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 /* On systems that accept any locale name, the real underlying \
642 * locale is often returned by this internal function, so we \
644 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
646 /* Internal glibc for querylocale(), but doesn't handle
647 * empty-string ("") locale properly; who knows what other
648 * glitches. Check for it now, under debug. */
650 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
651 uselocale((locale_t) 0));
653 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
654 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
655 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
657 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
658 if ( strNE(PL_curlocales[index], temp_name)
659 && ! ( isNAME_C_OR_POSIX(temp_name)
660 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
662 # ifdef USE_C_BACKTRACE
664 dump_c_backtrace(Perl_debug_log, 20, 1);
668 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
669 " (%s) and what internal glibc thinks"
670 " (%s)\n", category_names[index],
671 PL_curlocales[index], temp_name);
680 /* Without querylocale(), we have to use our record-keeping we've
683 if (category != LC_ALL) {
687 if (DEBUG_Lv_TEST || debug_initialization) {
688 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
693 return PL_curlocales[index];
695 else { /* For LC_ALL */
697 Size_t names_len = 0;
699 bool are_all_categories_the_same_locale = TRUE;
701 /* If we have a valid LC_ALL value, just return it */
702 if (PL_curlocales[LC_ALL_INDEX]) {
706 if (DEBUG_Lv_TEST || debug_initialization) {
707 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
712 return PL_curlocales[LC_ALL_INDEX];
715 /* Otherwise, we need to construct a string of name=value pairs.
716 * We use the glibc syntax, like
717 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
718 * First calculate the needed size. Along the way, check if all
719 * the locale names are the same */
720 for (i = 0; i < LC_ALL_INDEX; i++) {
724 if (DEBUG_Lv_TEST || debug_initialization) {
725 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]);
730 names_len += strlen(category_names[i])
732 + strlen(PL_curlocales[i])
735 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
736 are_all_categories_the_same_locale = FALSE;
740 /* If they are the same, we don't actually have to construct the
741 * string; we just make the entry in LC_ALL_INDEX valid, and be
742 * that single name */
743 if (are_all_categories_the_same_locale) {
744 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
745 return PL_curlocales[LC_ALL_INDEX];
748 names_len++; /* Trailing '\0' */
749 SAVEFREEPV(Newx(all_string, names_len, char));
752 /* Then fill in the string */
753 for (i = 0; i < LC_ALL_INDEX; i++) {
757 if (DEBUG_Lv_TEST || debug_initialization) {
758 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]);
763 my_strlcat(all_string, category_names[i], names_len);
764 my_strlcat(all_string, "=", names_len);
765 my_strlcat(all_string, PL_curlocales[i], names_len);
766 my_strlcat(all_string, ";", names_len);
771 if (DEBUG_L_TEST || debug_initialization) {
772 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
782 SETERRNO(EINVAL, LIB_INVARG);
790 } /* End of this being setlocale(LC_foo, NULL) */
792 /* Here, we are switching locales. */
794 # ifndef HAS_QUERYLOCALE
796 if (strEQ(locale, "")) {
798 /* For non-querylocale() systems, we do the setting of "" ourselves to
799 * be sure that we really know what's going on. We follow the Linux
800 * documented behavior (but if that differs from the actual behavior,
801 * this won't work exactly as the OS implements). We go out and
802 * examine the environment based on our understanding of how the system
803 * works, and use that to figure things out */
805 const char * const lc_all = PerlEnv_getenv("LC_ALL");
807 /* Use any "LC_ALL" environment variable, as it overrides everything
809 if (lc_all && strNE(lc_all, "")) {
814 /* Otherwise, we need to dig deeper. Unless overridden, the
815 * default is the LANG environment variable; if it doesn't exist,
818 const char * default_name;
820 default_name = PerlEnv_getenv("LANG");
822 if (! default_name || strEQ(default_name, "")) {
826 if (category != LC_ALL) {
827 const char * const name = PerlEnv_getenv(category_names[index]);
829 /* Here we are setting a single category. Assume will have the
831 locale = default_name;
833 /* But then look for an overriding environment variable */
834 if (name && strNE(name, "")) {
839 bool did_override = FALSE;
842 /* Here, we are getting LC_ALL. Any categories that don't have
843 * a corresponding environment variable set should be set to
844 * LANG, or to "C" if there is no LANG. If no individual
845 * categories differ from this, we can just set LC_ALL. This
846 * is buggy on systems that have extra categories that we don't
847 * know about. If there is an environment variable that sets
848 * that category, we won't know to look for it, and so our use
849 * of LANG or "C" improperly overrides it. On the other hand,
850 * if we don't do what is done here, and there is no
851 * environment variable, the category's locale should be set to
852 * LANG or "C". So there is no good solution. khw thinks the
853 * best is to look at systems to see what categories they have,
854 * and include them, and then to assume that we know the
857 for (i = 0; i < LC_ALL_INDEX; i++) {
858 const char * const env_override
859 = PerlEnv_getenv(category_names[i]);
860 const char * this_locale = ( env_override
861 && strNE(env_override, ""))
864 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
869 if (strNE(this_locale, default_name)) {
874 /* If all the categories are the same, we can set LC_ALL to
876 if (! did_override) {
877 locale = default_name;
881 /* Here, LC_ALL is no longer valid, as some individual
882 * categories don't match it. We call ourselves
883 * recursively, as that will execute the code that
884 * generates the proper locale string for this situation.
885 * We don't do the remainder of this function, as that is
886 * to update our records, and we've just done that for the
887 * individual categories in the loop above, and doing so
888 * would cause LC_ALL to be done as well */
889 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
893 } /* End of this being setlocale(LC_foo, "") */
894 else if (strchr(locale, ';')) {
896 /* LC_ALL may actually incude a conglomeration of various categories.
897 * Without querylocale, this code uses the glibc (as of this writing)
898 * syntax for representing that, but that is not a stable API, and
899 * other platforms do it differently, so we have to handle all cases
903 const char * s = locale;
904 const char * e = locale + strlen(locale);
906 const char * category_end;
907 const char * name_start;
908 const char * name_end;
910 /* If the string that gives what to set doesn't include all categories,
911 * the omitted ones get set to "C". To get this behavior, first set
912 * all the individual categories to "C", and override the furnished
914 for (i = 0; i < LC_ALL_INDEX; i++) {
915 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
922 /* Parse through the category */
923 while (isWORDCHAR(*p)) {
930 "panic: %s: %d: Unexpected character in locale name '%02X",
931 __FILE__, __LINE__, *(p-1));
934 /* Parse through the locale name */
936 while (p < e && *p != ';') {
939 "panic: %s: %d: Unexpected character in locale name '%02X",
940 __FILE__, __LINE__, *(p-1));
946 /* Space past the semi-colon */
951 /* Find the index of the category name in our lists */
952 for (i = 0; i < LC_ALL_INDEX; i++) {
953 char * individ_locale;
955 /* Keep going if this isn't the index. The strnNE() avoids a
956 * Perl_form(), but would fail if ever a category name could be
957 * a substring of another one, like if there were a
959 if strnNE(s, category_names[i], category_end - s) {
963 /* If this index is for the single category we're changing, we
964 * have found the locale to set it to. */
965 if (category == categories[i]) {
966 locale = Perl_form(aTHX_ "%.*s",
967 (int) (name_end - name_start),
972 assert(category == LC_ALL);
973 individ_locale = Perl_form(aTHX_ "%.*s",
974 (int) (name_end - name_start), name_start);
975 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
984 /* Here we have set all the individual categories by recursive calls.
985 * These collectively should have fixed up LC_ALL, so can just query
986 * what that now is */
987 assert(category == LC_ALL);
989 return do_setlocale_c(LC_ALL, NULL);
990 } /* End of this being setlocale(LC_ALL,
991 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
995 /* Here at the end of having to deal with the absence of querylocale().
996 * Some cases have already been fully handled by recursive calls to this
997 * function. But at this point, we haven't dealt with those, but are now
998 * prepared to, knowing what the locale name to set this category to is.
999 * This would have come for free if this system had had querylocale() */
1001 # endif /* end of ! querylocale */
1003 assert(PL_C_locale_obj);
1005 /* Switching locales generally entails freeing the current one's space (at
1006 * the C library's discretion). We need to stop using that locale before
1007 * the switch. So switch to a known locale object that we don't otherwise
1008 * mess with. This returns the locale object in effect at the time of the
1010 old_obj = uselocale(PL_C_locale_obj);
1014 if (DEBUG_Lv_TEST || debug_initialization) {
1015 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
1024 if (DEBUG_L_TEST || debug_initialization) {
1026 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1037 if (DEBUG_Lv_TEST || debug_initialization) {
1038 PerlIO_printf(Perl_debug_log,
1039 "%s:%d: emulate_setlocale now using %p\n",
1040 __FILE__, __LINE__, PL_C_locale_obj);
1045 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1046 * in fact, we already have switched to it (in preparation for what
1047 * normally is to come). But since we're already there, continue to use
1048 * it instead of trying to create a new locale */
1049 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1053 if (DEBUG_Lv_TEST || debug_initialization) {
1054 PerlIO_printf(Perl_debug_log,
1055 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1060 new_obj = PL_C_locale_obj;
1062 /* We already had switched to the C locale in preparation for freeing
1064 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1065 freelocale(old_obj);
1069 /* If we weren't in a thread safe locale, set so that newlocale() below
1070 * which uses 'old_obj', uses an empty one. Same for our reserved C
1071 * object. The latter is defensive coding, so that, even if there is
1072 * some bug, we will never end up trying to modify either of these, as
1073 * if passed to newlocale(), they can be. */
1074 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1075 old_obj = (locale_t) 0;
1078 /* Ready to create a new locale by modification of the exising one */
1079 new_obj = newlocale(mask, locale, old_obj);
1086 if (DEBUG_L_TEST || debug_initialization) {
1087 PerlIO_printf(Perl_debug_log,
1088 "%s:%d: emulate_setlocale creating new object"
1089 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1094 if (! uselocale(old_obj)) {
1098 if (DEBUG_L_TEST || debug_initialization) {
1099 PerlIO_printf(Perl_debug_log,
1100 "%s:%d: switching back failed: %d\n",
1101 __FILE__, __LINE__, GET_ERRNO);
1113 if (DEBUG_Lv_TEST || debug_initialization) {
1114 PerlIO_printf(Perl_debug_log,
1115 "%s:%d: emulate_setlocale created %p",
1116 __FILE__, __LINE__, new_obj);
1118 PerlIO_printf(Perl_debug_log,
1119 "; should have freed %p", old_obj);
1121 PerlIO_printf(Perl_debug_log, "\n");
1126 /* And switch into it */
1127 if (! uselocale(new_obj)) {
1132 if (DEBUG_L_TEST || debug_initialization) {
1133 PerlIO_printf(Perl_debug_log,
1134 "%s:%d: emulate_setlocale switching to new object"
1135 " failed\n", __FILE__, __LINE__);
1140 if (! uselocale(old_obj)) {
1144 if (DEBUG_L_TEST || debug_initialization) {
1145 PerlIO_printf(Perl_debug_log,
1146 "%s:%d: switching back failed: %d\n",
1147 __FILE__, __LINE__, GET_ERRNO);
1153 freelocale(new_obj);
1161 if (DEBUG_Lv_TEST || debug_initialization) {
1162 PerlIO_printf(Perl_debug_log,
1163 "%s:%d: emulate_setlocale now using %p\n",
1164 __FILE__, __LINE__, new_obj);
1169 /* We are done, except for updating our records (if the system doesn't keep
1170 * them) and in the case of locale "", we don't actually know what the
1171 * locale that got switched to is, as it came from the environment. So
1172 * have to find it */
1174 # ifdef HAS_QUERYLOCALE
1176 if (strEQ(locale, "")) {
1177 locale = querylocale(mask, new_obj);
1182 /* Here, 'locale' is the return value */
1184 /* Without querylocale(), we have to update our records */
1186 if (category == LC_ALL) {
1189 /* For LC_ALL, we change all individual categories to correspond */
1190 /* PL_curlocales is a parallel array, so has same
1191 * length as 'categories' */
1192 for (i = 0; i <= LC_ALL_INDEX; i++) {
1193 Safefree(PL_curlocales[i]);
1194 PL_curlocales[i] = savepv(locale);
1197 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
1201 /* For a single category, if it's not the same as the one in LC_ALL, we
1204 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1205 Safefree(PL_curlocales[LC_ALL_INDEX]);
1206 PL_curlocales[LC_ALL_INDEX] = NULL;
1209 /* Then update the category's record */
1210 Safefree(PL_curlocales[index]);
1211 PL_curlocales[index] = savepv(locale);
1213 FIX_GLIBC_LC_MESSAGES_BUG(index);
1221 #endif /* USE_POSIX_2008_LOCALE */
1223 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1224 didn't natively support them */
1226 /* The way this would work is that we would keep a per-thread list of the
1227 * correct locale for that thread. Any operation that was locale-sensitive
1228 * would have to be changed so that it would look like this:
1231 * setlocale to the correct locale for this operation
1235 * This leaves the global locale in the most recently used operation's, but it
1236 * was locked long enough to get the result. If that result is static, it
1237 * needs to be copied before the unlock.
1239 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1240 * the setup, but are no-ops when not needed, and similarly,
1241 * END_LOCALE_DEPENDENT_OP for the tear-down
1243 * But every call to a locale-sensitive function would have to be changed, and
1244 * if a module didn't cooperate by using the mutex, things would break.
1246 * This code was abandoned before being completed or tested, and is left as-is
1249 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1250 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1253 S_locking_setlocale(pTHX_
1255 const char * locale,
1257 const bool is_index_valid
1260 /* This function kind of performs a setlocale() on just the current thread;
1261 * thus it is kind of thread-safe. It does this by keeping a thread-level
1262 * array of the current locales for each category. Every time a locale is
1263 * switched to, it does the switch globally, but updates the thread's
1264 * array. A query as to what the current locale is just returns the
1265 * appropriate element from the array, and doesn't actually call the system
1266 * setlocale(). The saving into the array is done in an uninterruptible
1267 * section of code, so is unaffected by whatever any other threads might be
1270 * All locale-sensitive operations must work by first starting a critical
1271 * section, then switching to the thread's locale as kept by this function,
1272 * and then doing the operation, then ending the critical section. Thus,
1273 * each gets done in the appropriate locale. simulating thread-safety.
1275 * This function takes the same parameters, 'category' and 'locale', that
1276 * the regular setlocale() function does, but it also takes two additional
1277 * ones. This is because as described earlier. If we know on input the
1278 * index corresponding to the category into the array where we store the
1279 * current locales, we don't have to calculate it. If the caller knows at
1280 * compile time what the index is, it can pass it, setting
1281 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1285 /* If the input index might be incorrect, calculate the correct one */
1286 if (! is_index_valid) {
1289 if (DEBUG_Lv_TEST || debug_initialization) {
1290 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1293 for (i = 0; i <= LC_ALL_INDEX; i++) {
1294 if (category == categories[i]) {
1300 /* Here, we don't know about this category, so can't handle it.
1301 * XXX best we can do is to unsafely set this
1304 return my_setlocale(category, locale);
1308 if (DEBUG_Lv_TEST || debug_initialization) {
1309 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1313 /* For a query, just return what's in our records */
1314 if (new_locale == NULL) {
1315 return curlocales[index];
1319 /* Otherwise, we need to do the switch, and save the result, all in a
1320 * critical section */
1322 Safefree(curlocales[[index]]);
1324 /* It might be that this is called from an already-locked section of code.
1325 * We would have to detect and skip the LOCK/UNLOCK if so */
1328 curlocales[index] = savepv(my_setlocale(category, new_locale));
1330 if (strEQ(new_locale, "")) {
1334 /* The locale values come from the environment, and may not all be the
1335 * same, so for LC_ALL, we have to update all the others, while the
1336 * mutex is still locked */
1338 if (category == LC_ALL) {
1340 for (i = 0; i < LC_ALL_INDEX) {
1341 curlocales[i] = my_setlocale(categories[i], NULL);
1350 return curlocales[index];
1357 S_set_numeric_radix(pTHX_ const bool use_locale)
1359 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1360 * TRUE, use the radix character derived from the current locale */
1362 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1363 || defined(HAS_NL_LANGINFO))
1365 const char * radix = (use_locale)
1366 ? my_nl_langinfo(RADIXCHAR, FALSE)
1367 /* FALSE => already in dest locale */
1370 sv_setpv(PL_numeric_radix_sv, radix);
1372 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1373 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1374 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1375 SvCUR(PL_numeric_radix_sv))
1376 && _is_cur_LC_category_utf8(LC_NUMERIC))
1378 SvUTF8_on(PL_numeric_radix_sv);
1383 if (DEBUG_L_TEST || debug_initialization) {
1384 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1385 SvPVX(PL_numeric_radix_sv),
1386 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1392 PERL_UNUSED_ARG(use_locale);
1394 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1399 S_new_numeric(pTHX_ const char *newnum)
1402 #ifndef USE_LOCALE_NUMERIC
1404 PERL_UNUSED_ARG(newnum);
1408 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1409 * core Perl this and that 'newnum' is the name of the new locale.
1410 * It installs this locale as the current underlying default.
1412 * The default locale and the C locale can be toggled between by use of the
1413 * set_numeric_underlying() and set_numeric_standard() functions, which
1414 * should probably not be called directly, but only via macros like
1415 * SET_NUMERIC_STANDARD() in perl.h.
1417 * The toggling is necessary mainly so that a non-dot radix decimal point
1418 * character can be output, while allowing internal calculations to use a
1421 * This sets several interpreter-level variables:
1422 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1423 * PL_numeric_underlying A boolean indicating if the toggled state is such
1424 * that the current locale is the program's underlying
1426 * PL_numeric_standard An int indicating if the toggled state is such
1427 * that the current locale is the C locale or
1428 * indistinguishable from the C locale. If non-zero, it
1429 * is in C; if > 1, it means it may not be toggled away
1431 * PL_numeric_underlying_is_standard A bool kept by this function
1432 * indicating that the underlying locale and the standard
1433 * C locale are indistinguishable for the purposes of
1434 * LC_NUMERIC. This happens when both of the above two
1435 * variables are true at the same time. (Toggling is a
1436 * no-op under these circumstances.) This variable is
1437 * used to avoid having to recalculate.
1443 Safefree(PL_numeric_name);
1444 PL_numeric_name = NULL;
1445 PL_numeric_standard = TRUE;
1446 PL_numeric_underlying = TRUE;
1447 PL_numeric_underlying_is_standard = TRUE;
1451 save_newnum = stdize_locale(savepv(newnum));
1452 PL_numeric_underlying = TRUE;
1453 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1455 #ifndef TS_W32_BROKEN_LOCALECONV
1457 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1458 * them. But on broken Windows systems calling my_nl_langinfo() for
1459 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1460 * and just always change the locale if not C nor POSIX on those systems */
1461 if (! PL_numeric_standard) {
1462 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1463 FALSE /* Don't toggle locale */ ))
1464 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1469 /* Save the new name if it isn't the same as the previous one, if any */
1470 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1471 Safefree(PL_numeric_name);
1472 PL_numeric_name = save_newnum;
1475 Safefree(save_newnum);
1478 PL_numeric_underlying_is_standard = PL_numeric_standard;
1480 # ifdef HAS_POSIX_2008_LOCALE
1482 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1484 PL_underlying_numeric_obj);
1488 if (DEBUG_L_TEST || debug_initialization) {
1489 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1492 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1493 * have to worry about the radix being a non-dot. (Core operations that
1494 * need the underlying locale change to it temporarily). */
1495 if (PL_numeric_standard) {
1496 set_numeric_radix(0);
1499 set_numeric_standard();
1502 #endif /* USE_LOCALE_NUMERIC */
1507 Perl_set_numeric_standard(pTHX)
1510 #ifdef USE_LOCALE_NUMERIC
1512 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1513 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1514 * macro avoids calling this routine if toggling isn't necessary according
1515 * to our records (which could be wrong if some XS code has changed the
1516 * locale behind our back) */
1520 if (DEBUG_L_TEST || debug_initialization) {
1521 PerlIO_printf(Perl_debug_log,
1522 "Setting LC_NUMERIC locale to standard C\n");
1527 do_setlocale_c(LC_NUMERIC, "C");
1528 PL_numeric_standard = TRUE;
1529 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1530 set_numeric_radix(0);
1532 #endif /* USE_LOCALE_NUMERIC */
1537 Perl_set_numeric_underlying(pTHX)
1540 #ifdef USE_LOCALE_NUMERIC
1542 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1543 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1544 * instead of calling this directly. The macro avoids calling this routine
1545 * if toggling isn't necessary according to our records (which could be
1546 * wrong if some XS code has changed the locale behind our back) */
1550 if (DEBUG_L_TEST || debug_initialization) {
1551 PerlIO_printf(Perl_debug_log,
1552 "Setting LC_NUMERIC locale to %s\n",
1558 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1559 PL_numeric_standard = PL_numeric_underlying_is_standard;
1560 PL_numeric_underlying = TRUE;
1561 set_numeric_radix(! PL_numeric_standard);
1563 #endif /* USE_LOCALE_NUMERIC */
1568 * Set up for a new ctype locale.
1571 S_new_ctype(pTHX_ const char *newctype)
1574 #ifndef USE_LOCALE_CTYPE
1576 PERL_UNUSED_ARG(newctype);
1577 PERL_UNUSED_CONTEXT;
1581 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1582 * core Perl this and that 'newctype' is the name of the new locale.
1584 * This function sets up the folding arrays for all 256 bytes, assuming
1585 * that tofold() is tolc() since fold case is not a concept in POSIX,
1587 * Any code changing the locale (outside this file) should use
1588 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1589 * this function should be called directly only from this file and from
1590 * POSIX::setlocale() */
1594 /* Don't check for problems if we are suppressing the warnings */
1595 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1596 bool maybe_utf8_turkic = FALSE;
1598 PERL_ARGS_ASSERT_NEW_CTYPE;
1600 /* We will replace any bad locale warning with 1) nothing if the new one is
1601 * ok; or 2) a new warning for the bad new locale */
1602 if (PL_warn_locale) {
1603 SvREFCNT_dec_NN(PL_warn_locale);
1604 PL_warn_locale = NULL;
1607 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1609 /* A UTF-8 locale gets standard rules. But note that code still has to
1610 * handle this specially because of the three problematic code points */
1611 if (PL_in_utf8_CTYPE_locale) {
1612 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1614 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1615 * Turkic. Make sure these two are the only anomalies. (We don't use
1616 * towupper and towlower because they aren't in C89.) */
1618 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1620 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1624 if (toupper('i') == 'i' && tolower('I') == 'I') {
1627 check_for_problems = TRUE;
1628 maybe_utf8_turkic = TRUE;
1632 /* We don't populate the other lists if a UTF-8 locale, but do check that
1633 * everything works as expected, unless checking turned off */
1634 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1635 /* Assume enough space for every character being bad. 4 spaces each
1636 * for the 94 printable characters that are output like "'x' "; and 5
1637 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1639 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1640 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1642 unsigned int bad_count = 0; /* Count of bad characters */
1644 for (i = 0; i < 256; i++) {
1645 if (! PL_in_utf8_CTYPE_locale) {
1647 PL_fold_locale[i] = (U8) tolower(i);
1648 else if (islower(i))
1649 PL_fold_locale[i] = (U8) toupper(i);
1651 PL_fold_locale[i] = (U8) i;
1654 /* If checking for locale problems, see if the native ASCII-range
1655 * printables plus \n and \t are in their expected categories in
1656 * the new locale. If not, this could mean big trouble, upending
1657 * Perl's and most programs' assumptions, like having a
1658 * metacharacter with special meaning become a \w. Fortunately,
1659 * it's very rare to find locales that aren't supersets of ASCII
1660 * nowadays. It isn't a problem for most controls to be changed
1661 * into something else; we check only \n and \t, though perhaps \r
1662 * could be an issue as well. */
1663 if ( check_for_problems
1664 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1666 bool is_bad = FALSE;
1667 char name[4] = { '\0' };
1669 /* Convert the name into a string */
1674 else if (i == '\n') {
1675 my_strlcpy(name, "\\n", sizeof(name));
1677 else if (i == '\t') {
1678 my_strlcpy(name, "\\t", sizeof(name));
1682 my_strlcpy(name, "' '", sizeof(name));
1685 /* Check each possibe class */
1686 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1688 DEBUG_L(PerlIO_printf(Perl_debug_log,
1689 "isalnum('%s') unexpectedly is %d\n",
1690 name, cBOOL(isalnum(i))));
1692 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1694 DEBUG_L(PerlIO_printf(Perl_debug_log,
1695 "isalpha('%s') unexpectedly is %d\n",
1696 name, cBOOL(isalpha(i))));
1698 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1700 DEBUG_L(PerlIO_printf(Perl_debug_log,
1701 "isdigit('%s') unexpectedly is %d\n",
1702 name, cBOOL(isdigit(i))));
1704 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1706 DEBUG_L(PerlIO_printf(Perl_debug_log,
1707 "isgraph('%s') unexpectedly is %d\n",
1708 name, cBOOL(isgraph(i))));
1710 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1712 DEBUG_L(PerlIO_printf(Perl_debug_log,
1713 "islower('%s') unexpectedly is %d\n",
1714 name, cBOOL(islower(i))));
1716 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1718 DEBUG_L(PerlIO_printf(Perl_debug_log,
1719 "isprint('%s') unexpectedly is %d\n",
1720 name, cBOOL(isprint(i))));
1722 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1724 DEBUG_L(PerlIO_printf(Perl_debug_log,
1725 "ispunct('%s') unexpectedly is %d\n",
1726 name, cBOOL(ispunct(i))));
1728 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1730 DEBUG_L(PerlIO_printf(Perl_debug_log,
1731 "isspace('%s') unexpectedly is %d\n",
1732 name, cBOOL(isspace(i))));
1734 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1736 DEBUG_L(PerlIO_printf(Perl_debug_log,
1737 "isupper('%s') unexpectedly is %d\n",
1738 name, cBOOL(isupper(i))));
1740 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1742 DEBUG_L(PerlIO_printf(Perl_debug_log,
1743 "isxdigit('%s') unexpectedly is %d\n",
1744 name, cBOOL(isxdigit(i))));
1746 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1748 DEBUG_L(PerlIO_printf(Perl_debug_log,
1749 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1750 name, tolower(i), (int) toLOWER_A(i)));
1752 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1754 DEBUG_L(PerlIO_printf(Perl_debug_log,
1755 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1756 name, toupper(i), (int) toUPPER_A(i)));
1758 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1760 DEBUG_L(PerlIO_printf(Perl_debug_log,
1761 "'\\n' (=%02X) is not a control\n", (int) i));
1764 /* Add to the list; Separate multiple entries with a blank */
1767 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1769 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1775 if (bad_count == 2 && maybe_utf8_turkic) {
1777 *bad_chars_list = '\0';
1778 PL_fold_locale['I'] = 'I';
1779 PL_fold_locale['i'] = 'i';
1780 PL_in_utf8_turkic_locale = TRUE;
1781 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1782 __FILE__, __LINE__, newctype));
1785 PL_in_utf8_turkic_locale = FALSE;
1790 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1791 * this locale requires more than one byte, there are going to be
1793 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1794 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1795 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1797 if ( check_for_problems && MB_CUR_MAX > 1
1798 && ! PL_in_utf8_CTYPE_locale
1800 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1801 * locale. Just assume that the implementation for them (plus
1802 * for POSIX) is correct and the > 1 value is spurious. (Since
1803 * these are specially handled to never be considered UTF-8
1804 * locales, as long as this is the only problem, everything
1805 * should work fine */
1806 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1808 multi_byte_locale = TRUE;
1813 /* If we found problems and we want them output, do so */
1814 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1815 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1817 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1818 PL_warn_locale = Perl_newSVpvf(aTHX_
1819 "Locale '%s' contains (at least) the following characters"
1820 " which have\nunexpected meanings: %s\nThe Perl program"
1821 " will use the expected meanings",
1822 newctype, bad_chars_list);
1825 PL_warn_locale = Perl_newSVpvf(aTHX_
1826 "Locale '%s' may not work well.%s%s%s\n",
1829 ? " Some characters in it are not recognized by"
1833 ? "\nThe following characters (and maybe others)"
1834 " may not have the same meaning as the Perl"
1835 " program expects:\n"
1843 # ifdef HAS_NL_LANGINFO
1845 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1846 /* parameter FALSE is a don't care here */
1847 my_nl_langinfo(CODESET, FALSE));
1851 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1853 /* If we are actually in the scope of the locale or are debugging,
1854 * output the message now. If not in that scope, we save the
1855 * message to be output at the first operation using this locale,
1856 * if that actually happens. Most programs don't use locales, so
1857 * they are immune to bad ones. */
1858 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1860 /* The '0' below suppresses a bogus gcc compiler warning */
1861 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1863 if (IN_LC(LC_CTYPE)) {
1864 SvREFCNT_dec_NN(PL_warn_locale);
1865 PL_warn_locale = NULL;
1871 #endif /* USE_LOCALE_CTYPE */
1876 Perl__warn_problematic_locale()
1879 #ifdef USE_LOCALE_CTYPE
1883 /* Internal-to-core function that outputs the message in PL_warn_locale,
1884 * and then NULLS it. Should be called only through the macro
1885 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1887 if (PL_warn_locale) {
1888 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1889 SvPVX(PL_warn_locale),
1890 0 /* dummy to avoid compiler warning */ );
1891 SvREFCNT_dec_NN(PL_warn_locale);
1892 PL_warn_locale = NULL;
1900 S_new_collate(pTHX_ const char *newcoll)
1903 #ifndef USE_LOCALE_COLLATE
1905 PERL_UNUSED_ARG(newcoll);
1906 PERL_UNUSED_CONTEXT;
1910 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1911 * core Perl this and that 'newcoll' is the name of the new locale.
1913 * The design of locale collation is that every locale change is given an
1914 * index 'PL_collation_ix'. The first time a string particpates in an
1915 * operation that requires collation while locale collation is active, it
1916 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1917 * magic includes the collation index, and the transformation of the string
1918 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1919 * instead of the string itself. If a string changes, the magic is
1920 * cleared. The next time the locale changes, the index is incremented,
1921 * and so we know during a comparison that the transformation is not
1922 * necessarily still valid, and so is recomputed. Note that if the locale
1923 * changes enough times, the index could wrap (a U32), and it is possible
1924 * that a transformation would improperly be considered valid, leading to
1925 * an unlikely bug */
1928 if (PL_collation_name) {
1930 Safefree(PL_collation_name);
1931 PL_collation_name = NULL;
1933 PL_collation_standard = TRUE;
1934 is_standard_collation:
1935 PL_collxfrm_base = 0;
1936 PL_collxfrm_mult = 2;
1937 PL_in_utf8_COLLATE_locale = FALSE;
1938 PL_strxfrm_NUL_replacement = '\0';
1939 PL_strxfrm_max_cp = 0;
1943 /* If this is not the same locale as currently, set the new one up */
1944 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1946 Safefree(PL_collation_name);
1947 PL_collation_name = stdize_locale(savepv(newcoll));
1948 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1949 if (PL_collation_standard) {
1950 goto is_standard_collation;
1953 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1954 PL_strxfrm_NUL_replacement = '\0';
1955 PL_strxfrm_max_cp = 0;
1957 /* A locale collation definition includes primary, secondary, tertiary,
1958 * etc. weights for each character. To sort, the primary weights are
1959 * used, and only if they compare equal, then the secondary weights are
1960 * used, and only if they compare equal, then the tertiary, etc.
1962 * strxfrm() works by taking the input string, say ABC, and creating an
1963 * output transformed string consisting of first the primary weights,
1964 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1965 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1966 * may not have weights at every level. In our example, let's say B
1967 * doesn't have a tertiary weight, and A doesn't have a secondary
1968 * weight. The constructed string is then going to be
1969 * A¹B¹C¹ B²C² A³C³ ....
1970 * This has the desired effect that strcmp() will look at the secondary
1971 * or tertiary weights only if the strings compare equal at all higher
1972 * priority weights. The spaces shown here, like in
1974 * are not just for readability. In the general case, these must
1975 * actually be bytes, which we will call here 'separator weights'; and
1976 * they must be smaller than any other weight value, but since these
1977 * are C strings, only the terminating one can be a NUL (some
1978 * implementations may include a non-NUL separator weight just before
1979 * the NUL). Implementations tend to reserve 01 for the separator
1980 * weights. They are needed so that a shorter string's secondary
1981 * weights won't be misconstrued as primary weights of a longer string,
1982 * etc. By making them smaller than any other weight, the shorter
1983 * string will sort first. (Actually, if all secondary weights are
1984 * smaller than all primary ones, there is no need for a separator
1985 * weight between those two levels, etc.)
1987 * The length of the transformed string is roughly a linear function of
1988 * the input string. It's not exactly linear because some characters
1989 * don't have weights at all levels. When we call strxfrm() we have to
1990 * allocate some memory to hold the transformed string. The
1991 * calculations below try to find coefficients 'm' and 'b' for this
1992 * locale so that m*x + b equals how much space we need, given the size
1993 * of the input string in 'x'. If we calculate too small, we increase
1994 * the size as needed, and call strxfrm() again, but it is better to
1995 * get it right the first time to avoid wasted expensive string
1996 * transformations. */
1999 /* We use the string below to find how long the tranformation of it
2000 * is. Almost all locales are supersets of ASCII, or at least the
2001 * ASCII letters. We use all of them, half upper half lower,
2002 * because if we used fewer, we might hit just the ones that are
2003 * outliers in a particular locale. Most of the strings being
2004 * collated will contain a preponderance of letters, and even if
2005 * they are above-ASCII, they are likely to have the same number of
2006 * weight levels as the ASCII ones. It turns out that digits tend
2007 * to have fewer levels, and some punctuation has more, but those
2008 * are relatively sparse in text, and khw believes this gives a
2009 * reasonable result, but it could be changed if experience so
2011 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
2012 char * x_longer; /* Transformed 'longer' */
2013 Size_t x_len_longer; /* Length of 'x_longer' */
2015 char * x_shorter; /* We also transform a substring of 'longer' */
2016 Size_t x_len_shorter;
2018 /* _mem_collxfrm() is used get the transformation (though here we
2019 * are interested only in its length). It is used because it has
2020 * the intelligence to handle all cases, but to work, it needs some
2021 * values of 'm' and 'b' to get it started. For the purposes of
2022 * this calculation we use a very conservative estimate of 'm' and
2023 * 'b'. This assumes a weight can be multiple bytes, enough to
2024 * hold any UV on the platform, and there are 5 levels, 4 weight
2025 * bytes, and a trailing NUL. */
2026 PL_collxfrm_base = 5;
2027 PL_collxfrm_mult = 5 * sizeof(UV);
2029 /* Find out how long the transformation really is */
2030 x_longer = _mem_collxfrm(longer,
2034 /* We avoid converting to UTF-8 in the
2035 * called function by telling it the
2036 * string is in UTF-8 if the locale is a
2037 * UTF-8 one. Since the string passed
2038 * here is invariant under UTF-8, we can
2039 * claim it's UTF-8 even though it isn't.
2041 PL_in_utf8_COLLATE_locale);
2044 /* Find out how long the transformation of a substring of 'longer'
2045 * is. Together the lengths of these transformations are
2046 * sufficient to calculate 'm' and 'b'. The substring is all of
2047 * 'longer' except the first character. This minimizes the chances
2048 * of being swayed by outliers */
2049 x_shorter = _mem_collxfrm(longer + 1,
2052 PL_in_utf8_COLLATE_locale);
2053 Safefree(x_shorter);
2055 /* If the results are nonsensical for this simple test, the whole
2056 * locale definition is suspect. Mark it so that locale collation
2057 * is not active at all for it. XXX Should we warn? */
2058 if ( x_len_shorter == 0
2059 || x_len_longer == 0
2060 || x_len_shorter >= x_len_longer)
2062 PL_collxfrm_mult = 0;
2063 PL_collxfrm_base = 0;
2066 SSize_t base; /* Temporary */
2068 /* We have both: m * strlen(longer) + b = x_len_longer
2069 * m * strlen(shorter) + b = x_len_shorter;
2070 * subtracting yields:
2071 * m * (strlen(longer) - strlen(shorter))
2072 * = x_len_longer - x_len_shorter
2073 * But we have set things up so that 'shorter' is 1 byte smaller
2074 * than 'longer'. Hence:
2075 * m = x_len_longer - x_len_shorter
2077 * But if something went wrong, make sure the multiplier is at
2080 if (x_len_longer > x_len_shorter) {
2081 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2084 PL_collxfrm_mult = 1;
2089 * but in case something has gone wrong, make sure it is
2091 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2096 /* Add 1 for the trailing NUL */
2097 PL_collxfrm_base = base + 1;
2102 if (DEBUG_L_TEST || debug_initialization) {
2103 PerlIO_printf(Perl_debug_log,
2104 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2106 " collate multipler=%zu, collate base=%zu\n",
2108 PL_in_utf8_COLLATE_locale,
2109 x_len_shorter, x_len_longer,
2110 PL_collxfrm_mult, PL_collxfrm_base);
2117 #endif /* USE_LOCALE_COLLATE */
2125 #define USE_WSETLOCALE
2127 #ifdef USE_WSETLOCALE
2130 S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2137 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2144 Newx(wlocale, req_size, wchar_t);
2145 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2154 wresult = _wsetlocale(category, wlocale);
2158 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2159 Newx(result, req_size, char);
2160 SAVEFREEPV(result); /* is there something better we can do here? */
2161 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2162 result, req_size, NULL, NULL)) {
2177 S_win32_setlocale(pTHX_ int category, const char* locale)
2179 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2180 * difference between the two unless the input locale is "", which normally
2181 * means on Windows to get the machine default, which is set via the
2182 * computer's "Regional and Language Options" (or its current equivalent).
2183 * In POSIX, it instead means to find the locale from the user's
2184 * environment. This routine changes the Windows behavior to first look in
2185 * the environment, and, if anything is found, use that instead of going to
2186 * the machine default. If there is no environment override, the machine
2187 * default is used, by calling the real setlocale() with "".
2189 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2190 * use the particular category's variable if set; otherwise to use the LANG
2193 bool override_LC_ALL = FALSE;
2197 if (locale && strEQ(locale, "")) {
2201 locale = PerlEnv_getenv("LC_ALL");
2203 if (category == LC_ALL) {
2204 override_LC_ALL = TRUE;
2210 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2211 if (category == categories[i]) {
2212 locale = PerlEnv_getenv(category_names[i]);
2217 locale = PerlEnv_getenv("LANG");
2233 #ifdef USE_WSETLOCALE
2234 result = S_wrap_wsetlocale(aTHX_ category, locale);
2236 result = setlocale(category, locale);
2238 DEBUG_L(STMT_START {
2240 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2241 setlocale_debug_string(category, locale, result));
2245 if (! override_LC_ALL) {
2249 /* Here the input category was LC_ALL, and we have set it to what is in the
2250 * LANG variable or the system default if there is no LANG. But these have
2251 * lower priority than the other LC_foo variables, so override it for each
2252 * one that is set. (If they are set to "", it means to use the same thing
2253 * we just set LC_ALL to, so can skip) */
2255 for (i = 0; i < LC_ALL_INDEX; i++) {
2256 result = PerlEnv_getenv(category_names[i]);
2257 if (result && strNE(result, "")) {
2258 #ifdef USE_WSETLOCALE
2259 S_wrap_wsetlocale(aTHX_ categories[i], result);
2261 setlocale(categories[i], result);
2263 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2265 setlocale_debug_string(categories[i], result, "not captured")));
2269 result = setlocale(LC_ALL, NULL);
2270 DEBUG_L(STMT_START {
2272 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2274 setlocale_debug_string(LC_ALL, NULL, result));
2284 =for apidoc Perl_setlocale
2286 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2287 taking the same parameters, and returning the same information, except that it
2288 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2289 instead return C<C> if the underlying locale has a non-dot decimal point
2290 character, or a non-empty thousands separator for displaying floating point
2291 numbers. This is because perl keeps that locale category such that it has a
2292 dot and empty separator, changing the locale briefly during the operations
2293 where the underlying one is required. C<Perl_setlocale> knows about this, and
2294 compensates; regular C<setlocale> doesn't.
2296 Another reason it isn't completely a drop-in replacement is that it is
2297 declared to return S<C<const char *>>, whereas the system setlocale omits the
2298 C<const> (presumably because its API was specified long ago, and can't be
2299 updated; it is illegal to change the information C<setlocale> returns; doing
2300 so leads to segfaults.)
2302 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2303 C<setlocale> can be completely ineffective on some platforms under some
2306 C<Perl_setlocale> should not be used to change the locale except on systems
2307 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2308 the system C<setlocale()> is ineffective, returning the wrong information, and
2309 failing to actually change the locale. C<Perl_setlocale>, however works
2310 properly in all circumstances.
2312 The return points to a per-thread static buffer, which is overwritten the next
2313 time C<Perl_setlocale> is called from the same thread.
2320 Perl_setlocale(const int category, const char * locale)
2322 /* This wraps POSIX::setlocale() */
2326 PERL_UNUSED_ARG(category);
2327 PERL_UNUSED_ARG(locale);
2333 const char * retval;
2334 const char * newlocale;
2337 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2339 #ifdef USE_LOCALE_NUMERIC
2341 /* A NULL locale means only query what the current one is. We have the
2342 * LC_NUMERIC name saved, because we are normally switched into the C
2343 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2344 * the correct results. All other categories don't require special
2346 if (locale == NULL) {
2347 if (category == LC_NUMERIC) {
2349 /* We don't have to copy this return value, as it is a per-thread
2350 * variable, and won't change until a future setlocale */
2351 return PL_numeric_name;
2356 else if (category == LC_ALL) {
2357 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2366 retval = save_to_buffer(do_setlocale_r(category, locale),
2367 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2370 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2372 if (locale == NULL && category == LC_ALL) {
2373 RESTORE_LC_NUMERIC();
2378 DEBUG_L(PerlIO_printf(Perl_debug_log,
2379 "%s:%d: %s\n", __FILE__, __LINE__,
2380 setlocale_debug_string(category, locale, retval)));
2388 /* If locale == NULL, we are just querying the state */
2389 if (locale == NULL) {
2393 /* Now that have switched locales, we have to update our records to
2398 #ifdef USE_LOCALE_CTYPE
2405 #ifdef USE_LOCALE_COLLATE
2408 new_collate(retval);
2412 #ifdef USE_LOCALE_NUMERIC
2415 new_numeric(retval);
2423 /* LC_ALL updates all the things we care about. The values may not
2424 * be the same as 'retval', as the locale "" may have set things
2427 # ifdef USE_LOCALE_CTYPE
2429 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2430 new_ctype(newlocale);
2431 Safefree(newlocale);
2433 # endif /* USE_LOCALE_CTYPE */
2434 # ifdef USE_LOCALE_COLLATE
2436 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2437 new_collate(newlocale);
2438 Safefree(newlocale);
2441 # ifdef USE_LOCALE_NUMERIC
2443 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2444 new_numeric(newlocale);
2445 Safefree(newlocale);
2447 # endif /* USE_LOCALE_NUMERIC */
2460 PERL_STATIC_INLINE const char *
2461 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2463 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2464 * growing it if necessary */
2468 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2474 string_size = strlen(string) + offset + 1;
2476 if (*buf_size == 0) {
2477 Newx(*buf, string_size, char);
2478 *buf_size = string_size;
2480 else if (string_size > *buf_size) {
2481 Renew(*buf, string_size, char);
2482 *buf_size = string_size;
2485 Copy(string, *buf + offset, string_size - offset, char);
2491 =for apidoc Perl_langinfo
2493 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2494 taking the same C<item> parameter values, and returning the same information.
2495 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2496 of Perl's locale handling from your code, and can be used on systems that lack
2497 a native C<nl_langinfo>.
2505 The reason it isn't quite a drop-in replacement is actually an advantage. The
2506 only difference is that it returns S<C<const char *>>, whereas plain
2507 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2508 forbidden to write into the buffer. By declaring this C<const>, the compiler
2509 enforces this restriction, so if it is violated, you know at compilation time,
2510 rather than getting segfaults at runtime.
2514 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2515 without you having to write extra code. The reason for the extra code would be
2516 because these are from the C<LC_NUMERIC> locale category, which is normally
2517 kept set by Perl so that the radix is a dot, and the separator is the empty
2518 string, no matter what the underlying locale is supposed to be, and so to get
2519 the expected results, you have to temporarily toggle into the underlying
2520 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2521 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2522 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2523 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2524 (decimal point) character to be a dot.)
2528 The system function it replaces can have its static return buffer trashed,
2529 not only by a subsequent call to that function, but by a C<freelocale>,
2530 C<setlocale>, or other locale change. The returned buffer of this function is
2531 not changed until the next call to it, so the buffer is never in a trashed
2536 Its return buffer is per-thread, so it also is never overwritten by a call to
2537 this function from another thread; unlike the function it replaces.
2541 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2542 as Windows, hence makes your code more portable. Of the fifty-some possible
2543 items specified by the POSIX 2008 standard,
2544 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2545 only one is completely unimplemented, though on non-Windows platforms, another
2546 significant one is also not implemented). It uses various techniques to
2547 recover the other items, including calling C<L<localeconv(3)>>, and
2548 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2549 available. Later C<strftime()> versions have additional capabilities; C<""> is
2550 returned for those not available on your system.
2552 It is important to note that when called with an item that is recovered by
2553 using C<localeconv>, the buffer from any previous explicit call to
2554 C<localeconv> will be overwritten. This means you must save that buffer's
2555 contents if you need to access them after a call to this function. (But note
2556 that you might not want to be using C<localeconv()> directly anyway, because of
2557 issues like the ones listed in the second item of this list (above) for
2558 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2559 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2562 The details for those items which may deviate from what this emulation returns
2563 and what a native C<nl_langinfo()> would return are specified in
2568 When using C<Perl_langinfo> on systems that don't have a native
2569 C<nl_langinfo()>, you must
2571 #include "perl_langinfo.h"
2573 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2574 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2575 C<langinfo.h> would try to import into the namespace for code that doesn't need
2578 The original impetus for C<Perl_langinfo()> was so that code that needs to
2579 find out the current currency symbol, floating point radix character, or digit
2580 grouping separator can use, on all systems, the simpler and more
2581 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2582 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2583 is better to use the methods given in L<perlcall> to call
2584 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2591 #ifdef HAS_NL_LANGINFO
2592 Perl_langinfo(const nl_item item)
2594 Perl_langinfo(const int item)
2597 return my_nl_langinfo(item, TRUE);
2601 #ifdef HAS_NL_LANGINFO
2602 S_my_nl_langinfo(const nl_item item, bool toggle)
2604 S_my_nl_langinfo(const int item, bool toggle)
2608 const char * retval;
2610 #ifdef USE_LOCALE_NUMERIC
2612 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2613 * two items, and only if not already there */
2614 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2615 || PL_numeric_underlying))
2617 #endif /* No toggling needed if not using LC_NUMERIC */
2621 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2622 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2623 || ! defined(HAS_POSIX_2008_LOCALE) \
2624 || ! defined(DUPLOCALE)
2626 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2627 * for those items dependent on it. This must be copied to a buffer before
2628 * switching back, as some systems destroy the buffer when setlocale() is
2632 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2635 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2638 LOCALE_LOCK; /* Prevent interference from another thread executing
2639 this code section (the only call to nl_langinfo in
2643 /* Copy to a per-thread buffer, which is also one that won't be
2644 * destroyed by a subsequent setlocale(), such as the
2645 * RESTORE_LC_NUMERIC may do just below. */
2646 retval = save_to_buffer(nl_langinfo(item),
2647 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2652 RESTORE_LC_NUMERIC();
2656 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2659 bool do_free = FALSE;
2660 locale_t cur = uselocale((locale_t) 0);
2662 if (cur == LC_GLOBAL_LOCALE) {
2663 cur = duplocale(LC_GLOBAL_LOCALE);
2667 # ifdef USE_LOCALE_NUMERIC
2670 if (PL_underlying_numeric_obj) {
2671 cur = PL_underlying_numeric_obj;
2674 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2681 /* We have to save it to a buffer, because the freelocale() just below
2682 * can invalidate the internal one */
2683 retval = save_to_buffer(nl_langinfo_l(item, cur),
2684 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2693 if (strEQ(retval, "")) {
2694 if (item == YESSTR) {
2697 if (item == NOSTR) {
2704 #else /* Below, emulate nl_langinfo as best we can */
2708 # ifdef HAS_LOCALECONV
2710 const struct lconv* lc;
2712 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2714 # ifdef TS_W32_BROKEN_LOCALECONV
2716 const char * save_global;
2717 const char * save_thread;
2725 # ifdef HAS_STRFTIME
2728 bool return_format = FALSE; /* Return the %format, not the value */
2729 const char * format;
2733 /* We copy the results to a per-thread buffer, even if not
2734 * multi-threaded. This is in part to simplify this code, and partly
2735 * because we need a buffer anyway for strftime(), and partly because a
2736 * call of localeconv() could otherwise wipe out the buffer, and the
2737 * programmer would not be expecting this, as this is a nl_langinfo()
2738 * substitute after all, so s/he might be thinking their localeconv()
2739 * is safe until another localeconv() call. */
2744 /* This is unimplemented */
2745 case ERA: /* For use with strftime() %E modifier */
2750 /* We use only an English set, since we don't know any more */
2751 case YESEXPR: return "^[+1yY]";
2752 case YESSTR: return "yes";
2753 case NOEXPR: return "^[-0nN]";
2754 case NOSTR: return "no";
2760 /* On non-windows, this is unimplemented, in part because of
2761 * inconsistencies between vendors. The Darwin native
2762 * nl_langinfo() implementation simply looks at everything past
2763 * any dot in the name, but that doesn't work for other
2764 * vendors. Many Linux locales that don't have UTF-8 in their
2765 * names really are UTF-8, for example; z/OS locales that do
2766 * have UTF-8 in their names, aren't really UTF-8 */
2771 { /* But on Windows, the name does seem to be consistent, so
2776 const char * name = my_setlocale(LC_CTYPE, NULL);
2778 if (isNAME_C_OR_POSIX(name)) {
2779 return "ANSI_X3.4-1968";
2782 /* Find the dot in the locale name */
2783 first = (const char *) strchr(name, '.');
2789 /* Look at everything past the dot */
2794 if (! isDIGIT(*p)) {
2801 /* Here everything past the dot is a digit. Treat it as a
2803 retval = save_to_buffer("CP", &PL_langinfo_buf,
2804 &PL_langinfo_bufsize, 0);
2805 offset = STRLENs("CP");
2809 retval = save_to_buffer(first, &PL_langinfo_buf,
2810 &PL_langinfo_bufsize, offset);
2816 # ifdef HAS_LOCALECONV
2820 /* We don't bother with localeconv_l() because any system that
2821 * has it is likely to also have nl_langinfo() */
2823 LOCALE_LOCK_V; /* Prevent interference with other threads
2824 using localeconv() */
2826 # ifdef TS_W32_BROKEN_LOCALECONV
2828 /* This is a workaround for a Windows bug prior to VS 15.
2829 * What we do here is, while locked, switch to the global
2830 * locale so localeconv() works; then switch back just before
2831 * the unlock. This can screw things up if some thread is
2832 * already using the global locale while assuming no other is.
2833 * A different workaround would be to call GetCurrencyFormat on
2834 * a known value, and parse it; patches welcome
2836 * We have to use LC_ALL instead of LC_MONETARY because of
2837 * another bug in Windows */
2839 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2840 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2841 save_global= savepv(my_setlocale(LC_ALL, NULL));
2842 my_setlocale(LC_ALL, save_thread);
2848 || ! lc->currency_symbol
2849 || strEQ("", lc->currency_symbol))
2855 /* Leave the first spot empty to be filled in below */
2856 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2857 &PL_langinfo_bufsize, 1);
2858 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2859 { /* khw couldn't figure out how the localedef specifications
2860 would show that the $ should replace the radix; this is
2861 just a guess as to how it might work.*/
2862 PL_langinfo_buf[0] = '.';
2864 else if (lc->p_cs_precedes) {
2865 PL_langinfo_buf[0] = '-';
2868 PL_langinfo_buf[0] = '+';
2871 # ifdef TS_W32_BROKEN_LOCALECONV
2873 my_setlocale(LC_ALL, save_global);
2874 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2875 my_setlocale(LC_ALL, save_thread);
2876 Safefree(save_global);
2877 Safefree(save_thread);
2884 # ifdef TS_W32_BROKEN_LOCALECONV
2888 /* For this, we output a known simple floating point number to
2889 * a buffer, and parse it, looking for the radix */
2892 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2895 if (PL_langinfo_bufsize < 10) {
2896 PL_langinfo_bufsize = 10;
2897 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2900 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2902 if (needed_size >= (int) PL_langinfo_bufsize) {
2903 PL_langinfo_bufsize = needed_size + 1;
2904 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2905 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2907 assert(needed_size < (int) PL_langinfo_bufsize);
2910 ptr = PL_langinfo_buf;
2911 e = PL_langinfo_buf + PL_langinfo_bufsize;
2914 while (ptr < e && *ptr != '1') {
2921 while (ptr < e && *ptr != '5') {
2925 /* Everything in between is the radix string */
2927 PL_langinfo_buf[0] = '?';
2928 PL_langinfo_buf[1] = '\0';
2932 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2936 RESTORE_LC_NUMERIC();
2939 retval = PL_langinfo_buf;
2944 case RADIXCHAR: /* No special handling needed */
2951 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2954 LOCALE_LOCK_V; /* Prevent interference with other threads
2955 using localeconv() */
2957 # ifdef TS_W32_BROKEN_LOCALECONV
2959 /* This should only be for the thousands separator. A
2960 * different work around would be to use GetNumberFormat on a
2961 * known value and parse the result to find the separator */
2962 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2963 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2964 save_global = savepv(my_setlocale(LC_ALL, NULL));
2965 my_setlocale(LC_ALL, save_thread);
2967 /* This is the start of code that for broken Windows replaces
2968 * the above and below code, and instead calls
2969 * GetNumberFormat() and then would parse that to find the
2970 * thousands separator. It needs to handle UTF-16 vs -8
2973 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2974 DEBUG_L(PerlIO_printf(Perl_debug_log,
2975 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2976 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2986 temp = (item == RADIXCHAR)
2988 : lc->thousands_sep;
2994 retval = save_to_buffer(temp, &PL_langinfo_buf,
2995 &PL_langinfo_bufsize, 0);
2997 # ifdef TS_W32_BROKEN_LOCALECONV
2999 my_setlocale(LC_ALL, save_global);
3000 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3001 my_setlocale(LC_ALL, save_thread);
3002 Safefree(save_global);
3003 Safefree(save_thread);
3010 RESTORE_LC_NUMERIC();
3016 # ifdef HAS_STRFTIME
3018 /* These are defined by C89, so we assume that strftime supports
3019 * them, and so are returned unconditionally; they may not be what
3020 * the locale actually says, but should give good enough results
3021 * for someone using them as formats (as opposed to trying to parse
3022 * them to figure out what the locale says). The other format
3023 * items are actually tested to verify they work on the platform */
3024 case D_FMT: return "%x";
3025 case T_FMT: return "%X";
3026 case D_T_FMT: return "%c";
3028 /* These formats are only available in later strfmtime's */
3029 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
3031 /* The rest can be gotten from most versions of strftime(). */
3032 case ABDAY_1: case ABDAY_2: case ABDAY_3:
3033 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
3035 case AM_STR: case PM_STR:
3036 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
3037 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
3038 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
3039 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
3040 case DAY_5: case DAY_6: case DAY_7:
3041 case MON_1: case MON_2: case MON_3: case MON_4:
3042 case MON_5: case MON_6: case MON_7: case MON_8:
3043 case MON_9: case MON_10: case MON_11: case MON_12:
3047 init_tm(&tm); /* Precaution against core dumps */
3051 tm.tm_year = 2017 - 1900;
3058 "panic: %s: %d: switch case: %d problem",
3059 __FILE__, __LINE__, item);
3060 NOT_REACHED; /* NOTREACHED */
3062 case PM_STR: tm.tm_hour = 18;
3067 case ABDAY_7: tm.tm_wday++;
3068 case ABDAY_6: tm.tm_wday++;
3069 case ABDAY_5: tm.tm_wday++;
3070 case ABDAY_4: tm.tm_wday++;
3071 case ABDAY_3: tm.tm_wday++;
3072 case ABDAY_2: tm.tm_wday++;
3077 case DAY_7: tm.tm_wday++;
3078 case DAY_6: tm.tm_wday++;
3079 case DAY_5: tm.tm_wday++;
3080 case DAY_4: tm.tm_wday++;
3081 case DAY_3: tm.tm_wday++;
3082 case DAY_2: tm.tm_wday++;
3087 case ABMON_12: tm.tm_mon++;
3088 case ABMON_11: tm.tm_mon++;
3089 case ABMON_10: tm.tm_mon++;
3090 case ABMON_9: tm.tm_mon++;
3091 case ABMON_8: tm.tm_mon++;
3092 case ABMON_7: tm.tm_mon++;
3093 case ABMON_6: tm.tm_mon++;
3094 case ABMON_5: tm.tm_mon++;
3095 case ABMON_4: tm.tm_mon++;
3096 case ABMON_3: tm.tm_mon++;
3097 case ABMON_2: tm.tm_mon++;
3102 case MON_12: tm.tm_mon++;
3103 case MON_11: tm.tm_mon++;
3104 case MON_10: tm.tm_mon++;
3105 case MON_9: tm.tm_mon++;
3106 case MON_8: tm.tm_mon++;
3107 case MON_7: tm.tm_mon++;
3108 case MON_6: tm.tm_mon++;
3109 case MON_5: tm.tm_mon++;
3110 case MON_4: tm.tm_mon++;
3111 case MON_3: tm.tm_mon++;
3112 case MON_2: tm.tm_mon++;
3119 return_format = TRUE;
3124 return_format = TRUE;
3129 return_format = TRUE;
3134 return_format = TRUE;
3139 format = "%Ow"; /* Find the alternate digit for 0 */
3143 /* We can't use my_strftime() because it doesn't look at
3145 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3148 /* A zero return means one of:
3149 * a) there wasn't enough space in PL_langinfo_buf
3150 * b) the format, like a plain %p, returns empty
3151 * c) it was an illegal format, though some
3152 * implementations of strftime will just return the
3153 * illegal format as a plain character sequence.
3155 * To quickly test for case 'b)', try again but precede
3156 * the format with a plain character. If that result is
3157 * still empty, the problem is either 'a)' or 'c)' */
3159 Size_t format_size = strlen(format) + 1;
3160 Size_t mod_size = format_size + 1;
3164 Newx(mod_format, mod_size, char);
3165 Newx(temp_result, PL_langinfo_bufsize, char);
3167 my_strlcpy(mod_format + 1, format, mod_size);
3168 len = strftime(temp_result,
3169 PL_langinfo_bufsize,
3171 Safefree(mod_format);
3172 Safefree(temp_result);
3174 /* If 'len' is non-zero, it means that we had a case like
3175 * %p which means the current locale doesn't use a.m. or
3176 * p.m., and that is valid */
3179 /* Here, still didn't work. If we get well beyond a
3180 * reasonable size, bail out to prevent an infinite
3183 if (PL_langinfo_bufsize > 100 * format_size) {
3184 *PL_langinfo_buf = '\0';
3187 /* Double the buffer size to retry; Add 1 in case
3188 * original was 0, so we aren't stuck at 0. */
3189 PL_langinfo_bufsize *= 2;
3190 PL_langinfo_bufsize++;
3191 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3199 /* Here, we got a result.
3201 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3202 * alternate format for wday 0. If the value is the same as
3203 * the normal 0, there isn't an alternate, so clear the buffer.
3205 if ( item == ALT_DIGITS
3206 && strEQ(PL_langinfo_buf, "0"))
3208 *PL_langinfo_buf = '\0';
3211 /* ALT_DIGITS is problematic. Experiments on it showed that
3212 * strftime() did not always work properly when going from
3213 * alt-9 to alt-10. Only a few locales have this item defined,
3214 * and in all of them on Linux that khw was able to find,
3215 * nl_langinfo() merely returned the alt-0 character, possibly
3216 * doubled. Most Unicode digits are in blocks of 10
3217 * consecutive code points, so that is sufficient information
3218 * for those scripts, as we can infer alt-1, alt-2, .... But
3219 * for a Japanese locale, a CJK ideographic 0 is returned, and
3220 * the CJK digits are not in code point order, so you can't
3221 * really infer anything. The localedef for this locale did
3222 * specify the succeeding digits, so that strftime() works
3223 * properly on them, without needing to infer anything. But
3224 * the nl_langinfo() return did not give sufficient information
3225 * for the caller to understand what's going on. So until
3226 * there is evidence that it should work differently, this
3227 * returns the alt-0 string for ALT_DIGITS.
3229 * wday was chosen because its range is all a single digit.
3230 * Things like tm_sec have two digits as the minimum: '00' */
3234 retval = PL_langinfo_buf;
3236 /* If to return the format, not the value, overwrite the buffer
3237 * with it. But some strftime()s will keep the original format
3238 * if illegal, so change those to "" */
3239 if (return_format) {
3240 if (strEQ(PL_langinfo_buf, format)) {
3241 *PL_langinfo_buf = '\0';
3244 retval = save_to_buffer(format, &PL_langinfo_buf,
3245 &PL_langinfo_bufsize, 0);
3263 * Initialize locale awareness.
3266 Perl_init_i18nl10n(pTHX_ int printwarn)
3270 * 0 if not to output warning when setup locale is bad
3271 * 1 if to output warning based on value of PERL_BADLANG
3272 * >1 if to output regardless of PERL_BADLANG
3275 * 1 = set ok or not applicable,
3276 * 0 = fallback to a locale of lower priority
3277 * -1 = fallback to all locales failed, not even to the C locale
3279 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3280 * set, debugging information is output.
3282 * This looks more complicated than it is, mainly due to the #ifdefs.
3284 * We try to set LC_ALL to the value determined by the environment. If
3285 * there is no LC_ALL on this platform, we try the individual categories we
3286 * know about. If this works, we are done.
3288 * But if it doesn't work, we have to do something else. We search the
3289 * environment variables ourselves instead of relying on the system to do
3290 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3291 * think there is one), and the ultimate fallback "C". This is all done in
3292 * the same loop as above to avoid duplicating code, but it makes things
3293 * more complex. The 'trial_locales' array is initialized with just one
3294 * element; it causes the behavior described in the paragraph above this to
3295 * happen. If that fails, we add elements to 'trial_locales', and do extra
3296 * loop iterations to cause the behavior described in this paragraph.
3298 * On Ultrix, the locale MUST come from the environment, so there is
3299 * preliminary code to set it. I (khw) am not sure that it is necessary,
3300 * and that this couldn't be folded into the loop, but barring any real
3301 * platforms to test on, it's staying as-is
3303 * A slight complication is that in embedded Perls, the locale may already
3304 * be set-up, and we don't want to get it from the normal environment
3305 * variables. This is handled by having a special environment variable
3306 * indicate we're in this situation. We simply set setlocale's 2nd
3307 * parameter to be a NULL instead of "". That indicates to setlocale that
3308 * it is not to change anything, but to return the current value,
3309 * effectively initializing perl's db to what the locale already is.
3311 * We play the same trick with NULL if a LC_ALL succeeds. We call
3312 * setlocale() on the individual categores with NULL to get their existing
3313 * values for our db, instead of trying to change them.
3321 PERL_UNUSED_ARG(printwarn);
3323 #else /* USE_LOCALE */
3326 const char * const language = PerlEnv_getenv("LANGUAGE");
3330 /* NULL uses the existing already set up locale */
3331 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3334 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3335 unsigned int trial_locales_count;
3336 const char * const lc_all = PerlEnv_getenv("LC_ALL");
3337 const char * const lang = PerlEnv_getenv("LANG");
3338 bool setlocale_failure = FALSE;
3341 /* A later getenv() could zap this, so only use here */
3342 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3344 const bool locwarn = (printwarn > 1
3346 && ( ! bad_lang_use_once
3348 /* disallow with "" or "0" */
3350 && strNE("0", bad_lang_use_once)))));
3352 /* setlocale() return vals; not copied so must be looked at immediately */
3353 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3355 /* current locale for given category; should have been copied so aren't
3357 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3361 /* In some systems you can find out the system default locale
3362 * and use that as the fallback locale. */
3363 # define SYSTEM_DEFAULT_LOCALE
3365 # ifdef SYSTEM_DEFAULT_LOCALE
3367 const char *system_default_locale = NULL;
3372 # define DEBUG_LOCALE_INIT(a,b,c)
3375 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3377 # define DEBUG_LOCALE_INIT(category, locale, result) \
3379 if (debug_initialization) { \
3380 PerlIO_printf(Perl_debug_log, \
3382 __FILE__, __LINE__, \
3383 setlocale_debug_string(category, \
3389 /* Make sure the parallel arrays are properly set up */
3390 # ifdef USE_LOCALE_NUMERIC
3391 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3392 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3393 # ifdef USE_POSIX_2008_LOCALE
3394 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3397 # ifdef USE_LOCALE_CTYPE
3398 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3399 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3400 # ifdef USE_POSIX_2008_LOCALE
3401 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3404 # ifdef USE_LOCALE_COLLATE
3405 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3406 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3407 # ifdef USE_POSIX_2008_LOCALE
3408 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3411 # ifdef USE_LOCALE_TIME
3412 assert(categories[LC_TIME_INDEX] == LC_TIME);
3413 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3414 # ifdef USE_POSIX_2008_LOCALE
3415 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3418 # ifdef USE_LOCALE_MESSAGES
3419 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3420 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3421 # ifdef USE_POSIX_2008_LOCALE
3422 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3425 # ifdef USE_LOCALE_MONETARY
3426 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3427 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3428 # ifdef USE_POSIX_2008_LOCALE
3429 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3432 # ifdef USE_LOCALE_ADDRESS
3433 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3434 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3435 # ifdef USE_POSIX_2008_LOCALE
3436 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3439 # ifdef USE_LOCALE_IDENTIFICATION
3440 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3441 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3442 # ifdef USE_POSIX_2008_LOCALE
3443 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3446 # ifdef USE_LOCALE_MEASUREMENT
3447 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3448 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3449 # ifdef USE_POSIX_2008_LOCALE
3450 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3453 # ifdef USE_LOCALE_PAPER
3454 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3455 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3456 # ifdef USE_POSIX_2008_LOCALE
3457 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3460 # ifdef USE_LOCALE_TELEPHONE
3461 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3462 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3463 # ifdef USE_POSIX_2008_LOCALE
3464 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3467 # ifdef USE_LOCALE_SYNTAX
3468 assert(categories[LC_SYNTAX_INDEX] == LC_SYNTAX);
3469 assert(strEQ(category_names[LC_SYNTAX_INDEX], "LC_SYNTAX"));
3470 # ifdef USE_POSIX_2008_LOCALE
3471 assert(category_masks[LC_SYNTAX_INDEX] == LC_SYNTAX_MASK);
3474 # ifdef USE_LOCALE_TOD
3475 assert(categories[LC_TOD_INDEX] == LC_TOD);
3476 assert(strEQ(category_names[LC_TOD_INDEX], "LC_TOD"));
3477 # ifdef USE_POSIX_2008_LOCALE
3478 assert(category_masks[LC_TOD_INDEX] == LC_TOD_MASK);
3482 assert(categories[LC_ALL_INDEX] == LC_ALL);
3483 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3484 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3485 # ifdef USE_POSIX_2008_LOCALE
3486 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3489 # endif /* DEBUGGING */
3491 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
3492 * why these particular incantations are used. */
3494 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
3497 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
3500 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
3503 /* Initialize the cache of the program's UTF-8ness for the always known
3504 * locales C and POSIX */
3505 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3506 sizeof(PL_locale_utf8ness));
3508 # ifdef USE_THREAD_SAFE_LOCALE
3511 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3515 # ifdef USE_POSIX_2008_LOCALE
3517 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3518 if (! PL_C_locale_obj) {
3519 Perl_croak_nocontext(
3520 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3522 if (DEBUG_Lv_TEST || debug_initialization) {
3523 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3528 # ifdef USE_LOCALE_NUMERIC
3530 PL_numeric_radix_sv = newSVpvs(".");
3534 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3536 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3537 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3540 # ifdef LOCALE_ENVIRON_REQUIRED
3543 * Ultrix setlocale(..., "") fails if there are no environment
3544 * variables from which to get a locale name.
3548 # error Ultrix without LC_ALL not implemented
3554 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3555 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3556 if (sl_result[LC_ALL_INDEX])
3559 setlocale_failure = TRUE;
3561 if (! setlocale_failure) {
3562 const char * locale_param;
3563 for (i = 0; i < LC_ALL_INDEX; i++) {
3564 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3567 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3568 if (! sl_result[i]) {
3569 setlocale_failure = TRUE;
3571 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3576 # endif /* LC_ALL */
3577 # endif /* LOCALE_ENVIRON_REQUIRED */
3579 /* We try each locale in the list until we get one that works, or exhaust
3580 * the list. Normally the loop is executed just once. But if setting the
3581 * locale fails, inside the loop we add fallback trials to the array and so
3582 * will execute the loop multiple times */
3583 trial_locales[0] = setlocale_init;
3584 trial_locales_count = 1;
3586 for (i= 0; i < trial_locales_count; i++) {
3587 const char * trial_locale = trial_locales[i];
3591 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3592 * when i==0, but I (khw) don't think that behavior makes much
3594 setlocale_failure = FALSE;
3596 # ifdef SYSTEM_DEFAULT_LOCALE
3597 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3599 /* On Windows machines, an entry of "" after the 0th means to use
3600 * the system default locale, which we now proceed to get. */
3601 if (strEQ(trial_locale, "")) {
3604 /* Note that this may change the locale, but we are going to do
3605 * that anyway just below */
3606 system_default_locale = do_setlocale_c(LC_ALL, "");
3607 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3609 /* Skip if invalid or if it's already on the list of locales to
3611 if (! system_default_locale) {
3612 goto next_iteration;
3614 for (j = 0; j < trial_locales_count; j++) {
3615 if (strEQ(system_default_locale, trial_locales[j])) {
3616 goto next_iteration;
3620 trial_locale = system_default_locale;
3623 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3625 # endif /* SYSTEM_DEFAULT_LOCALE */
3631 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3632 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3633 if (! sl_result[LC_ALL_INDEX]) {
3634 setlocale_failure = TRUE;
3637 /* Since LC_ALL succeeded, it should have changed all the other
3638 * categories it can to its value; so we massage things so that the
3639 * setlocales below just return their category's current values.
3640 * This adequately handles the case in NetBSD where LC_COLLATE may
3641 * not be defined for a locale, and setting it individually will
3642 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3643 * the POSIX locale. */
3644 trial_locale = NULL;
3647 # endif /* LC_ALL */
3649 if (! setlocale_failure) {
3651 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3653 = savepv(do_setlocale_r(categories[j], trial_locale));
3654 if (! curlocales[j]) {
3655 setlocale_failure = TRUE;
3657 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3660 if (! setlocale_failure) { /* All succeeded */
3661 break; /* Exit trial_locales loop */
3665 /* Here, something failed; will need to try a fallback. */
3671 if (locwarn) { /* Output failure info only on the first one */
3675 PerlIO_printf(Perl_error_log,
3676 "perl: warning: Setting locale failed.\n");
3678 # else /* !LC_ALL */
3680 PerlIO_printf(Perl_error_log,
3681 "perl: warning: Setting locale failed for the categories:\n\t");
3683 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3684 if (! curlocales[j]) {
3685 PerlIO_printf(Perl_error_log, category_names[j]);
3688 Safefree(curlocales[j]);
3692 # endif /* LC_ALL */
3694 PerlIO_printf(Perl_error_log,
3695 "perl: warning: Please check that your locale settings:\n");
3699 PerlIO_printf(Perl_error_log,
3700 "\tLANGUAGE = %c%s%c,\n",
3701 language ? '"' : '(',
3702 language ? language : "unset",
3703 language ? '"' : ')');
3706 PerlIO_printf(Perl_error_log,
3707 "\tLC_ALL = %c%s%c,\n",
3709 lc_all ? lc_all : "unset",
3710 lc_all ? '"' : ')');
3712 # if defined(USE_ENVIRON_ARRAY)
3717 /* Look through the environment for any variables of the
3718 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3719 * already handled above. These are assumed to be locale
3720 * settings. Output them and their values. */
3721 for (e = environ; *e; e++) {
3722 const STRLEN prefix_len = sizeof("LC_") - 1;
3725 if ( strBEGINs(*e, "LC_")
3726 && ! strBEGINs(*e, "LC_ALL=")
3727 && (uppers_len = strspn(*e + prefix_len,
3728 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3729 && ((*e)[prefix_len + uppers_len] == '='))
3731 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3732 (int) (prefix_len + uppers_len), *e,
3733 *e + prefix_len + uppers_len + 1);
3740 PerlIO_printf(Perl_error_log,
3741 "\t(possibly more locale environment variables)\n");
3745 PerlIO_printf(Perl_error_log,
3746 "\tLANG = %c%s%c\n",
3748 lang ? lang : "unset",
3751 PerlIO_printf(Perl_error_log,
3752 " are supported and installed on your system.\n");
3755 /* Calculate what fallback locales to try. We have avoided this
3756 * until we have to, because failure is quite unlikely. This will
3757 * usually change the upper bound of the loop we are in.
3759 * Since the system's default way of setting the locale has not
3760 * found one that works, We use Perl's defined ordering: LC_ALL,
3761 * LANG, and the C locale. We don't try the same locale twice, so
3762 * don't add to the list if already there. (On POSIX systems, the
3763 * LC_ALL element will likely be a repeat of the 0th element "",
3764 * but there's no harm done by doing it explicitly.
3766 * Note that this tries the LC_ALL environment variable even on
3767 * systems which have no LC_ALL locale setting. This may or may
3768 * not have been originally intentional, but there's no real need
3769 * to change the behavior. */
3771 for (j = 0; j < trial_locales_count; j++) {
3772 if (strEQ(lc_all, trial_locales[j])) {
3776 trial_locales[trial_locales_count++] = lc_all;
3781 for (j = 0; j < trial_locales_count; j++) {
3782 if (strEQ(lang, trial_locales[j])) {
3786 trial_locales[trial_locales_count++] = lang;
3790 # if defined(WIN32) && defined(LC_ALL)
3792 /* For Windows, we also try the system default locale before "C".
3793 * (If there exists a Windows without LC_ALL we skip this because
3794 * it gets too complicated. For those, the "C" is the next
3795 * fallback possibility). The "" is the same as the 0th element of
3796 * the array, but the code at the loop above knows to treat it
3797 * differently when not the 0th */
3798 trial_locales[trial_locales_count++] = "";
3802 for (j = 0; j < trial_locales_count; j++) {
3803 if (strEQ("C", trial_locales[j])) {
3807 trial_locales[trial_locales_count++] = "C";
3810 } /* end of first time through the loop */
3818 } /* end of looping through the trial locales */
3820 if (ok < 1) { /* If we tried to fallback */
3822 if (! setlocale_failure) { /* fallback succeeded */
3823 msg = "Falling back to";
3825 else { /* fallback failed */
3828 /* We dropped off the end of the loop, so have to decrement i to
3829 * get back to the value the last time through */
3833 msg = "Failed to fall back to";
3835 /* To continue, we should use whatever values we've got */
3837 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3838 Safefree(curlocales[j]);
3839 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3840 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3845 const char * description;
3846 const char * name = "";
3847 if (strEQ(trial_locales[i], "C")) {
3848 description = "the standard locale";
3852 # ifdef SYSTEM_DEFAULT_LOCALE
3854 else if (strEQ(trial_locales[i], "")) {
3855 description = "the system default locale";
3856 if (system_default_locale) {
3857 name = system_default_locale;
3861 # endif /* SYSTEM_DEFAULT_LOCALE */
3864 description = "a fallback locale";
3865 name = trial_locales[i];
3867 if (name && strNE(name, "")) {
3868 PerlIO_printf(Perl_error_log,
3869 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3872 PerlIO_printf(Perl_error_log,
3873 "perl: warning: %s %s.\n", msg, description);
3876 } /* End of tried to fallback */
3878 /* Done with finding the locales; update our records */
3880 # ifdef USE_LOCALE_CTYPE
3882 new_ctype(curlocales[LC_CTYPE_INDEX]);
3885 # ifdef USE_LOCALE_COLLATE
3887 new_collate(curlocales[LC_COLLATE_INDEX]);
3890 # ifdef USE_LOCALE_NUMERIC
3892 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3896 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3898 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3900 /* This caches whether each category's locale is UTF-8 or not. This
3901 * may involve changing the locale. It is ok to do this at
3902 * initialization time before any threads have started, but not later
3903 * unless thread-safe operations are used.
3904 * Caching means that if the program heeds our dictate not to change
3905 * locales in threaded applications, this data will remain valid, and
3906 * it may get queried without having to change locales. If the
3907 * environment is such that all categories have the same locale, this
3908 * isn't needed, as the code will not change the locale; but this
3909 * handles the uncommon case where the environment has disparate
3910 * locales for the categories */
3911 (void) _is_cur_LC_category_utf8(categories[i]);
3915 Safefree(curlocales[i]);
3918 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3920 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3921 * locale is UTF-8. The call to new_ctype() just above has already
3922 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3923 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3924 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3925 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3926 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3928 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3929 This is an alternative to using the -C command line switch
3930 (the -C if present will override this). */
3932 const char *p = PerlEnv_getenv("PERL_UNICODE");
3933 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3934 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3939 #endif /* USE_LOCALE */
3942 /* So won't continue to output stuff */
3943 DEBUG_INITIALIZATION_set(FALSE);
3950 #ifdef USE_LOCALE_COLLATE
3953 Perl__mem_collxfrm(pTHX_ const char *input_string,
3954 STRLEN len, /* Length of 'input_string' */
3955 STRLEN *xlen, /* Set to length of returned string
3956 (not including the collation index
3958 bool utf8 /* Is the input in UTF-8? */
3962 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3963 * differences. First, it handles embedded NULs. Second, it allocates a bit
3964 * more memory than needed for the transformed data itself. The real
3965 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3966 * the length of that, and doesn't include the collation index size.
3967 * Please see sv_collxfrm() to see how this is used. */
3969 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3971 char * s = (char *) input_string;
3972 STRLEN s_strlen = strlen(input_string);
3974 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3975 STRLEN length_in_chars;
3976 bool first_time = TRUE; /* Cleared after first loop iteration */
3978 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3980 /* Must be NUL-terminated */
3981 assert(*(input_string + len) == '\0');
3983 /* If this locale has defective collation, skip */
3984 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3985 DEBUG_L(PerlIO_printf(Perl_debug_log,
3986 "_mem_collxfrm: locale's collation is defective\n"));
3990 /* Replace any embedded NULs with the control that sorts before any others.
3991 * This will give as good as possible results on strings that don't
3992 * otherwise contain that character, but otherwise there may be
3993 * less-than-perfect results with that character and NUL. This is
3994 * unavoidable unless we replace strxfrm with our own implementation. */
3995 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3999 STRLEN sans_nuls_len;
4000 int try_non_controls;
4001 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
4002 making sure 2nd byte is NUL.
4004 STRLEN this_replacement_len;
4006 /* If we don't know what non-NUL control character sorts lowest for
4007 * this locale, find it */
4008 if (PL_strxfrm_NUL_replacement == '\0') {
4010 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
4011 includes the collation index
4014 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
4016 /* Unlikely, but it may be that no control will work to replace
4017 * NUL, in which case we instead look for any character. Controls
4018 * are preferred because collation order is, in general, context
4019 * sensitive, with adjoining characters affecting the order, and
4020 * controls are less likely to have such interactions, allowing the
4021 * NUL-replacement to stand on its own. (Another way to look at it
4022 * is to imagine what would happen if the NUL were replaced by a
4023 * combining character; it wouldn't work out all that well.) */
4024 for (try_non_controls = 0;
4025 try_non_controls < 2;
4028 /* Look through all legal code points (NUL isn't) */
4029 for (j = 1; j < 256; j++) {
4030 char * x; /* j's xfrm plus collation index */
4031 STRLEN x_len; /* length of 'x' */
4032 STRLEN trial_len = 1;
4033 char cur_source[] = { '\0', '\0' };
4035 /* Skip non-controls the first time through the loop. The
4036 * controls in a UTF-8 locale are the L1 ones */
4037 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
4044 /* Create a 1-char string of the current code point */
4045 cur_source[0] = (char) j;
4047 /* Then transform it */
4048 x = _mem_collxfrm(cur_source, trial_len, &x_len,
4049 0 /* The string is not in UTF-8 */);
4051 /* Ignore any character that didn't successfully transform.
4057 /* If this character's transformation is lower than
4058 * the current lowest, this one becomes the lowest */
4059 if ( cur_min_x == NULL
4060 || strLT(x + COLLXFRM_HDR_LEN,
4061 cur_min_x + COLLXFRM_HDR_LEN))
4063 PL_strxfrm_NUL_replacement = j;
4064 Safefree(cur_min_x);
4070 } /* end of loop through all 255 characters */
4072 /* Stop looking if found */
4077 /* Unlikely, but possible, if there aren't any controls that
4078 * work in the locale, repeat the loop, looking for any
4079 * character that works */
4080 DEBUG_L(PerlIO_printf(Perl_debug_log,
4081 "_mem_collxfrm: No control worked. Trying non-controls\n"));
4082 } /* End of loop to try first the controls, then any char */
4085 DEBUG_L(PerlIO_printf(Perl_debug_log,
4086 "_mem_collxfrm: Couldn't find any character to replace"
4087 " embedded NULs in locale %s with", PL_collation_name));
4091 DEBUG_L(PerlIO_printf(Perl_debug_log,
4092 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
4093 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
4095 Safefree(cur_min_x);
4096 } /* End of determining the character that is to replace NULs */
4098 /* If the replacement is variant under UTF-8, it must match the
4099 * UTF8-ness of the original */
4100 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
4101 this_replacement_char[0] =
4102 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
4103 this_replacement_char[1] =
4104 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
4105 this_replacement_len = 2;
4108 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
4109 /* this_replacement_char[1] = '\0' was done at initialization */
4110 this_replacement_len = 1;
4113 /* The worst case length for the replaced string would be if every
4114 * character in it is NUL. Multiply that by the length of each
4115 * replacement, and allow for a trailing NUL */
4116 sans_nuls_len = (len * this_replacement_len) + 1;
4117 Newx(sans_nuls, sans_nuls_len, char);
4120 /* Replace each NUL with the lowest collating control. Loop until have
4121 * exhausted all the NULs */
4122 while (s + s_strlen < e) {
4123 my_strlcat(sans_nuls, s, sans_nuls_len);
4125 /* Do the actual replacement */
4126 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
4128 /* Move past the input NUL */
4130 s_strlen = strlen(s);
4133 /* And add anything that trails the final NUL */
4134 my_strlcat(sans_nuls, s, sans_nuls_len);
4136 /* Switch so below we transform this modified string */
4139 } /* End of replacing NULs */
4141 /* Make sure the UTF8ness of the string and locale match */
4142 if (utf8 != PL_in_utf8_COLLATE_locale) {
4143 /* XXX convert above Unicode to 10FFFF? */
4144 const char * const t = s; /* Temporary so we can later find where the
4147 /* Here they don't match. Change the string's to be what the locale is
4150 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
4151 s = (char *) bytes_to_utf8((const U8 *) s, &len);
4154 else { /* locale is not UTF-8; but input is; downgrade the input */
4156 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
4158 /* If the downgrade was successful we are done, but if the input
4159 * contains things that require UTF-8 to represent, have to do
4160 * damage control ... */
4161 if (UNLIKELY(utf8)) {
4163 /* What we do is construct a non-UTF-8 string with
4164 * 1) the characters representable by a single byte converted
4165 * to be so (if necessary);
4166 * 2) and the rest converted to collate the same as the
4167 * highest collating representable character. That makes
4168 * them collate at the end. This is similar to how we
4169 * handle embedded NULs, but we use the highest collating
4170 * code point instead of the smallest. Like the NUL case,
4171 * this isn't perfect, but is the best we can reasonably
4172 * do. Every above-255 code point will sort the same as
4173 * the highest-sorting 0-255 code point. If that code
4174 * point can combine in a sequence with some other code
4175 * points for weight calculations, us changing something to
4176 * be it can adversely affect the results. But in most
4177 * cases, it should work reasonably. And note that this is
4178 * really an illegal situation: using code points above 255
4179 * on a locale where only 0-255 are valid. If two strings
4180 * sort entirely equal, then the sort order for the
4181 * above-255 code points will be in code point order. */
4185 /* If we haven't calculated the code point with the maximum
4186 * collating order for this locale, do so now */
4187 if (! PL_strxfrm_max_cp) {
4190 /* The current transformed string that collates the
4191 * highest (except it also includes the prefixed collation
4193 char * cur_max_x = NULL;
4195 /* Look through all legal code points (NUL isn't) */
4196 for (j = 1; j < 256; j++) {
4199 char cur_source[] = { '\0', '\0' };
4201 /* Create a 1-char string of the current code point */
4202 cur_source[0] = (char) j;
4204 /* Then transform it */
4205 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4207 /* If something went wrong (which it shouldn't), just
4208 * ignore this code point */
4213 /* If this character's transformation is higher than
4214 * the current highest, this one becomes the highest */
4215 if ( cur_max_x == NULL
4216 || strGT(x + COLLXFRM_HDR_LEN,
4217 cur_max_x + COLLXFRM_HDR_LEN))
4219 PL_strxfrm_max_cp = j;
4220 Safefree(cur_max_x);
4229 DEBUG_L(PerlIO_printf(Perl_debug_log,
4230 "_mem_collxfrm: Couldn't find any character to"
4231 " replace above-Latin1 chars in locale %s with",
4232 PL_collation_name));
4236 DEBUG_L(PerlIO_printf(Perl_debug_log,
4237 "_mem_collxfrm: highest 1-byte collating character"
4238 " in locale %s is 0x%02X\n",
4240 PL_strxfrm_max_cp));
4242 Safefree(cur_max_x);
4245 /* Here we know which legal code point collates the highest.
4246 * We are ready to construct the non-UTF-8 string. The length
4247 * will be at least 1 byte smaller than the input string
4248 * (because we changed at least one 2-byte character into a
4249 * single byte), but that is eaten up by the trailing NUL */
4255 char * e = (char *) t + len;
4257 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4259 if (UTF8_IS_INVARIANT(cur_char)) {
4262 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4263 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4265 else { /* Replace illegal cp with highest collating
4267 s[d++] = PL_strxfrm_max_cp;
4271 Renew(s, d, char); /* Free up unused space */
4276 /* Here, we have constructed a modified version of the input. It could
4277 * be that we already had a modified copy before we did this version.
4278 * If so, that copy is no longer needed */
4279 if (t != input_string) {
4284 length_in_chars = (utf8)
4285 ? utf8_length((U8 *) s, (U8 *) s + len)
4288 /* The first element in the output is the collation id, used by
4289 * sv_collxfrm(); then comes the space for the transformed string. The
4290 * equation should give us a good estimate as to how much is needed */
4291 xAlloc = COLLXFRM_HDR_LEN
4293 + (PL_collxfrm_mult * length_in_chars);
4294 Newx(xbuf, xAlloc, char);
4295 if (UNLIKELY(! xbuf)) {
4296 DEBUG_L(PerlIO_printf(Perl_debug_log,
4297 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4301 /* Store the collation id */
4302 *(U32*)xbuf = PL_collation_ix;
4304 /* Then the transformation of the input. We loop until successful, or we
https://perl5.git.perl.org / perl5.git / blob