm_base);
2116 #endif /* USE_LOCALE_COLLATE */
2124 #define USE_WSETLOCALE
2126 #ifdef USE_WSETLOCALE
2129 S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2136 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2143 Newx(wlocale, req_size, wchar_t);
2144 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2153 wresult = _wsetlocale(category, wlocale);
2157 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2158 Newx(result, req_size, char);
2159 SAVEFREEPV(result); /* is there something better we can do here? */
2160 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2161 result, req_size, NULL, NULL)) {
2176 S_win32_setlocale(pTHX_ int category, const char* locale)
2178 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2179 * difference between the two unless the input locale is "", which normally
2180 * means on Windows to get the machine default, which is set via the
2181 * computer's "Regional and Language Options" (or its current equivalent).
2182 * In POSIX, it instead means to find the locale from the user's
2183 * environment. This routine changes the Windows behavior to first look in
2184 * the environment, and, if anything is found, use that instead of going to
2185 * the machine default. If there is no environment override, the machine
2186 * default is used, by calling the real setlocale() with "".
2188 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2189 * use the particular category's variable if set; otherwise to use the LANG
2192 bool override_LC_ALL = FALSE;
2196 if (locale && strEQ(locale, "")) {
2200 locale = PerlEnv_getenv("LC_ALL");
2202 if (category == LC_ALL) {
2203 override_LC_ALL = TRUE;
2209 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2210 if (category == categories[i]) {
2211 locale = PerlEnv_getenv(category_names[i]);
2216 locale = PerlEnv_getenv("LANG");
2232 #ifdef USE_WSETLOCALE
2233 result = S_wrap_wsetlocale(aTHX_ category, locale);
2235 result = setlocale(category, locale);
2237 DEBUG_L(STMT_START {
2239 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2240 setlocale_debug_string(category, locale, result));
2244 if (! override_LC_ALL) {
2248 /* Here the input category was LC_ALL, and we have set it to what is in the
2249 * LANG variable or the system default if there is no LANG. But these have
2250 * lower priority than the other LC_foo variables, so override it for each
2251 * one that is set. (If they are set to "", it means to use the same thing
2252 * we just set LC_ALL to, so can skip) */
2254 for (i = 0; i < LC_ALL_INDEX; i++) {
2255 result = PerlEnv_getenv(category_names[i]);
2256 if (result && strNE(result, "")) {
2257 #ifdef USE_WSETLOCALE
2258 S_wrap_wsetlocale(aTHX_ categories[i], result);
2260 setlocale(categories[i], result);
2262 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2264 setlocale_debug_string(categories[i], result, "not captured")));
2268 result = setlocale(LC_ALL, NULL);
2269 DEBUG_L(STMT_START {
2271 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2273 setlocale_debug_string(LC_ALL, NULL, result));
2283 =for apidoc Perl_setlocale
2285 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2286 taking the same parameters, and returning the same information, except that it
2287 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2288 instead return C<C> if the underlying locale has a non-dot decimal point
2289 character, or a non-empty thousands separator for displaying floating point
2290 numbers. This is because perl keeps that locale category such that it has a
2291 dot and empty separator, changing the locale briefly during the operations
2292 where the underlying one is required. C<Perl_setlocale> knows about this, and
2293 compensates; regular C<setlocale> doesn't.
2295 Another reason it isn't completely a drop-in replacement is that it is
2296 declared to return S<C<const char *>>, whereas the system setlocale omits the
2297 C<const> (presumably because its API was specified long ago, and can't be
2298 updated; it is illegal to change the information C<setlocale> returns; doing
2299 so leads to segfaults.)
2301 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2302 C<setlocale> can be completely ineffective on some platforms under some
2305 C<Perl_setlocale> should not be used to change the locale except on systems
2306 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2307 the system C<setlocale()> is ineffective, returning the wrong information, and
2308 failing to actually change the locale. C<Perl_setlocale>, however works
2309 properly in all circumstances.
2311 The return points to a per-thread static buffer, which is overwritten the next
2312 time C<Perl_setlocale> is called from the same thread.
2319 Perl_setlocale(const int category, const char * locale)
2321 /* This wraps POSIX::setlocale() */
2325 PERL_UNUSED_ARG(category);
2326 PERL_UNUSED_ARG(locale);
2332 const char * retval;
2333 const char * newlocale;
2336 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2338 #ifdef USE_LOCALE_NUMERIC
2340 /* A NULL locale means only query what the current one is. We have the
2341 * LC_NUMERIC name saved, because we are normally switched into the C
2342 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2343 * the correct results. All other categories don't require special
2345 if (locale == NULL) {
2346 if (category == LC_NUMERIC) {
2348 /* We don't have to copy this return value, as it is a per-thread
2349 * variable, and won't change until a future setlocale */
2350 return PL_numeric_name;
2355 else if (category == LC_ALL) {
2356 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2365 retval = save_to_buffer(do_setlocale_r(category, locale),
2366 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2369 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2371 if (locale == NULL && category == LC_ALL) {
2372 RESTORE_LC_NUMERIC();
2377 DEBUG_L(PerlIO_printf(Perl_debug_log,
2378 "%s:%d: %s\n", __FILE__, __LINE__,
2379 setlocale_debug_string(category, locale, retval)));
2387 /* If locale == NULL, we are just querying the state */
2388 if (locale == NULL) {
2392 /* Now that have switched locales, we have to update our records to
2397 #ifdef USE_LOCALE_CTYPE
2404 #ifdef USE_LOCALE_COLLATE
2407 new_collate(retval);
2411 #ifdef USE_LOCALE_NUMERIC
2414 new_numeric(retval);
2422 /* LC_ALL updates all the things we care about. The values may not
2423 * be the same as 'retval', as the locale "" may have set things
2426 # ifdef USE_LOCALE_CTYPE
2428 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2429 new_ctype(newlocale);
2430 Safefree(newlocale);
2432 # endif /* USE_LOCALE_CTYPE */
2433 # ifdef USE_LOCALE_COLLATE
2435 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2436 new_collate(newlocale);
2437 Safefree(newlocale);
2440 # ifdef USE_LOCALE_NUMERIC
2442 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2443 new_numeric(newlocale);
2444 Safefree(newlocale);
2446 # endif /* USE_LOCALE_NUMERIC */
2459 PERL_STATIC_INLINE const char *
2460 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2462 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2463 * growing it if necessary */
2467 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2473 string_size = strlen(string) + offset + 1;
2475 if (*buf_size == 0) {
2476 Newx(*buf, string_size, char);
2477 *buf_size = string_size;
2479 else if (string_size > *buf_size) {
2480 Renew(*buf, string_size, char);
2481 *buf_size = string_size;
2484 Copy(string, *buf + offset, string_size - offset, char);
2490 =for apidoc Perl_langinfo
2492 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2493 taking the same C<item> parameter values, and returning the same information.
2494 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2495 of Perl's locale handling from your code, and can be used on systems that lack
2496 a native C<nl_langinfo>.
2504 The reason it isn't quite a drop-in replacement is actually an advantage. The
2505 only difference is that it returns S<C<const char *>>, whereas plain
2506 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2507 forbidden to write into the buffer. By declaring this C<const>, the compiler
2508 enforces this restriction, so if it is violated, you know at compilation time,
2509 rather than getting segfaults at runtime.
2513 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2514 without you having to write extra code. The reason for the extra code would be
2515 because these are from the C<LC_NUMERIC> locale category, which is normally
2516 kept set by Perl so that the radix is a dot, and the separator is the empty
2517 string, no matter what the underlying locale is supposed to be, and so to get
2518 the expected results, you have to temporarily toggle into the underlying
2519 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2520 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2521 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2522 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2523 (decimal point) character to be a dot.)
2527 The system function it replaces can have its static return buffer trashed,
2528 not only by a subsequent call to that function, but by a C<freelocale>,
2529 C<setlocale>, or other locale change. The returned buffer of this function is
2530 not changed until the next call to it, so the buffer is never in a trashed
2535 Its return buffer is per-thread, so it also is never overwritten by a call to
2536 this function from another thread; unlike the function it replaces.
2540 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2541 as Windows, hence makes your code more portable. Of the fifty-some possible
2542 items specified by the POSIX 2008 standard,
2543 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2544 only one is completely unimplemented, though on non-Windows platforms, another
2545 significant one is also not implemented). It uses various techniques to
2546 recover the other items, including calling C<L<localeconv(3)>>, and
2547 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2548 available. Later C<strftime()> versions have additional capabilities; C<""> is
2549 returned for those not available on your system.
2551 It is important to note that when called with an item that is recovered by
2552 using C<localeconv>, the buffer from any previous explicit call to
2553 C<localeconv> will be overwritten. This means you must save that buffer's
2554 contents if you need to access them after a call to this function. (But note
2555 that you might not want to be using C<localeconv()> directly anyway, because of
2556 issues like the ones listed in the second item of this list (above) for
2557 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2558 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2561 The details for those items which may deviate from what this emulation returns
2562 and what a native C<nl_langinfo()> would return are specified in
2567 When using C<Perl_langinfo> on systems that don't have a native
2568 C<nl_langinfo()>, you must
2570 #include "perl_langinfo.h"
2572 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2573 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2574 C<langinfo.h> would try to import into the namespace for code that doesn't need
2577 The original impetus for C<Perl_langinfo()> was so that code that needs to
2578 find out the current currency symbol, floating point radix character, or digit
2579 grouping separator can use, on all systems, the simpler and more
2580 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2581 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2582 is better to use the methods given in L<perlcall> to call
2583 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2590 #ifdef HAS_NL_LANGINFO
2591 Perl_langinfo(const nl_item item)
2593 Perl_langinfo(const int item)
2596 return my_nl_langinfo(item, TRUE);
2600 #ifdef HAS_NL_LANGINFO
2601 S_my_nl_langinfo(const nl_item item, bool toggle)
2603 S_my_nl_langinfo(const int item, bool toggle)
2607 const char * retval;
2609 #ifdef USE_LOCALE_NUMERIC
2611 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2612 * two items, and only if not already there */
2613 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2614 || PL_numeric_underlying))
2616 #endif /* No toggling needed if not using LC_NUMERIC */
2620 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2621 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2622 || ! defined(HAS_POSIX_2008_LOCALE) \
2623 || ! defined(DUPLOCALE)
2625 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2626 * for those items dependent on it. This must be copied to a buffer before
2627 * switching back, as some systems destroy the buffer when setlocale() is
2631 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2634 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2637 LOCALE_LOCK; /* Prevent interference from another thread executing
2638 this code section (the only call to nl_langinfo in
2642 /* Copy to a per-thread buffer, which is also one that won't be
2643 * destroyed by a subsequent setlocale(), such as the
2644 * RESTORE_LC_NUMERIC may do just below. */
2645 retval = save_to_buffer(nl_langinfo(item),
2646 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2651 RESTORE_LC_NUMERIC();
2655 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2658 bool do_free = FALSE;
2659 locale_t cur = uselocale((locale_t) 0);
2661 if (cur == LC_GLOBAL_LOCALE) {
2662 cur = duplocale(LC_GLOBAL_LOCALE);
2666 # ifdef USE_LOCALE_NUMERIC
2669 if (PL_underlying_numeric_obj) {
2670 cur = PL_underlying_numeric_obj;
2673 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2680 /* We have to save it to a buffer, because the freelocale() just below
2681 * can invalidate the internal one */
2682 retval = save_to_buffer(nl_langinfo_l(item, cur),
2683 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2692 if (strEQ(retval, "")) {
2693 if (item == YESSTR) {
2696 if (item == NOSTR) {
2703 #else /* Below, emulate nl_langinfo as best we can */
2707 # ifdef HAS_LOCALECONV
2709 const struct lconv* lc;
2711 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2713 # ifdef TS_W32_BROKEN_LOCALECONV
2715 const char * save_global;
2716 const char * save_thread;
2724 # ifdef HAS_STRFTIME
2727 bool return_format = FALSE; /* Return the %format, not the value */
2728 const char * format;
2732 /* We copy the results to a per-thread buffer, even if not
2733 * multi-threaded. This is in part to simplify this code, and partly
2734 * because we need a buffer anyway for strftime(), and partly because a
2735 * call of localeconv() could otherwise wipe out the buffer, and the
2736 * programmer would not be expecting this, as this is a nl_langinfo()
2737 * substitute after all, so s/he might be thinking their localeconv()
2738 * is safe until another localeconv() call. */
2743 /* This is unimplemented */
2744 case ERA: /* For use with strftime() %E modifier */
2749 /* We use only an English set, since we don't know any more */
2750 case YESEXPR: return "^[+1yY]";
2751 case YESSTR: return "yes";
2752 case NOEXPR: return "^[-0nN]";
2753 case NOSTR: return "no";
2759 /* On non-windows, this is unimplemented, in part because of
2760 * inconsistencies between vendors. The Darwin native
2761 * nl_langinfo() implementation simply looks at everything past
2762 * any dot in the name, but that doesn't work for other
2763 * vendors. Many Linux locales that don't have UTF-8 in their
2764 * names really are UTF-8, for example; z/OS locales that do
2765 * have UTF-8 in their names, aren't really UTF-8 */
2770 { /* But on Windows, the name does seem to be consistent, so
2775 const char * name = my_setlocale(LC_CTYPE, NULL);
2777 if (isNAME_C_OR_POSIX(name)) {
2778 return "ANSI_X3.4-1968";
2781 /* Find the dot in the locale name */
2782 first = (const char *) strchr(name, '.');
2788 /* Look at everything past the dot */
2793 if (! isDIGIT(*p)) {
2800 /* Here everything past the dot is a digit. Treat it as a
2802 retval = save_to_buffer("CP", &PL_langinfo_buf,
2803 &PL_langinfo_bufsize, 0);
2804 offset = STRLENs("CP");
2808 retval = save_to_buffer(first, &PL_langinfo_buf,
2809 &PL_langinfo_bufsize, offset);
2815 # ifdef HAS_LOCALECONV
2819 /* We don't bother with localeconv_l() because any system that
2820 * has it is likely to also have nl_langinfo() */
2822 LOCALE_LOCK_V; /* Prevent interference with other threads
2823 using localeconv() */
2825 # ifdef TS_W32_BROKEN_LOCALECONV
2827 /* This is a workaround for a Windows bug prior to VS 15.
2828 * What we do here is, while locked, switch to the global
2829 * locale so localeconv() works; then switch back just before
2830 * the unlock. This can screw things up if some thread is
2831 * already using the global locale while assuming no other is.
2832 * A different workaround would be to call GetCurrencyFormat on
2833 * a known value, and parse it; patches welcome
2835 * We have to use LC_ALL instead of LC_MONETARY because of
2836 * another bug in Windows */
2838 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2839 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2840 save_global= savepv(my_setlocale(LC_ALL, NULL));
2841 my_setlocale(LC_ALL, save_thread);
2847 || ! lc->currency_symbol
2848 || strEQ("", lc->currency_symbol))
2854 /* Leave the first spot empty to be filled in below */
2855 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2856 &PL_langinfo_bufsize, 1);
2857 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2858 { /* khw couldn't figure out how the localedef specifications
2859 would show that the $ should replace the radix; this is
2860 just a guess as to how it might work.*/
2861 PL_langinfo_buf[0] = '.';
2863 else if (lc->p_cs_precedes) {
2864 PL_langinfo_buf[0] = '-';
2867 PL_langinfo_buf[0] = '+';
2870 # ifdef TS_W32_BROKEN_LOCALECONV
2872 my_setlocale(LC_ALL, save_global);
2873 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2874 my_setlocale(LC_ALL, save_thread);
2875 Safefree(save_global);
2876 Safefree(save_thread);
2883 # ifdef TS_W32_BROKEN_LOCALECONV
2887 /* For this, we output a known simple floating point number to
2888 * a buffer, and parse it, looking for the radix */
2891 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2894 if (PL_langinfo_bufsize < 10) {
2895 PL_langinfo_bufsize = 10;
2896 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2899 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2901 if (needed_size >= (int) PL_langinfo_bufsize) {
2902 PL_langinfo_bufsize = needed_size + 1;
2903 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2904 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2906 assert(needed_size < (int) PL_langinfo_bufsize);
2909 ptr = PL_langinfo_buf;
2910 e = PL_langinfo_buf + PL_langinfo_bufsize;
2913 while (ptr < e && *ptr != '1') {
2920 while (ptr < e && *ptr != '5') {
2924 /* Everything in between is the radix string */
2926 PL_langinfo_buf[0] = '?';
2927 PL_langinfo_buf[1] = '\0';
2931 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2935 RESTORE_LC_NUMERIC();
2938 retval = PL_langinfo_buf;
2943 case RADIXCHAR: /* No special handling needed */
2950 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2953 LOCALE_LOCK_V; /* Prevent interference with other threads
2954 using localeconv() */
2956 # ifdef TS_W32_BROKEN_LOCALECONV
2958 /* This should only be for the thousands separator. A
2959 * different work around would be to use GetNumberFormat on a
2960 * known value and parse the result to find the separator */
2961 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2962 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2963 save_global = savepv(my_setlocale(LC_ALL, NULL));
2964 my_setlocale(LC_ALL, save_thread);
2966 /* This is the start of code that for broken Windows replaces
2967 * the above and below code, and instead calls
2968 * GetNumberFormat() and then would parse that to find the
2969 * thousands separator. It needs to handle UTF-16 vs -8
2972 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2973 DEBUG_L(PerlIO_printf(Perl_debug_log,
2974 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2975 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2985 temp = (item == RADIXCHAR)
2987 : lc->thousands_sep;
2993 retval = save_to_buffer(temp, &PL_langinfo_buf,
2994 &PL_langinfo_bufsize, 0);
2996 # ifdef TS_W32_BROKEN_LOCALECONV
2998 my_setlocale(LC_ALL, save_global);
2999 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3000 my_setlocale(LC_ALL, save_thread);
3001 Safefree(save_global);
3002 Safefree(save_thread);
3009 RESTORE_LC_NUMERIC();
3015 # ifdef HAS_STRFTIME
3017 /* These are defined by C89, so we assume that strftime supports
3018 * them, and so are returned unconditionally; they may not be what
3019 * the locale actually says, but should give good enough results
3020 * for someone using them as formats (as opposed to trying to parse
3021 * them to figure out what the locale says). The other format
3022 * items are actually tested to verify they work on the platform */
3023 case D_FMT: return "%x";
3024 case T_FMT: return "%X";
3025 case D_T_FMT: return "%c";
3027 /* These formats are only available in later strfmtime's */
3028 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
3030 /* The rest can be gotten from most versions of strftime(). */
3031 case ABDAY_1: case ABDAY_2: case ABDAY_3:
3032 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
3034 case AM_STR: case PM_STR:
3035 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
3036 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
3037 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
3038 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
3039 case DAY_5: case DAY_6: case DAY_7:
3040 case MON_1: case MON_2: case MON_3: case MON_4:
3041 case MON_5: case MON_6: case MON_7: case MON_8:
3042 case MON_9: case MON_10: case MON_11: case MON_12:
3046 init_tm(&tm); /* Precaution against core dumps */
3050 tm.tm_year = 2017 - 1900;
3057 "panic: %s: %d: switch case: %d problem",
3058 __FILE__, __LINE__, item);
3059 NOT_REACHED; /* NOTREACHED */
3061 case PM_STR: tm.tm_hour = 18;
3066 case ABDAY_7: tm.tm_wday++;
3067 case ABDAY_6: tm.tm_wday++;
3068 case ABDAY_5: tm.tm_wday++;
3069 case ABDAY_4: tm.tm_wday++;
3070 case ABDAY_3: tm.tm_wday++;
3071 case ABDAY_2: tm.tm_wday++;
3076 case DAY_7: tm.tm_wday++;
3077 case DAY_6: tm.tm_wday++;
3078 case DAY_5: tm.tm_wday++;
3079 case DAY_4: tm.tm_wday++;
3080 case DAY_3: tm.tm_wday++;
3081 case DAY_2: tm.tm_wday++;
3086 case ABMON_12: tm.tm_mon++;
3087 case ABMON_11: tm.tm_mon++;
3088 case ABMON_10: tm.tm_mon++;
3089 case ABMON_9: tm.tm_mon++;
3090 case ABMON_8: tm.tm_mon++;
3091 case ABMON_7: tm.tm_mon++;
3092 case ABMON_6: tm.tm_mon++;
3093 case ABMON_5: tm.tm_mon++;
3094 case ABMON_4: tm.tm_mon++;
3095 case ABMON_3: tm.tm_mon++;
3096 case ABMON_2: tm.tm_mon++;
3101 case MON_12: tm.tm_mon++;
3102 case MON_11: tm.tm_mon++;
3103 case MON_10: tm.tm_mon++;
3104 case MON_9: tm.tm_mon++;
3105 case MON_8: tm.tm_mon++;
3106 case MON_7: tm.tm_mon++;
3107 case MON_6: tm.tm_mon++;
3108 case MON_5: tm.tm_mon++;
3109 case MON_4: tm.tm_mon++;
3110 case MON_3: tm.tm_mon++;
3111 case MON_2: tm.tm_mon++;
3118 return_format = TRUE;
3123 return_format = TRUE;
3128 return_format = TRUE;
3133 return_format = TRUE;
3138 format = "%Ow"; /* Find the alternate digit for 0 */
3142 /* We can't use my_strftime() because it doesn't look at
3144 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3147 /* A zero return means one of:
3148 * a) there wasn't enough space in PL_langinfo_buf
3149 * b) the format, like a plain %p, returns empty
3150 * c) it was an illegal format, though some
3151 * implementations of strftime will just return the
3152 * illegal format as a plain character sequence.
3154 * To quickly test for case 'b)', try again but precede
3155 * the format with a plain character. If that result is
3156 * still empty, the problem is either 'a)' or 'c)' */
3158 Size_t format_size = strlen(format) + 1;
3159 Size_t mod_size = format_size + 1;
3163 Newx(mod_format, mod_size, char);
3164 Newx(temp_result, PL_langinfo_bufsize, char);
3166 my_strlcpy(mod_format + 1, format, mod_size);
3167 len = strftime(temp_result,
3168 PL_langinfo_bufsize,
3170 Safefree(mod_format);
3171 Safefree(temp_result);
3173 /* If 'len' is non-zero, it means that we had a case like
3174 * %p which means the current locale doesn't use a.m. or
3175 * p.m., and that is valid */
3178 /* Here, still didn't work. If we get well beyond a
3179 * reasonable size, bail out to prevent an infinite
3182 if (PL_langinfo_bufsize > 100 * format_size) {
3183 *PL_langinfo_buf = '\0';
3186 /* Double the buffer size to retry; Add 1 in case
3187 * original was 0, so we aren't stuck at 0. */
3188 PL_langinfo_bufsize *= 2;
3189 PL_langinfo_bufsize++;
3190 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3198 /* Here, we got a result.
3200 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3201 * alternate format for wday 0. If the value is the same as
3202 * the normal 0, there isn't an alternate, so clear the buffer.
3204 if ( item == ALT_DIGITS
3205 && strEQ(PL_langinfo_buf, "0"))
3207 *PL_langinfo_buf = '\0';
3210 /* ALT_DIGITS is problematic. Experiments on it showed that
3211 * strftime() did not always work properly when going from
3212 * alt-9 to alt-10. Only a few locales have this item defined,
3213 * and in all of them on Linux that khw was able to find,
3214 * nl_langinfo() merely returned the alt-0 character, possibly
3215 * doubled. Most Unicode digits are in blocks of 10
3216 * consecutive code points, so that is sufficient information
3217 * for those scripts, as we can infer alt-1, alt-2, .... But
3218 * for a Japanese locale, a CJK ideographic 0 is returned, and
3219 * the CJK digits are not in code point order, so you can't
3220 * really infer anything. The localedef for this locale did
3221 * specify the succeeding digits, so that strftime() works
3222 * properly on them, without needing to infer anything. But
3223 * the nl_langinfo() return did not give sufficient information
3224 * for the caller to understand what's going on. So until
3225 * there is evidence that it should work differently, this
3226 * returns the alt-0 string for ALT_DIGITS.
3228 * wday was chosen because its range is all a single digit.
3229 * Things like tm_sec have two digits as the minimum: '00' */
3233 retval = PL_langinfo_buf;
3235 /* If to return the format, not the value, overwrite the buffer
3236 * with it. But some strftime()s will keep the original format
3237 * if illegal, so change those to "" */
3238 if (return_format) {
3239 if (strEQ(PL_langinfo_buf, format)) {
3240 *PL_langinfo_buf = '\0';
3243 retval = save_to_buffer(format, &PL_langinfo_buf,
3244 &PL_langinfo_bufsize, 0);
3262 * Initialize locale awareness.
3265 Perl_init_i18nl10n(pTHX_ int printwarn)
3269 * 0 if not to output warning when setup locale is bad
3270 * 1 if to output warning based on value of PERL_BADLANG
3271 * >1 if to output regardless of PERL_BADLANG
3274 * 1 = set ok or not applicable,
3275 * 0 = fallback to a locale of lower priority
3276 * -1 = fallback to all locales failed, not even to the C locale
3278 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3279 * set, debugging information is output.
3281 * This looks more complicated than it is, mainly due to the #ifdefs.
3283 * We try to set LC_ALL to the value determined by the environment. If
3284 * there is no LC_ALL on this platform, we try the individual categories we
3285 * know about. If this works, we are done.
3287 * But if it doesn't work, we have to do something else. We search the
3288 * environment variables ourselves instead of relying on the system to do
3289 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3290 * think there is one), and the ultimate fallback "C". This is all done in
3291 * the same loop as above to avoid duplicating code, but it makes things
3292 * more complex. The 'trial_locales' array is initialized with just one
3293 * element; it causes the behavior described in the paragraph above this to
3294 * happen. If that fails, we add elements to 'trial_locales', and do extra
3295 * loop iterations to cause the behavior described in this paragraph.
3297 * On Ultrix, the locale MUST come from the environment, so there is
3298 * preliminary code to set it. I (khw) am not sure that it is necessary,
3299 * and that this couldn't be folded into the loop, but barring any real
3300 * platforms to test on, it's staying as-is
3302 * A slight complication is that in embedded Perls, the locale may already
3303 * be set-up, and we don't want to get it from the normal environment
3304 * variables. This is handled by having a special environment variable
3305 * indicate we're in this situation. We simply set setlocale's 2nd
3306 * parameter to be a NULL instead of "". That indicates to setlocale that
3307 * it is not to change anything, but to return the current value,
3308 * effectively initializing perl's db to what the locale already is.
3310 * We play the same trick with NULL if a LC_ALL succeeds. We call
3311 * setlocale() on the individual categores with NULL to get their existing
3312 * values for our db, instead of trying to change them.
3320 PERL_UNUSED_ARG(printwarn);
3322 #else /* USE_LOCALE */
3325 const char * const language = PerlEnv_getenv("LANGUAGE");
3329 /* NULL uses the existing already set up locale */
3330 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3333 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3334 unsigned int trial_locales_count;
3335 const char * const lc_all = PerlEnv_getenv("LC_ALL");
3336 const char * const lang = PerlEnv_getenv("LANG");
3337 bool setlocale_failure = FALSE;
3340 /* A later getenv() could zap this, so only use here */
3341 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3343 const bool locwarn = (printwarn > 1
3345 && ( ! bad_lang_use_once
3347 /* disallow with "" or "0" */
3349 && strNE("0", bad_lang_use_once)))));
3351 /* setlocale() return vals; not copied so must be looked at immediately */
3352 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3354 /* current locale for given category; should have been copied so aren't
3356 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3360 /* In some systems you can find out the system default locale
3361 * and use that as the fallback locale. */
3362 # define SYSTEM_DEFAULT_LOCALE
3364 # ifdef SYSTEM_DEFAULT_LOCALE
3366 const char *system_default_locale = NULL;
3371 # define DEBUG_LOCALE_INIT(a,b,c)
3374 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3376 # define DEBUG_LOCALE_INIT(category, locale, result) \
3378 if (debug_initialization) { \
3379 PerlIO_printf(Perl_debug_log, \
3381 __FILE__, __LINE__, \
3382 setlocale_debug_string(category, \
3388 /* Make sure the parallel arrays are properly set up */
3389 # ifdef USE_LOCALE_NUMERIC
3390 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3391 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3392 # ifdef USE_POSIX_2008_LOCALE
3393 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3396 # ifdef USE_LOCALE_CTYPE
3397 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3398 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3399 # ifdef USE_POSIX_2008_LOCALE
3400 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3403 # ifdef USE_LOCALE_COLLATE
3404 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3405 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3406 # ifdef USE_POSIX_2008_LOCALE
3407 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3410 # ifdef USE_LOCALE_TIME
3411 assert(categories[LC_TIME_INDEX] == LC_TIME);
3412 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3413 # ifdef USE_POSIX_2008_LOCALE
3414 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3417 # ifdef USE_LOCALE_MESSAGES
3418 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3419 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3420 # ifdef USE_POSIX_2008_LOCALE
3421 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3424 # ifdef USE_LOCALE_MONETARY
3425 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3426 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3427 # ifdef USE_POSIX_2008_LOCALE
3428 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3431 # ifdef USE_LOCALE_ADDRESS
3432 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3433 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3434 # ifdef USE_POSIX_2008_LOCALE
3435 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3438 # ifdef USE_LOCALE_IDENTIFICATION
3439 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3440 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3441 # ifdef USE_POSIX_2008_LOCALE
3442 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3445 # ifdef USE_LOCALE_MEASUREMENT
3446 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3447 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3448 # ifdef USE_POSIX_2008_LOCALE
3449 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3452 # ifdef USE_LOCALE_PAPER
3453 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3454 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3455 # ifdef USE_POSIX_2008_LOCALE
3456 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3459 # ifdef USE_LOCALE_TELEPHONE
3460 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3461 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3462 # ifdef USE_POSIX_2008_LOCALE
3463 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3466 # ifdef USE_LOCALE_SYNTAX
3467 assert(categories[LC_SYNTAX_INDEX] == LC_SYNTAX);
3468 assert(strEQ(category_names[LC_SYNTAX_INDEX], "LC_SYNTAX"));
3469 # ifdef USE_POSIX_2008_LOCALE
3470 assert(category_masks[LC_SYNTAX_INDEX] == LC_SYNTAX_MASK);
3473 # ifdef USE_LOCALE_TOD
3474 assert(categories[LC_TOD_INDEX] == LC_TOD);
3475 assert(strEQ(category_names[LC_TOD_INDEX], "LC_TOD"));
3476 # ifdef USE_POSIX_2008_LOCALE
3477 assert(category_masks[LC_TOD_INDEX] == LC_TOD_MASK);
3481 assert(categories[LC_ALL_INDEX] == LC_ALL);
3482 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3483 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3484 # ifdef USE_POSIX_2008_LOCALE
3485 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3488 # endif /* DEBUGGING */
3490 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
3491 * why these particular incantations are used. */
3493 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
3496 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
3499 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
3502 /* Initialize the cache of the program's UTF-8ness for the always known
3503 * locales C and POSIX */
3504 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3505 sizeof(PL_locale_utf8ness));
3507 # ifdef USE_THREAD_SAFE_LOCALE
3510 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3514 # ifdef USE_POSIX_2008_LOCALE
3516 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3517 if (! PL_C_locale_obj) {
3518 Perl_croak_nocontext(
3519 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3521 if (DEBUG_Lv_TEST || debug_initialization) {
3522 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3527 # ifdef USE_LOCALE_NUMERIC
3529 PL_numeric_radix_sv = newSVpvs(".");
3533 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3535 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3536 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3539 # ifdef LOCALE_ENVIRON_REQUIRED
3542 * Ultrix setlocale(..., "") fails if there are no environment
3543 * variables from which to get a locale name.
3547 # error Ultrix without LC_ALL not implemented
3553 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3554 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3555 if (sl_result[LC_ALL_INDEX])
3558 setlocale_failure = TRUE;
3560 if (! setlocale_failure) {
3561 const char * locale_param;
3562 for (i = 0; i < LC_ALL_INDEX; i++) {
3563 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3566 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3567 if (! sl_result[i]) {
3568 setlocale_failure = TRUE;
3570 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3575 # endif /* LC_ALL */
3576 # endif /* LOCALE_ENVIRON_REQUIRED */
3578 /* We try each locale in the list until we get one that works, or exhaust
3579 * the list. Normally the loop is executed just once. But if setting the
3580 * locale fails, inside the loop we add fallback trials to the array and so
3581 * will execute the loop multiple times */
3582 trial_locales[0] = setlocale_init;
3583 trial_locales_count = 1;
3585 for (i= 0; i < trial_locales_count; i++) {
3586 const char * trial_locale = trial_locales[i];
3590 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3591 * when i==0, but I (khw) don't think that behavior makes much
3593 setlocale_failure = FALSE;
3595 # ifdef SYSTEM_DEFAULT_LOCALE
3596 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3598 /* On Windows machines, an entry of "" after the 0th means to use
3599 * the system default locale, which we now proceed to get. */
3600 if (strEQ(trial_locale, "")) {
3603 /* Note that this may change the locale, but we are going to do
3604 * that anyway just below */
3605 system_default_locale = do_setlocale_c(LC_ALL, "");
3606 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3608 /* Skip if invalid or if it's already on the list of locales to
3610 if (! system_default_locale) {
3611 goto next_iteration;
3613 for (j = 0; j < trial_locales_count; j++) {
3614 if (strEQ(system_default_locale, trial_locales[j])) {
3615 goto next_iteration;
3619 trial_locale = system_default_locale;
3622 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3624 # endif /* SYSTEM_DEFAULT_LOCALE */
3630 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3631 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3632 if (! sl_result[LC_ALL_INDEX]) {
3633 setlocale_failure = TRUE;
3636 /* Since LC_ALL succeeded, it should have changed all the other
3637 * categories it can to its value; so we massage things so that the
3638 * setlocales below just return their category's current values.
3639 * This adequately handles the case in NetBSD where LC_COLLATE may
3640 * not be defined for a locale, and setting it individually will
3641 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3642 * the POSIX locale. */
3643 trial_locale = NULL;
3646 # endif /* LC_ALL */
3648 if (! setlocale_failure) {
3650 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3652 = savepv(do_setlocale_r(categories[j], trial_locale));
3653 if (! curlocales[j]) {
3654 setlocale_failure = TRUE;
3656 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3659 if (! setlocale_failure) { /* All succeeded */
3660 break; /* Exit trial_locales loop */
3664 /* Here, something failed; will need to try a fallback. */
3670 if (locwarn) { /* Output failure info only on the first one */
3674 PerlIO_printf(Perl_error_log,
3675 "perl: warning: Setting locale failed.\n");
3677 # else /* !LC_ALL */
3679 PerlIO_printf(Perl_error_log,
3680 "perl: warning: Setting locale failed for the categories:\n\t");
3682 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3683 if (! curlocales[j]) {
3684 PerlIO_printf(Perl_error_log, category_names[j]);
3687 Safefree(curlocales[j]);
3691 # endif /* LC_ALL */
3693 PerlIO_printf(Perl_error_log,
3694 "perl: warning: Please check that your locale settings:\n");
3698 PerlIO_printf(Perl_error_log,
3699 "\tLANGUAGE = %c%s%c,\n",
3700 language ? '"' : '(',
3701 language ? language : "unset",
3702 language ? '"' : ')');
3705 PerlIO_printf(Perl_error_log,
3706 "\tLC_ALL = %c%s%c,\n",
3708 lc_all ? lc_all : "unset",
3709 lc_all ? '"' : ')');
3711 # if defined(USE_ENVIRON_ARRAY)
3716 /* Look through the environment for any variables of the
3717 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3718 * already handled above. These are assumed to be locale
3719 * settings. Output them and their values. */
3720 for (e = environ; *e; e++) {
3721 const STRLEN prefix_len = sizeof("LC_") - 1;
3724 if ( strBEGINs(*e, "LC_")
3725 && ! strBEGINs(*e, "LC_ALL=")
3726 && (uppers_len = strspn(*e + prefix_len,
3727 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3728 && ((*e)[prefix_len + uppers_len] == '='))
3730 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3731 (int) (prefix_len + uppers_len), *e,
3732 *e + prefix_len + uppers_len + 1);
3739 PerlIO_printf(Perl_error_log,
3740 "\t(possibly more locale environment variables)\n");
3744 PerlIO_printf(Perl_error_log,
3745 "\tLANG = %c%s%c\n",
3747 lang ? lang : "unset",
3750 PerlIO_printf(Perl_error_log,
3751 " are supported and installed on your system.\n");
3754 /* Calculate what fallback locales to try. We have avoided this
3755 * until we have to, because failure is quite unlikely. This will
3756 * usually change the upper bound of the loop we are in.
3758 * Since the system's default way of setting the locale has not
3759 * found one that works, We use Perl's defined ordering: LC_ALL,
3760 * LANG, and the C locale. We don't try the same locale twice, so
3761 * don't add to the list if already there. (On POSIX systems, the
3762 * LC_ALL element will likely be a repeat of the 0th element "",
3763 * but there's no harm done by doing it explicitly.
3765 * Note that this tries the LC_ALL environment variable even on
3766 * systems which have no LC_ALL locale setting. This may or may
3767 * not have been originally intentional, but there's no real need
3768 * to change the behavior. */
3770 for (j = 0; j < trial_locales_count; j++) {
3771 if (strEQ(lc_all, trial_locales[j])) {
3775 trial_locales[trial_locales_count++] = lc_all;
3780 for (j = 0; j < trial_locales_count; j++) {
3781 if (strEQ(lang, trial_locales[j])) {
3785 trial_locales[trial_locales_count++] = lang;
3789 # if defined(WIN32) && defined(LC_ALL)
3791 /* For Windows, we also try the system default locale before "C".
3792 * (If there exists a Windows without LC_ALL we skip this because
3793 * it gets too complicated. For those, the "C" is the next
3794 * fallback possibility). The "" is the same as the 0th element of
3795 * the array, but the code at the loop above knows to treat it
3796 * differently when not the 0th */
3797 trial_locales[trial_locales_count++] = "";
3801 for (j = 0; j < trial_locales_count; j++) {
3802 if (strEQ("C", trial_locales[j])) {
3806 trial_locales[trial_locales_count++] = "C";
3809 } /* end of first time through the loop */
3817 } /* end of looping through the trial locales */
3819 if (ok < 1) { /* If we tried to fallback */
3821 if (! setlocale_failure) { /* fallback succeeded */
3822 msg = "Falling back to";
3824 else { /* fallback failed */
3827 /* We dropped off the end of the loop, so have to decrement i to
3828 * get back to the value the last time through */
3832 msg = "Failed to fall back to";
3834 /* To continue, we should use whatever values we've got */
3836 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3837 Safefree(curlocales[j]);
3838 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3839 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3844 const char * description;
3845 const char * name = "";
3846 if (strEQ(trial_locales[i], "C")) {
3847 description = "the standard locale";
3851 # ifdef SYSTEM_DEFAULT_LOCALE
3853 else if (strEQ(trial_locales[i], "")) {
3854 description = "the system default locale";
3855 if (system_default_locale) {
3856 name = system_default_locale;
3860 # endif /* SYSTEM_DEFAULT_LOCALE */
3863 description = "a fallback locale";
3864 name = trial_locales[i];
3866 if (name && strNE(name, "")) {
3867 PerlIO_printf(Perl_error_log,
3868 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3871 PerlIO_printf(Perl_error_log,
3872 "perl: warning: %s %s.\n", msg, description);
3875 } /* End of tried to fallback */
3877 /* Done with finding the locales; update our records */
3879 # ifdef USE_LOCALE_CTYPE
3881 new_ctype(curlocales[LC_CTYPE_INDEX]);
3884 # ifdef USE_LOCALE_COLLATE
3886 new_collate(curlocales[LC_COLLATE_INDEX]);
3889 # ifdef USE_LOCALE_NUMERIC
3891 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3895 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3897 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3899 /* This caches whether each category's locale is UTF-8 or not. This
3900 * may involve changing the locale. It is ok to do this at
3901 * initialization time before any threads have started, but not later
3902 * unless thread-safe operations are used.
3903 * Caching means that if the program heeds our dictate not to change
3904 * locales in threaded applications, this data will remain valid, and
3905 * it may get queried without having to change locales. If the
3906 * environment is such that all categories have the same locale, this
3907 * isn't needed, as the code will not change the locale; but this
3908 * handles the uncommon case where the environment has disparate
3909 * locales for the categories */
3910 (void) _is_cur_LC_category_utf8(categories[i]);
3914 Safefree(curlocales[i]);
3917 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3919 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3920 * locale is UTF-8. The call to new_ctype() just above has already
3921 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3922 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3923 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3924 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3925 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3927 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3928 This is an alternative to using the -C command line switch
3929 (the -C if present will override this). */
3931 const char *p = PerlEnv_getenv("PERL_UNICODE");
3932 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3933 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3938 #endif /* USE_LOCALE */
3941 /* So won't continue to output stuff */
3942 DEBUG_INITIALIZATION_set(FALSE);
3949 #ifdef USE_LOCALE_COLLATE
3952 Perl__mem_collxfrm(pTHX_ const char *input_string,
3953 STRLEN len, /* Length of 'input_string' */
3954 STRLEN *xlen, /* Set to length of returned string
3955 (not including the collation index
3957 bool utf8 /* Is the input in UTF-8? */
3961 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3962 * differences. First, it handles embedded NULs. Second, it allocates a bit
3963 * more memory than needed for the transformed data itself. The real
3964 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3965 * the length of that, and doesn't include the collation index size.
3966 * Please see sv_collxfrm() to see how this is used. */
3968 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3970 char * s = (char *) input_string;
3971 STRLEN s_strlen = strlen(input_string);
3973 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3974 STRLEN length_in_chars;
3975 bool first_time = TRUE; /* Cleared after first loop iteration */
3977 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3979 /* Must be NUL-terminated */
3980 assert(*(input_string + len) == '\0');
3982 /* If this locale has defective collation, skip */
3983 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3984 DEBUG_L(PerlIO_printf(Perl_debug_log,
3985 "_mem_collxfrm: locale's collation is defective\n"));
3989 /* Replace any embedded NULs with the control that sorts before any others.
3990 * This will give as good as possible results on strings that don't
3991 * otherwise contain that character, but otherwise there may be
3992 * less-than-perfect results with that character and NUL. This is
3993 * unavoidable unless we replace strxfrm with our own implementation. */
3994 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3998 STRLEN sans_nuls_len;
3999 int try_non_controls;
4000 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
4001 making sure 2nd byte is NUL.
4003 STRLEN this_replacement_len;
4005 /* If we don't know what non-NUL control character sorts lowest for
4006 * this locale, find it */
4007 if (PL_strxfrm_NUL_replacement == '\0') {
4009 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
4010 includes the collation index
4013 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
4015 /* Unlikely, but it may be that no control will work to replace
4016 * NUL, in which case we instead look for any character. Controls
4017 * are preferred because collation order is, in general, context
4018 * sensitive, with adjoining characters affecting the order, and
4019 * controls are less likely to have such interactions, allowing the
4020 * NUL-replacement to stand on its own. (Another way to look at it
4021 * is to imagine what would happen if the NUL were replaced by a
4022 * combining character; it wouldn't work out all that well.) */
4023 for (try_non_controls = 0;
4024 try_non_controls < 2;
4027 /* Look through all legal code points (NUL isn't) */
4028 for (j = 1; j < 256; j++) {
4029 char * x; /* j's xfrm plus collation index */
4030 STRLEN x_len; /* length of 'x' */
4031 STRLEN trial_len = 1;
4032 char cur_source[] = { '\0', '\0' };
4034 /* Skip non-controls the first time through the loop. The
4035 * controls in a UTF-8 locale are the L1 ones */
4036 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
4043 /* Create a 1-char string of the current code point */
4044 cur_source[0] = (char) j;
4046 /* Then transform it */
4047 x = _mem_collxfrm(cur_source, trial_len, &x_len,
4048 0 /* The string is not in UTF-8 */);
4050 /* Ignore any character that didn't successfully transform.
4056 /* If this character's transformation is lower than
4057 * the current lowest, this one becomes the lowest */
4058 if ( cur_min_x == NULL
4059 || strLT(x + COLLXFRM_HDR_LEN,
4060 cur_min_x + COLLXFRM_HDR_LEN))
4062 PL_strxfrm_NUL_replacement = j;
4063 Safefree(cur_min_x);
4069 } /* end of loop through all 255 characters */
4071 /* Stop looking if found */
4076 /* Unlikely, but possible, if there aren't any controls that
4077 * work in the locale, repeat the loop, looking for any
4078 * character that works */
4079 DEBUG_L(PerlIO_printf(Perl_debug_log,
4080 "_mem_collxfrm: No control worked. Trying non-controls\n"));
4081 } /* End of loop to try first the controls, then any char */
4084 DEBUG_L(PerlIO_printf(Perl_debug_log,
4085 "_mem_collxfrm: Couldn't find any character to replace"
4086 " embedded NULs in locale %s with", PL_collation_name));
4090 DEBUG_L(PerlIO_printf(Perl_debug_log,
4091 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
4092 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
4094 Safefree(cur_min_x);
4095 } /* End of determining the character that is to replace NULs */
4097 /* If the replacement is variant under UTF-8, it must match the
4098 * UTF8-ness of the original */
4099 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
4100 this_replacement_char[0] =
4101 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
4102 this_replacement_char[1] =
4103 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
4104 this_replacement_len = 2;
4107 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
4108 /* this_replacement_char[1] = '\0' was done at initialization */
4109 this_replacement_len = 1;
4112 /* The worst case length for the replaced string would be if every
4113 * character in it is NUL. Multiply that by the length of each
4114 * replacement, and allow for a trailing NUL */
4115 sans_nuls_len = (len * this_replacement_len) + 1;
4116 Newx(sans_nuls, sans_nuls_len, char);
4119 /* Replace each NUL with the lowest collating control. Loop until have
4120 * exhausted all the NULs */
4121 while (s + s_strlen < e) {
4122 my_strlcat(sans_nuls, s, sans_nuls_len);
4124 /* Do the actual replacement */
4125 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
4127 /* Move past the input NUL */
4129 s_strlen = strlen(s);
4132 /* And add anything that trails the final NUL */
4133 my_strlcat(sans_nuls, s, sans_nuls_len);
4135 /* Switch so below we transform this modified string */
4138 } /* End of replacing NULs */
4140 /* Make sure the UTF8ness of the string and locale match */
4141 if (utf8 != PL_in_utf8_COLLATE_locale) {
4142 /* XXX convert above Unicode to 10FFFF? */
4143 const char * const t = s; /* Temporary so we can later find where the
4146 /* Here they don't match. Change the string's to be what the locale is
4149 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
4150 s = (char *) bytes_to_utf8((const U8 *) s, &len);
4153 else { /* locale is not UTF-8; but input is; downgrade the input */
4155 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
4157 /* If the downgrade was successful we are done, but if the input
4158 * contains things that require UTF-8 to represent, have to do
4159 * damage control ... */
4160 if (UNLIKELY(utf8)) {
4162 /* What we do is construct a non-UTF-8 string with
4163 * 1) the characters representable by a single byte converted
4164 * to be so (if necessary);
4165 * 2) and the rest converted to collate the same as the
4166 * highest collating representable character. That makes
4167 * them collate at the end. This is similar to how we
4168 * handle embedded NULs, but we use the highest collating
4169 * code point instead of the smallest. Like the NUL case,
4170 * this isn't perfect, but is the best we can reasonably
4171 * do. Every above-255 code point will sort the same as
4172 * the highest-sorting 0-255 code point. If that code
4173 * point can combine in a sequence with some other code
4174 * points for weight calculations, us changing something to
4175 * be it can adversely affect the results. But in most
4176 * cases, it should work reasonably. And note that this is
4177 * really an illegal situation: using code points above 255
4178 * on a locale where only 0-255 are valid. If two strings
4179 * sort entirely equal, then the sort order for the
4180 * above-255 code points will be in code point order. */
4184 /* If we haven't calculated the code point with the maximum
4185 * collating order for this locale, do so now */
4186 if (! PL_strxfrm_max_cp) {
4189 /* The current transformed string that collates the
4190 * highest (except it also includes the prefixed collation
4192 char * cur_max_x = NULL;
4194 /* Look through all legal code points (NUL isn't) */
4195 for (j = 1; j < 256; j++) {
4198 char cur_source[] = { '\0', '\0' };
4200 /* Create a 1-char string of the current code point */
4201 cur_source[0] = (char) j;
4203 /* Then transform it */
4204 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4206 /* If something went wrong (which it shouldn't), just
4207 * ignore this code point */
4212 /* If this character's transformation is higher than
4213 * the current highest, this one becomes the highest */
4214 if ( cur_max_x == NULL
4215 || strGT(x + COLLXFRM_HDR_LEN,
4216 cur_max_x + COLLXFRM_HDR_LEN))
4218 PL_strxfrm_max_cp = j;
4219 Safefree(cur_max_x);
4228 DEBUG_L(PerlIO_printf(Perl_debug_log,
4229 "_mem_collxfrm: Couldn't find any character to"
4230 " replace above-Latin1 chars in locale %s with",
4231 PL_collation_name));
4235 DEBUG_L(PerlIO_printf(Perl_debug_log,
4236 "_mem_collxfrm: highest 1-byte collating character"
4237 " in locale %s is 0x%02X\n",
4239 PL_strxfrm_max_cp));
4241 Safefree(cur_max_x);
4244 /* Here we know which legal code point collates the highest.
4245 * We are ready to construct the non-UTF-8 string. The length
4246 * will be at least 1 byte smaller than the input string
4247 * (because we changed at least one 2-byte character into a
4248 * single byte), but that is eaten up by the trailing NUL */
4254 char * e = (char *) t + len;
4256 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4258 if (UTF8_IS_INVARIANT(cur_char)) {
4261 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4262 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4264 else { /* Replace illegal cp with highest collating
4266 s[d++] = PL_strxfrm_max_cp;
4270 Renew(s, d, char); /* Free up unused space */
4275 /* Here, we have constructed a modified version of the input. It could
4276 * be that we already had a modified copy before we did this version.
4277 * If so, that copy is no longer needed */
4278 if (t != input_string) {
4283 length_in_chars = (utf8)
4284 ? utf8_length((U8 *) s, (U8 *) s + len)
4287 /* The first element in the output is the collation id, used by
4288 * sv_collxfrm(); then comes the space for the transformed string. The
4289 * equation should give us a good estimate as to how much is needed */
4290 xAlloc = COLLXFRM_HDR_LEN
4292 + (PL_collxfrm_mult * length_in_chars);
4293 Newx(xbuf, xAlloc, char);
4294 if (UNLIKELY(! xbuf)) {
4295 DEBUG_L(PerlIO_printf(Perl_debug_log,
4296 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4300 /* Store the collation id */
4301 *(U32*)xbuf = PL_collation_ix;
4303 /* Then the transformation of the input. We loop until successful, or we
4307 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
https://perl5.git.perl.org / perl5.git / blob