3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * A Elbereth Gilthoniel,
13 * silivren penna míriel
14 * o menel aglar elenath!
15 * Na-chaered palan-díriel
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
23 /* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
47 #define PERL_IN_LOCALE_C
48 #include "perl_langinfo.h"
60 /* If the environment says to, we can output debugging information during
61 * initialization. This is done before option parsing, and before any thread
62 * creation, so can be a file-level static */
63 #if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
64 # define debug_initialization 0
65 # define DEBUG_INITIALIZATION_set(v)
67 static bool debug_initialization = FALSE;
68 # define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
72 /* Returns the Unix errno portion; ignoring any others. This is a macro here
73 * instead of putting it into perl.h, because unclear to khw what should be
75 #define GET_ERRNO saved_errno
77 /* strlen() of a literal string constant. We might want this more general,
78 * but using it in just this file for now. A problem with more generality is
79 * the compiler warnings about comparing unlike signs */
80 #define STRLENs(s) (sizeof("" s "") - 1)
82 /* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
83 * return of setlocale(), then this is extremely likely to be the C or POSIX
84 * locale. However, the output of setlocale() is documented to be opaque, but
85 * the odds are extremely small that it would return these two strings for some
86 * other locale. Note that VMS in these two locales includes many non-ASCII
87 * characters as controls and punctuation (below are hex bytes):
89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
90 * Oddly, none there are listed as alphas, though some represent alphabetics
91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
92 #define isNAME_C_OR_POSIX(name) \
94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
95 || strEQ((name), "POSIX")))
99 /* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
100 * looked up. This is in the form of a C string: */
102 #define UTF8NESS_SEP "\v"
103 #define UTF8NESS_PREFIX "\f"
105 /* So, the string looks like:
107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
109 * where the digit 0 after the \a indicates that the locale starting just
110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
112 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
113 STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
115 #define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
118 /* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
119 * kept there always. The remining portion of the cache is LRU, with the
120 * oldest looked-up locale at the tail end */
123 S_stdize_locale(pTHX_ char *locs)
125 /* Standardize the locale name from a string returned by 'setlocale',
126 * possibly modifying that string.
128 * The typical return value of setlocale() is either
129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
131 * (the space-separated values represent the various sublocales,
132 * in some unspecified order). This is not handled by this function.
134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
135 * which is harmful for further use of the string in setlocale(). This
136 * function removes the trailing new line and everything up through the '='
139 const char * const s = strchr(locs, '=');
142 PERL_ARGS_ASSERT_STDIZE_LOCALE;
145 const char * const t = strchr(s, '.');
148 const char * const u = strchr(t, '\n');
149 if (u && (u[1] == 0)) {
150 const STRLEN len = u - s;
151 Move(s + 1, locs, len, char);
159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
164 /* Two parallel arrays; first the locale categories Perl uses on this system;
165 * the second array is their names. These arrays are in mostly arbitrary
168 const int categories[] = {
170 # ifdef USE_LOCALE_NUMERIC
173 # ifdef USE_LOCALE_CTYPE
176 # ifdef USE_LOCALE_COLLATE
179 # ifdef USE_LOCALE_TIME
182 # ifdef USE_LOCALE_MESSAGES
185 # ifdef USE_LOCALE_MONETARY
188 # ifdef USE_LOCALE_ADDRESS
191 # ifdef USE_LOCALE_IDENTIFICATION
194 # ifdef USE_LOCALE_MEASUREMENT
197 # ifdef USE_LOCALE_PAPER
200 # ifdef USE_LOCALE_TELEPHONE
206 -1 /* Placeholder because C doesn't allow a
207 trailing comma, and it would get complicated
208 with all the #ifdef's */
211 /* The top-most real element is LC_ALL */
213 const char * const category_names[] = {
215 # ifdef USE_LOCALE_NUMERIC
218 # ifdef USE_LOCALE_CTYPE
221 # ifdef USE_LOCALE_COLLATE
224 # ifdef USE_LOCALE_TIME
227 # ifdef USE_LOCALE_MESSAGES
230 # ifdef USE_LOCALE_MONETARY
233 # ifdef USE_LOCALE_ADDRESS
236 # ifdef USE_LOCALE_IDENTIFICATION
239 # ifdef USE_LOCALE_MEASUREMENT
242 # ifdef USE_LOCALE_PAPER
245 # ifdef USE_LOCALE_TELEPHONE
251 NULL /* Placeholder */
256 /* On systems with LC_ALL, it is kept in the highest index position. (-2
257 * to account for the final unused placeholder element.) */
258 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
262 /* On systems without LC_ALL, we pretend it is there, one beyond the real
263 * top element, hence in the unused placeholder element. */
264 # define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
268 /* Pretending there is an LC_ALL element just above allows us to avoid most
269 * special cases. Most loops through these arrays in the code below are
270 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
271 * on either type of system. But the code must be written to not access the
272 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
273 * checked for at compile time by using the #define LC_ALL_INDEX which is only
274 * defined if we do have LC_ALL. */
277 S_category_name(const int category)
283 if (category == LC_ALL) {
289 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
290 if (category == categories[i]) {
291 return category_names[i];
296 const char suffix[] = " (unknown)";
298 Size_t length = sizeof(suffix) + 1;
307 /* Calculate the number of digits */
313 Newx(unknown, length, char);
314 my_snprintf(unknown, length, "%d%s", category, suffix);
320 /* Now create LC_foo_INDEX #defines for just those categories on this system */
321 # ifdef USE_LOCALE_NUMERIC
322 # define LC_NUMERIC_INDEX 0
323 # define _DUMMY_NUMERIC LC_NUMERIC_INDEX
325 # define _DUMMY_NUMERIC -1
327 # ifdef USE_LOCALE_CTYPE
328 # define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
329 # define _DUMMY_CTYPE LC_CTYPE_INDEX
331 # define _DUMMY_CTYPE _DUMMY_NUMERIC
333 # ifdef USE_LOCALE_COLLATE
334 # define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
335 # define _DUMMY_COLLATE LC_COLLATE_INDEX
337 # define _DUMMY_COLLATE _DUMMY_CTYPE
339 # ifdef USE_LOCALE_TIME
340 # define LC_TIME_INDEX _DUMMY_COLLATE + 1
341 # define _DUMMY_TIME LC_TIME_INDEX
343 # define _DUMMY_TIME _DUMMY_COLLATE
345 # ifdef USE_LOCALE_MESSAGES
346 # define LC_MESSAGES_INDEX _DUMMY_TIME + 1
347 # define _DUMMY_MESSAGES LC_MESSAGES_INDEX
349 # define _DUMMY_MESSAGES _DUMMY_TIME
351 # ifdef USE_LOCALE_MONETARY
352 # define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
353 # define _DUMMY_MONETARY LC_MONETARY_INDEX
355 # define _DUMMY_MONETARY _DUMMY_MESSAGES
357 # ifdef USE_LOCALE_ADDRESS
358 # define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
359 # define _DUMMY_ADDRESS LC_ADDRESS_INDEX
361 # define _DUMMY_ADDRESS _DUMMY_MONETARY
363 # ifdef USE_LOCALE_IDENTIFICATION
364 # define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
365 # define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
367 # define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
369 # ifdef USE_LOCALE_MEASUREMENT
370 # define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
371 # define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
373 # define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
375 # ifdef USE_LOCALE_PAPER
376 # define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
377 # define _DUMMY_PAPER LC_PAPER_INDEX
379 # define _DUMMY_PAPER _DUMMY_MEASUREMENT
381 # ifdef USE_LOCALE_TELEPHONE
382 # define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
383 # define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
385 # define _DUMMY_TELEPHONE _DUMMY_PAPER
388 # define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
390 #endif /* ifdef USE_LOCALE */
392 /* Windows requres a customized base-level setlocale() */
394 # define my_setlocale(cat, locale) win32_setlocale(cat, locale)
396 # define my_setlocale(cat, locale) setlocale(cat, locale)
399 #ifndef USE_POSIX_2008_LOCALE
401 /* "do_setlocale_c" is intended to be called when the category is a constant
402 * known at compile time; "do_setlocale_r", not known until run time */
403 # define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
404 # define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
406 #else /* Below uses POSIX 2008 */
408 /* We emulate setlocale with our own function. LC_foo is not valid for the
409 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
410 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
411 * proper offset into the array 'category_masks[]'. At runtime, we have to
412 * search through the array (as the actual numbers may not be small contiguous
413 * positive integers which would lend themselves to array lookup). */
414 # define do_setlocale_c(cat, locale) \
415 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
416 # define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
418 /* A third array, parallel to the ones above to map from category to its
420 const int category_masks[] = {
421 # ifdef USE_LOCALE_NUMERIC
424 # ifdef USE_LOCALE_CTYPE
427 # ifdef USE_LOCALE_COLLATE
430 # ifdef USE_LOCALE_TIME
433 # ifdef USE_LOCALE_MESSAGES
436 # ifdef USE_LOCALE_MONETARY
439 # ifdef USE_LOCALE_ADDRESS
442 # ifdef USE_LOCALE_IDENTIFICATION
443 LC_IDENTIFICATION_MASK,
445 # ifdef USE_LOCALE_MEASUREMENT
448 # ifdef USE_LOCALE_PAPER
451 # ifdef USE_LOCALE_TELEPHONE
454 /* LC_ALL can't be turned off by a Configure
455 * option, and in Posix 2008, should always be
456 * here, so compile it in unconditionally.
457 * This could catch some glitches at compile
463 S_emulate_setlocale(const int category,
466 const bool is_index_valid
469 /* This function effectively performs a setlocale() on just the current
470 * thread; thus it is thread-safe. It does this by using the POSIX 2008
471 * locale functions to emulate the behavior of setlocale(). Similar to
472 * regular setlocale(), the return from this function points to memory that
473 * can be overwritten by other system calls, so needs to be copied
474 * immediately if you need to retain it. The difference here is that
475 * system calls besides another setlocale() can overwrite it.
477 * By doing this, most locale-sensitive functions become thread-safe. The
478 * exceptions are mostly those that return a pointer to static memory.
480 * This function takes the same parameters, 'category' and 'locale', that
481 * the regular setlocale() function does, but it also takes two additional
482 * ones. This is because the 2008 functions don't use a category; instead
483 * they use a corresponding mask. Because this function operates in both
484 * worlds, it may need one or the other or both. This function can
485 * calculate the mask from the input category, but to avoid this
486 * calculation, if the caller knows at compile time what the mask is, it
487 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
488 * parameter is ignored.
490 * POSIX 2008, for some sick reason, chose not to provide a method to find
491 * the category name of a locale. Some vendors have created a
492 * querylocale() function to do just that. This function is a lot simpler
493 * to implement on systems that have this. Otherwise, we have to keep
494 * track of what the locale has been set to, so that we can return its
495 * name to emulate setlocale(). It's also possible for C code in some
496 * library to change the locale without us knowing it, though as of
497 * September 2017, there are no occurrences in CPAN of uselocale(). Some
498 * libraries do use setlocale(), but that changes the global locale, and
499 * threads using per-thread locales will just ignore those changes.
500 * Another problem is that without querylocale(), we have to guess at what
501 * was meant by setting a locale of "". We handle this by not actually
502 * ever setting to "" (unless querylocale exists), but to emulate what we
503 * think should happen for "".
513 if (DEBUG_Lv_TEST || debug_initialization) {
514 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);
519 /* If the input mask might be incorrect, calculate the correct one */
520 if (! is_index_valid) {
525 if (DEBUG_Lv_TEST || debug_initialization) {
526 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
531 for (i = 0; i <= LC_ALL_INDEX; i++) {
532 if (category == categories[i]) {
538 /* Here, we don't know about this category, so can't handle it.
539 * Fallback to the early POSIX usages */
540 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
541 "Unknown locale category %d; can't set it to %s\n",
549 if (DEBUG_Lv_TEST || debug_initialization) {
550 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
557 mask = category_masks[index];
561 if (DEBUG_Lv_TEST || debug_initialization) {
562 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
567 /* If just querying what the existing locale is ... */
568 if (locale == NULL) {
569 locale_t cur_obj = uselocale((locale_t) 0);
573 if (DEBUG_Lv_TEST || debug_initialization) {
574 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
579 if (cur_obj == LC_GLOBAL_LOCALE) {
580 return my_setlocale(category, NULL);
583 # ifdef HAS_QUERYLOCALE
585 return (char *) querylocale(mask, cur_obj);
589 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
590 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
592 # if defined(_NL_LOCALE_NAME) \
593 && defined(DEBUGGING) \
594 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
595 /* On systems that accept any locale name, the real underlying locale
596 * is often returned by this internal function, so we can't use it */
598 /* Internal glibc for querylocale(), but doesn't handle
599 * empty-string ("") locale properly; who knows what other
600 * glitches. Check for it now, under debug. */
602 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
603 uselocale((locale_t) 0));
605 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
606 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
607 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
609 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
610 if ( strNE(PL_curlocales[index], temp_name)
611 && ! ( isNAME_C_OR_POSIX(temp_name)
612 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
614 # ifdef USE_C_BACKTRACE
616 dump_c_backtrace(Perl_debug_log, 20, 1);
620 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
621 " (%s) and what internal glibc thinks"
622 " (%s)\n", category_names[index],
623 PL_curlocales[index], temp_name);
632 /* Without querylocale(), we have to use our record-keeping we've
635 if (category != LC_ALL) {
639 if (DEBUG_Lv_TEST || debug_initialization) {
640 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
645 return PL_curlocales[index];
647 else { /* For LC_ALL */
649 Size_t names_len = 0;
651 bool are_all_categories_the_same_locale = TRUE;
653 /* If we have a valid LC_ALL value, just return it */
654 if (PL_curlocales[LC_ALL_INDEX]) {
658 if (DEBUG_Lv_TEST || debug_initialization) {
659 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
664 return PL_curlocales[LC_ALL_INDEX];
667 /* Otherwise, we need to construct a string of name=value pairs.
668 * We use the glibc syntax, like
669 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
670 * First calculate the needed size. Along the way, check if all
671 * the locale names are the same */
672 for (i = 0; i < LC_ALL_INDEX; i++) {
676 if (DEBUG_Lv_TEST || debug_initialization) {
677 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]);
682 names_len += strlen(category_names[i])
684 + strlen(PL_curlocales[i])
687 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
688 are_all_categories_the_same_locale = FALSE;
692 /* If they are the same, we don't actually have to construct the
693 * string; we just make the entry in LC_ALL_INDEX valid, and be
694 * that single name */
695 if (are_all_categories_the_same_locale) {
696 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
697 return PL_curlocales[LC_ALL_INDEX];
700 names_len++; /* Trailing '\0' */
701 SAVEFREEPV(Newx(all_string, names_len, char));
704 /* Then fill in the string */
705 for (i = 0; i < LC_ALL_INDEX; i++) {
709 if (DEBUG_Lv_TEST || debug_initialization) {
710 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]);
715 my_strlcat(all_string, category_names[i], names_len);
716 my_strlcat(all_string, "=", names_len);
717 my_strlcat(all_string, PL_curlocales[i], names_len);
718 my_strlcat(all_string, ";", names_len);
723 if (DEBUG_L_TEST || debug_initialization) {
724 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
734 SETERRNO(EINVAL, LIB_INVARG);
744 /* Here, we are switching locales. */
746 # ifndef HAS_QUERYLOCALE
748 if (strEQ(locale, "")) {
750 /* For non-querylocale() systems, we do the setting of "" ourselves to
751 * be sure that we really know what's going on. We follow the Linux
752 * documented behavior (but if that differs from the actual behavior,
753 * this won't work exactly as the OS implements). We go out and
754 * examine the environment based on our understanding of how the system
755 * works, and use that to figure things out */
757 const char * const lc_all = PerlEnv_getenv("LC_ALL");
759 /* Use any "LC_ALL" environment variable, as it overrides everything
761 if (lc_all && strNE(lc_all, "")) {
766 /* Otherwise, we need to dig deeper. Unless overridden, the
767 * default is the LANG environment variable; if it doesn't exist,
770 const char * default_name;
772 /* To minimize other threads messing with the environment, we copy
773 * the variable, making it a temporary. But this doesn't work upon
774 * program initialization before any scopes are created, and at
775 * this time, there's nothing else going on that would interfere.
776 * So skip the copy in that case */
777 if (PL_scopestack_ix == 0) {
778 default_name = PerlEnv_getenv("LANG");
781 default_name = savepv(PerlEnv_getenv("LANG"));
784 if (! default_name || strEQ(default_name, "")) {
787 else if (PL_scopestack_ix != 0) {
788 SAVEFREEPV(default_name);
791 if (category != LC_ALL) {
792 const char * const name = PerlEnv_getenv(category_names[index]);
794 /* Here we are setting a single category. Assume will have the
796 locale = default_name;
798 /* But then look for an overriding environment variable */
799 if (name && strNE(name, "")) {
804 bool did_override = FALSE;
807 /* Here, we are getting LC_ALL. Any categories that don't have
808 * a corresponding environment variable set should be set to
809 * LANG, or to "C" if there is no LANG. If no individual
810 * categories differ from this, we can just set LC_ALL. This
811 * is buggy on systems that have extra categories that we don't
812 * know about. If there is an environment variable that sets
813 * that category, we won't know to look for it, and so our use
814 * of LANG or "C" improperly overrides it. On the other hand,
815 * if we don't do what is done here, and there is no
816 * environment variable, the category's locale should be set to
817 * LANG or "C". So there is no good solution. khw thinks the
818 * best is to look at systems to see what categories they have,
819 * and include them, and then to assume that we know the
822 for (i = 0; i < LC_ALL_INDEX; i++) {
823 const char * const env_override
824 = savepv(PerlEnv_getenv(category_names[i]));
825 const char * this_locale = ( env_override
826 && strNE(env_override, ""))
829 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
831 Safefree(env_override);
835 if (strNE(this_locale, default_name)) {
839 Safefree(env_override);
842 /* If all the categories are the same, we can set LC_ALL to
844 if (! did_override) {
845 locale = default_name;
849 /* Here, LC_ALL is no longer valid, as some individual
850 * categories don't match it. We call ourselves
851 * recursively, as that will execute the code that
852 * generates the proper locale string for this situation.
853 * We don't do the remainder of this function, as that is
854 * to update our records, and we've just done that for the
855 * individual categories in the loop above, and doing so
856 * would cause LC_ALL to be done as well */
857 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
862 else if (strchr(locale, ';')) {
864 /* LC_ALL may actually incude a conglomeration of various categories.
865 * Without querylocale, this code uses the glibc (as of this writing)
866 * syntax for representing that, but that is not a stable API, and
867 * other platforms do it differently, so we have to handle all cases
871 const char * s = locale;
872 const char * e = locale + strlen(locale);
874 const char * category_end;
875 const char * name_start;
876 const char * name_end;
878 /* If the string that gives what to set doesn't include all categories,
879 * the omitted ones get set to "C". To get this behavior, first set
880 * all the individual categories to "C", and override the furnished
882 for (i = 0; i < LC_ALL_INDEX; i++) {
883 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
890 /* Parse through the category */
891 while (isWORDCHAR(*p)) {
898 "panic: %s: %d: Unexpected character in locale name '%02X",
899 __FILE__, __LINE__, *(p-1));
902 /* Parse through the locale name */
904 while (p < e && *p != ';') {
907 "panic: %s: %d: Unexpected character in locale name '%02X",
908 __FILE__, __LINE__, *(p-1));
914 /* Space past the semi-colon */
919 /* Find the index of the category name in our lists */
920 for (i = 0; i < LC_ALL_INDEX; i++) {
921 char * individ_locale;
923 /* Keep going if this isn't the index. The strnNE() avoids a
924 * Perl_form(), but would fail if ever a category name could be
925 * a substring of another one, like if there were a
927 if strnNE(s, category_names[i], category_end - s) {
931 /* If this index is for the single category we're changing, we
932 * have found the locale to set it to. */
933 if (category == categories[i]) {
934 locale = Perl_form(aTHX_ "%.*s",
935 (int) (name_end - name_start),
940 assert(category == LC_ALL);
941 individ_locale = Perl_form(aTHX_ "%.*s",
942 (int) (name_end - name_start), name_start);
943 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
952 /* Here we have set all the individual categories by recursive calls.
953 * These collectively should have fixed up LC_ALL, so can just query
954 * what that now is */
955 assert(category == LC_ALL);
957 return do_setlocale_c(LC_ALL, NULL);
962 /* Here at the end of having to deal with the absence of querylocale().
963 * Some cases have already been fully handled by recursive calls to this
964 * function. But at this point, we haven't dealt with those, but are now
965 * prepared to, knowing what the locale name to set this category to is.
966 * This would have come for free if this system had had querylocale() */
968 # endif /* end of ! querylocale */
970 assert(PL_C_locale_obj);
972 /* Switching locales generally entails freeing the current one's space (at
973 * the C library's discretion). We need to stop using that locale before
974 * the switch. So switch to a known locale object that we don't otherwise
975 * mess with. This returns the locale object in effect at the time of the
977 old_obj = uselocale(PL_C_locale_obj);
981 if (DEBUG_Lv_TEST || debug_initialization) {
982 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
991 if (DEBUG_L_TEST || debug_initialization) {
993 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1004 if (DEBUG_Lv_TEST || debug_initialization) {
1005 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1010 /* If we are switching to the LC_ALL C locale, it already exists. Use
1011 * it instead of trying to create a new locale */
1012 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1016 if (DEBUG_Lv_TEST || debug_initialization) {
1017 PerlIO_printf(Perl_debug_log,
1018 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1023 new_obj = PL_C_locale_obj;
1025 /* We already had switched to the C locale in preparation for freeing
1027 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1028 freelocale(old_obj);
1032 /* If we weren't in a thread safe locale, set so that newlocale() below
1033 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1034 The latter is defensive coding, so that, even if there is some bug, we
1035 will never end up trying to modify either of these, as if passed to
1036 newlocale(), they can be. */
1037 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1038 old_obj = (locale_t) 0;
1041 /* Ready to create a new locale by modification of the exising one */
1042 new_obj = newlocale(mask, locale, old_obj);
1049 if (DEBUG_L_TEST || debug_initialization) {
1050 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1055 if (! uselocale(old_obj)) {
1059 if (DEBUG_L_TEST || debug_initialization) {
1060 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1072 if (DEBUG_Lv_TEST || debug_initialization) {
1073 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p", __FILE__, __LINE__, new_obj);
1075 PerlIO_printf(Perl_debug_log, "; should have freed %p", old_obj);
1077 PerlIO_printf(Perl_debug_log, "\n");
1082 /* And switch into it */
1083 if (! uselocale(new_obj)) {
1088 if (DEBUG_L_TEST || debug_initialization) {
1089 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1094 if (! uselocale(old_obj)) {
1098 if (DEBUG_L_TEST || debug_initialization) {
1099 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1105 freelocale(new_obj);
1113 if (DEBUG_Lv_TEST || debug_initialization) {
1114 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1119 /* We are done, except for updating our records (if the system doesn't keep
1120 * them) and in the case of locale "", we don't actually know what the
1121 * locale that got switched to is, as it came from the environment. So
1122 * have to find it */
1124 # ifdef HAS_QUERYLOCALE
1126 if (strEQ(locale, "")) {
1127 locale = querylocale(mask, new_obj);
1132 /* Here, 'locale' is the return value */
1134 /* Without querylocale(), we have to update our records */
1136 if (category == LC_ALL) {
1139 /* For LC_ALL, we change all individual categories to correspond */
1140 /* PL_curlocales is a parallel array, so has same
1141 * length as 'categories' */
1142 for (i = 0; i <= LC_ALL_INDEX; i++) {
1143 Safefree(PL_curlocales[i]);
1144 PL_curlocales[i] = savepv(locale);
1149 /* For a single category, if it's not the same as the one in LC_ALL, we
1152 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1153 Safefree(PL_curlocales[LC_ALL_INDEX]);
1154 PL_curlocales[LC_ALL_INDEX] = NULL;
1157 /* Then update the category's record */
1158 Safefree(PL_curlocales[index]);
1159 PL_curlocales[index] = savepv(locale);
1167 #endif /* USE_POSIX_2008_LOCALE */
1169 #if 0 /* Code that was to emulate thread-safe locales on platforms that
1170 didn't natively support them */
1172 /* The way this would work is that we would keep a per-thread list of the
1173 * correct locale for that thread. Any operation that was locale-sensitive
1174 * would have to be changed so that it would look like this:
1177 * setlocale to the correct locale for this operation
1181 * This leaves the global locale in the most recently used operation's, but it
1182 * was locked long enough to get the result. If that result is static, it
1183 * needs to be copied before the unlock.
1185 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1186 * the setup, but are no-ops when not needed, and similarly,
1187 * END_LOCALE_DEPENDENT_OP for the tear-down
1189 * But every call to a locale-sensitive function would have to be changed, and
1190 * if a module didn't cooperate by using the mutex, things would break.
1192 * This code was abandoned before being completed or tested, and is left as-is
1195 # define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1196 # define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1199 S_locking_setlocale(pTHX_
1201 const char * locale,
1203 const bool is_index_valid
1206 /* This function kind of performs a setlocale() on just the current thread;
1207 * thus it is kind of thread-safe. It does this by keeping a thread-level
1208 * array of the current locales for each category. Every time a locale is
1209 * switched to, it does the switch globally, but updates the thread's
1210 * array. A query as to what the current locale is just returns the
1211 * appropriate element from the array, and doesn't actually call the system
1212 * setlocale(). The saving into the array is done in an uninterruptible
1213 * section of code, so is unaffected by whatever any other threads might be
1216 * All locale-sensitive operations must work by first starting a critical
1217 * section, then switching to the thread's locale as kept by this function,
1218 * and then doing the operation, then ending the critical section. Thus,
1219 * each gets done in the appropriate locale. simulating thread-safety.
1221 * This function takes the same parameters, 'category' and 'locale', that
1222 * the regular setlocale() function does, but it also takes two additional
1223 * ones. This is because as described earlier. If we know on input the
1224 * index corresponding to the category into the array where we store the
1225 * current locales, we don't have to calculate it. If the caller knows at
1226 * compile time what the index is, it it can pass it, setting
1227 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1231 /* If the input index might be incorrect, calculate the correct one */
1232 if (! is_index_valid) {
1235 if (DEBUG_Lv_TEST || debug_initialization) {
1236 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1239 for (i = 0; i <= LC_ALL_INDEX; i++) {
1240 if (category == categories[i]) {
1246 /* Here, we don't know about this category, so can't handle it.
1247 * XXX best we can do is to unsafely set this
1250 return my_setlocale(category, locale);
1254 if (DEBUG_Lv_TEST || debug_initialization) {
1255 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1259 /* For a query, just return what's in our records */
1260 if (new_locale == NULL) {
1261 return curlocales[index];
1265 /* Otherwise, we need to do the switch, and save the result, all in a
1266 * critical section */
1268 Safefree(curlocales[[index]]);
1270 /* It might be that this is called from an already-locked section of code.
1271 * We would have to detect and skip the LOCK/UNLOCK if so */
1274 curlocales[index] = savepv(my_setlocale(category, new_locale));
1276 if (strEQ(new_locale, "")) {
1280 /* The locale values come from the environment, and may not all be the
1281 * same, so for LC_ALL, we have to update all the others, while the
1282 * mutex is still locked */
1284 if (category == LC_ALL) {
1286 for (i = 0; i < LC_ALL_INDEX) {
1287 curlocales[i] = my_setlocale(categories[i], NULL);
1296 return curlocales[index];
1303 S_set_numeric_radix(pTHX_ const bool use_locale)
1305 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1306 * TRUE, use the radix character derived from the current locale */
1308 #if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1309 || defined(HAS_NL_LANGINFO))
1311 const char * radix = (use_locale)
1312 ? my_nl_langinfo(RADIXCHAR, FALSE)
1313 /* FALSE => already in dest locale */
1316 sv_setpv(PL_numeric_radix_sv, radix);
1318 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1319 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1320 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
1321 SvCUR(PL_numeric_radix_sv))
1322 && _is_cur_LC_category_utf8(LC_NUMERIC))
1324 SvUTF8_on(PL_numeric_radix_sv);
1329 if (DEBUG_L_TEST || debug_initialization) {
1330 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
1331 SvPVX(PL_numeric_radix_sv),
1332 cBOOL(SvUTF8(PL_numeric_radix_sv)));
1338 PERL_UNUSED_ARG(use_locale);
1340 #endif /* USE_LOCALE_NUMERIC and can find the radix char */
1345 S_new_numeric(pTHX_ const char *newnum)
1348 #ifndef USE_LOCALE_NUMERIC
1350 PERL_UNUSED_ARG(newnum);
1354 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
1355 * core Perl this and that 'newnum' is the name of the new locale.
1356 * It installs this locale as the current underlying default.
1358 * The default locale and the C locale can be toggled between by use of the
1359 * set_numeric_underlying() and set_numeric_standard() functions, which
1360 * should probably not be called directly, but only via macros like
1361 * SET_NUMERIC_STANDARD() in perl.h.
1363 * The toggling is necessary mainly so that a non-dot radix decimal point
1364 * character can be output, while allowing internal calculations to use a
1367 * This sets several interpreter-level variables:
1368 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
1369 * PL_numeric_underlying A boolean indicating if the toggled state is such
1370 * that the current locale is the program's underlying
1372 * PL_numeric_standard An int indicating if the toggled state is such
1373 * that the current locale is the C locale or
1374 * indistinguishable from the C locale. If non-zero, it
1375 * is in C; if > 1, it means it may not be toggled away
1377 * PL_numeric_underlying_is_standard A bool kept by this function
1378 * indicating that the underlying locale and the standard
1379 * C locale are indistinguishable for the purposes of
1380 * LC_NUMERIC. This happens when both of the above two
1381 * variables are true at the same time. (Toggling is a
1382 * no-op under these circumstances.) This variable is
1383 * used to avoid having to recalculate.
1389 Safefree(PL_numeric_name);
1390 PL_numeric_name = NULL;
1391 PL_numeric_standard = TRUE;
1392 PL_numeric_underlying = TRUE;
1393 PL_numeric_underlying_is_standard = TRUE;
1397 save_newnum = stdize_locale(savepv(newnum));
1398 PL_numeric_underlying = TRUE;
1399 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1401 #ifndef TS_W32_BROKEN_LOCALECONV
1403 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1404 * them. But on broken Windows systems calling my_nl_langinfo() for
1405 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1406 * and just always change the locale if not C nor POSIX on those systems */
1407 if (! PL_numeric_standard) {
1408 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
1409 FALSE /* Don't toggle locale */ ))
1410 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
1415 /* Save the new name if it isn't the same as the previous one, if any */
1416 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
1417 Safefree(PL_numeric_name);
1418 PL_numeric_name = save_newnum;
1421 Safefree(save_newnum);
1424 PL_numeric_underlying_is_standard = PL_numeric_standard;
1426 # ifdef HAS_POSIX_2008_LOCALE
1428 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1430 PL_underlying_numeric_obj);
1434 if (DEBUG_L_TEST || debug_initialization) {
1435 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1438 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1439 * have to worry about the radix being a non-dot. (Core operations that
1440 * need the underlying locale change to it temporarily). */
1441 if (PL_numeric_standard) {
1442 set_numeric_radix(0);
1445 set_numeric_standard();
1448 #endif /* USE_LOCALE_NUMERIC */
1453 Perl_set_numeric_standard(pTHX)
1456 #ifdef USE_LOCALE_NUMERIC
1458 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1459 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1460 * macro avoids calling this routine if toggling isn't necessary according
1461 * to our records (which could be wrong if some XS code has changed the
1462 * locale behind our back) */
1466 if (DEBUG_L_TEST || debug_initialization) {
1467 PerlIO_printf(Perl_debug_log,
1468 "Setting LC_NUMERIC locale to standard C\n");
1473 do_setlocale_c(LC_NUMERIC, "C");
1474 PL_numeric_standard = TRUE;
1475 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1476 set_numeric_radix(0);
1478 #endif /* USE_LOCALE_NUMERIC */
1483 Perl_set_numeric_underlying(pTHX)
1486 #ifdef USE_LOCALE_NUMERIC
1488 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
1489 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1490 * instead of calling this directly. The macro avoids calling this routine
1491 * if toggling isn't necessary according to our records (which could be
1492 * wrong if some XS code has changed the locale behind our back) */
1496 if (DEBUG_L_TEST || debug_initialization) {
1497 PerlIO_printf(Perl_debug_log,
1498 "Setting LC_NUMERIC locale to %s\n",
1504 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1505 PL_numeric_standard = PL_numeric_underlying_is_standard;
1506 PL_numeric_underlying = TRUE;
1507 set_numeric_radix(! PL_numeric_standard);
1509 #endif /* USE_LOCALE_NUMERIC */
1514 * Set up for a new ctype locale.
1517 S_new_ctype(pTHX_ const char *newctype)
1520 #ifndef USE_LOCALE_CTYPE
1522 PERL_UNUSED_ARG(newctype);
1523 PERL_UNUSED_CONTEXT;
1527 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
1528 * core Perl this and that 'newctype' is the name of the new locale.
1530 * This function sets up the folding arrays for all 256 bytes, assuming
1531 * that tofold() is tolc() since fold case is not a concept in POSIX,
1533 * Any code changing the locale (outside this file) should use
1534 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1535 * this function should be called directly only from this file and from
1536 * POSIX::setlocale() */
1541 /* Don't check for problems if we are suppressing the warnings */
1542 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1543 bool maybe_utf8_turkic = FALSE;
1545 PERL_ARGS_ASSERT_NEW_CTYPE;
1547 /* We will replace any bad locale warning with 1) nothing if the new one is
1548 * ok; or 2) a new warning for the bad new locale */
1549 if (PL_warn_locale) {
1550 SvREFCNT_dec_NN(PL_warn_locale);
1551 PL_warn_locale = NULL;
1554 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
1556 /* A UTF-8 locale gets standard rules. But note that code still has to
1557 * handle this specially because of the three problematic code points */
1558 if (PL_in_utf8_CTYPE_locale) {
1559 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1561 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1562 * Turkic. Make sure these two are the only anomalies. (We don't use
1563 * towupper and towlower because they aren't in C89.) */
1565 #if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1567 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1571 if (toupper('i') == 'i' && tolower('I') == 'I') {
1574 check_for_problems = TRUE;
1575 maybe_utf8_turkic = TRUE;
1579 /* We don't populate the other lists if a UTF-8 locale, but do check that
1580 * everything works as expected, unless checking turned off */
1581 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
1582 /* Assume enough space for every character being bad. 4 spaces each
1583 * for the 94 printable characters that are output like "'x' "; and 5
1584 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1586 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
1587 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1589 unsigned int bad_count = 0; /* Count of bad characters */
1591 for (i = 0; i < 256; i++) {
1592 if (! PL_in_utf8_CTYPE_locale) {
1594 PL_fold_locale[i] = (U8) tolower(i);
1595 else if (islower(i))
1596 PL_fold_locale[i] = (U8) toupper(i);
1598 PL_fold_locale[i] = (U8) i;
1601 /* If checking for locale problems, see if the native ASCII-range
1602 * printables plus \n and \t are in their expected categories in
1603 * the new locale. If not, this could mean big trouble, upending
1604 * Perl's and most programs' assumptions, like having a
1605 * metacharacter with special meaning become a \w. Fortunately,
1606 * it's very rare to find locales that aren't supersets of ASCII
1607 * nowadays. It isn't a problem for most controls to be changed
1608 * into something else; we check only \n and \t, though perhaps \r
1609 * could be an issue as well. */
1610 if ( check_for_problems
1611 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1613 bool is_bad = FALSE;
1614 char name[4] = { '\0' };
1616 /* Convert the name into a string */
1621 else if (i == '\n') {
1622 my_strlcpy(name, "\\n", sizeof(name));
1624 else if (i == '\t') {
1625 my_strlcpy(name, "\\t", sizeof(name));
1629 my_strlcpy(name, "' '", sizeof(name));
1632 /* Check each possibe class */
1633 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1635 DEBUG_L(PerlIO_printf(Perl_debug_log,
1636 "isalnum('%s') unexpectedly is %d\n",
1637 name, cBOOL(isalnum(i))));
1639 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1641 DEBUG_L(PerlIO_printf(Perl_debug_log,
1642 "isalpha('%s') unexpectedly is %d\n",
1643 name, cBOOL(isalpha(i))));
1645 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1647 DEBUG_L(PerlIO_printf(Perl_debug_log,
1648 "isdigit('%s') unexpectedly is %d\n",
1649 name, cBOOL(isdigit(i))));
1651 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1653 DEBUG_L(PerlIO_printf(Perl_debug_log,
1654 "isgraph('%s') unexpectedly is %d\n",
1655 name, cBOOL(isgraph(i))));
1657 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1659 DEBUG_L(PerlIO_printf(Perl_debug_log,
1660 "islower('%s') unexpectedly is %d\n",
1661 name, cBOOL(islower(i))));
1663 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1665 DEBUG_L(PerlIO_printf(Perl_debug_log,
1666 "isprint('%s') unexpectedly is %d\n",
1667 name, cBOOL(isprint(i))));
1669 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1671 DEBUG_L(PerlIO_printf(Perl_debug_log,
1672 "ispunct('%s') unexpectedly is %d\n",
1673 name, cBOOL(ispunct(i))));
1675 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1677 DEBUG_L(PerlIO_printf(Perl_debug_log,
1678 "isspace('%s') unexpectedly is %d\n",
1679 name, cBOOL(isspace(i))));
1681 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1683 DEBUG_L(PerlIO_printf(Perl_debug_log,
1684 "isupper('%s') unexpectedly is %d\n",
1685 name, cBOOL(isupper(i))));
1687 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1689 DEBUG_L(PerlIO_printf(Perl_debug_log,
1690 "isxdigit('%s') unexpectedly is %d\n",
1691 name, cBOOL(isxdigit(i))));
1693 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1695 DEBUG_L(PerlIO_printf(Perl_debug_log,
1696 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1697 name, tolower(i), (int) toLOWER_A(i)));
1699 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1701 DEBUG_L(PerlIO_printf(Perl_debug_log,
1702 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1703 name, toupper(i), (int) toUPPER_A(i)));
1705 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1707 DEBUG_L(PerlIO_printf(Perl_debug_log,
1708 "'\\n' (=%02X) is not a control\n", (int) i));
1711 /* Add to the list; Separate multiple entries with a blank */
1714 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
1716 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1722 if (bad_count == 2 && maybe_utf8_turkic) {
1724 *bad_chars_list = '\0';
1725 PL_fold_locale['I'] = 'I';
1726 PL_fold_locale['i'] = 'i';
1727 PL_in_utf8_turkic_locale = TRUE;
1728 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1729 __FILE__, __LINE__, newctype));
1732 PL_in_utf8_turkic_locale = FALSE;
1737 /* We only handle single-byte locales (outside of UTF-8 ones; so if
1738 * this locale requires more than one byte, there are going to be
1740 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1741 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1742 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1744 if ( check_for_problems && MB_CUR_MAX > 1
1745 && ! PL_in_utf8_CTYPE_locale
1747 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1748 * locale. Just assume that the implementation for them (plus
1749 * for POSIX) is correct and the > 1 value is spurious. (Since
1750 * these are specially handled to never be considered UTF-8
1751 * locales, as long as this is the only problem, everything
1752 * should work fine */
1753 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1755 multi_byte_locale = TRUE;
1760 /* If we found problems and we want them output, do so */
1761 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1762 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1764 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1765 PL_warn_locale = Perl_newSVpvf(aTHX_
1766 "Locale '%s' contains (at least) the following characters"
1767 " which have\nunexpected meanings: %s\nThe Perl program"
1768 " will use the expected meanings",
1769 newctype, bad_chars_list);
1772 PL_warn_locale = Perl_newSVpvf(aTHX_
1773 "Locale '%s' may not work well.%s%s%s\n",
1776 ? " Some characters in it are not recognized by"
1780 ? "\nThe following characters (and maybe others)"
1781 " may not have the same meaning as the Perl"
1782 " program expects:\n"
1790 # ifdef HAS_NL_LANGINFO
1792 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1793 /* parameter FALSE is a don't care here */
1794 my_nl_langinfo(CODESET, FALSE));
1798 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1800 /* If we are actually in the scope of the locale or are debugging,
1801 * output the message now. If not in that scope, we save the
1802 * message to be output at the first operation using this locale,
1803 * if that actually happens. Most programs don't use locales, so
1804 * they are immune to bad ones. */
1805 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
1807 /* The '0' below suppresses a bogus gcc compiler warning */
1808 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
1810 if (IN_LC(LC_CTYPE)) {
1811 SvREFCNT_dec_NN(PL_warn_locale);
1812 PL_warn_locale = NULL;
1818 #endif /* USE_LOCALE_CTYPE */
1823 Perl__warn_problematic_locale()
1826 #ifdef USE_LOCALE_CTYPE
1830 /* Internal-to-core function that outputs the message in PL_warn_locale,
1831 * and then NULLS it. Should be called only through the macro
1832 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1834 if (PL_warn_locale) {
1835 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1836 SvPVX(PL_warn_locale),
1837 0 /* dummy to avoid compiler warning */ );
1838 SvREFCNT_dec_NN(PL_warn_locale);
1839 PL_warn_locale = NULL;
1847 S_new_collate(pTHX_ const char *newcoll)
1850 #ifndef USE_LOCALE_COLLATE
1852 PERL_UNUSED_ARG(newcoll);
1853 PERL_UNUSED_CONTEXT;
1857 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
1858 * core Perl this and that 'newcoll' is the name of the new locale.
1860 * The design of locale collation is that every locale change is given an
1861 * index 'PL_collation_ix'. The first time a string particpates in an
1862 * operation that requires collation while locale collation is active, it
1863 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1864 * magic includes the collation index, and the transformation of the string
1865 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1866 * instead of the string itself. If a string changes, the magic is
1867 * cleared. The next time the locale changes, the index is incremented,
1868 * and so we know during a comparison that the transformation is not
1869 * necessarily still valid, and so is recomputed. Note that if the locale
1870 * changes enough times, the index could wrap (a U32), and it is possible
1871 * that a transformation would improperly be considered valid, leading to
1872 * an unlikely bug */
1875 if (PL_collation_name) {
1877 Safefree(PL_collation_name);
1878 PL_collation_name = NULL;
1880 PL_collation_standard = TRUE;
1881 is_standard_collation:
1882 PL_collxfrm_base = 0;
1883 PL_collxfrm_mult = 2;
1884 PL_in_utf8_COLLATE_locale = FALSE;
1885 PL_strxfrm_NUL_replacement = '\0';
1886 PL_strxfrm_max_cp = 0;
1890 /* If this is not the same locale as currently, set the new one up */
1891 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1893 Safefree(PL_collation_name);
1894 PL_collation_name = stdize_locale(savepv(newcoll));
1895 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
1896 if (PL_collation_standard) {
1897 goto is_standard_collation;
1900 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
1901 PL_strxfrm_NUL_replacement = '\0';
1902 PL_strxfrm_max_cp = 0;
1904 /* A locale collation definition includes primary, secondary, tertiary,
1905 * etc. weights for each character. To sort, the primary weights are
1906 * used, and only if they compare equal, then the secondary weights are
1907 * used, and only if they compare equal, then the tertiary, etc.
1909 * strxfrm() works by taking the input string, say ABC, and creating an
1910 * output transformed string consisting of first the primary weights,
1911 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1912 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1913 * may not have weights at every level. In our example, let's say B
1914 * doesn't have a tertiary weight, and A doesn't have a secondary
1915 * weight. The constructed string is then going to be
1916 * A¹B¹C¹ B²C² A³C³ ....
1917 * This has the desired effect that strcmp() will look at the secondary
1918 * or tertiary weights only if the strings compare equal at all higher
1919 * priority weights. The spaces shown here, like in
1921 * are not just for readability. In the general case, these must
1922 * actually be bytes, which we will call here 'separator weights'; and
1923 * they must be smaller than any other weight value, but since these
1924 * are C strings, only the terminating one can be a NUL (some
1925 * implementations may include a non-NUL separator weight just before
1926 * the NUL). Implementations tend to reserve 01 for the separator
1927 * weights. They are needed so that a shorter string's secondary
1928 * weights won't be misconstrued as primary weights of a longer string,
1929 * etc. By making them smaller than any other weight, the shorter
1930 * string will sort first. (Actually, if all secondary weights are
1931 * smaller than all primary ones, there is no need for a separator
1932 * weight between those two levels, etc.)
1934 * The length of the transformed string is roughly a linear function of
1935 * the input string. It's not exactly linear because some characters
1936 * don't have weights at all levels. When we call strxfrm() we have to
1937 * allocate some memory to hold the transformed string. The
1938 * calculations below try to find coefficients 'm' and 'b' for this
1939 * locale so that m*x + b equals how much space we need, given the size
1940 * of the input string in 'x'. If we calculate too small, we increase
1941 * the size as needed, and call strxfrm() again, but it is better to
1942 * get it right the first time to avoid wasted expensive string
1943 * transformations. */
1946 /* We use the string below to find how long the tranformation of it
1947 * is. Almost all locales are supersets of ASCII, or at least the
1948 * ASCII letters. We use all of them, half upper half lower,
1949 * because if we used fewer, we might hit just the ones that are
1950 * outliers in a particular locale. Most of the strings being
1951 * collated will contain a preponderance of letters, and even if
1952 * they are above-ASCII, they are likely to have the same number of
1953 * weight levels as the ASCII ones. It turns out that digits tend
1954 * to have fewer levels, and some punctuation has more, but those
1955 * are relatively sparse in text, and khw believes this gives a
1956 * reasonable result, but it could be changed if experience so
1958 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1959 char * x_longer; /* Transformed 'longer' */
1960 Size_t x_len_longer; /* Length of 'x_longer' */
1962 char * x_shorter; /* We also transform a substring of 'longer' */
1963 Size_t x_len_shorter;
1965 /* _mem_collxfrm() is used get the transformation (though here we
1966 * are interested only in its length). It is used because it has
1967 * the intelligence to handle all cases, but to work, it needs some
1968 * values of 'm' and 'b' to get it started. For the purposes of
1969 * this calculation we use a very conservative estimate of 'm' and
1970 * 'b'. This assumes a weight can be multiple bytes, enough to
1971 * hold any UV on the platform, and there are 5 levels, 4 weight
1972 * bytes, and a trailing NUL. */
1973 PL_collxfrm_base = 5;
1974 PL_collxfrm_mult = 5 * sizeof(UV);
1976 /* Find out how long the transformation really is */
1977 x_longer = _mem_collxfrm(longer,
1981 /* We avoid converting to UTF-8 in the
1982 * called function by telling it the
1983 * string is in UTF-8 if the locale is a
1984 * UTF-8 one. Since the string passed
1985 * here is invariant under UTF-8, we can
1986 * claim it's UTF-8 even though it isn't.
1988 PL_in_utf8_COLLATE_locale);
1991 /* Find out how long the transformation of a substring of 'longer'
1992 * is. Together the lengths of these transformations are
1993 * sufficient to calculate 'm' and 'b'. The substring is all of
1994 * 'longer' except the first character. This minimizes the chances
1995 * of being swayed by outliers */
1996 x_shorter = _mem_collxfrm(longer + 1,
1999 PL_in_utf8_COLLATE_locale);
2000 Safefree(x_shorter);
2002 /* If the results are nonsensical for this simple test, the whole
2003 * locale definition is suspect. Mark it so that locale collation
2004 * is not active at all for it. XXX Should we warn? */
2005 if ( x_len_shorter == 0
2006 || x_len_longer == 0
2007 || x_len_shorter >= x_len_longer)
2009 PL_collxfrm_mult = 0;
2010 PL_collxfrm_base = 0;
2013 SSize_t base; /* Temporary */
2015 /* We have both: m * strlen(longer) + b = x_len_longer
2016 * m * strlen(shorter) + b = x_len_shorter;
2017 * subtracting yields:
2018 * m * (strlen(longer) - strlen(shorter))
2019 * = x_len_longer - x_len_shorter
2020 * But we have set things up so that 'shorter' is 1 byte smaller
2021 * than 'longer'. Hence:
2022 * m = x_len_longer - x_len_shorter
2024 * But if something went wrong, make sure the multiplier is at
2027 if (x_len_longer > x_len_shorter) {
2028 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2031 PL_collxfrm_mult = 1;
2036 * but in case something has gone wrong, make sure it is
2038 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2043 /* Add 1 for the trailing NUL */
2044 PL_collxfrm_base = base + 1;
2049 if (DEBUG_L_TEST || debug_initialization) {
2050 PerlIO_printf(Perl_debug_log,
2051 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2053 " collate multipler=%zu, collate base=%zu\n",
2055 PL_in_utf8_COLLATE_locale,
2056 x_len_shorter, x_len_longer,
2057 PL_collxfrm_mult, PL_collxfrm_base);
2064 #endif /* USE_LOCALE_COLLATE */
2073 S_win32_setlocale(pTHX_ int category, const char* locale)
2075 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
2076 * difference between the two unless the input locale is "", which normally
2077 * means on Windows to get the machine default, which is set via the
2078 * computer's "Regional and Language Options" (or its current equivalent).
2079 * In POSIX, it instead means to find the locale from the user's
2080 * environment. This routine changes the Windows behavior to first look in
2081 * the environment, and, if anything is found, use that instead of going to
2082 * the machine default. If there is no environment override, the machine
2083 * default is used, by calling the real setlocale() with "".
2085 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2086 * use the particular category's variable if set; otherwise to use the LANG
2089 bool override_LC_ALL = FALSE;
2093 if (locale && strEQ(locale, "")) {
2097 locale = PerlEnv_getenv("LC_ALL");
2099 if (category == LC_ALL) {
2100 override_LC_ALL = TRUE;
2106 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2107 if (category == categories[i]) {
2108 locale = PerlEnv_getenv(category_names[i]);
2113 locale = PerlEnv_getenv("LANG");
2129 result = setlocale(category, locale);
2130 DEBUG_L(STMT_START {
2132 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2133 setlocale_debug_string(category, locale, result));
2137 if (! override_LC_ALL) {
2141 /* Here the input category was LC_ALL, and we have set it to what is in the
2142 * LANG variable or the system default if there is no LANG. But these have
2143 * lower priority than the other LC_foo variables, so override it for each
2144 * one that is set. (If they are set to "", it means to use the same thing
2145 * we just set LC_ALL to, so can skip) */
2147 for (i = 0; i < LC_ALL_INDEX; i++) {
2148 result = PerlEnv_getenv(category_names[i]);
2149 if (result && strNE(result, "")) {
2150 setlocale(categories[i], result);
2151 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2153 setlocale_debug_string(categories[i], result, "not captured")));
2157 result = setlocale(LC_ALL, NULL);
2158 DEBUG_L(STMT_START {
2160 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2162 setlocale_debug_string(LC_ALL, NULL, result));
2173 =head1 Locale-related functions and macros
2175 =for apidoc Perl_setlocale
2177 This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2178 taking the same parameters, and returning the same information, except that it
2179 returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2180 instead return C<C> if the underlying locale has a non-dot decimal point
2181 character, or a non-empty thousands separator for displaying floating point
2182 numbers. This is because perl keeps that locale category such that it has a
2183 dot and empty separator, changing the locale briefly during the operations
2184 where the underlying one is required. C<Perl_setlocale> knows about this, and
2185 compensates; regular C<setlocale> doesn't.
2187 Another reason it isn't completely a drop-in replacement is that it is
2188 declared to return S<C<const char *>>, whereas the system setlocale omits the
2189 C<const> (presumably because its API was specified long ago, and can't be
2190 updated; it is illegal to change the information C<setlocale> returns; doing
2191 so leads to segfaults.)
2193 Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2194 C<setlocale> can be completely ineffective on some platforms under some
2197 C<Perl_setlocale> should not be used to change the locale except on systems
2198 where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2199 the system C<setlocale()> is ineffective, returning the wrong information, and
2200 failing to actually change the locale. C<Perl_setlocale>, however works
2201 properly in all circumstances.
2203 The return points to a per-thread static buffer, which is overwritten the next
2204 time C<Perl_setlocale> is called from the same thread.
2211 Perl_setlocale(const int category, const char * locale)
2213 /* This wraps POSIX::setlocale() */
2217 PERL_UNUSED_ARG(category);
2218 PERL_UNUSED_ARG(locale);
2224 const char * retval;
2225 const char * newlocale;
2228 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2230 #ifdef USE_LOCALE_NUMERIC
2232 /* A NULL locale means only query what the current one is. We have the
2233 * LC_NUMERIC name saved, because we are normally switched into the C
2234 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2235 * the correct results. All other categories don't require special
2237 if (locale == NULL) {
2238 if (category == LC_NUMERIC) {
2240 /* We don't have to copy this return value, as it is a per-thread
2241 * variable, and won't change until a future setlocale */
2242 return PL_numeric_name;
2247 else if (category == LC_ALL) {
2248 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2257 retval = save_to_buffer(do_setlocale_r(category, locale),
2258 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2261 #if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2263 if (locale == NULL && category == LC_ALL) {
2264 RESTORE_LC_NUMERIC();
2269 DEBUG_L(PerlIO_printf(Perl_debug_log,
2270 "%s:%d: %s\n", __FILE__, __LINE__,
2271 setlocale_debug_string(category, locale, retval)));
2279 /* If locale == NULL, we are just querying the state */
2280 if (locale == NULL) {
2284 /* Now that have switched locales, we have to update our records to
2289 #ifdef USE_LOCALE_CTYPE
2296 #ifdef USE_LOCALE_COLLATE
2299 new_collate(retval);
2303 #ifdef USE_LOCALE_NUMERIC
2306 new_numeric(retval);
2314 /* LC_ALL updates all the things we care about. The values may not
2315 * be the same as 'retval', as the locale "" may have set things
2318 # ifdef USE_LOCALE_CTYPE
2320 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
2321 new_ctype(newlocale);
2322 Safefree(newlocale);
2324 # endif /* USE_LOCALE_CTYPE */
2325 # ifdef USE_LOCALE_COLLATE
2327 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
2328 new_collate(newlocale);
2329 Safefree(newlocale);
2332 # ifdef USE_LOCALE_NUMERIC
2334 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
2335 new_numeric(newlocale);
2336 Safefree(newlocale);
2338 # endif /* USE_LOCALE_NUMERIC */
2351 PERL_STATIC_INLINE const char *
2352 S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2354 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2355 * growing it if necessary */
2359 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2365 string_size = strlen(string) + offset + 1;
2367 if (*buf_size == 0) {
2368 Newx(*buf, string_size, char);
2369 *buf_size = string_size;
2371 else if (string_size > *buf_size) {
2372 Renew(*buf, string_size, char);
2373 *buf_size = string_size;
2376 Copy(string, *buf + offset, string_size - offset, char);
2382 =for apidoc Perl_langinfo
2384 This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
2385 taking the same C<item> parameter values, and returning the same information.
2386 But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2387 of Perl's locale handling from your code, and can be used on systems that lack
2388 a native C<nl_langinfo>.
2396 The reason it isn't quite a drop-in replacement is actually an advantage. The
2397 only difference is that it returns S<C<const char *>>, whereas plain
2398 C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2399 forbidden to write into the buffer. By declaring this C<const>, the compiler
2400 enforces this restriction, so if it is violated, you know at compilation time,
2401 rather than getting segfaults at runtime.
2405 It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
2406 without you having to write extra code. The reason for the extra code would be
2407 because these are from the C<LC_NUMERIC> locale category, which is normally
2408 kept set by Perl so that the radix is a dot, and the separator is the empty
2409 string, no matter what the underlying locale is supposed to be, and so to get
2410 the expected results, you have to temporarily toggle into the underlying
2411 locale, and later toggle back. (You could use plain C<nl_langinfo> and
2412 C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2413 the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2414 (or equivalent) locale would break a lot of CPAN, which is expecting the radix
2415 (decimal point) character to be a dot.)
2419 The system function it replaces can have its static return buffer trashed,
2420 not only by a subesequent call to that function, but by a C<freelocale>,
2421 C<setlocale>, or other locale change. The returned buffer of this function is
2422 not changed until the next call to it, so the buffer is never in a trashed
2427 Its return buffer is per-thread, so it also is never overwritten by a call to
2428 this function from another thread; unlike the function it replaces.
2432 But most importantly, it works on systems that don't have C<nl_langinfo>, such
2433 as Windows, hence makes your code more portable. Of the fifty-some possible
2434 items specified by the POSIX 2008 standard,
2435 L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2436 only one is completely unimplemented, though on non-Windows platforms, another
2437 significant one is also not implemented). It uses various techniques to
2438 recover the other items, including calling C<L<localeconv(3)>>, and
2439 C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2440 available. Later C<strftime()> versions have additional capabilities; C<""> is
2441 returned for those not available on your system.
2443 It is important to note that when called with an item that is recovered by
2444 using C<localeconv>, the buffer from any previous explicit call to
2445 C<localeconv> will be overwritten. This means you must save that buffer's
2446 contents if you need to access them after a call to this function. (But note
2447 that you might not want to be using C<localeconv()> directly anyway, because of
2448 issues like the ones listed in the second item of this list (above) for
2449 C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2450 call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2453 The details for those items which may deviate from what this emulation returns
2454 and what a native C<nl_langinfo()> would return are specified in
2459 When using C<Perl_langinfo> on systems that don't have a native
2460 C<nl_langinfo()>, you must
2462 #include "perl_langinfo.h"
2464 before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2465 C<#include> with this one. (Doing it this way keeps out the symbols that plain
2466 C<langinfo.h> would try to import into the namespace for code that doesn't need
2469 The original impetus for C<Perl_langinfo()> was so that code that needs to
2470 find out the current currency symbol, floating point radix character, or digit
2471 grouping separator can use, on all systems, the simpler and more
2472 thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2473 pain to make thread-friendly. For other fields returned by C<localeconv>, it
2474 is better to use the methods given in L<perlcall> to call
2475 L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2482 #ifdef HAS_NL_LANGINFO
2483 Perl_langinfo(const nl_item item)
2485 Perl_langinfo(const int item)
2488 return my_nl_langinfo(item, TRUE);
2492 #ifdef HAS_NL_LANGINFO
2493 S_my_nl_langinfo(const nl_item item, bool toggle)
2495 S_my_nl_langinfo(const int item, bool toggle)
2499 const char * retval;
2501 #ifdef USE_LOCALE_NUMERIC
2503 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2504 * two items, and only if not already there */
2505 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
2506 || PL_numeric_underlying))
2508 #endif /* No toggling needed if not using LC_NUMERIC */
2512 #if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2513 # if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2514 || ! defined(HAS_POSIX_2008_LOCALE) \
2515 || ! defined(DUPLOCALE)
2517 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
2518 * for those items dependent on it. This must be copied to a buffer before
2519 * switching back, as some systems destroy the buffer when setlocale() is
2523 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2526 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2529 LOCALE_LOCK; /* Prevent interference from another thread executing
2530 this code section (the only call to nl_langinfo in
2534 /* Copy to a per-thread buffer, which is also one that won't be
2535 * destroyed by a subsequent setlocale(), such as the
2536 * RESTORE_LC_NUMERIC may do just below. */
2537 retval = save_to_buffer(nl_langinfo(item),
2538 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2543 RESTORE_LC_NUMERIC();
2547 # else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2550 bool do_free = FALSE;
2551 locale_t cur = uselocale((locale_t) 0);
2553 if (cur == LC_GLOBAL_LOCALE) {
2554 cur = duplocale(LC_GLOBAL_LOCALE);
2558 # ifdef USE_LOCALE_NUMERIC
2561 if (PL_underlying_numeric_obj) {
2562 cur = PL_underlying_numeric_obj;
2565 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2572 /* We have to save it to a buffer, because the freelocale() just below
2573 * can invalidate the internal one */
2574 retval = save_to_buffer(nl_langinfo_l(item, cur),
2575 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2584 if (strEQ(retval, "")) {
2585 if (item == YESSTR) {
2588 if (item == NOSTR) {
2595 #else /* Below, emulate nl_langinfo as best we can */
2599 # ifdef HAS_LOCALECONV
2601 const struct lconv* lc;
2603 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2605 # ifdef TS_W32_BROKEN_LOCALECONV
2607 const char * save_global;
2608 const char * save_thread;
2616 # ifdef HAS_STRFTIME
2619 bool return_format = FALSE; /* Return the %format, not the value */
2620 const char * format;
2624 /* We copy the results to a per-thread buffer, even if not
2625 * multi-threaded. This is in part to simplify this code, and partly
2626 * because we need a buffer anyway for strftime(), and partly because a
2627 * call of localeconv() could otherwise wipe out the buffer, and the
2628 * programmer would not be expecting this, as this is a nl_langinfo()
2629 * substitute after all, so s/he might be thinking their localeconv()
2630 * is safe until another localeconv() call. */
2635 /* This is unimplemented */
2636 case ERA: /* For use with strftime() %E modifier */
2641 /* We use only an English set, since we don't know any more */
2642 case YESEXPR: return "^[+1yY]";
2643 case YESSTR: return "yes";
2644 case NOEXPR: return "^[-0nN]";
2645 case NOSTR: return "no";
2651 /* On non-windows, this is unimplemented, in part because of
2652 * inconsistencies between vendors. The Darwin native
2653 * nl_langinfo() implementation simply looks at everything past
2654 * any dot in the name, but that doesn't work for other
2655 * vendors. Many Linux locales that don't have UTF-8 in their
2656 * names really are UTF-8, for example; z/OS locales that do
2657 * have UTF-8 in their names, aren't really UTF-8 */
2662 { /* But on Windows, the name does seem to be consistent, so
2667 const char * name = my_setlocale(LC_CTYPE, NULL);
2669 if (isNAME_C_OR_POSIX(name)) {
2670 return "ANSI_X3.4-1968";
2673 /* Find the dot in the locale name */
2674 first = (const char *) strchr(name, '.');
2680 /* Look at everything past the dot */
2685 if (! isDIGIT(*p)) {
2692 /* Here everything past the dot is a digit. Treat it as a
2694 retval = save_to_buffer("CP", &PL_langinfo_buf,
2695 &PL_langinfo_bufsize, 0);
2696 offset = STRLENs("CP");
2700 retval = save_to_buffer(first, &PL_langinfo_buf,
2701 &PL_langinfo_bufsize, offset);
2707 # ifdef HAS_LOCALECONV
2711 /* We don't bother with localeconv_l() because any system that
2712 * has it is likely to also have nl_langinfo() */
2714 LOCALE_LOCK_V; /* Prevent interference with other threads
2715 using localeconv() */
2717 # ifdef TS_W32_BROKEN_LOCALECONV
2719 /* This is a workaround for a Windows bug prior to VS 15.
2720 * What we do here is, while locked, switch to the global
2721 * locale so localeconv() works; then switch back just before
2722 * the unlock. This can screw things up if some thread is
2723 * already using the global locale while assuming no other is.
2724 * A different workaround would be to call GetCurrencyFormat on
2725 * a known value, and parse it; patches welcome
2727 * We have to use LC_ALL instead of LC_MONETARY because of
2728 * another bug in Windows */
2730 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2731 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2732 save_global= savepv(my_setlocale(LC_ALL, NULL));
2733 my_setlocale(LC_ALL, save_thread);
2739 || ! lc->currency_symbol
2740 || strEQ("", lc->currency_symbol))
2746 /* Leave the first spot empty to be filled in below */
2747 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2748 &PL_langinfo_bufsize, 1);
2749 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2750 { /* khw couldn't figure out how the localedef specifications
2751 would show that the $ should replace the radix; this is
2752 just a guess as to how it might work.*/
2753 PL_langinfo_buf[0] = '.';
2755 else if (lc->p_cs_precedes) {
2756 PL_langinfo_buf[0] = '-';
2759 PL_langinfo_buf[0] = '+';
2762 # ifdef TS_W32_BROKEN_LOCALECONV
2764 my_setlocale(LC_ALL, save_global);
2765 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2766 my_setlocale(LC_ALL, save_thread);
2767 Safefree(save_global);
2768 Safefree(save_thread);
2775 # ifdef TS_W32_BROKEN_LOCALECONV
2779 /* For this, we output a known simple floating point number to
2780 * a buffer, and parse it, looking for the radix */
2783 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2786 if (PL_langinfo_bufsize < 10) {
2787 PL_langinfo_bufsize = 10;
2788 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2791 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2793 if (needed_size >= (int) PL_langinfo_bufsize) {
2794 PL_langinfo_bufsize = needed_size + 1;
2795 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2796 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2798 assert(needed_size < (int) PL_langinfo_bufsize);
2801 ptr = PL_langinfo_buf;
2802 e = PL_langinfo_buf + PL_langinfo_bufsize;
2805 while (ptr < e && *ptr != '1') {
2812 while (ptr < e && *ptr != '5') {
2816 /* Everything in between is the radix string */
2818 PL_langinfo_buf[0] = '?';
2819 PL_langinfo_buf[1] = '\0';
2823 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2827 RESTORE_LC_NUMERIC();
2830 retval = PL_langinfo_buf;
2835 case RADIXCHAR: /* No special handling needed */
2842 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2845 LOCALE_LOCK_V; /* Prevent interference with other threads
2846 using localeconv() */
2848 # ifdef TS_W32_BROKEN_LOCALECONV
2850 /* This should only be for the thousands separator. A
2851 * different work around would be to use GetNumberFormat on a
2852 * known value and parse the result to find the separator */
2853 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2854 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2855 save_global = savepv(my_setlocale(LC_ALL, NULL));
2856 my_setlocale(LC_ALL, save_thread);
2858 /* This is the start of code that for broken Windows replaces
2859 * the above and below code, and instead calls
2860 * GetNumberFormat() and then would parse that to find the
2861 * thousands separator. It needs to handle UTF-16 vs -8
2864 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2865 DEBUG_L(PerlIO_printf(Perl_debug_log,
2866 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2867 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2877 temp = (item == RADIXCHAR)
2879 : lc->thousands_sep;
2885 retval = save_to_buffer(temp, &PL_langinfo_buf,
2886 &PL_langinfo_bufsize, 0);
2888 # ifdef TS_W32_BROKEN_LOCALECONV
2890 my_setlocale(LC_ALL, save_global);
2891 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2892 my_setlocale(LC_ALL, save_thread);
2893 Safefree(save_global);
2894 Safefree(save_thread);
2901 RESTORE_LC_NUMERIC();
2907 # ifdef HAS_STRFTIME
2909 /* These are defined by C89, so we assume that strftime supports
2910 * them, and so are returned unconditionally; they may not be what
2911 * the locale actually says, but should give good enough results
2912 * for someone using them as formats (as opposed to trying to parse
2913 * them to figure out what the locale says). The other format
2914 * items are actually tested to verify they work on the platform */
2915 case D_FMT: return "%x";
2916 case T_FMT: return "%X";
2917 case D_T_FMT: return "%c";
2919 /* These formats are only available in later strfmtime's */
2920 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
2922 /* The rest can be gotten from most versions of strftime(). */
2923 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2924 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2926 case AM_STR: case PM_STR:
2927 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2928 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2929 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2930 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2931 case DAY_5: case DAY_6: case DAY_7:
2932 case MON_1: case MON_2: case MON_3: case MON_4:
2933 case MON_5: case MON_6: case MON_7: case MON_8:
2934 case MON_9: case MON_10: case MON_11: case MON_12:
2938 init_tm(&tm); /* Precaution against core dumps */
2942 tm.tm_year = 2017 - 1900;
2949 "panic: %s: %d: switch case: %d problem",
2950 __FILE__, __LINE__, item);
2951 NOT_REACHED; /* NOTREACHED */
2953 case PM_STR: tm.tm_hour = 18;
2958 case ABDAY_7: tm.tm_wday++;
2959 case ABDAY_6: tm.tm_wday++;
2960 case ABDAY_5: tm.tm_wday++;
2961 case ABDAY_4: tm.tm_wday++;
2962 case ABDAY_3: tm.tm_wday++;
2963 case ABDAY_2: tm.tm_wday++;
2968 case DAY_7: tm.tm_wday++;
2969 case DAY_6: tm.tm_wday++;
2970 case DAY_5: tm.tm_wday++;
2971 case DAY_4: tm.tm_wday++;
2972 case DAY_3: tm.tm_wday++;
2973 case DAY_2: tm.tm_wday++;
2978 case ABMON_12: tm.tm_mon++;
2979 case ABMON_11: tm.tm_mon++;
2980 case ABMON_10: tm.tm_mon++;
2981 case ABMON_9: tm.tm_mon++;
2982 case ABMON_8: tm.tm_mon++;
2983 case ABMON_7: tm.tm_mon++;
2984 case ABMON_6: tm.tm_mon++;
2985 case ABMON_5: tm.tm_mon++;
2986 case ABMON_4: tm.tm_mon++;
2987 case ABMON_3: tm.tm_mon++;
2988 case ABMON_2: tm.tm_mon++;
2993 case MON_12: tm.tm_mon++;
2994 case MON_11: tm.tm_mon++;
2995 case MON_10: tm.tm_mon++;
2996 case MON_9: tm.tm_mon++;
2997 case MON_8: tm.tm_mon++;
2998 case MON_7: tm.tm_mon++;
2999 case MON_6: tm.tm_mon++;
3000 case MON_5: tm.tm_mon++;
3001 case MON_4: tm.tm_mon++;
3002 case MON_3: tm.tm_mon++;
3003 case MON_2: tm.tm_mon++;
3010 return_format = TRUE;
3015 return_format = TRUE;
3020 return_format = TRUE;
3025 return_format = TRUE;
3030 format = "%Ow"; /* Find the alternate digit for 0 */
3034 /* We can't use my_strftime() because it doesn't look at
3036 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3039 /* A zero return means one of:
3040 * a) there wasn't enough space in PL_langinfo_buf
3041 * b) the format, like a plain %p, returns empty
3042 * c) it was an illegal format, though some
3043 * implementations of strftime will just return the
3044 * illegal format as a plain character sequence.
3046 * To quickly test for case 'b)', try again but precede
3047 * the format with a plain character. If that result is
3048 * still empty, the problem is either 'a)' or 'c)' */
3050 Size_t format_size = strlen(format) + 1;
3051 Size_t mod_size = format_size + 1;
3055 Newx(mod_format, mod_size, char);
3056 Newx(temp_result, PL_langinfo_bufsize, char);
3058 my_strlcpy(mod_format + 1, format, mod_size);
3059 len = strftime(temp_result,
3060 PL_langinfo_bufsize,
3062 Safefree(mod_format);
3063 Safefree(temp_result);
3065 /* If 'len' is non-zero, it means that we had a case like
3066 * %p which means the current locale doesn't use a.m. or
3067 * p.m., and that is valid */
3070 /* Here, still didn't work. If we get well beyond a
3071 * reasonable size, bail out to prevent an infinite
3074 if (PL_langinfo_bufsize > 100 * format_size) {
3075 *PL_langinfo_buf = '\0';
3078 /* Double the buffer size to retry; Add 1 in case
3079 * original was 0, so we aren't stuck at 0. */
3080 PL_langinfo_bufsize *= 2;
3081 PL_langinfo_bufsize++;
3082 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3090 /* Here, we got a result.
3092 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3093 * alternate format for wday 0. If the value is the same as
3094 * the normal 0, there isn't an alternate, so clear the buffer.
3096 if ( item == ALT_DIGITS
3097 && strEQ(PL_langinfo_buf, "0"))
3099 *PL_langinfo_buf = '\0';
3102 /* ALT_DIGITS is problematic. Experiments on it showed that
3103 * strftime() did not always work properly when going from
3104 * alt-9 to alt-10. Only a few locales have this item defined,
3105 * and in all of them on Linux that khw was able to find,
3106 * nl_langinfo() merely returned the alt-0 character, possibly
3107 * doubled. Most Unicode digits are in blocks of 10
3108 * consecutive code points, so that is sufficient information
3109 * for those scripts, as we can infer alt-1, alt-2, .... But
3110 * for a Japanese locale, a CJK ideographic 0 is returned, and
3111 * the CJK digits are not in code point order, so you can't
3112 * really infer anything. The localedef for this locale did
3113 * specify the succeeding digits, so that strftime() works
3114 * properly on them, without needing to infer anything. But
3115 * the nl_langinfo() return did not give sufficient information
3116 * for the caller to understand what's going on. So until
3117 * there is evidence that it should work differently, this
3118 * returns the alt-0 string for ALT_DIGITS.
3120 * wday was chosen because its range is all a single digit.
3121 * Things like tm_sec have two digits as the minimum: '00' */
3125 retval = PL_langinfo_buf;
3127 /* If to return the format, not the value, overwrite the buffer
3128 * with it. But some strftime()s will keep the original format
3129 * if illegal, so change those to "" */
3130 if (return_format) {
3131 if (strEQ(PL_langinfo_buf, format)) {
3132 *PL_langinfo_buf = '\0';
3135 retval = save_to_buffer(format, &PL_langinfo_buf,
3136 &PL_langinfo_bufsize, 0);
3154 * Initialize locale awareness.
3157 Perl_init_i18nl10n(pTHX_ int printwarn)
3161 * 0 if not to output warning when setup locale is bad
3162 * 1 if to output warning based on value of PERL_BADLANG
3163 * >1 if to output regardless of PERL_BADLANG
3166 * 1 = set ok or not applicable,
3167 * 0 = fallback to a locale of lower priority
3168 * -1 = fallback to all locales failed, not even to the C locale
3170 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3171 * set, debugging information is output.
3173 * This looks more complicated than it is, mainly due to the #ifdefs.
3175 * We try to set LC_ALL to the value determined by the environment. If
3176 * there is no LC_ALL on this platform, we try the individual categories we
3177 * know about. If this works, we are done.
3179 * But if it doesn't work, we have to do something else. We search the
3180 * environment variables ourselves instead of relying on the system to do
3181 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3182 * think there is one), and the ultimate fallback "C". This is all done in
3183 * the same loop as above to avoid duplicating code, but it makes things
3184 * more complex. The 'trial_locales' array is initialized with just one
3185 * element; it causes the behavior described in the paragraph above this to
3186 * happen. If that fails, we add elements to 'trial_locales', and do extra
3187 * loop iterations to cause the behavior described in this paragraph.
3189 * On Ultrix, the locale MUST come from the environment, so there is
3190 * preliminary code to set it. I (khw) am not sure that it is necessary,
3191 * and that this couldn't be folded into the loop, but barring any real
3192 * platforms to test on, it's staying as-is
3194 * A slight complication is that in embedded Perls, the locale may already
3195 * be set-up, and we don't want to get it from the normal environment
3196 * variables. This is handled by having a special environment variable
3197 * indicate we're in this situation. We simply set setlocale's 2nd
3198 * parameter to be a NULL instead of "". That indicates to setlocale that
3199 * it is not to change anything, but to return the current value,
3200 * effectively initializing perl's db to what the locale already is.
3202 * We play the same trick with NULL if a LC_ALL succeeds. We call
3203 * setlocale() on the individual categores with NULL to get their existing
3204 * values for our db, instead of trying to change them.
3213 PERL_UNUSED_ARG(printwarn);
3215 #else /* USE_LOCALE */
3218 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
3222 /* NULL uses the existing already set up locale */
3223 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3226 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3227 unsigned int trial_locales_count;
3228 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3229 const char * const lang = savepv(PerlEnv_getenv("LANG"));
3230 bool setlocale_failure = FALSE;
3233 /* A later getenv() could zap this, so only use here */
3234 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3236 const bool locwarn = (printwarn > 1
3238 && ( ! bad_lang_use_once
3240 /* disallow with "" or "0" */
3242 && strNE("0", bad_lang_use_once)))));
3244 /* setlocale() return vals; not copied so must be looked at immediately */
3245 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
3247 /* current locale for given category; should have been copied so aren't
3249 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
3253 /* In some systems you can find out the system default locale
3254 * and use that as the fallback locale. */
3255 # define SYSTEM_DEFAULT_LOCALE
3257 # ifdef SYSTEM_DEFAULT_LOCALE
3259 const char *system_default_locale = NULL;
3264 # define DEBUG_LOCALE_INIT(a,b,c)
3267 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
3269 # define DEBUG_LOCALE_INIT(category, locale, result) \
3271 if (debug_initialization) { \
3272 PerlIO_printf(Perl_debug_log, \
3274 __FILE__, __LINE__, \
3275 setlocale_debug_string(category, \
3281 /* Make sure the parallel arrays are properly set up */
3282 # ifdef USE_LOCALE_NUMERIC
3283 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3284 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
3285 # ifdef USE_POSIX_2008_LOCALE
3286 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3289 # ifdef USE_LOCALE_CTYPE
3290 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3291 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
3292 # ifdef USE_POSIX_2008_LOCALE
3293 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3296 # ifdef USE_LOCALE_COLLATE
3297 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3298 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
3299 # ifdef USE_POSIX_2008_LOCALE
3300 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3303 # ifdef USE_LOCALE_TIME
3304 assert(categories[LC_TIME_INDEX] == LC_TIME);
3305 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
3306 # ifdef USE_POSIX_2008_LOCALE
3307 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3310 # ifdef USE_LOCALE_MESSAGES
3311 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3312 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
3313 # ifdef USE_POSIX_2008_LOCALE
3314 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3317 # ifdef USE_LOCALE_MONETARY
3318 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3319 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
3320 # ifdef USE_POSIX_2008_LOCALE
3321 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3324 # ifdef USE_LOCALE_ADDRESS
3325 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3326 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
3327 # ifdef USE_POSIX_2008_LOCALE
3328 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3331 # ifdef USE_LOCALE_IDENTIFICATION
3332 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3333 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
3334 # ifdef USE_POSIX_2008_LOCALE
3335 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3338 # ifdef USE_LOCALE_MEASUREMENT
3339 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3340 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
3341 # ifdef USE_POSIX_2008_LOCALE
3342 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3345 # ifdef USE_LOCALE_PAPER
3346 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3347 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
3348 # ifdef USE_POSIX_2008_LOCALE
3349 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3352 # ifdef USE_LOCALE_TELEPHONE
3353 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3354 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
3355 # ifdef USE_POSIX_2008_LOCALE
3356 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3360 assert(categories[LC_ALL_INDEX] == LC_ALL);
3361 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3362 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
3363 # ifdef USE_POSIX_2008_LOCALE
3364 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3367 # endif /* DEBUGGING */
3369 /* Initialize the cache of the program's UTF-8ness for the always known
3370 * locales C and POSIX */
3371 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3372 sizeof(PL_locale_utf8ness));
3374 # ifdef USE_THREAD_SAFE_LOCALE
3377 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3381 # ifdef USE_POSIX_2008_LOCALE
3383 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3384 if (! PL_C_locale_obj) {
3385 Perl_croak_nocontext(
3386 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3388 if (DEBUG_Lv_TEST || debug_initialization) {
3389 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3394 # ifdef USE_LOCALE_NUMERIC
3396 PL_numeric_radix_sv = newSVpvs(".");
3400 # if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3402 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3403 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3406 # ifdef LOCALE_ENVIRON_REQUIRED
3409 * Ultrix setlocale(..., "") fails if there are no environment
3410 * variables from which to get a locale name.
3414 # error Ultrix without LC_ALL not implemented
3420 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3421 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3422 if (sl_result[LC_ALL_INDEX])
3425 setlocale_failure = TRUE;
3427 if (! setlocale_failure) {
3428 const char * locale_param;
3429 for (i = 0; i < LC_ALL_INDEX; i++) {
3430 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3433 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3434 if (! sl_result[i]) {
3435 setlocale_failure = TRUE;
3437 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
3442 # endif /* LC_ALL */
3443 # endif /* LOCALE_ENVIRON_REQUIRED */
3445 /* We try each locale in the list until we get one that works, or exhaust
3446 * the list. Normally the loop is executed just once. But if setting the
3447 * locale fails, inside the loop we add fallback trials to the array and so
3448 * will execute the loop multiple times */
3449 trial_locales[0] = setlocale_init;
3450 trial_locales_count = 1;
3452 for (i= 0; i < trial_locales_count; i++) {
3453 const char * trial_locale = trial_locales[i];
3457 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3458 * when i==0, but I (khw) don't think that behavior makes much
3460 setlocale_failure = FALSE;
3462 # ifdef SYSTEM_DEFAULT_LOCALE
3463 # ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
3465 /* On Windows machines, an entry of "" after the 0th means to use
3466 * the system default locale, which we now proceed to get. */
3467 if (strEQ(trial_locale, "")) {
3470 /* Note that this may change the locale, but we are going to do
3471 * that anyway just below */
3472 system_default_locale = do_setlocale_c(LC_ALL, "");
3473 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
3475 /* Skip if invalid or if it's already on the list of locales to
3477 if (! system_default_locale) {
3478 goto next_iteration;
3480 for (j = 0; j < trial_locales_count; j++) {
3481 if (strEQ(system_default_locale, trial_locales[j])) {
3482 goto next_iteration;
3486 trial_locale = system_default_locale;
3489 # error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3491 # endif /* SYSTEM_DEFAULT_LOCALE */
3497 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3498 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3499 if (! sl_result[LC_ALL_INDEX]) {
3500 setlocale_failure = TRUE;
3503 /* Since LC_ALL succeeded, it should have changed all the other
3504 * categories it can to its value; so we massage things so that the
3505 * setlocales below just return their category's current values.
3506 * This adequately handles the case in NetBSD where LC_COLLATE may
3507 * not be defined for a locale, and setting it individually will
3508 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
3509 * the POSIX locale. */
3510 trial_locale = NULL;
3513 # endif /* LC_ALL */
3515 if (! setlocale_failure) {
3517 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3519 = savepv(do_setlocale_r(categories[j], trial_locale));
3520 if (! curlocales[j]) {
3521 setlocale_failure = TRUE;
3523 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3526 if (! setlocale_failure) { /* All succeeded */
3527 break; /* Exit trial_locales loop */
3531 /* Here, something failed; will need to try a fallback. */
3537 if (locwarn) { /* Output failure info only on the first one */
3541 PerlIO_printf(Perl_error_log,
3542 "perl: warning: Setting locale failed.\n");
3544 # else /* !LC_ALL */
3546 PerlIO_printf(Perl_error_log,
3547 "perl: warning: Setting locale failed for the categories:\n\t");
3549 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3550 if (! curlocales[j]) {
3551 PerlIO_printf(Perl_error_log, category_names[j]);
3554 Safefree(curlocales[j]);
3558 # endif /* LC_ALL */
3560 PerlIO_printf(Perl_error_log,
3561 "perl: warning: Please check that your locale settings:\n");
3565 PerlIO_printf(Perl_error_log,
3566 "\tLANGUAGE = %c%s%c,\n",
3567 language ? '"' : '(',
3568 language ? language : "unset",
3569 language ? '"' : ')');
3572 PerlIO_printf(Perl_error_log,
3573 "\tLC_ALL = %c%s%c,\n",
3575 lc_all ? lc_all : "unset",
3576 lc_all ? '"' : ')');
3578 # if defined(USE_ENVIRON_ARRAY)
3583 /* Look through the environment for any variables of the
3584 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3585 * already handled above. These are assumed to be locale
3586 * settings. Output them and their values. */
3587 for (e = environ; *e; e++) {
3588 const STRLEN prefix_len = sizeof("LC_") - 1;
3591 if ( strBEGINs(*e, "LC_")
3592 && ! strBEGINs(*e, "LC_ALL=")
3593 && (uppers_len = strspn(*e + prefix_len,
3594 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3595 && ((*e)[prefix_len + uppers_len] == '='))
3597 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
3598 (int) (prefix_len + uppers_len), *e,
3599 *e + prefix_len + uppers_len + 1);
3606 PerlIO_printf(Perl_error_log,
3607 "\t(possibly more locale environment variables)\n");
3611 PerlIO_printf(Perl_error_log,
3612 "\tLANG = %c%s%c\n",
3614 lang ? lang : "unset",
3617 PerlIO_printf(Perl_error_log,
3618 " are supported and installed on your system.\n");
3621 /* Calculate what fallback locales to try. We have avoided this
3622 * until we have to, because failure is quite unlikely. This will
3623 * usually change the upper bound of the loop we are in.
3625 * Since the system's default way of setting the locale has not
3626 * found one that works, We use Perl's defined ordering: LC_ALL,
3627 * LANG, and the C locale. We don't try the same locale twice, so
3628 * don't add to the list if already there. (On POSIX systems, the
3629 * LC_ALL element will likely be a repeat of the 0th element "",
3630 * but there's no harm done by doing it explicitly.
3632 * Note that this tries the LC_ALL environment variable even on
3633 * systems which have no LC_ALL locale setting. This may or may
3634 * not have been originally intentional, but there's no real need
3635 * to change the behavior. */
3637 for (j = 0; j < trial_locales_count; j++) {
3638 if (strEQ(lc_all, trial_locales[j])) {
3642 trial_locales[trial_locales_count++] = lc_all;
3647 for (j = 0; j < trial_locales_count; j++) {
3648 if (strEQ(lang, trial_locales[j])) {
3652 trial_locales[trial_locales_count++] = lang;
3656 # if defined(WIN32) && defined(LC_ALL)
3658 /* For Windows, we also try the system default locale before "C".
3659 * (If there exists a Windows without LC_ALL we skip this because
3660 * it gets too complicated. For those, the "C" is the next
3661 * fallback possibility). The "" is the same as the 0th element of
3662 * the array, but the code at the loop above knows to treat it
3663 * differently when not the 0th */
3664 trial_locales[trial_locales_count++] = "";
3668 for (j = 0; j < trial_locales_count; j++) {
3669 if (strEQ("C", trial_locales[j])) {
3673 trial_locales[trial_locales_count++] = "C";
3676 } /* end of first time through the loop */
3684 } /* end of looping through the trial locales */
3686 if (ok < 1) { /* If we tried to fallback */
3688 if (! setlocale_failure) { /* fallback succeeded */
3689 msg = "Falling back to";
3691 else { /* fallback failed */
3694 /* We dropped off the end of the loop, so have to decrement i to
3695 * get back to the value the last time through */
3699 msg = "Failed to fall back to";
3701 /* To continue, we should use whatever values we've got */
3703 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3704 Safefree(curlocales[j]);
3705 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3706 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3711 const char * description;
3712 const char * name = "";
3713 if (strEQ(trial_locales[i], "C")) {
3714 description = "the standard locale";
3718 # ifdef SYSTEM_DEFAULT_LOCALE
3720 else if (strEQ(trial_locales[i], "")) {
3721 description = "the system default locale";
3722 if (system_default_locale) {
3723 name = system_default_locale;
3727 # endif /* SYSTEM_DEFAULT_LOCALE */
3730 description = "a fallback locale";
3731 name = trial_locales[i];
3733 if (name && strNE(name, "")) {
3734 PerlIO_printf(Perl_error_log,
3735 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3738 PerlIO_printf(Perl_error_log,
3739 "perl: warning: %s %s.\n", msg, description);
3742 } /* End of tried to fallback */
3744 /* Done with finding the locales; update our records */
3746 # ifdef USE_LOCALE_CTYPE
3748 new_ctype(curlocales[LC_CTYPE_INDEX]);
3751 # ifdef USE_LOCALE_COLLATE
3753 new_collate(curlocales[LC_COLLATE_INDEX]);
3756 # ifdef USE_LOCALE_NUMERIC
3758 new_numeric(curlocales[LC_NUMERIC_INDEX]);
3762 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
3764 # if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
3766 /* This caches whether each category's locale is UTF-8 or not. This
3767 * may involve changing the locale. It is ok to do this at
3768 * initialization time before any threads have started, but not later
3769 * unless thread-safe operations are used.
3770 * Caching means that if the program heeds our dictate not to change
3771 * locales in threaded applications, this data will remain valid, and
3772 * it may get queried without having to change locales. If the
3773 * environment is such that all categories have the same locale, this
3774 * isn't needed, as the code will not change the locale; but this
3775 * handles the uncommon case where the environment has disparate
3776 * locales for the categories */
3777 (void) _is_cur_LC_category_utf8(categories[i]);
3781 Safefree(curlocales[i]);
3784 # if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3786 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
3787 * locale is UTF-8. The call to new_ctype() just above has already
3788 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3789 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3790 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3791 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3792 PL_utf8locale = PL_in_utf8_CTYPE_locale;
3794 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
3795 This is an alternative to using the -C command line switch
3796 (the -C if present will override this). */
3798 const char *p = PerlEnv_getenv("PERL_UNICODE");
3799 PL_unicode = p ? parse_unicode_opts(&p) : 0;
3800 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3814 #endif /* USE_LOCALE */
3817 /* So won't continue to output stuff */
3818 DEBUG_INITIALIZATION_set(FALSE);
3825 #ifdef USE_LOCALE_COLLATE
3828 Perl__mem_collxfrm(pTHX_ const char *input_string,
3829 STRLEN len, /* Length of 'input_string' */
3830 STRLEN *xlen, /* Set to length of returned string
3831 (not including the collation index
3833 bool utf8 /* Is the input in UTF-8? */
3837 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3838 * differences. First, it handles embedded NULs. Second, it allocates a bit
3839 * more memory than needed for the transformed data itself. The real
3840 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
3841 * the length of that, and doesn't include the collation index size.
3842 * Please see sv_collxfrm() to see how this is used. */
3844 #define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3846 char * s = (char *) input_string;
3847 STRLEN s_strlen = strlen(input_string);
3849 STRLEN xAlloc; /* xalloc is a reserved word in VC */
3850 STRLEN length_in_chars;
3851 bool first_time = TRUE; /* Cleared after first loop iteration */
3853 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3855 /* Must be NUL-terminated */
3856 assert(*(input_string + len) == '\0');
3858 /* If this locale has defective collation, skip */
3859 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
3860 DEBUG_L(PerlIO_printf(Perl_debug_log,
3861 "_mem_collxfrm: locale's collation is defective\n"));
3865 /* Replace any embedded NULs with the control that sorts before any others.
3866 * This will give as good as possible results on strings that don't
3867 * otherwise contain that character, but otherwise there may be
3868 * less-than-perfect results with that character and NUL. This is
3869 * unavoidable unless we replace strxfrm with our own implementation. */
3870 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3874 STRLEN sans_nuls_len;
3875 int try_non_controls;
3876 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3877 making sure 2nd byte is NUL.
3879 STRLEN this_replacement_len;
3881 /* If we don't know what non-NUL control character sorts lowest for
3882 * this locale, find it */
3883 if (PL_strxfrm_NUL_replacement == '\0') {
3885 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
3886 includes the collation index
3889 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
3891 /* Unlikely, but it may be that no control will work to replace
3892 * NUL, in which case we instead look for any character. Controls
3893 * are preferred because collation order is, in general, context
3894 * sensitive, with adjoining characters affecting the order, and
3895 * controls are less likely to have such interactions, allowing the
3896 * NUL-replacement to stand on its own. (Another way to look at it
3897 * is to imagine what would happen if the NUL were replaced by a
3898 * combining character; it wouldn't work out all that well.) */
3899 for (try_non_controls = 0;
3900 try_non_controls < 2;
3903 /* Look through all legal code points (NUL isn't) */
3904 for (j = 1; j < 256; j++) {
3905 char * x; /* j's xfrm plus collation index */
3906 STRLEN x_len; /* length of 'x' */
3907 STRLEN trial_len = 1;
3908 char cur_source[] = { '\0', '\0' };
3910 /* Skip non-controls the first time through the loop. The
3911 * controls in a UTF-8 locale are the L1 ones */
3912 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3919 /* Create a 1-char string of the current code point */
3920 cur_source[0] = (char) j;
3922 /* Then transform it */
3923 x = _mem_collxfrm(cur_source, trial_len, &x_len,
3924 0 /* The string is not in UTF-8 */);
3926 /* Ignore any character that didn't successfully transform.
3932 /* If this character's transformation is lower than
3933 * the current lowest, this one becomes the lowest */
3934 if ( cur_min_x == NULL
3935 || strLT(x + COLLXFRM_HDR_LEN,
3936 cur_min_x + COLLXFRM_HDR_LEN))
3938 PL_strxfrm_NUL_replacement = j;
3944 } /* end of loop through all 255 characters */
3946 /* Stop looking if found */
3951 /* Unlikely, but possible, if there aren't any controls that
3952 * work in the locale, repeat the loop, looking for any
3953 * character that works */
3954 DEBUG_L(PerlIO_printf(Perl_debug_log,
3955 "_mem_collxfrm: No control worked. Trying non-controls\n"));
3956 } /* End of loop to try first the controls, then any char */
3959 DEBUG_L(PerlIO_printf(Perl_debug_log,
3960 "_mem_collxfrm: Couldn't find any character to replace"
3961 " embedded NULs in locale %s with", PL_collation_name));
3965 DEBUG_L(PerlIO_printf(Perl_debug_log,
3966 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
3967 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
3969 Safefree(cur_min_x);
3970 } /* End of determining the character that is to replace NULs */
3972 /* If the replacement is variant under UTF-8, it must match the
3973 * UTF8-ness of the original */
3974 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3975 this_replacement_char[0] =
3976 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3977 this_replacement_char[1] =
3978 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
3979 this_replacement_len = 2;
3982 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
3983 /* this_replacement_char[1] = '\0' was done at initialization */
3984 this_replacement_len = 1;
3987 /* The worst case length for the replaced string would be if every
3988 * character in it is NUL. Multiply that by the length of each
3989 * replacement, and allow for a trailing NUL */
3990 sans_nuls_len = (len * this_replacement_len) + 1;
3991 Newx(sans_nuls, sans_nuls_len, char);
3994 /* Replace each NUL with the lowest collating control. Loop until have
3995 * exhausted all the NULs */
3996 while (s + s_strlen < e) {
3997 my_strlcat(sans_nuls, s, sans_nuls_len);
3999 /* Do the actual replacement */
4000 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
4002 /* Move past the input NUL */
4004 s_strlen = strlen(s);
4007 /* And add anything that trails the final NUL */
4008 my_strlcat(sans_nuls, s, sans_nuls_len);
4010 /* Switch so below we transform this modified string */
4013 } /* End of replacing NULs */
4015 /* Make sure the UTF8ness of the string and locale match */
4016 if (utf8 != PL_in_utf8_COLLATE_locale) {
4017 /* XXX convert above Unicode to 10FFFF? */
4018 const char * const t = s; /* Temporary so we can later find where the
4021 /* Here they don't match. Change the string's to be what the locale is
4024 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
4025 s = (char *) bytes_to_utf8((const U8 *) s, &len);
4028 else { /* locale is not UTF-8; but input is; downgrade the input */
4030 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
4032 /* If the downgrade was successful we are done, but if the input
4033 * contains things that require UTF-8 to represent, have to do
4034 * damage control ... */
4035 if (UNLIKELY(utf8)) {
4037 /* What we do is construct a non-UTF-8 string with
4038 * 1) the characters representable by a single byte converted
4039 * to be so (if necessary);
4040 * 2) and the rest converted to collate the same as the
4041 * highest collating representable character. That makes
4042 * them collate at the end. This is similar to how we
4043 * handle embedded NULs, but we use the highest collating
4044 * code point instead of the smallest. Like the NUL case,
4045 * this isn't perfect, but is the best we can reasonably
4046 * do. Every above-255 code point will sort the same as
4047 * the highest-sorting 0-255 code point. If that code
4048 * point can combine in a sequence with some other code
4049 * points for weight calculations, us changing something to
4050 * be it can adversely affect the results. But in most
4051 * cases, it should work reasonably. And note that this is
4052 * really an illegal situation: using code points above 255
4053 * on a locale where only 0-255 are valid. If two strings
4054 * sort entirely equal, then the sort order for the
4055 * above-255 code points will be in code point order. */
4059 /* If we haven't calculated the code point with the maximum
4060 * collating order for this locale, do so now */
4061 if (! PL_strxfrm_max_cp) {
4064 /* The current transformed string that collates the
4065 * highest (except it also includes the prefixed collation
4067 char * cur_max_x = NULL;
4069 /* Look through all legal code points (NUL isn't) */
4070 for (j = 1; j < 256; j++) {
4073 char cur_source[] = { '\0', '\0' };
4075 /* Create a 1-char string of the current code point */
4076 cur_source[0] = (char) j;
4078 /* Then transform it */
4079 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
4081 /* If something went wrong (which it shouldn't), just
4082 * ignore this code point */
4087 /* If this character's transformation is higher than
4088 * the current highest, this one becomes the highest */
4089 if ( cur_max_x == NULL
4090 || strGT(x + COLLXFRM_HDR_LEN,
4091 cur_max_x + COLLXFRM_HDR_LEN))
4093 PL_strxfrm_max_cp = j;
4102 DEBUG_L(PerlIO_printf(Perl_debug_log,
4103 "_mem_collxfrm: Couldn't find any character to"
4104 " replace above-Latin1 chars in locale %s with",
4105 PL_collation_name));
4109 DEBUG_L(PerlIO_printf(Perl_debug_log,
4110 "_mem_collxfrm: highest 1-byte collating character"
4111 " in locale %s is 0x%02X\n",
4113 PL_strxfrm_max_cp));
4115 Safefree(cur_max_x);
4118 /* Here we know which legal code point collates the highest.
4119 * We are ready to construct the non-UTF-8 string. The length
4120 * will be at least 1 byte smaller than the input string
4121 * (because we changed at least one 2-byte character into a
4122 * single byte), but that is eaten up by the trailing NUL */
4128 char * e = (char *) t + len;
4130 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
4132 if (UTF8_IS_INVARIANT(cur_char)) {
4135 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
4136 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
4138 else { /* Replace illegal cp with highest collating
4140 s[d++] = PL_strxfrm_max_cp;
4144 Renew(s, d, char); /* Free up unused space */
4149 /* Here, we have constructed a modified version of the input. It could
4150 * be that we already had a modified copy before we did this version.
4151 * If so, that copy is no longer needed */
4152 if (t != input_string) {
4157 length_in_chars = (utf8)
4158 ? utf8_length((U8 *) s, (U8 *) s + len)
4161 /* The first element in the output is the collation id, used by
4162 * sv_collxfrm(); then comes the space for the transformed string. The
4163 * equation should give us a good estimate as to how much is needed */
4164 xAlloc = COLLXFRM_HDR_LEN
4166 + (PL_collxfrm_mult * length_in_chars);
4167 Newx(xbuf, xAlloc, char);
4168 if (UNLIKELY(! xbuf)) {
4169 DEBUG_L(PerlIO_printf(Perl_debug_log,
4170 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
4174 /* Store the collation id */
4175 *(U32*)xbuf = PL_collation_ix;
4177 /* Then the transformation of the input. We loop until successful, or we
4181 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4183 /* If the transformed string occupies less space than we told strxfrm()
4184 * was available, it means it successfully transformed the whole
4186 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
4188 /* Some systems include a trailing NUL in the returned length.
4189 * Ignore it, using a loop in case multiple trailing NULs are
4192 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
4197 /* If the first try didn't get it, it means our prediction was low.
4198 * Modify the coefficients so that we predict a larger value in any
4199 * future transformations */
4201 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
4202 STRLEN computed_guess = PL_collxfrm_base
4203 + (PL_collxfrm_mult * length_in_chars);
4205 /* On zero-length input, just keep current slope instead of
4207 const STRLEN new_m = (length_in_chars != 0)
4208 ? needed / length_in_chars
4211 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4212 "%s: %d: initial size of %zu bytes for a length "
4213 "%zu string was insufficient, %zu needed\n",
4215 computed_guess, length_in_chars, needed));
4217 /* If slope increased, use it, but discard this result for
4218 * length 1 strings, as we can't be sure that it's a real slope
4220 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
4224 STRLEN old_m = PL_collxfrm_mult;
4225 STRLEN old_b = PL_collxfrm_base;
4229 PL_collxfrm_mult = new_m;
4230 PL_collxfrm_base = 1; /* +1 For trailing NUL */
4231 computed_guess = PL_collxfrm_base
4232 + (PL_collxfrm_mult * length_in_chars);
4233 if (computed_guess < needed) {
4234 PL_collxfrm_base += needed - computed_guess;
4237 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4238 "%s: %d: slope is now %zu; was %zu, base "
4239 "is now %zu; was %zu\n",
4241 PL_collxfrm_mult, old_m,
4242 PL_collxfrm_base, old_b));
4244 else { /* Slope didn't change, but 'b' did */
4245 const STRLEN new_b = needed
4248 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
4249 "%s: %d: base is now %zu; was %zu\n",
4251 new_b, PL_collxfrm_base));
4252 PL_collxfrm_base = new_b;
4259 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
4260 DEBUG_L(PerlIO_printf(Perl_debug_log,
4261 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
4262 *xlen, PERL_INT_MAX));
4266 /* A well-behaved strxfrm() returns exactly how much space it needs
4267 * (usually not including the trailing NUL) when it fails due to not
4268 * enough space being provided. Assume that this is the case unless
4269 * it's been proven otherwise */
4270 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
4271 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
4273 else { /* Here, either:
4274 * 1) The strxfrm() has previously shown bad behavior; or
4275 * 2) It isn't the first time through the loop, which means
4276 * that the strxfrm() is now showing bad behavior, because
4277 * we gave it what it said was needed in the previous
4278 * iteration, and it came back saying it needed still more.
4279 * (Many versions of cygwin fit this. When the buffer size
4280 * isn't sufficient, they return the input size instead of
4281 * how much is needed.)
4282 * Increase the buffer size by a fixed percentage and try again.
4284 xAlloc += (xAlloc / 4) + 1;
4285 PL_strxfrm_is_behaved = FALSE;
4289 if (DEBUG_Lv_TEST || debug_initialization) {
4290 PerlIO_printf(Perl_debug_log,