3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * A Elbereth Gilthoniel,
13 * silivren penna míriel
14 * o menel aglar elenath!
15 * Na-chaered palan-díriel
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
47 #define PERL_IN_LOCALE_C
48 #include "perl_langinfo.h"
57 /* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
60 #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61 # define debug_initialization 0
62 # define DEBUG_INITIALIZATION_set(v)
64 static bool debug_initialization = FALSE;
65 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
69 /* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
72 #define GET_ERRNO saved_errno
74 /* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
77 #define STRLENs(s) (sizeof("" s "") - 1)
79 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89 #define isNAME_C_OR_POSIX(name) \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
96 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
99 #define UTF8NESS_SEP "\v"
100 #define UTF8NESS_PREFIX "\f"
102 /* So, the string looks like:
104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
109 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
112 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
115 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
120 S_stdize_locale(pTHX_ char *locs)
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
136 const char * const s = strchr(locs, '=');
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
142 const char * const t = strchr(s, '.');
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
161 /* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
165 const int categories[] = {
167 # ifdef USE_LOCALE_NUMERIC
170 # ifdef USE_LOCALE_CTYPE
173 # ifdef USE_LOCALE_COLLATE
176 # ifdef USE_LOCALE_TIME
179 # ifdef USE_LOCALE_MESSAGES
182 # ifdef USE_LOCALE_MONETARY
185 # ifdef USE_LOCALE_ADDRESS
188 # ifdef USE_LOCALE_IDENTIFICATION
191 # ifdef USE_LOCALE_MEASUREMENT
194 # ifdef USE_LOCALE_PAPER
197 # ifdef USE_LOCALE_TELEPHONE
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
208 /* The top-most real element is LC_ALL */
210 const char * category_names[] = {
212 # ifdef USE_LOCALE_NUMERIC
215 # ifdef USE_LOCALE_CTYPE
218 # ifdef USE_LOCALE_COLLATE
221 # ifdef USE_LOCALE_TIME
224 # ifdef USE_LOCALE_MESSAGES
227 # ifdef USE_LOCALE_MONETARY
230 # ifdef USE_LOCALE_ADDRESS
233 # ifdef USE_LOCALE_IDENTIFICATION
236 # ifdef USE_LOCALE_MEASUREMENT
239 # ifdef USE_LOCALE_PAPER
242 # ifdef USE_LOCALE_TELEPHONE
248 NULL /* Placeholder */
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
265 /* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
274 S_category_name(const int category)
280 if (category == LC_ALL) {
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
293 const char suffix[] = " (unknown)";
295 Size_t length = sizeof(suffix) + 1;
304 /* Calculate the number of digits */
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
317 /* Now create LC_foo_INDEX #defines for just those categories on this system */
318 # ifdef USE_LOCALE_NUMERIC
319 # define LC_NUMERIC_INDEX 0
320 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
322 # define _DUMMY_NUMERIC -1
324 # ifdef USE_LOCALE_CTYPE
325 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326 # define _DUMMY_CTYPE LC_CTYPE_INDEX
328 # define _DUMMY_CTYPE _DUMMY_NUMERIC
330 # ifdef USE_LOCALE_COLLATE
331 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332 # define _DUMMY_COLLATE LC_COLLATE_INDEX
334 # define _DUMMY_COLLATE _DUMMY_COLLATE
336 # ifdef USE_LOCALE_TIME
337 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
338 # define _DUMMY_TIME LC_TIME_INDEX
340 # define _DUMMY_TIME _DUMMY_COLLATE
342 # ifdef USE_LOCALE_MESSAGES
343 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
346 # define _DUMMY_MESSAGES _DUMMY_TIME
348 # ifdef USE_LOCALE_MONETARY
349 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350 # define _DUMMY_MONETARY LC_MONETARY_INDEX
352 # define _DUMMY_MONETARY _DUMMY_MESSAGES
354 # ifdef USE_LOCALE_ADDRESS
355 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
358 # define _DUMMY_ADDRESS _DUMMY_MONETARY
360 # ifdef USE_LOCALE_IDENTIFICATION
361 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
364 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
366 # ifdef USE_LOCALE_MEASUREMENT
367 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
370 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
372 # ifdef USE_LOCALE_PAPER
373 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374 # define _DUMMY_PAPER LC_PAPER_INDEX
376 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
378 # ifdef USE_LOCALE_TELEPHONE
379 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
382 # define _DUMMY_TELEPHONE _DUMMY_PAPER
385 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
387 #endif /* ifdef USE_LOCALE */
389 /* Windows requres a customized base-level setlocale() */
391 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
393 # define my_setlocale(cat, locale) setlocale(cat, locale)
396 #ifndef USE_POSIX_2008_LOCALE
398 /* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
403 #else /* Below uses POSIX 2008 */
405 /* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411 # define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
415 /* A third array, parallel to the ones above to map from category to its
417 const int category_masks[] = {
418 # ifdef USE_LOCALE_NUMERIC
421 # ifdef USE_LOCALE_CTYPE
424 # ifdef USE_LOCALE_COLLATE
427 # ifdef USE_LOCALE_TIME
430 # ifdef USE_LOCALE_MESSAGES
433 # ifdef USE_LOCALE_MONETARY
436 # ifdef USE_LOCALE_ADDRESS
439 # ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
442 # ifdef USE_LOCALE_MEASUREMENT
445 # ifdef USE_LOCALE_PAPER
448 # ifdef USE_LOCALE_TELEPHONE
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
460 S_emulate_setlocale(const int category,
463 const bool is_index_valid
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
554 mask = category_masks[index];
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
580 # ifdef HAS_QUERYLOCALE
582 return (char *) querylocale(mask, cur_obj);
585 # if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
588 /* Internal glibc for querylocale(), but doesn't handle
589 * empty-string ("") locale properly; who knows what other
590 * glitches. Check it for now, under debug. */
592 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
593 uselocale((locale_t) 0));
595 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
596 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
597 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
599 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
600 if ( strNE(PL_curlocales[index], temp_name)
601 && ! ( isNAME_C_OR_POSIX(temp_name)
602 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
604 # ifdef USE_C_BACKTRACE
606 dump_c_backtrace(Perl_debug_log, 20, 1);
610 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
611 " (%s) and what internal glibc thinks"
612 " (%s)\n", category_names[index],
613 PL_curlocales[index], temp_name);
622 /* Without querylocale(), we have to use our record-keeping we've
625 if (category != LC_ALL) {
629 if (DEBUG_Lv_TEST || debug_initialization) {
630 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 return PL_curlocales[index];
637 else { /* For LC_ALL */
639 Size_t names_len = 0;
642 /* If we have a valid LC_ALL value, just return it */
643 if (PL_curlocales[LC_ALL_INDEX]) {
647 if (DEBUG_Lv_TEST || debug_initialization) {
648 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
653 return PL_curlocales[LC_ALL_INDEX];
656 /* Otherwise, we need to construct a string of name=value pairs.
657 * We use the glibc syntax, like
658 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
659 * First calculate the needed size. */
660 for (i = 0; i < LC_ALL_INDEX; i++) {
664 if (DEBUG_Lv_TEST || debug_initialization) {
665 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]);
670 names_len += strlen(category_names[i])
672 + strlen(PL_curlocales[i])
675 names_len++; /* Trailing '\0' */
676 SAVEFREEPV(Newx(all_string, names_len, char));
679 /* Then fill in the string */
680 for (i = 0; i < LC_ALL_INDEX; i++) {
684 if (DEBUG_Lv_TEST || debug_initialization) {
685 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]);
690 my_strlcat(all_string, category_names[i], names_len);
691 my_strlcat(all_string, "=", names_len);
692 my_strlcat(all_string, PL_curlocales[i], names_len);
693 my_strlcat(all_string, ";", names_len);
698 if (DEBUG_L_TEST || debug_initialization) {
699 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
709 SETERRNO(EINVAL, LIB_INVARG);
719 assert(PL_C_locale_obj);
721 /* Otherwise, we are switching locales. This will generally entail freeing
722 * the current one's space (at the C library's discretion). We need to
723 * stop using that locale before the switch. So switch to a known locale
724 * object that we don't otherwise mess with. This returns the locale
725 * object in effect at the time of the switch. */
726 old_obj = uselocale(PL_C_locale_obj);
730 if (DEBUG_Lv_TEST || debug_initialization) {
731 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
740 if (DEBUG_L_TEST || debug_initialization) {
742 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
753 if (DEBUG_Lv_TEST || debug_initialization) {
754 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
759 /* If we weren't in a thread safe locale, set so that newlocale() below
760 which uses 'old_obj', uses an empty one. Same for our reserved C object.
761 The latter is defensive coding, so that, even if there is some bug, we
762 will never end up trying to modify either of these, as if passed to
763 newlocale(), they can be. */
764 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
765 old_obj = (locale_t) 0;
768 /* Create the new locale (it may actually modify the current one). */
770 # ifndef HAS_QUERYLOCALE
772 if (strEQ(locale, "")) {
774 /* For non-querylocale() systems, we do the setting of "" ourselves to
775 * be sure that we really know what's going on. We follow the Linux
776 * documented behavior (but if that differs from the actual behavior,
777 * this won't work exactly as the OS implements). We go out and
778 * examine the environment based on our understanding of how the system
779 * works, and use that to figure things out */
781 const char * const lc_all = PerlEnv_getenv("LC_ALL");
783 /* Use any "LC_ALL" environment variable, as it overrides everything
785 if (lc_all && strNE(lc_all, "")) {
790 /* Otherwise, we need to dig deeper. Unless overridden, the
791 * default is the LANG environment variable; if it doesn't exist,
794 const char * default_name;
796 /* To minimize other threads messing with the environment, we copy
797 * the variable, making it a temporary. But this doesn't work upon
798 * program initialization before any scopes are created, and at
799 * this time, there's nothing else going on that would interfere.
800 * So skip the copy in that case */
801 if (PL_scopestack_ix == 0) {
802 default_name = PerlEnv_getenv("LANG");
805 default_name = savepv(PerlEnv_getenv("LANG"));
808 if (! default_name || strEQ(default_name, "")) {
811 else if (PL_scopestack_ix != 0) {
812 SAVEFREEPV(default_name);
815 if (category != LC_ALL) {
816 const char * const name = PerlEnv_getenv(category_names[index]);
818 /* Here we are setting a single category. Assume will have the
820 locale = default_name;
822 /* But then look for an overriding environment variable */
823 if (name && strNE(name, "")) {
828 bool did_override = FALSE;
831 /* Here, we are getting LC_ALL. Any categories that don't have
832 * a corresponding environment variable set should be set to
833 * LANG, or to "C" if there is no LANG. If no individual
834 * categories differ from this, we can just set LC_ALL. This
835 * is buggy on systems that have extra categories that we don't
836 * know about. If there is an environment variable that sets
837 * that category, we won't know to look for it, and so our use
838 * of LANG or "C" improperly overrides it. On the other hand,
839 * if we don't do what is done here, and there is no
840 * environment variable, the category's locale should be set to
841 * LANG or "C". So there is no good solution. khw thinks the
842 * best is to look at systems to see what categories they have,
843 * and include them, and then to assume that we know the
846 for (i = 0; i < LC_ALL_INDEX; i++) {
847 const char * const env_override
848 = savepv(PerlEnv_getenv(category_names[i]));
849 const char * this_locale = ( env_override
850 && strNE(env_override, ""))
853 emulate_setlocale(categories[i], this_locale, i, TRUE);
855 if (strNE(this_locale, default_name)) {
859 Safefree(env_override);
862 /* If all the categories are the same, we can set LC_ALL to
864 if (! did_override) {
865 locale = default_name;
869 /* Here, LC_ALL is no longer valid, as some individual
870 * categories don't match it. We call ourselves
871 * recursively, as that will execute the code that
872 * generates the proper locale string for this situation.
873 * We don't do the remainder of this function, as that is
874 * to update our records, and we've just done that for the
875 * individual categories in the loop above, and doing so
876 * would cause LC_ALL to be done as well */
877 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
882 else if (strchr(locale, ';')) {
884 /* LC_ALL may actually incude a conglomeration of various categories.
885 * Without querylocale, this code uses the glibc (as of this writing)
886 * syntax for representing that, but that is not a stable API, and
887 * other platforms do it differently, so we have to handle all cases
890 const char * s = locale;
891 const char * e = locale + strlen(locale);
893 const char * category_end;
894 const char * name_start;
895 const char * name_end;
900 /* Parse through the category */
901 while (isWORDCHAR(*p)) {
908 "panic: %s: %d: Unexpected character in locale name '%02X",
909 __FILE__, __LINE__, *(p-1));
912 /* Parse through the locale name */
914 while (isGRAPH(*p) && *p != ';') {
921 "panic: %s: %d: Unexpected character in locale name '%02X",
922 __FILE__, __LINE__, *(p-1));
925 /* Find the index of the category name in our lists */
926 for (i = 0; i < LC_ALL_INDEX; i++) {
928 /* Keep going if this isn't the index. The strnNE() avoids a
929 * Perl_form(), but would fail if ever a category name could be
930 * a substring of another one, like if there were a
932 if strnNE(s, category_names[i], category_end - s) {
936 /* If this index is for the single category we're changing, we
937 * have found the locale to set it to. */
938 if (category == categories[i]) {
939 locale = Perl_form(aTHX_ "%.*s",
940 (int) (name_end - name_start),
945 if (category == LC_ALL) {
946 char * individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
947 emulate_setlocale(categories[i], individ_locale, i, TRUE);
948 Safefree(individ_locale);
955 /* Here we have set all the individual categories by recursive calls.
956 * These collectively should have fixed up LC_ALL, so can just query
957 * what that now is */
958 assert(category == LC_ALL);
960 return do_setlocale_c(LC_ALL, NULL);
965 # endif /* end of ! querylocale */
967 /* Ready to create a new locale by modification of the exising one */
968 new_obj = newlocale(mask, locale, old_obj);
975 if (DEBUG_L_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
981 if (! uselocale(old_obj)) {
986 if (DEBUG_L_TEST || debug_initialization) {
987 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
999 if (DEBUG_Lv_TEST || debug_initialization) {
1000 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1005 /* And switch into it */
1006 if (! uselocale(new_obj)) {
1011 if (DEBUG_L_TEST || debug_initialization) {
1012 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1017 if (! uselocale(old_obj)) {
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1028 freelocale(new_obj);
1035 if (DEBUG_Lv_TEST || debug_initialization) {
1036 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1041 /* We are done, except for updating our records (if the system doesn't keep
1042 * them) and in the case of locale "", we don't actually know what the
1043 * locale that got switched to is, as it came from the environment. So
1044 * have to find it */
1046 # ifdef HAS_QUERYLOCALE
1048 if (strEQ(locale, "")) {
1049 locale = querylocale(mask, new_obj);
1054 /* Here, 'locale' is the return value */
1056 /* Without querylocale(), we have to update our records */
1058 if (category == LC_ALL) {
1061 /* For LC_ALL, we change all individual categories to correspond */
1062 /* PL_curlocales is a parallel array, so has same
1063 * length as 'categories' */
1064 for (i = 0; i <= LC_ALL_INDEX; i++) {
1065 Safefree(PL_curlocales[i]);
1066 PL_curlocales[i] = savepv(locale);
1071 /* For a single category, if it's not the same as the one in LC_ALL, we
1074 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1075 Safefree(PL_curlocales[LC_ALL_INDEX]);
1076 PL_curlocales[LC_ALL_INDEX] = NULL;
1079 /* Then update the category's record */
1080 Safefree(PL_curlocales[index]);
1081 PL_curlocales[index] = savepv(locale);
1089 #endif /* USE_POSIX_2008_LOCALE */
1091 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1092 didn't natively support them */
1094 /* The way this would work is that we would keep a per-thread list of the
1095 * correct locale for that thread. Any operation that was locale-sensitive
1096 * would have to be changed so that it would look like this:
1099 * setlocale to the correct locale for this operation
1103 * This leaves the global locale in the most recently used operation's, but it
1104 * was locked long enough to get the result. If that result is static, it
1105 * needs to be copied before the unlock.
1107 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1108 * the setup, but are no-ops when not needed, and similarly,
1109 * END_LOCALE_DEPENDENT_OP for the tear-down
1111 * But every call to a locale-sensitive function would have to be changed, and
1112 * if a module didn't cooperate by using the mutex, things would break.
1114 * This code was abandoned before being completed or tested, and is left as-is
1117 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1118 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1121 S_locking_setlocale(pTHX_
1123 const char * locale,
1125 const bool is_index_valid
1128 /* This function kind of performs a setlocale() on just the current thread;
1129 * thus it is kind of thread-safe. It does this by keeping a thread-level
1130 * array of the current locales for each category. Every time a locale is
1131 * switched to, it does the switch globally, but updates the thread's
1132 * array. A query as to what the current locale is just returns the
1133 * appropriate element from the array, and doesn't actually call the system
1134 * setlocale(). The saving into the array is done in an uninterruptible
1135 * section of code, so is unaffected by whatever any other threads might be
1138 * All locale-sensitive operations must work by first starting a critical
1139 * section, then switching to the thread's locale as kept by this function,
1140 * and then doing the operation, then ending the critical section. Thus,
1141 * each gets done in the appropriate locale. simulating thread-safety.
1143 * This function takes the same parameters, 'category' and 'locale', that
1144 * the regular setlocale() function does, but it also takes two additional
1145 * ones. This is because as described earlier. If we know on input the
1146 * index corresponding to the category into the array where we store the
1147 * current locales, we don't have to calculate it. If the caller knows at
1148 * compile time what the index is, it it can pass it, setting
1149 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1153 /* If the input index might be incorrect, calculate the correct one */
1154 if (! is_index_valid) {
1157 if (DEBUG_Lv_TEST || debug_initialization) {
1158 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1161 for (i = 0; i <= LC_ALL_INDEX; i++) {
1162 if (category == categories[i]) {
1168 /* Here, we don't know about this category, so can't handle it.
1169 * XXX best we can do is to unsafely set this
1172 return my_setlocale(category, locale);
1176 if (DEBUG_Lv_TEST || debug_initialization) {
1177 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1181 /* For a query, just return what's in our records */
1182 if (new_locale == NULL) {
1183 return curlocales[index];
1187 /* Otherwise, we need to do the switch, and save the result, all in a
1188 * critical section */
1190 Safefree(curlocales[[index]]);
1192 /* It might be that this is called from an already-locked section of code.
1193 * We would have to detect and skip the LOCK/UNLOCK if so */
1196 curlocales[index] = savepv(my_setlocale(category, new_locale));
1198 if (strEQ(new_locale, "")) {
1202 /* The locale values come from the environment, and may not all be the
1203 * same, so for LC_ALL, we have to update all the others, while the
1204 * mutex is still locked */
1206 if (category == LC_ALL) {
1208 for (i = 0; i < LC_ALL_INDEX) {
1209 curlocales[i] = my_setlocale(categories[i], NULL);
1218 return curlocales[index];
1224 S_set_numeric_radix(pTHX_ const bool use_locale)
1226 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1227 * TRUE, use the radix character derived from the current locale */
1229 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1230 || defined(HAS_NL_LANGINFO))
1232 const char * radix = (use_locale)
1233 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1234 /* FALSE => already in dest locale */
1237 sv_setpv(PL_numeric_radix_sv, radix);
1239 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1240 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1241 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1242 SvCUR(PL_numeric_radix_sv))
1243 && _is_cur_LC_category_utf8(LC_NUMERIC))
1245 SvUTF8_on(PL_numeric_radix_sv);
1250 if (DEBUG_L_TEST || debug_initialization) {
1251 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1252 SvPVX(PL_numeric_radix_sv),
1253 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1257 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1262 S_new_numeric(pTHX_ const char *newnum)
1265 #ifndef USE_LOCALE_NUMERIC
1267 PERL_UNUSED_ARG(newnum);
1271 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1272 * core Perl this and that 'newnum' is the name of the new locale.
1273 * It installs this locale as the current underlying default.
1275 * The default locale and the C locale can be toggled between by use of the
1276 * set_numeric_underlying() and set_numeric_standard() functions, which
1277 * should probably not be called directly, but only via macros like
1278 * SET_NUMERIC_STANDARD() in perl.h.
1280 * The toggling is necessary mainly so that a non-dot radix decimal point
1281 * character can be output, while allowing internal calculations to use a
1284 * This sets several interpreter-level variables:
1285 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1286 * PL_numeric_underlying A boolean indicating if the toggled state is such
1287 * that the current locale is the program's underlying
1289 * PL_numeric_standard An int indicating if the toggled state is such
1290 * that the current locale is the C locale or
1291 * indistinguishable from the C locale. If non-zero, it
1292 * is in C; if > 1, it means it may not be toggled away
1294 * PL_numeric_underlying_is_standard A bool kept by this function
1295 * indicating that the underlying locale and the standard
1296 * C locale are indistinguishable for the purposes of
1297 * LC_NUMERIC. This happens when both of the above two
1298 * variables are true at the same time. (Toggling is a
1299 * no-op under these circumstances.) This variable is
1300 * used to avoid having to recalculate.
1306 Safefree(PL_numeric_name);
1307 PL_numeric_name = NULL;
1308 PL_numeric_standard = TRUE;
1309 PL_numeric_underlying = TRUE;
1310 PL_numeric_underlying_is_standard = TRUE;
1314 save_newnum = stdize_locale(savepv(newnum));
1315 PL_numeric_underlying = TRUE;
1316 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1318 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1320 if (! PL_numeric_standard) {
1321 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1322 FALSE /* Don't toggle locale */ ))
1323 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1327 /* Save the new name if it isn't the same as the previous one, if any */
1328 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1329 Safefree(PL_numeric_name);
1330 PL_numeric_name = save_newnum;
1333 Safefree(save_newnum);
1336 PL_numeric_underlying_is_standard = PL_numeric_standard;
1338 # ifdef HAS_POSIX_2008_LOCALE
1340 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1342 PL_underlying_numeric_obj);
1346 if (DEBUG_L_TEST || debug_initialization) {
1347 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1350 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1351 * have to worry about the radix being a non-dot. (Core operations that
1352 * need the underlying locale change to it temporarily). */
1353 set_numeric_standard();
1355 #endif /* USE_LOCALE_NUMERIC */
1360 Perl_set_numeric_standard(pTHX)
1363 #ifdef USE_LOCALE_NUMERIC
1365 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1366 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1367 * macro avoids calling this routine if toggling isn't necessary according
1368 * to our records (which could be wrong if some XS code has changed the
1369 * locale behind our back) */
1371 do_setlocale_c(LC_NUMERIC, "C");
1372 PL_numeric_standard = TRUE;
1373 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1374 set_numeric_radix(0);
1378 if (DEBUG_L_TEST || debug_initialization) {
1379 PerlIO_printf(Perl_debug_log,
1380 "LC_NUMERIC locale now is standard C\n");
1384 #endif /* USE_LOCALE_NUMERIC */
1389 Perl_set_numeric_underlying(pTHX)
1392 #ifdef USE_LOCALE_NUMERIC
1394 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1395 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1396 * instead of calling this directly. The macro avoids calling this routine
1397 * if toggling isn't necessary according to our records (which could be
1398 * wrong if some XS code has changed the locale behind our back) */
1400 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1401 PL_numeric_standard = PL_numeric_underlying_is_standard;
1402 PL_numeric_underlying = TRUE;
1403 set_numeric_radix(! PL_numeric_standard);
1407 if (DEBUG_L_TEST || debug_initialization) {
1408 PerlIO_printf(Perl_debug_log,
1409 "LC_NUMERIC locale now is %s\n",
1414 #endif /* USE_LOCALE_NUMERIC */
1419 * Set up for a new ctype locale.
1422 S_new_ctype(pTHX_ const char *newctype)
1425 #ifndef USE_LOCALE_CTYPE
1427 PERL_ARGS_ASSERT_NEW_CTYPE;
1428 PERL_UNUSED_ARG(newctype);
1429 PERL_UNUSED_CONTEXT;
1433 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1434 * core Perl this and that 'newctype' is the name of the new locale.
1436 * This function sets up the folding arrays for all 256 bytes, assuming
1437 * that tofold() is tolc() since fold case is not a concept in POSIX,
1439 * Any code changing the locale (outside this file) should use
1440 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1441 * this function should be called directly only from this file and from
1442 * POSIX::setlocale() */
1447 /* Don't check for problems if we are suppressing the warnings */
1448 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1450 PERL_ARGS_ASSERT_NEW_CTYPE;
1452 /* We will replace any bad locale warning with 1) nothing if the new one is
1453 * ok; or 2) a new warning for the bad new locale */
1454 if (PL_warn_locale) {
1455 SvREFCNT_dec_NN(PL_warn_locale);
1456 PL_warn_locale = NULL;
1459 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1461 /* A UTF-8 locale gets standard rules. But note that code still has to
1462 * handle this specially because of the three problematic code points */
1463 if (PL_in_utf8_CTYPE_locale) {
1464 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1467 /* We don't populate the other lists if a UTF-8 locale, but do check that
1468 * everything works as expected, unless checking turned off */
1469 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1470 /* Assume enough space for every character being bad. 4 spaces each
1471 * for the 94 printable characters that are output like "'x' "; and 5
1472 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1474 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1475 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1477 unsigned int bad_count = 0; /* Count of bad characters */
1479 for (i = 0; i < 256; i++) {
1480 if (! PL_in_utf8_CTYPE_locale) {
1482 PL_fold_locale[i] = (U8) tolower(i);
1483 else if (islower(i))
1484 PL_fold_locale[i] = (U8) toupper(i);
1486 PL_fold_locale[i] = (U8) i;
1489 /* If checking for locale problems, see if the native ASCII-range
1490 * printables plus \n and \t are in their expected categories in
1491 * the new locale. If not, this could mean big trouble, upending
1492 * Perl's and most programs' assumptions, like having a
1493 * metacharacter with special meaning become a \w. Fortunately,
1494 * it's very rare to find locales that aren't supersets of ASCII
1495 * nowadays. It isn't a problem for most controls to be changed
1496 * into something else; we check only \n and \t, though perhaps \r
1497 * could be an issue as well. */
1498 if ( check_for_problems
1499 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1501 bool is_bad = FALSE;
1502 char name[3] = { '\0' };
1504 /* Convert the name into a string */
1509 else if (i == '\n') {
1510 my_strlcpy(name, "\n", sizeof(name));
1513 my_strlcpy(name, "\t", sizeof(name));
1516 /* Check each possibe class */
1517 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1519 DEBUG_L(PerlIO_printf(Perl_debug_log,
1520 "isalnum('%s') unexpectedly is %d\n",
1521 name, cBOOL(isalnum(i))));
1523 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1525 DEBUG_L(PerlIO_printf(Perl_debug_log,
1526 "isalpha('%s') unexpectedly is %d\n",
1527 name, cBOOL(isalpha(i))));
1529 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1531 DEBUG_L(PerlIO_printf(Perl_debug_log,
1532 "isdigit('%s') unexpectedly is %d\n",
1533 name, cBOOL(isdigit(i))));
1535 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1537 DEBUG_L(PerlIO_printf(Perl_debug_log,
1538 "isgraph('%s') unexpectedly is %d\n",
1539 name, cBOOL(isgraph(i))));
1541 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1543 DEBUG_L(PerlIO_printf(Perl_debug_log,
1544 "islower('%s') unexpectedly is %d\n",
1545 name, cBOOL(islower(i))));
1547 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1549 DEBUG_L(PerlIO_printf(Perl_debug_log,
1550 "isprint('%s') unexpectedly is %d\n",
1551 name, cBOOL(isprint(i))));
1553 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1555 DEBUG_L(PerlIO_printf(Perl_debug_log,
1556 "ispunct('%s') unexpectedly is %d\n",
1557 name, cBOOL(ispunct(i))));
1559 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1561 DEBUG_L(PerlIO_printf(Perl_debug_log,
1562 "isspace('%s') unexpectedly is %d\n",
1563 name, cBOOL(isspace(i))));
1565 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1567 DEBUG_L(PerlIO_printf(Perl_debug_log,
1568 "isupper('%s') unexpectedly is %d\n",
1569 name, cBOOL(isupper(i))));
1571 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1573 DEBUG_L(PerlIO_printf(Perl_debug_log,
1574 "isxdigit('%s') unexpectedly is %d\n",
1575 name, cBOOL(isxdigit(i))));
1577 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1579 DEBUG_L(PerlIO_printf(Perl_debug_log,
1580 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1581 name, tolower(i), (int) toLOWER_A(i)));
1583 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1585 DEBUG_L(PerlIO_printf(Perl_debug_log,
1586 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1587 name, toupper(i), (int) toUPPER_A(i)));
1589 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1591 DEBUG_L(PerlIO_printf(Perl_debug_log,
1592 "'\\n' (=%02X) is not a control\n", (int) i));
1595 /* Add to the list; Separate multiple entries with a blank */
1598 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1600 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1608 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1609 * this locale requires more than one byte, there are going to be
1611 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1612 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1613 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1615 if ( check_for_problems && MB_CUR_MAX > 1
1616 && ! PL_in_utf8_CTYPE_locale
1618 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1619 * locale. Just assume that the implementation for them (plus
1620 * for POSIX) is correct and the > 1 value is spurious. (Since
1621 * these are specially handled to never be considered UTF-8
1622 * locales, as long as this is the only problem, everything
1623 * should work fine */
1624 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1626 multi_byte_locale = TRUE;
1631 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1632 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1633 PL_warn_locale = Perl_newSVpvf(aTHX_
1634 "Locale '%s' contains (at least) the following characters"
1635 " which have\nnon-standard meanings: %s\nThe Perl program"
1636 " will use the standard meanings",
1637 newctype, bad_chars_list);
1641 PL_warn_locale = Perl_newSVpvf(aTHX_
1642 "Locale '%s' may not work well.%s%s%s\n",
1645 ? " Some characters in it are not recognized by"
1649 ? "\nThe following characters (and maybe others)"
1650 " may not have the same meaning as the Perl"
1651 " program expects:\n"
1659 # ifdef HAS_NL_LANGINFO
1661 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1662 /* parameter FALSE is a don't care here */
1663 my_nl_langinfo(PERL_CODESET, FALSE));
1667 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1669 /* If we are actually in the scope of the locale or are debugging,
1670 * output the message now. If not in that scope, we save the
1671 * message to be output at the first operation using this locale,
1672 * if that actually happens. Most programs don't use locales, so
1673 * they are immune to bad ones. */
1674 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1676 /* The '0' below suppresses a bogus gcc compiler warning */
1677 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1679 if (IN_LC(LC_CTYPE)) {
1680 SvREFCNT_dec_NN(PL_warn_locale);
1681 PL_warn_locale = NULL;
1687 #endif /* USE_LOCALE_CTYPE */
1692 Perl__warn_problematic_locale()
1695 #ifdef USE_LOCALE_CTYPE
1699 /* Internal-to-core function that outputs the message in PL_warn_locale,
1700 * and then NULLS it. Should be called only through the macro
1701 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1703 if (PL_warn_locale) {
1704 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1705 SvPVX(PL_warn_locale),
1706 0 /* dummy to avoid compiler warning */ );
1707 SvREFCNT_dec_NN(PL_warn_locale);
1708 PL_warn_locale = NULL;
1716 S_new_collate(pTHX_ const char *newcoll)
1719 #ifndef USE_LOCALE_COLLATE
1721 PERL_UNUSED_ARG(newcoll);
1722 PERL_UNUSED_CONTEXT;
1726 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1727 * core Perl this and that 'newcoll' is the name of the new locale.
1729 * The design of locale collation is that every locale change is given an
1730 * index 'PL_collation_ix'. The first time a string particpates in an
1731 * operation that requires collation while locale collation is active, it
1732 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1733 * magic includes the collation index, and the transformation of the string
1734 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1735 * instead of the string itself. If a string changes, the magic is
1736 * cleared. The next time the locale changes, the index is incremented,
1737 * and so we know during a comparison that the transformation is not
1738 * necessarily still valid, and so is recomputed. Note that if the locale
1739 * changes enough times, the index could wrap (a U32), and it is possible
1740 * that a transformation would improperly be considered valid, leading to
1741 * an unlikely bug */
1744 if (PL_collation_name) {
1746 Safefree(PL_collation_name);
1747 PL_collation_name = NULL;
1749 PL_collation_standard = TRUE;
1750 is_standard_collation:
1751 PL_collxfrm_base = 0;
1752 PL_collxfrm_mult = 2;
1753 PL_in_utf8_COLLATE_locale = FALSE;
1754 PL_strxfrm_NUL_replacement = '\0';
1755 PL_strxfrm_max_cp = 0;
1759 /* If this is not the same locale as currently, set the new one up */
1760 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1762 Safefree(PL_collation_name);
1763 PL_collation_name = stdize_locale(savepv(newcoll));
1764 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1765 if (PL_collation_standard) {
1766 goto is_standard_collation;
1769 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1770 PL_strxfrm_NUL_replacement = '\0';
1771 PL_strxfrm_max_cp = 0;
1773 /* A locale collation definition includes primary, secondary, tertiary,
1774 * etc. weights for each character. To sort, the primary weights are
1775 * used, and only if they compare equal, then the secondary weights are
1776 * used, and only if they compare equal, then the tertiary, etc.
1778 * strxfrm() works by taking the input string, say ABC, and creating an
1779 * output transformed string consisting of first the primary weights,
1780 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1781 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1782 * may not have weights at every level. In our example, let's say B
1783 * doesn't have a tertiary weight, and A doesn't have a secondary
1784 * weight. The constructed string is then going to be
1785 * A¹B¹C¹ B²C² A³C³ ....
1786 * This has the desired effect that strcmp() will look at the secondary
1787 * or tertiary weights only if the strings compare equal at all higher
1788 * priority weights. The spaces shown here, like in
1790 * are not just for readability. In the general case, these must
1791 * actually be bytes, which we will call here 'separator weights'; and
1792 * they must be smaller than any other weight value, but since these
1793 * are C strings, only the terminating one can be a NUL (some
1794 * implementations may include a non-NUL separator weight just before
1795 * the NUL). Implementations tend to reserve 01 for the separator
1796 * weights. They are needed so that a shorter string's secondary
1797 * weights won't be misconstrued as primary weights of a longer string,
1798 * etc. By making them smaller than any other weight, the shorter
1799 * string will sort first. (Actually, if all secondary weights are
1800 * smaller than all primary ones, there is no need for a separator
1801 * weight between those two levels, etc.)
1803 * The length of the transformed string is roughly a linear function of
1804 * the input string. It's not exactly linear because some characters
1805 * don't have weights at all levels. When we call strxfrm() we have to
1806 * allocate some memory to hold the transformed string. The
1807 * calculations below try to find coefficients 'm' and 'b' for this
1808 * locale so that m*x + b equals how much space we need, given the size
1809 * of the input string in 'x'. If we calculate too small, we increase
1810 * the size as needed, and call strxfrm() again, but it is better to
1811 * get it right the first time to avoid wasted expensive string
1812 * transformations. */
1815 /* We use the string below to find how long the tranformation of it
1816 * is. Almost all locales are supersets of ASCII, or at least the
1817 * ASCII letters. We use all of them, half upper half lower,
1818 * because if we used fewer, we might hit just the ones that are
1819 * outliers in a particular locale. Most of the strings being
1820 * collated will contain a preponderance of letters, and even if
1821 * they are above-ASCII, they are likely to have the same number of
1822 * weight levels as the ASCII ones. It turns out that digits tend
1823 * to have fewer levels, and some punctuation has more, but those
1824 * are relatively sparse in text, and khw believes this gives a
1825 * reasonable result, but it could be changed if experience so
1827 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1828 char * x_longer; /* Transformed 'longer' */
1829 Size_t x_len_longer; /* Length of 'x_longer' */
1831 char * x_shorter; /* We also transform a substring of 'longer' */
1832 Size_t x_len_shorter;
1834 /* _mem_collxfrm() is used get the transformation (though here we
1835 * are interested only in its length). It is used because it has
1836 * the intelligence to handle all cases, but to work, it needs some
1837 * values of 'm' and 'b' to get it started. For the purposes of
1838 * this calculation we use a very conservative estimate of 'm' and
1839 * 'b'. This assumes a weight can be multiple bytes, enough to
1840 * hold any UV on the platform, and there are 5 levels, 4 weight
1841 * bytes, and a trailing NUL. */
1842 PL_collxfrm_base = 5;
1843 PL_collxfrm_mult = 5 * sizeof(UV);
1845 /* Find out how long the transformation really is */
1846 x_longer = _mem_collxfrm(longer,
1850 /* We avoid converting to UTF-8 in the
1851 * called function by telling it the
1852 * string is in UTF-8 if the locale is a
1853 * UTF-8 one. Since the string passed
1854 * here is invariant under UTF-8, we can
1855 * claim it's UTF-8 even though it isn't.
1857 PL_in_utf8_COLLATE_locale);
1860 /* Find out how long the transformation of a substring of 'longer'
1861 * is. Together the lengths of these transformations are
1862 * sufficient to calculate 'm' and 'b'. The substring is all of
1863 * 'longer' except the first character. This minimizes the chances
1864 * of being swayed by outliers */
1865 x_shorter = _mem_collxfrm(longer + 1,
1868 PL_in_utf8_COLLATE_locale);
1869 Safefree(x_shorter);
1871 /* If the results are nonsensical for this simple test, the whole
1872 * locale definition is suspect. Mark it so that locale collation
1873 * is not active at all for it. XXX Should we warn? */
1874 if ( x_len_shorter == 0
1875 || x_len_longer == 0
1876 || x_len_shorter >= x_len_longer)
1878 PL_collxfrm_mult = 0;
1879 PL_collxfrm_base = 0;
1882 SSize_t base; /* Temporary */
1884 /* We have both: m * strlen(longer) + b = x_len_longer
1885 * m * strlen(shorter) + b = x_len_shorter;
1886 * subtracting yields:
1887 * m * (strlen(longer) - strlen(shorter))
1888 * = x_len_longer - x_len_shorter
1889 * But we have set things up so that 'shorter' is 1 byte smaller
1890 * than 'longer'. Hence:
1891 * m = x_len_longer - x_len_shorter
1893 * But if something went wrong, make sure the multiplier is at
1896 if (x_len_longer > x_len_shorter) {
1897 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1900 PL_collxfrm_mult = 1;
1905 * but in case something has gone wrong, make sure it is
1907 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1912 /* Add 1 for the trailing NUL */
1913 PL_collxfrm_base = base + 1;
1918 if (DEBUG_L_TEST || debug_initialization) {
1919 PerlIO_printf(Perl_debug_log,
1920 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1922 " collate multipler=%zu, collate base=%zu\n",
1924 PL_in_utf8_COLLATE_locale,
1925 x_len_shorter, x_len_longer,
1926 PL_collxfrm_mult, PL_collxfrm_base);
1933 #endif /* USE_LOCALE_COLLATE */
1940 S_win32_setlocale(pTHX_ int category, const char* locale)
1942 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
1943 * difference between the two unless the input locale is "", which normally
1944 * means on Windows to get the machine default, which is set via the
1945 * computer's "Regional and Language Options" (or its current equivalent).
1946 * In POSIX, it instead means to find the locale from the user's
1947 * environment. This routine changes the Windows behavior to first look in
1948 * the environment, and, if anything is found, use that instead of going to
1949 * the machine default. If there is no environment override, the machine
1950 * default is used, by calling the real setlocale() with "".
1952 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1953 * use the particular category's variable if set; otherwise to use the LANG
1956 bool override_LC_ALL = FALSE;
1960 if (locale && strEQ(locale, "")) {
1964 locale = PerlEnv_getenv("LC_ALL");
1966 if (category == LC_ALL) {
1967 override_LC_ALL = TRUE;
1973 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1974 if (category == categories[i]) {
1975 locale = PerlEnv_getenv(category_names[i]);
1980 locale = PerlEnv_getenv("LANG");
1996 result = setlocale(category, locale);
1997 DEBUG_L(STMT_START {
1999 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2000 setlocale_debug_string(category, locale, result));
2004 if (! override_LC_ALL) {
2008 /* Here the input category was LC_ALL, and we have set it to what is in the
2009 * LANG variable or the system default if there is no LANG. But these have
2010 * lower priority than the other LC_foo variables, so override it for each
2011 * one that is set. (If they are set to "", it means to use the same thing
2012 * we just set LC_ALL to, so can skip) */
2014 for (i = 0; i < LC_ALL_INDEX; i++) {
2015 result = PerlEnv_getenv(category_names[i]);
2016 if (result && strNE(result, "")) {
2017 setlocale(categories[i], result);
2018 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2020 setlocale_debug_string(categories[i], result, "not captured")));
2024 result = setlocale(LC_ALL, NULL);
2025 DEBUG_L(STMT_START {
2027 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2029 setlocale_debug_string(LC_ALL, NULL, result));
2040 =head1 Locale-related functions and macros
2042 =for apidoc Perl_setlocale
2044 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2045 taking the same parameters, and returning the same information, except that it
2046 returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2047 perl keeps that locale category as C<C>, changing it briefly during the
2048 operations where the underlying one is required.
2050 Another reason it isn't completely a drop-in replacement is that it is
2051 declared to return S<C<const char *>>, whereas the system setlocale omits the
2052 C<const>. (If it were being written today, plain setlocale would be declared
2053 const, since it is illegal to change the information it returns; doing so leads
2056 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2057 C<setlocale> can be completely ineffective on some platforms under some
2060 C<Perl_setlocale> should not be used to change the locale except on systems
2061 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2062 the system C<setlocale()> is ineffective, returning the wrong information, and
2063 failing to actually change the locale. C<Perl_setlocale>, however works
2064 properly in all circumstances.
2066 The return points to a per-thread static buffer, which is overwritten the next
2067 time C<Perl_setlocale> is called from the same thread.
2074 Perl_setlocale(const int category, const char * locale)
2076 /* This wraps POSIX::setlocale() */
2078 const char * retval;
2079 const char * newlocale;
2081 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2084 #ifdef USE_LOCALE_NUMERIC
2086 /* A NULL locale means only query what the current one is. We have the
2087 * LC_NUMERIC name saved, because we are normally switched into the C
2088 * locale for it. For an LC_ALL query, switch back to get the correct
2089 * results. All other categories don't require special handling */
2090 if (locale == NULL) {
2091 if (category == LC_NUMERIC) {
2093 /* We don't have to copy this return value, as it is a per-thread
2094 * variable, and won't change until a future setlocale */
2095 return PL_numeric_name;
2100 else if (category == LC_ALL) {
2101 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2110 retval = do_setlocale_r(category, locale);
2113 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2115 if (locale == NULL && category == LC_ALL) {
2116 RESTORE_LC_NUMERIC();
2121 DEBUG_L(PerlIO_printf(Perl_debug_log,
2122 "%s:%d: %s\n", __FILE__, __LINE__,
2123 setlocale_debug_string(category, locale, retval)));
2131 save_to_buffer(retval, &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2132 retval = PL_setlocale_buf;
2134 /* If locale == NULL, we are just querying the state */
2135 if (locale == NULL) {
2139 /* Now that have switched locales, we have to update our records to
2144 #ifdef USE_LOCALE_CTYPE
2151 #ifdef USE_LOCALE_COLLATE
2154 new_collate(retval);
2158 #ifdef USE_LOCALE_NUMERIC
2161 new_numeric(retval);
2169 /* LC_ALL updates all the things we care about. The values may not
2170 * be the same as 'retval', as the locale "" may have set things
2173 # ifdef USE_LOCALE_CTYPE
2175 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2176 new_ctype(newlocale);
2178 # endif /* USE_LOCALE_CTYPE */
2179 # ifdef USE_LOCALE_COLLATE
2181 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2182 new_collate(newlocale);
2185 # ifdef USE_LOCALE_NUMERIC
2187 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2188 new_numeric(newlocale);
2190 # endif /* USE_LOCALE_NUMERIC */
2201 PERL_STATIC_INLINE const char *
2202 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2204 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2205 * growing it if necessary */
2207 const Size_t string_size = strlen(string) + offset + 1;
2209 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2211 if (*buf_size == 0) {
2212 Newx(*buf, string_size, char);
2213 *buf_size = string_size;
2215 else if (string_size > *buf_size) {
2216 Renew(*buf, string_size, char);
2217 *buf_size = string_size;
2220 Copy(string, *buf + offset, string_size - offset, char);
2226 =for apidoc Perl_langinfo
2228 This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2229 taking the same C<item> parameter values, and returning the same information.
2230 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2231 of Perl's locale handling from your code, and can be used on systems that lack
2232 a native C<nl_langinfo>.
2240 It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
2241 without you having to write extra code. The reason for the extra code would be
2242 because these are from the C<LC_NUMERIC> locale category, which is normally
2243 kept set to the C locale by Perl, no matter what the underlying locale is
2244 supposed to be, and so to get the expected results, you have to temporarily
2245 toggle into the underlying locale, and later toggle back. (You could use
2246 plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
2247 but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
2248 keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
2249 expecting the radix (decimal point) character to be a dot.)
2253 Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
2254 makes your code more portable. Of the fifty-some possible items specified by
2255 the POSIX 2008 standard,
2256 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2257 only two are completely unimplemented. It uses various techniques to recover
2258 the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
2259 both of which are specified in C89, so should be always be available. Later
2260 C<strftime()> versions have additional capabilities; C<""> is returned for
2261 those not available on your system.
2263 It is important to note that on such systems, this calls C<localeconv>, and so
2264 overwrites the static buffer returned from previous explicit calls to that
2265 function. Thus, if the program doesn't use or save the information from an
2266 explicit C<localeconv> call (which good practice suggests should be done
2267 anyway), use of this function can break it.
2269 The details for those items which may differ from what this emulation returns
2270 and what a native C<nl_langinfo()> would return are:
2278 Unimplemented, so returns C<"">.
2288 Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
2289 removed from POSIX 2008, and are retained for backwards compatibility. Your
2290 platform's C<nl_langinfo> may not support them.
2294 Always evaluates to C<%x>, the locale's appropriate date representation.
2298 Always evaluates to C<%X>, the locale's appropriate time representation.
2302 Always evaluates to C<%c>, the locale's appropriate date and time
2307 The return may be incorrect for those rare locales where the currency symbol
2308 replaces the radix character.
2309 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2310 to work differently.
2314 Currently this gives the same results as Linux does.
2315 Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2316 to work differently.
2322 =item C<ERA_D_T_FMT>
2326 These are derived by using C<strftime()>, and not all versions of that function
2327 know about them. C<""> is returned for these on such systems.
2331 When using C<Perl_langinfo> on systems that don't have a native
2332 C<nl_langinfo()>, you must
2334 #include "perl_langinfo.h"
2336 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2337 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2338 C<langinfo.h> imports into the namespace for code that doesn't need it.)
2340 You also should not use the bare C<langinfo.h> item names, but should preface
2341 them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2342 The C<PERL_I<foo>> versions will also work for this function on systems that do
2343 have a native C<nl_langinfo>.
2347 It is thread-friendly, returning its result in a buffer that won't be
2348 overwritten by another thread, so you don't have to code for that possibility.
2349 The buffer can be overwritten by the next call to C<nl_langinfo> or
2350 C<Perl_langinfo> in the same thread.
2354 ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
2355 *>>, but you are (only by documentation) forbidden to write into the buffer.
2356 By declaring this C<const>, the compiler enforces this restriction. The extra
2357 C<const> is why this isn't an unequivocal drop-in replacement for
2362 The original impetus for C<Perl_langinfo()> was so that code that needs to
2363 find out the current currency symbol, floating point radix character, or digit
2364 grouping separator can use, on all systems, the simpler and more
2365 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2366 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2367 is better to use the methods given in L<perlcall> to call
2368 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2375 #ifdef HAS_NL_LANGINFO
2376 Perl_langinfo(const nl_item item)
2378 Perl_langinfo(const int item)
2381 return my_nl_langinfo(item, TRUE);
2385 #ifdef HAS_NL_LANGINFO
2386 S_my_nl_langinfo(const nl_item item, bool toggle)
2388 S_my_nl_langinfo(const int item, bool toggle)
2391 const char * retval;
2394 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2395 * two items, and only if not already there */
2396 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2397 || PL_numeric_underlying))
2402 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2403 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2404 || ! defined(HAS_POSIX_2008_LOCALE)
2406 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2407 * for those items dependent on it. This must be copied to a buffer before
2408 * switching back, as some systems destroy the buffer when setlocale() is
2412 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2415 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2418 LOCALE_LOCK; /* Prevent interference from another thread executing
2419 this code section (the only call to nl_langinfo in
2422 retval = nl_langinfo(item);
2424 # ifdef USE_ITHREADS
2426 /* Copy to a per-thread buffer */
2427 save_to_buffer(retval, &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2428 retval = PL_langinfo_buf;
2435 RESTORE_LC_NUMERIC();
2439 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2442 bool do_free = FALSE;
2443 locale_t cur = uselocale((locale_t) 0);
2445 if (cur == LC_GLOBAL_LOCALE) {
2446 cur = duplocale(LC_GLOBAL_LOCALE);
2451 if (PL_underlying_numeric_obj) {
2452 cur = PL_underlying_numeric_obj;
2455 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2460 /* We don't have to copy it to a buffer, as this is a thread-safe
2461 * function which Configure has made sure of */
2462 retval = nl_langinfo_l(item, cur);
2471 if (strEQ(retval, "")) {
2472 if (item == PERL_YESSTR) {
2475 if (item == PERL_NOSTR) {
2482 #else /* Below, emulate nl_langinfo as best we can */
2486 # ifdef HAS_LOCALECONV
2488 const struct lconv* lc;
2489 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2492 # ifdef HAS_STRFTIME
2495 bool return_format = FALSE; /* Return the %format, not the value */
2496 const char * format;
2500 /* We copy the results to a per-thread buffer, even if not
2501 * multi-threaded. This is in part to simplify this code, and partly
2502 * because we need a buffer anyway for strftime(), and partly because a
2503 * call of localeconv() could otherwise wipe out the buffer, and the
2504 * programmer would not be expecting this, as this is a nl_langinfo()
2505 * substitute after all, so s/he might be thinking their localeconv()
2506 * is safe until another localeconv() call. */
2511 /* These 2 are unimplemented */
2513 case PERL_ERA: /* For use with strftime() %E modifier */
2518 /* We use only an English set, since we don't know any more */
2519 case PERL_YESEXPR: return "^[+1yY]";
2520 case PERL_YESSTR: return "yes";
2521 case PERL_NOEXPR: return "^[-0nN]";
2522 case PERL_NOSTR: return "no";
2524 # ifdef HAS_LOCALECONV
2528 /* We don't bother with localeconv_l() because any system that
2529 * has it is likely to also have nl_langinfo() */
2531 LOCALE_LOCK; /* Prevent interference with other threads
2532 using localeconv() */
2536 || ! lc->currency_symbol
2537 || strEQ("", lc->currency_symbol))
2543 /* Leave the first spot empty to be filled in below */
2544 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2545 &PL_langinfo_bufsize, 1);
2546 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2547 { /* khw couldn't figure out how the localedef specifications
2548 would show that the $ should replace the radix; this is
2549 just a guess as to how it might work.*/
2550 *PL_langinfo_buf = '.';
2552 else if (lc->p_cs_precedes) {
2553 *PL_langinfo_buf = '-';
2556 *PL_langinfo_buf = '+';
2562 case PERL_RADIXCHAR:
2566 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2569 LOCALE_LOCK; /* Prevent interference with other threads
2570 using localeconv() */
2577 retval = (item == PERL_RADIXCHAR)
2579 : lc->thousands_sep;
2585 save_to_buffer(retval, &PL_langinfo_buf,
2586 &PL_langinfo_bufsize, 0);
2591 RESTORE_LC_NUMERIC();
2597 # ifdef HAS_STRFTIME
2599 /* These are defined by C89, so we assume that strftime supports
2600 * them, and so are returned unconditionally; they may not be what
2601 * the locale actually says, but should give good enough results
2602 * for someone using them as formats (as opposed to trying to parse
2603 * them to figure out what the locale says). The other format
2604 * items are actually tested to verify they work on the platform */
2605 case PERL_D_FMT: return "%x";
2606 case PERL_T_FMT: return "%X";
2607 case PERL_D_T_FMT: return "%c";
2609 /* These formats are only available in later strfmtime's */
2610 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2611 case PERL_T_FMT_AMPM:
2613 /* The rest can be gotten from most versions of strftime(). */
2614 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2615 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2617 case PERL_ALT_DIGITS:
2618 case PERL_AM_STR: case PERL_PM_STR:
2619 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2620 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2621 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2622 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2623 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2624 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2625 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2626 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2627 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2632 init_tm(&tm); /* Precaution against core dumps */
2636 tm.tm_year = 2017 - 1900;
2643 "panic: %s: %d: switch case: %d problem",
2644 __FILE__, __LINE__, item);
2645 NOT_REACHED; /* NOTREACHED */
2647 case PERL_PM_STR: tm.tm_hour = 18;
2652 case PERL_ABDAY_7: tm.tm_wday++;
2653 case PERL_ABDAY_6: tm.tm_wday++;
2654 case PERL_ABDAY_5: tm.tm_wday++;
2655 case PERL_ABDAY_4: tm.tm_wday++;
2656 case PERL_ABDAY_3: tm.tm_wday++;
2657 case PERL_ABDAY_2: tm.tm_wday++;
2662 case PERL_DAY_7: tm.tm_wday++;
2663 case PERL_DAY_6: tm.tm_wday++;
2664 case PERL_DAY_5: tm.tm_wday++;
2665 case PERL_DAY_4: tm.tm_wday++;
2666 case PERL_DAY_3: tm.tm_wday++;
2667 case PERL_DAY_2: tm.tm_wday++;
2672 case PERL_ABMON_12: tm.tm_mon++;
2673 case PERL_ABMON_11: tm.tm_mon++;
2674 case PERL_ABMON_10: tm.tm_mon++;
2675 case PERL_ABMON_9: tm.tm_mon++;
2676 case PERL_ABMON_8: tm.tm_mon++;
2677 case PERL_ABMON_7: tm.tm_mon++;
2678 case PERL_ABMON_6: tm.tm_mon++;
2679 case PERL_ABMON_5: tm.tm_mon++;
2680 case PERL_ABMON_4: tm.tm_mon++;
2681 case PERL_ABMON_3: tm.tm_mon++;
2682 case PERL_ABMON_2: tm.tm_mon++;
2687 case PERL_MON_12: tm.tm_mon++;
2688 case PERL_MON_11: tm.tm_mon++;
2689 case PERL_MON_10: tm.tm_mon++;
2690 case PERL_MON_9: tm.tm_mon++;
2691 case PERL_MON_8: tm.tm_mon++;
2692 case PERL_MON_7: tm.tm_mon++;
2693 case PERL_MON_6: tm.tm_mon++;
2694 case PERL_MON_5: tm.tm_mon++;
2695 case PERL_MON_4: tm.tm_mon++;
2696 case PERL_MON_3: tm.tm_mon++;
2697 case PERL_MON_2: tm.tm_mon++;
2702 case PERL_T_FMT_AMPM:
2704 return_format = TRUE;
2707 case PERL_ERA_D_FMT:
2709 return_format = TRUE;
2712 case PERL_ERA_T_FMT:
2714 return_format = TRUE;
2717 case PERL_ERA_D_T_FMT:
2719 return_format = TRUE;
2722 case PERL_ALT_DIGITS:
2724 format = "%Ow"; /* Find the alternate digit for 0 */
2728 /* We can't use my_strftime() because it doesn't look at
2730 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2733 /* A zero return means one of:
2734 * a) there wasn't enough space in PL_langinfo_buf
2735 * b) the format, like a plain %p, returns empty
2736 * c) it was an illegal format, though some
2737 * implementations of strftime will just return the
2738 * illegal format as a plain character sequence.
2740 * To quickly test for case 'b)', try again but precede
2741 * the format with a plain character. If that result is
2742 * still empty, the problem is either 'a)' or 'c)' */
2744 Size_t format_size = strlen(format) + 1;
2745 Size_t mod_size = format_size + 1;
2749 Newx(mod_format, mod_size, char);
2750 Newx(temp_result, PL_langinfo_bufsize, char);
2752 my_strlcpy(mod_format + 1, format, mod_size);
2753 len = strftime(temp_result,
2754 PL_langinfo_bufsize,
2756 Safefree(mod_format);
2757 Safefree(temp_result);
2759 /* If 'len' is non-zero, it means that we had a case like
2760 * %p which means the current locale doesn't use a.m. or
2761 * p.m., and that is valid */
2764 /* Here, still didn't work. If we get well beyond a
2765 * reasonable size, bail out to prevent an infinite
2768 if (PL_langinfo_bufsize > 100 * format_size) {
2769 *PL_langinfo_buf = '\0';
2772 /* Double the buffer size to retry; Add 1 in case
2773 * original was 0, so we aren't stuck at 0. */
2774 PL_langinfo_bufsize *= 2;
2775 PL_langinfo_bufsize++;
2776 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2784 /* Here, we got a result.
2786 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2787 * alternate format for wday 0. If the value is the same as
2788 * the normal 0, there isn't an alternate, so clear the buffer.
2790 if ( item == PERL_ALT_DIGITS
2791 && strEQ(PL_langinfo_buf, "0"))
2793 *PL_langinfo_buf = '\0';
2796 /* ALT_DIGITS is problematic. Experiments on it showed that
2797 * strftime() did not always work properly when going from
2798 * alt-9 to alt-10. Only a few locales have this item defined,
2799 * and in all of them on Linux that khw was able to find,
2800 * nl_langinfo() merely returned the alt-0 character, possibly
2801 * doubled. Most Unicode digits are in blocks of 10
2802 * consecutive code points, so that is sufficient information
2803 * for those scripts, as we can infer alt-1, alt-2, .... But
2804 * for a Japanese locale, a CJK ideographic 0 is returned, and
2805 * the CJK digits are not in code point order, so you can't
2806 * really infer anything. The localedef for this locale did
2807 * specify the succeeding digits, so that strftime() works
2808 * properly on them, without needing to infer anything. But
2809 * the nl_langinfo() return did not give sufficient information
2810 * for the caller to understand what's going on. So until
2811 * there is evidence that it should work differently, this
2812 * returns the alt-0 string for ALT_DIGITS.
2814 * wday was chosen because its range is all a single digit.
2815 * Things like tm_sec have two digits as the minimum: '00' */
2819 /* If to return the format, not the value, overwrite the buffer
2820 * with it. But some strftime()s will keep the original format
2821 * if illegal, so change those to "" */
2822 if (return_format) {
2823 if (strEQ(PL_langinfo_buf, format)) {
2824 *PL_langinfo_buf = '\0';
2827 save_to_buffer(format, &PL_langinfo_buf,
2828 &PL_langinfo_bufsize, 0);
2839 return PL_langinfo_buf;
2846 * Initialize locale awareness.
2849 Perl_init_i18nl10n(pTHX_ int printwarn)
2853 * 0 if not to output warning when setup locale is bad
2854 * 1 if to output warning based on value of PERL_BADLANG
2855 * >1 if to output regardless of PERL_BADLANG
2858 * 1 = set ok or not applicable,
2859 * 0 = fallback to a locale of lower priority
2860 * -1 = fallback to all locales failed, not even to the C locale
2862 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2863 * set, debugging information is output.
2865 * This looks more complicated than it is, mainly due to the #ifdefs.
2867 * We try to set LC_ALL to the value determined by the environment. If
2868 * there is no LC_ALL on this platform, we try the individual categories we
2869 * know about. If this works, we are done.
2871 * But if it doesn't work, we have to do something else. We search the
2872 * environment variables ourselves instead of relying on the system to do
2873 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2874 * think there is one), and the ultimate fallback "C". This is all done in
2875 * the same loop as above to avoid duplicating code, but it makes things
2876 * more complex. The 'trial_locales' array is initialized with just one
2877 * element; it causes the behavior described in the paragraph above this to
2878 * happen. If that fails, we add elements to 'trial_locales', and do extra
2879 * loop iterations to cause the behavior described in this paragraph.
2881 * On Ultrix, the locale MUST come from the environment, so there is
2882 * preliminary code to set it. I (khw) am not sure that it is necessary,
2883 * and that this couldn't be folded into the loop, but barring any real
2884 * platforms to test on, it's staying as-is
2886 * A slight complication is that in embedded Perls, the locale may already
2887 * be set-up, and we don't want to get it from the normal environment
2888 * variables. This is handled by having a special environment variable
2889 * indicate we're in this situation. We simply set setlocale's 2nd
2890 * parameter to be a NULL instead of "". That indicates to setlocale that
2891 * it is not to change anything, but to return the current value,
2892 * effectively initializing perl's db to what the locale already is.
2894 * We play the same trick with NULL if a LC_ALL succeeds. We call
2895 * setlocale() on the individual categores with NULL to get their existing
2896 * values for our db, instead of trying to change them.
2903 PERL_UNUSED_ARG(printwarn);
2905 #else /* USE_LOCALE */
2908 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
2912 /* NULL uses the existing already set up locale */
2913 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2916 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2917 unsigned int trial_locales_count;
2918 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2919 const char * const lang = savepv(PerlEnv_getenv("LANG"));
2920 bool setlocale_failure = FALSE;
2923 /* A later getenv() could zap this, so only use here */
2924 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2926 const bool locwarn = (printwarn > 1
2928 && ( ! bad_lang_use_once
2930 /* disallow with "" or "0" */
2932 && strNE("0", bad_lang_use_once)))));
2934 /* setlocale() return vals; not copied so must be looked at immediately */
2935 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
2937 /* current locale for given category; should have been copied so aren't
2939 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
2943 /* In some systems you can find out the system default locale
2944 * and use that as the fallback locale. */
2945 # define SYSTEM_DEFAULT_LOCALE
2947 # ifdef SYSTEM_DEFAULT_LOCALE
2949 const char *system_default_locale = NULL;
2954 # define DEBUG_LOCALE_INIT(a,b,c)
2957 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
2959 # define DEBUG_LOCALE_INIT(category, locale, result) \
2961 if (debug_initialization) { \
2962 PerlIO_printf(Perl_debug_log, \
2964 __FILE__, __LINE__, \
2965 setlocale_debug_string(category, \
2971 /* Make sure the parallel arrays are properly set up */
2972 # ifdef USE_LOCALE_NUMERIC
2973 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2974 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
2975 # ifdef USE_POSIX_2008_LOCALE
2976 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
2979 # ifdef USE_LOCALE_CTYPE
2980 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2981 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
2982 # ifdef USE_POSIX_2008_LOCALE
2983 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
2986 # ifdef USE_LOCALE_COLLATE
2987 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2988 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
2989 # ifdef USE_POSIX_2008_LOCALE
2990 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
2993 # ifdef USE_LOCALE_TIME
2994 assert(categories[LC_TIME_INDEX] == LC_TIME);
2995 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2996 # ifdef USE_POSIX_2008_LOCALE
2997 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3000 # ifdef USE_LOCALE_MESSAGES
3001 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3002 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3003 # ifdef USE_POSIX_2008_LOCALE
3004 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3007 # ifdef USE_LOCALE_MONETARY
3008 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3009 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3010 # ifdef USE_POSIX_2008_LOCALE
3011 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3014 # ifdef USE_LOCALE_ADDRESS
3015 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3016 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3017 # ifdef USE_POSIX_2008_LOCALE
3018 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3021 # ifdef USE_LOCALE_IDENTIFICATION
3022 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3023 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3024 # ifdef USE_POSIX_2008_LOCALE
3025 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3028 # ifdef USE_LOCALE_MEASUREMENT
3029 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3030 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3031 # ifdef USE_POSIX_2008_LOCALE
3032 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3035 # ifdef USE_LOCALE_PAPER
3036 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3037 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3038 # ifdef USE_POSIX_2008_LOCALE
3039 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3042 # ifdef USE_LOCALE_TELEPHONE
3043 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3044 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3045 # ifdef USE_POSIX_2008_LOCALE
3046 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3050 assert(categories[LC_ALL_INDEX] == LC_ALL);
3051 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3052 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3053 # ifdef USE_POSIX_2008_LOCALE
3054 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3057 # endif /* DEBUGGING */
3059 /* Initialize the cache of the program's UTF-8ness for the always known
3060 * locales C and POSIX */
3061 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3062 sizeof(PL_locale_utf8ness));
3064 # ifdef USE_THREAD_SAFE_LOCALE
3067 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3071 # if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3073 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3074 if (! PL_C_locale_obj) {
3075 Perl_croak_nocontext(
3076 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3078 if (DEBUG_Lv_TEST || debug_initialization) {
3079 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3084 PL_numeric_radix_sv = newSVpvs(".");
3086 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3088 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3089 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3092 # ifdef LOCALE_ENVIRON_REQUIRED
3095 * Ultrix setlocale(..., "") fails if there are no environment
3096 * variables from which to get a locale name.
3100 # error Ultrix without LC_ALL not implemented
3106 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3107 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3108 if (sl_result[LC_ALL_INDEX])
3111 setlocale_failure = TRUE;
3113 if (! setlocale_failure) {
3114 const char * locale_param;
3115 for (i = 0; i < LC_ALL_INDEX; i++) {
3116 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3119 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3120 if (! sl_result[i]) {
3121 setlocale_failure = TRUE;
3123 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3128 # endif /* LC_ALL */
3129 # endif /* LOCALE_ENVIRON_REQUIRED */
3131 /* We try each locale in the list until we get one that works, or exhaust
3132 * the list. Normally the loop is executed just once. But if setting the
3133 * locale fails, inside the loop we add fallback trials to the array and so
3134 * will execute the loop multiple times */
3135 trial_locales[0] = setlocale_init;
3136 trial_locales_count = 1;
3138 for (i= 0; i < trial_locales_count; i++) {
3139 const char * trial_locale = trial_locales[i];
3143 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3144 * when i==0, but I (khw) don't think that behavior makes much
3146 setlocale_failure = FALSE;
3148 # ifdef SYSTEM_DEFAULT_LOCALE
3149 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3151 /* On Windows machines, an entry of "" after the 0th means to use
3152 * the system default locale, which we now proceed to get. */
3153 if (strEQ(trial_locale, "")) {
3156 /* Note that this may change the locale, but we are going to do
3157 * that anyway just below */
3158 system_default_locale = do_setlocale_c(LC_ALL, "");
3159 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3161 /* Skip if invalid or if it's already on the list of locales to
3163 if (! system_default_locale) {
3164 goto next_iteration;
3166 for (j = 0; j < trial_locales_count; j++) {
3167 if (strEQ(system_default_locale, trial_locales[j])) {
3168 goto next_iteration;
3172 trial_locale = system_default_locale;
3175 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3177 # endif /* SYSTEM_DEFAULT_LOCALE */
3183 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3184 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3185 if (! sl_result[LC_ALL_INDEX]) {
3186 setlocale_failure = TRUE;
3189 /* Since LC_ALL succeeded, it should have changed all the other
3190 * categories it can to its value; so we massage things so that the
3191 * setlocales below just return their category's current values.
3192 * This adequately handles the case in NetBSD where LC_COLLATE may
3193 * not be defined for a locale, and setting it individually will
3194 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3195 * the POSIX locale. */
3196 trial_locale = NULL;
3199 # endif /* LC_ALL */
3201 if (! setlocale_failure) {
3203 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3205 = savepv(do_setlocale_r(categories[j], trial_locale));
3206 if (! curlocales[j]) {
3207 setlocale_failure = TRUE;
3209 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3212 if (! setlocale_failure) { /* All succeeded */
3213 break; /* Exit trial_locales loop */
3217 /* Here, something failed; will need to try a fallback. */
3223 if (locwarn) { /* Output failure info only on the first one */
3227 PerlIO_printf(Perl_error_log,
3228 "perl: warning: Setting locale failed.\n");
3230 # else /* !LC_ALL */
3232 PerlIO_printf(Perl_error_log,
3233 "perl: warning: Setting locale failed for the categories:\n\t");
3235 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3236 if (! curlocales[j]) {
3237 PerlIO_printf(Perl_error_log, category_names[j]);
3240 Safefree(curlocales[j]);
3244 # endif /* LC_ALL */
3246 PerlIO_printf(Perl_error_log,
3247 "perl: warning: Please check that your locale settings:\n");
3251 PerlIO_printf(Perl_error_log,
3252 "\tLANGUAGE = %c%s%c,\n",
3253 language ? '"' : '(',
3254 language ? language : "unset",
3255 language ? '"' : ')');
3258 PerlIO_printf(Perl_error_log,
3259 "\tLC_ALL = %c%s%c,\n",
3261 lc_all ? lc_all : "unset",
3262 lc_all ? '"' : ')');
3264 # if defined(USE_ENVIRON_ARRAY)
3269 /* Look through the environment for any variables of the
3270 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3271 * already handled above. These are assumed to be locale
3272 * settings. Output them and their values. */
3273 for (e = environ; *e; e++) {
3274 const STRLEN prefix_len = sizeof("LC_") - 1;
3277 if ( strBEGINs(*e, "LC_")
3278 && ! strBEGINs(*e, "LC_ALL=")
3279 && (uppers_len = strspn(*e + prefix_len,
3280 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3281 && ((*e)[prefix_len + uppers_len] == '='))
3283 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3284 (int) (prefix_len + uppers_len), *e,
3285 *e + prefix_len + uppers_len + 1);
3292 PerlIO_printf(Perl_error_log,
3293 "\t(possibly more locale environment variables)\n");
3297 PerlIO_printf(Perl_error_log,
3298 "\tLANG = %c%s%c\n",
3300 lang ? lang : "unset",
3303 PerlIO_printf(Perl_error_log,
3304 " are supported and installed on your system.\n");
3307 /* Calculate what fallback locales to try. We have avoided this
3308 * until we have to, because failure is quite unlikely. This will
3309 * usually change the upper bound of the loop we are in.
3311 * Since the system's default way of setting the locale has not
3312 * found one that works, We use Perl's defined ordering: LC_ALL,
3313 * LANG, and the C locale. We don't try the same locale twice, so
3314 * don't add to the list if already there. (On POSIX systems, the
3315 * LC_ALL element will likely be a repeat of the 0th element "",
3316 * but there's no harm done by doing it explicitly.
3318 * Note that this tries the LC_ALL environment variable even on
3319 * systems which have no LC_ALL locale setting. This may or may
3320 * not have been originally intentional, but there's no real need
3321 * to change the behavior. */
3323 for (j = 0; j < trial_locales_count; j++) {
3324 if (strEQ(lc_all, trial_locales[j])) {
3328 trial_locales[trial_locales_count++] = lc_all;
3333 for (j = 0; j < trial_locales_count; j++) {
3334 if (strEQ(lang, trial_locales[j])) {
3338 trial_locales[trial_locales_count++] = lang;
3342 # if defined(WIN32) && defined(LC_ALL)
3344 /* For Windows, we also try the system default locale before "C".
3345 * (If there exists a Windows without LC_ALL we skip this because
3346 * it gets too complicated. For those, the "C" is the next
3347 * fallback possibility). The "" is the same as the 0th element of
3348 * the array, but the code at the loop above knows to treat it
3349 * differently when not the 0th */
3350 trial_locales[trial_locales_count++] = "";
3354 for (j = 0; j < trial_locales_count; j++) {
3355 if (strEQ("C", trial_locales[j])) {
3359 trial_locales[trial_locales_count++] = "C";
3362 } /* end of first time through the loop */
3370 } /* end of looping through the trial locales */
3372 if (ok < 1) { /* If we tried to fallback */
3374 if (! setlocale_failure) { /* fallback succeeded */
3375 msg = "Falling back to";
3377 else { /* fallback failed */
3380 /* We dropped off the end of the loop, so have to decrement i to
3381 * get back to the value the last time through */
3385 msg = "Failed to fall back to";
3387 /* To continue, we should use whatever values we've got */
3389 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3390 Safefree(curlocales[j]);
3391 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3392 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3397 const char * description;
3398 const char * name = "";
3399 if (strEQ(trial_locales[i], "C")) {
3400 description = "the standard locale";
3404 # ifdef SYSTEM_DEFAULT_LOCALE
3406 else if (strEQ(trial_locales[i], "")) {
3407 description = "the system default locale";
3408 if (system_default_locale) {
3409 name = system_default_locale;
3413 # endif /* SYSTEM_DEFAULT_LOCALE */
3416 description = "a fallback locale";
3417 name = trial_locales[i];
3419 if (name && strNE(name, "")) {
3420 PerlIO_printf(Perl_error_log,
3421 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3424 PerlIO_printf(Perl_error_log,
3425 "perl: warning: %s %s.\n", msg, description);
3428 } /* End of tried to fallback */
3430 /* Done with finding the locales; update our records */
3432 # ifdef USE_LOCALE_CTYPE
3434 new_ctype(curlocales[LC_CTYPE_INDEX]);
3437 # ifdef USE_LOCALE_COLLATE
3439 new_collate(curlocales[LC_COLLATE_INDEX]);
3442 # ifdef USE_LOCALE_NUMERIC
3444 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3448 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3450 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3452 /* This caches whether each category's locale is UTF-8 or not. This
3453 * may involve changing the locale. It is ok to do this at
3454 * initialization time before any threads have started, but not later
3455 * unless thread-safe operations are used.
3456 * Caching means that if the program heeds our dictate not to change
3457 * locales in threaded applications, this data will remain valid, and
3458 * it may get queried without having to change locales. If the
3459 * environment is such that all categories have the same locale, this
3460 * isn't needed, as the code will not change the locale; but this
3461 * handles the uncommon case where the environment has disparate
3462 * locales for the categories */
3463 (void) _is_cur_LC_category_utf8(categories[i]);
3467 Safefree(curlocales[i]);
3470 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3472 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3473 * locale is UTF-8. The call to new_ctype() just above has already
3474 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3475 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3476 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3477 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3478 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3480 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3481 This is an alternative to using the -C command line switch
3482 (the -C if present will override this). */
3484 const char *p = PerlEnv_getenv("PERL_UNICODE");
3485 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3486 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3500 #endif /* USE_LOCALE */
3503 /* So won't continue to output stuff */
3504 DEBUG_INITIALIZATION_set(FALSE);
3511 #ifdef USE_LOCALE_COLLATE
3514 Perl__mem_collxfrm(pTHX_ const char *input_string,
3515 STRLEN len, /* Length of 'input_string' */
3516 STRLEN *xlen, /* Set to length of returned string
3517 (not including the collation index
3519 bool utf8 /* Is the input in UTF-8? */
3523 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3524 * differences. First, it handles embedded NULs. Second, it allocates a bit
3525 * more memory than needed for the transformed data itself. The real
3526 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3527 * the length of that, and doesn't include the collation index size.
3528 * Please see sv_collxfrm() to see how this is used. */
3530 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3532 char * s = (char *) input_string;
3533 STRLEN s_strlen = strlen(input_string);
3535 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3536 STRLEN length_in_chars;
3537 bool first_time = TRUE; /* Cleared after first loop iteration */
3539 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3541 /* Must be NUL-terminated */
3542 assert(*(input_string + len) == '\0');
3544 /* If this locale has defective collation, skip */
3545 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3546 DEBUG_L(PerlIO_printf(Perl_debug_log,
3547 "_mem_collxfrm: locale's collation is defective\n"));
3551 /* Replace any embedded NULs with the control that sorts before any others.
3552 * This will give as good as possible results on strings that don't
3553 * otherwise contain that character, but otherwise there may be
3554 * less-than-perfect results with that character and NUL. This is
3555 * unavoidable unless we replace strxfrm with our own implementation. */
3556 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3560 STRLEN sans_nuls_len;
3561 int try_non_controls;
3562 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3563 making sure 2nd byte is NUL.
3565 STRLEN this_replacement_len;
3567 /* If we don't know what non-NUL control character sorts lowest for
3568 * this locale, find it */
3569 if (PL_strxfrm_NUL_replacement == '\0') {
3571 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3572 includes the collation index
3575 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3577 /* Unlikely, but it may be that no control will work to replace
3578 * NUL, in which case we instead look for any character. Controls
3579 * are preferred because collation order is, in general, context
3580 * sensitive, with adjoining characters affecting the order, and
3581 * controls are less likely to have such interactions, allowing the
3582 * NUL-replacement to stand on its own. (Another way to look at it
3583 * is to imagine what would happen if the NUL were replaced by a
3584 * combining character; it wouldn't work out all that well.) */
3585 for (try_non_controls = 0;
3586 try_non_controls < 2;
3589 /* Look through all legal code points (NUL isn't) */
3590 for (j = 1; j < 256; j++) {
3591 char * x; /* j's xfrm plus collation index */
3592 STRLEN x_len; /* length of 'x' */
3593 STRLEN trial_len = 1;
3594 char cur_source[] = { '\0', '\0' };
3596 /* Skip non-controls the first time through the loop. The
3597 * controls in a UTF-8 locale are the L1 ones */
3598 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3605 /* Create a 1-char string of the current code point */
3606 cur_source[0] = (char) j;
3608 /* Then transform it */
3609 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3610 0 /* The string is not in UTF-8 */);
3612 /* Ignore any character that didn't successfully transform.
3618 /* If this character's transformation is lower than
3619 * the current lowest, this one becomes the lowest */
3620 if ( cur_min_x == NULL
3621 || strLT(x + COLLXFRM_HDR_LEN,
3622 cur_min_x + COLLXFRM_HDR_LEN))
3624 PL_strxfrm_NUL_replacement = j;
3630 } /* end of loop through all 255 characters */
3632 /* Stop looking if found */
3637 /* Unlikely, but possible, if there aren't any controls that
3638 * work in the locale, repeat the loop, looking for any
3639 * character that works */
3640 DEBUG_L(PerlIO_printf(Perl_debug_log,
3641 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3642 } /* End of loop to try first the controls, then any char */
3645 DEBUG_L(PerlIO_printf(Perl_debug_log,
3646 "_mem_collxfrm: Couldn't find any character to replace"
3647 " embedded NULs in locale %s with", PL_collation_name));
3651 DEBUG_L(PerlIO_printf(Perl_debug_log,
3652 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3653 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3655 Safefree(cur_min_x);
3656 } /* End of determining the character that is to replace NULs */
3658 /* If the replacement is variant under UTF-8, it must match the
3659 * UTF8-ness of the original */
3660 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3661 this_replacement_char[0] =
3662 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3663 this_replacement_char[1] =
3664 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3665 this_replacement_len = 2;
3668 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3669 /* this_replacement_char[1] = '\0' was done at initialization */
3670 this_replacement_len = 1;
3673 /* The worst case length for the replaced string would be if every
3674 * character in it is NUL. Multiply that by the length of each
3675 * replacement, and allow for a trailing NUL */
3676 sans_nuls_len = (len * this_replacement_len) + 1;
3677 Newx(sans_nuls, sans_nuls_len, char);
3680 /* Replace each NUL with the lowest collating control. Loop until have
3681 * exhausted all the NULs */
3682 while (s + s_strlen < e) {
3683 my_strlcat(sans_nuls, s, sans_nuls_len);
3685 /* Do the actual replacement */
3686 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
3688 /* Move past the input NUL */
3690 s_strlen = strlen(s);
3693 /* And add anything that trails the final NUL */
3694 my_strlcat(sans_nuls, s, sans_nuls_len);
3696 /* Switch so below we transform this modified string */
3699 } /* End of replacing NULs */
3701 /* Make sure the UTF8ness of the string and locale match */
3702 if (utf8 != PL_in_utf8_COLLATE_locale) {
3703 /* XXX convert above Unicode to 10FFFF? */
3704 const char * const t = s; /* Temporary so we can later find where the
3707 /* Here they don't match. Change the string's to be what the locale is
3710 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3711 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3714 else { /* locale is not UTF-8; but input is; downgrade the input */
3716 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3718 /* If the downgrade was successful we are done, but if the input
3719 * contains things that require UTF-8 to represent, have to do
3720 * damage control ... */
3721 if (UNLIKELY(utf8)) {
3723 /* What we do is construct a non-UTF-8 string with
3724 * 1) the characters representable by a single byte converted
3725 * to be so (if necessary);
3726 * 2) and the rest converted to collate the same as the
3727 * highest collating representable character. That makes
3728 * them collate at the end. This is similar to how we
3729 * handle embedded NULs, but we use the highest collating
3730 * code point instead of the smallest. Like the NUL case,
3731 * this isn't perfect, but is the best we can reasonably
3732 * do. Every above-255 code point will sort the same as
3733 * the highest-sorting 0-255 code point. If that code
3734 * point can combine in a sequence with some other code
3735 * points for weight calculations, us changing something to
3736 * be it can adversely affect the results. But in most
3737 * cases, it should work reasonably. And note that this is
3738 * really an illegal situation: using code points above 255
3739 * on a locale where only 0-255 are valid. If two strings
3740 * sort entirely equal, then the sort order for the
3741 * above-255 code points will be in code point order. */
3745 /* If we haven't calculated the code point with the maximum
3746 * collating order for this locale, do so now */
3747 if (! PL_strxfrm_max_cp) {
3750 /* The current transformed string that collates the
3751 * highest (except it also includes the prefixed collation
3753 char * cur_max_x = NULL;
3755 /* Look through all legal code points (NUL isn't) */
3756 for (j = 1; j < 256; j++) {
3759 char cur_source[] = { '\0', '\0' };
3761 /* Create a 1-char string of the current code point */
3762 cur_source[0] = (char) j;
3764 /* Then transform it */
3765 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3767 /* If something went wrong (which it shouldn't), just
3768 * ignore this code point */
3773 /* If this character's transformation is higher than
3774 * the current highest, this one becomes the highest */
3775 if ( cur_max_x == NULL
3776 || strGT(x + COLLXFRM_HDR_LEN,
3777 cur_max_x + COLLXFRM_HDR_LEN))
3779 PL_strxfrm_max_cp = j;
3788 DEBUG_L(PerlIO_printf(Perl_debug_log,
3789 "_mem_collxfrm: Couldn't find any character to"
3790 " replace above-Latin1 chars in locale %s with",
3791 PL_collation_name));
3795 DEBUG_L(PerlIO_printf(Perl_debug_log,
3796 "_mem_collxfrm: highest 1-byte collating character"
3797 " in locale %s is 0x%02X\n",
3799 PL_strxfrm_max_cp));
3801 Safefree(cur_max_x);
3804 /* Here we know which legal code point collates the highest.
3805 * We are ready to construct the non-UTF-8 string. The length
3806 * will be at least 1 byte smaller than the input string
3807 * (because we changed at least one 2-byte character into a
3808 * single byte), but that is eaten up by the trailing NUL */
3814 char * e = (char *) t + len;
3816 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3818 if (UTF8_IS_INVARIANT(cur_char)) {
3821 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
3822 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3824 else { /* Replace illegal cp with highest collating
3826 s[d++] = PL_strxfrm_max_cp;
3830 Renew(s, d, char); /* Free up unused space */
3835 /* Here, we have constructed a modified version of the input. It could
3836 * be that we already had a modified copy before we did this version.
3837 * If so, that copy is no longer needed */
3838 if (t != input_string) {
3843 length_in_chars = (utf8)
3844 ? utf8_length((U8 *) s, (U8 *) s + len)
3847 /* The first element in the output is the collation id, used by
3848 * sv_collxfrm(); then comes the space for the transformed string. The
3849 * equation should give us a good estimate as to how much is needed */
3850 xAlloc = COLLXFRM_HDR_LEN
3852 + (PL_collxfrm_mult * length_in_chars);
3853 Newx(xbuf, xAlloc, char);
3854 if (UNLIKELY(! xbuf)) {
3855 DEBUG_L(PerlIO_printf(Perl_debug_log,
3856 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
3860 /* Store the collation id */
3861 *(U32*)xbuf = PL_collation_ix;
3863 /* Then the transformation of the input. We loop until successful, or we
3867 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
3869 /* If the transformed string occupies less space than we told strxfrm()
3870 * was available, it means it successfully transformed the whole
3872 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
3874 /* Some systems include a trailing NUL in the returned length.
3875 * Ignore it, using a loop in case multiple trailing NULs are
3878 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3883 /* If the first try didn't get it, it means our prediction was low.
3884 * Modify the coefficients so that we predict a larger value in any
3885 * future transformations */
3887 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3888 STRLEN computed_guess = PL_collxfrm_base
3889 + (PL_collxfrm_mult * length_in_chars);
3891 /* On zero-length input, just keep current slope instead of
3893 const STRLEN new_m = (length_in_chars != 0)
3894 ? needed / length_in_chars
3897 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3898 "%s: %d: initial size of %zu bytes for a length "
3899 "%zu string was insufficient, %zu needed\n",
3901 computed_guess, length_in_chars, needed));
3903 /* If slope increased, use it, but discard this result for
3904 * length 1 strings, as we can't be sure that it's a real slope
3906 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
3910 STRLEN old_m = PL_collxfrm_mult;
3911 STRLEN old_b = PL_collxfrm_base;
3915 PL_collxfrm_mult = new_m;
3916 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3917 computed_guess = PL_collxfrm_base
3918 + (PL_collxfrm_mult * length_in_chars);
3919 if (computed_guess < needed) {
3920 PL_collxfrm_base += needed - computed_guess;
3923 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3924 "%s: %d: slope is now %zu; was %zu, base "
3925 "is now %zu; was %zu\n",
3927 PL_collxfrm_mult, old_m,
3928 PL_collxfrm_base, old_b));
3930 else { /* Slope didn't change, but 'b' did */
3931 const STRLEN new_b = needed
3934 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
3935 "%s: %d: base is now %zu; was %zu\n",
3937 new_b, PL_collxfrm_base));
3938 PL_collxfrm_base = new_b;
3945 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
3946 DEBUG_L(PerlIO_printf(Perl_debug_log,
3947 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
3948 *xlen, PERL_INT_MAX));
3952 /* A well-behaved strxfrm() returns exactly how much space it needs
3953 * (usually not including the trailing NUL) when it fails due to not
3954 * enough space being provided. Assume that this is the case unless
3955 * it's been proven otherwise */
3956 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
3957 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
3959 else { /* Here, either:
3960 * 1) The strxfrm() has previously shown bad behavior; or
3961 * 2) It isn't the first time through the loop, which means
3962 * that the strxfrm() is now showing bad behavior, because
3963 * we gave it what it said was needed in the previous
3964 * iteration, and it came back saying it needed still more.
3965 * (Many versions of cygwin fit this. When the buffer size
3966 * isn't sufficient, they return the input size instead of
3967 * how much is needed.)
3968 * Increase the buffer size by a fixed percentage and try again.
3970 xAlloc += (xAlloc / 4) + 1;
3971 PL_strxfrm_is_behaved = FALSE;
3975 if (DEBUG_Lv_TEST || debug_initialization) {
3976 PerlIO_printf(Perl_debug_log,
3977 "_mem_collxfrm required more space than previously calculated"
3978 " for locale %s, trying again with new guess=%d+%zu\n",
3979 PL_collation_name, (int) COLLXFRM_HDR_LEN,
3980 xAlloc - COLLXFRM_HDR_LEN);
3987 Renew(xbuf, xAlloc, char);
3988 if (UNLIKELY(! xbuf)) {
3989 DEBUG_L(PerlIO_printf(Perl_debug_log,
3990 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4000 if (DEBUG_Lv_TEST || debug_initialization) {
4002 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
4003 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
4004 PerlIO_printf(Perl_debug_log, "%s\n",
4005 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
4011 /* Free up unneeded space; retain ehough for trailing NUL */
4012 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
4014 if (s != input_string) {
4022 if (s != input_string) {
4029 if (DEBUG_Lv_TEST || debug_initialization) {
4030 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
4041 S_print_collxfrm_input_and_return(pTHX_
4042 const char * const s,
4043 const char * const e,
4044 const STRLEN * const xlen,
4048 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
4050 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
4051 (UV)PL_collation_ix);
4053 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
4056 PerlIO_printf(Perl_debug_log, "NULL");
4058 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
4060 print_bytes_for_locale(s, e, is_utf8);
4062 PerlIO_printf(Perl_debug_log, "'\n");
4066 S_print_bytes_for_locale(pTHX_
4067 const char * const s,
4068 const char * const e,
4072 bool prev_was_printable = TRUE;
4073 bool first_time = TRUE;
4075 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
4079 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
4082 if (! prev_was_printable) {
4083 PerlIO_printf(Perl_debug_log, " ");
4085 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
4086 prev_was_printable = TRUE;
4090 PerlIO_printf(Perl_debug_log, " ");
4092 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
4093 prev_was_printable = FALSE;
4095 t += (is_utf8) ? UTF8SKIP(t) : 1;
4100 # endif /* #ifdef DEBUGGING */
4101 #endif /* USE_LOCALE_COLLATE */
4106 S_switch_category_locale_to_template(pTHX_ const int switch_category, const int template_category, const char * template_locale)
4108 /* Changes the locale for LC_'switch_category" to that of
4109 * LC_'template_category', if they aren't already the same. If not NULL,
4110 * 'template_locale' is the locale that 'template_category' is in.
4112 * Returns a copy of the name of the original locale for 'switch_category'
4113 * so can be switched back to with the companion function
4114 * restore_switched_locale(), (NULL if no restoral is necessary.) */
4116 char * restore_to_locale = NULL;
4118 if (switch_category == template_category) { /* No changes needed */
4122 /* Find the original locale of the category we may need to change, so that
4123 * it can be restored to later */
4124 restore_to_locale = stdize_locale(savepv(do_setlocale_r(switch_category,
4126 if (! restore_to_locale) {
4128 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4129 __FILE__, __LINE__, category_name(switch_category), errno);
4132 /* If the locale of the template category wasn't passed in, find it now */
4133 if (template_locale == NULL) {
4134 template_locale = do_setlocale_r(template_category, NULL);
4135 if (! template_locale) {
4137 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4138 __FILE__, __LINE__, category_name(template_category), errno);
4142 /* It the locales are the same, there's nothing to do */
4143 if (strEQ(restore_to_locale, template_locale)) {
4144 Safefree(restore_to_locale);
4146 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale unchanged as %s\n",
4147 category_name(switch_category), restore_to_locale));
4152 /* Finally, change the locale to the template one */
4153 if (! do_setlocale_r(switch_category, template_locale)) {
4155 "panic: %s: %d: Could not change %s locale to %s, errno=%d\n",
4156 __FILE__, __LINE__, category_name(switch_category),
4157 template_locale, errno);
4160 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s locale switched to %s\n",
4161 category_name(switch_category), template_locale));
4163 return restore_to_locale;
4167 S_restore_switched_locale(pTHX_ const int category, const char * const original_locale)
4169 /* Restores the locale for LC_'category' to 'original_locale' (which is a
4170 * copy that will be freed by this function), or do nothing if the latter
4171 * parameter is NULL */
4173 if (original_locale == NULL) {
4177 if (! do_setlocale_r(category, original_locale)) {
4179 "panic: %s: %d: setlocale %s restore to %s failed, errno=%d\n",
4181 category_name(category), original_locale, errno);
4184 Safefree(original_locale);
4188 Perl__is_cur_LC_category_utf8(pTHX_ int category)
4190 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
4191 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
4192 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
4193 * could give the wrong result. The result will very likely be correct for
4194 * languages that have commonly used non-ASCII characters, but for notably
4195 * English, it comes down to if the locale's name ends in something like
4196 * "UTF-8". It errs on the side of not being a UTF-8 locale.
4198 * If the platform is early C89, not containing mbtowc(), or we are
4199 * compiled to not pay attention to LC_CTYPE, this employs heuristics.
4200 * These work very well for non-Latin locales or those whose currency
4201 * symbol isn't a '$' nor plain ASCII text. But without LC_CTYPE and at
4202 * least MB_CUR_MAX, English locales with an ASCII currency symbol depend
4203 * on the name containing UTF-8 or not. */
4205 /* Name of current locale corresponding to the input category */
4206 const char *save_input_locale = NULL;
4208 bool is_utf8 = FALSE; /* The return value */
4210 /* The variables below are for the cache of previous lookups using this
4211 * function. The cache is a C string, described at the definition for
4212 * 'C_and_POSIX_utf8ness'.
4214 * The first part of the cache is fixed, for the C and POSIX locales. The
4215 * varying part starts just after them. */
4216 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
4218 Size_t utf8ness_cache_size; /* Size of the varying portion */
4219 Size_t input_name_len; /* Length in bytes of save_input_locale */
4220 Size_t input_name_len_with_overhead; /* plus extra chars used to store
4221 the name in the cache */
4222 char * delimited; /* The name plus the delimiters used to store
4224 char * name_pos; /* position of 'delimited' in the cache, or 0
4230 assert(category != LC_ALL);
4234 /* Get the desired category's locale */
4235 save_input_locale = stdize_locale(savepv(do_setlocale_r(category, NULL)));
4236 if (! save_input_locale) {
4238 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
4239 __FILE__, __LINE__, category_name(category), errno);
4242 DEBUG_L(PerlIO_printf(Perl_debug_log,
4243 "Current locale for %s is %s\n",
4244 category_name(category), save_input_locale));
4246 input_name_len = strlen(save_input_locale);
4248 /* In our cache, each name is accompanied by two delimiters and a single
4250 input_name_len_with_overhead = input_name_len + 3;
4252 /* Allocate and populate space for a copy of the name surrounded by the
4254 Newx(delimited, input_name_len_with_overhead, char);
4255 delimited[0] = UTF8NESS_SEP[0];
4256 Copy(save_input_locale, delimited + 1, input_name_len, char);
4257 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
4258 delimited[input_name_len+2] = '\0';
4260 /* And see if that is in the cache */
4261 name_pos = instr(PL_locale_utf8ness, delimited);
4263 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
4267 if (DEBUG_Lv_TEST || debug_initialization) {
4268 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
4269 save_input_locale, is_utf8);
4274 /* And, if not already in that position, move it to the beginning of
4275 * the non-constant portion of the list, since it is the most recently
4276 * used. (We don't have to worry about overflow, since just moving
4277 * existing names around) */
4278 if (name_pos > utf8ness_cache) {
4279 Move(utf8ness_cache,
4280 utf8ness_cache + input_name_len_with_overhead,
4281 name_pos - utf8ness_cache, char);
4284 input_name_len_with_overhead - 1, char);
4285 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
4288 Safefree(delimited);
4289 Safefree(save_input_locale);
4293 /* Here we don't have stored the utf8ness for the input locale. We have to
4296 # if defined(USE_LOCALE_CTYPE) \
4297 && ( (defined(HAS_NL_LANGINFO) && defined(CODESET)) \
4298 || (defined(HAS_MBTOWC) || defined(HAS_MBRTOWC)))
4301 const char *original_ctype_locale
4302 = switch_category_locale_to_template(LC_CTYPE,
4306 /* Here the current LC_CTYPE is set to the locale of the category whose
4307 * information is desired. This means that nl_langinfo() and mbtowc()
4308 * should give the correct results */
4310 # ifdef MB_CUR_MAX /* But we can potentially rule out UTF-8ness, avoiding
4311 calling the functions if we have this */
4313 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
4314 * Unicode code point. */
4316 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s: %d: MB_CUR_MAX=%d\n",
4317 __FILE__, __LINE__, (int) MB_CUR_MAX));
4318 if ((unsigned) MB_CUR_MAX < STRLENs(MAX_UNICODE_UTF8)) {
4320 restore_switched_locale(LC_CTYPE, original_ctype_locale);