This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Update File::Glob to 1.31 to account for changes in 5.27.10
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
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.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
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
0d071d52
KW
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
a9ad02a8
KW
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
34 * of 'use locale'.
e9bc6d6b
KW
35 *
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.
166f8a29
DM
44 */
45
98994639
HS
46#include "EXTERN.h"
47#define PERL_IN_LOCALE_C
f7416781 48#include "perl_langinfo.h"
98994639
HS
49#include "perl.h"
50
a4af207c
JH
51#include "reentr.h"
52
0dec74cd
KW
53#ifdef I_WCHAR
54# include <wchar.h>
55#endif
56
2fcc0ca9
KW
57/* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
5a4b0634
KW
60#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61# define debug_initialization 0
62# define DEBUG_INITIALIZATION_set(v)
63#else
2fcc0ca9 64static bool debug_initialization = FALSE;
5a4b0634 65# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
66#endif
67
48015184
KW
68
69/* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
71 * done generally. */
72#define GET_ERRNO saved_errno
73
291a84fb
KW
74/* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
77#define STRLENs(s) (sizeof("" s "") - 1)
78
63e5b0d7
KW
79/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
85 * cntrl: 84-97 9B-9F
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89#define isNAME_C_OR_POSIX(name) \
90 ( (name) != NULL \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
93
8ef6e574
KW
94#ifdef USE_LOCALE
95
47280b20
KW
96/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
98
99#define UTF8NESS_SEP "\v"
100#define UTF8NESS_PREFIX "\f"
101
102/* So, the string looks like:
98994639 103 *
47280b20 104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 105 *
47280b20
KW
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
108
109STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
111
112#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
114
115/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
118
98994639
HS
119STATIC char *
120S_stdize_locale(pTHX_ char *locs)
121{
47280b20
KW
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
124 *
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
130 *
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
134 * */
135
7452cf6a 136 const char * const s = strchr(locs, '=');
98994639
HS
137 bool okay = TRUE;
138
7918f24d
NC
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
140
8772537c
AL
141 if (s) {
142 const char * const t = strchr(s, '.');
98994639 143 okay = FALSE;
8772537c
AL
144 if (t) {
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
149 locs[len] = 0;
150 okay = TRUE;
98994639
HS
151 }
152 }
153 }
154
155 if (!okay)
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
157
158 return locs;
159}
160
e5f10d49
KW
161/* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
163 * order. */
164
165const int categories[] = {
166
167# ifdef USE_LOCALE_NUMERIC
168 LC_NUMERIC,
169# endif
170# ifdef USE_LOCALE_CTYPE
171 LC_CTYPE,
172# endif
173# ifdef USE_LOCALE_COLLATE
174 LC_COLLATE,
175# endif
176# ifdef USE_LOCALE_TIME
177 LC_TIME,
178# endif
179# ifdef USE_LOCALE_MESSAGES
180 LC_MESSAGES,
181# endif
182# ifdef USE_LOCALE_MONETARY
183 LC_MONETARY,
184# endif
9821811f
KW
185# ifdef USE_LOCALE_ADDRESS
186 LC_ADDRESS,
187# endif
188# ifdef USE_LOCALE_IDENTIFICATION
189 LC_IDENTIFICATION,
190# endif
191# ifdef USE_LOCALE_MEASUREMENT
192 LC_MEASUREMENT,
193# endif
194# ifdef USE_LOCALE_PAPER
195 LC_PAPER,
196# endif
197# ifdef USE_LOCALE_TELEPHONE
198 LC_TELEPHONE,
199# endif
e5f10d49
KW
200# ifdef LC_ALL
201 LC_ALL,
202# endif
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
206};
207
208/* The top-most real element is LC_ALL */
209
210const char * category_names[] = {
211
212# ifdef USE_LOCALE_NUMERIC
213 "LC_NUMERIC",
214# endif
215# ifdef USE_LOCALE_CTYPE
216 "LC_CTYPE",
217# endif
218# ifdef USE_LOCALE_COLLATE
219 "LC_COLLATE",
220# endif
221# ifdef USE_LOCALE_TIME
222 "LC_TIME",
223# endif
224# ifdef USE_LOCALE_MESSAGES
225 "LC_MESSAGES",
226# endif
227# ifdef USE_LOCALE_MONETARY
228 "LC_MONETARY",
229# endif
9821811f
KW
230# ifdef USE_LOCALE_ADDRESS
231 "LC_ADDRESS",
232# endif
233# ifdef USE_LOCALE_IDENTIFICATION
234 "LC_IDENTIFICATION",
235# endif
236# ifdef USE_LOCALE_MEASUREMENT
237 "LC_MEASUREMENT",
238# endif
239# ifdef USE_LOCALE_PAPER
240 "LC_PAPER",
241# endif
242# ifdef USE_LOCALE_TELEPHONE
243 "LC_TELEPHONE",
244# endif
e5f10d49
KW
245# ifdef LC_ALL
246 "LC_ALL",
247# endif
248 NULL /* Placeholder */
249 };
250
251# ifdef LC_ALL
252
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
256
257# else
258
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
262
263# endif
264
265/* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
948523db
KW
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
e5f10d49 272
b09aaf40
KW
273STATIC const char *
274S_category_name(const int category)
275{
276 unsigned int i;
277
278#ifdef LC_ALL
279
280 if (category == LC_ALL) {
281 return "LC_ALL";
282 }
283
284#endif
285
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
289 }
290 }
291
292 {
293 const char suffix[] = " (unknown)";
294 int temp = category;
295 Size_t length = sizeof(suffix) + 1;
296 char * unknown;
297 dTHX;
298
299 if (temp < 0) {
300 length++;
301 temp = - temp;
302 }
303
304 /* Calculate the number of digits */
305 while (temp >= 10) {
306 temp /= 10;
307 length++;
308 }
309
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
312 SAVEFREEPV(unknown);
313 return unknown;
314 }
315}
316
948523db
KW
317/* Now create LC_foo_INDEX #defines for just those categories on this system */
318# ifdef USE_LOCALE_NUMERIC
319# define LC_NUMERIC_INDEX 0
320# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
321# else
322# define _DUMMY_NUMERIC -1
323# endif
324# ifdef USE_LOCALE_CTYPE
325# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326# define _DUMMY_CTYPE LC_CTYPE_INDEX
327# else
328# define _DUMMY_CTYPE _DUMMY_NUMERIC
329# endif
330# ifdef USE_LOCALE_COLLATE
331# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332# define _DUMMY_COLLATE LC_COLLATE_INDEX
333# else
334# define _DUMMY_COLLATE _DUMMY_COLLATE
335# endif
336# ifdef USE_LOCALE_TIME
337# define LC_TIME_INDEX _DUMMY_COLLATE + 1
338# define _DUMMY_TIME LC_TIME_INDEX
339# else
340# define _DUMMY_TIME _DUMMY_COLLATE
341# endif
342# ifdef USE_LOCALE_MESSAGES
343# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
345# else
346# define _DUMMY_MESSAGES _DUMMY_TIME
347# endif
348# ifdef USE_LOCALE_MONETARY
349# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350# define _DUMMY_MONETARY LC_MONETARY_INDEX
351# else
352# define _DUMMY_MONETARY _DUMMY_MESSAGES
353# endif
9821811f
KW
354# ifdef USE_LOCALE_ADDRESS
355# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
357# else
358# define _DUMMY_ADDRESS _DUMMY_MONETARY
359# endif
360# ifdef USE_LOCALE_IDENTIFICATION
361# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
363# else
364# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
365# endif
366# ifdef USE_LOCALE_MEASUREMENT
367# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
369# else
370# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
371# endif
372# ifdef USE_LOCALE_PAPER
373# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374# define _DUMMY_PAPER LC_PAPER_INDEX
375# else
376# define _DUMMY_PAPER _DUMMY_MEASUREMENT
377# endif
378# ifdef USE_LOCALE_TELEPHONE
379# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
381# else
382# define _DUMMY_TELEPHONE _DUMMY_PAPER
383# endif
948523db 384# ifdef LC_ALL
9821811f 385# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
386# endif
387#endif /* ifdef USE_LOCALE */
8ef6e574 388
d2b24094 389/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
390#ifdef WIN32
391# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
392#else
393# define my_setlocale(cat, locale) setlocale(cat, locale)
394#endif
395
396#ifndef USE_POSIX_2008_LOCALE
397
398/* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
402
403#else /* Below uses POSIX 2008 */
404
405/* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411# define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
414
415/* A third array, parallel to the ones above to map from category to its
416 * equivalent mask */
417const int category_masks[] = {
418# ifdef USE_LOCALE_NUMERIC
419 LC_NUMERIC_MASK,
420# endif
421# ifdef USE_LOCALE_CTYPE
422 LC_CTYPE_MASK,
423# endif
424# ifdef USE_LOCALE_COLLATE
425 LC_COLLATE_MASK,
426# endif
427# ifdef USE_LOCALE_TIME
428 LC_TIME_MASK,
429# endif
430# ifdef USE_LOCALE_MESSAGES
431 LC_MESSAGES_MASK,
432# endif
433# ifdef USE_LOCALE_MONETARY
434 LC_MONETARY_MASK,
435# endif
436# ifdef USE_LOCALE_ADDRESS
437 LC_ADDRESS_MASK,
438# endif
439# ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
441# endif
442# ifdef USE_LOCALE_MEASUREMENT
443 LC_MEASUREMENT_MASK,
444# endif
445# ifdef USE_LOCALE_PAPER
446 LC_PAPER_MASK,
447# endif
448# ifdef USE_LOCALE_TELEPHONE
449 LC_TELEPHONE_MASK,
450# endif
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
455 * time */
456 LC_ALL_MASK
457 };
458
459STATIC const char *
460S_emulate_setlocale(const int category,
461 const char * locale,
462 unsigned int index,
463 const bool is_index_valid
464 )
465{
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
473 *
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
476 *
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
486 *
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
501 */
502
503 int mask;
504 locale_t old_obj;
505 locale_t new_obj;
506 dTHX;
507
508# ifdef DEBUGGING
509
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
512 }
513
514# endif
515
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
518 unsigned int i;
519
520# ifdef DEBUGGING
521
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
524 }
525
526# endif
527
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
530 index = i;
531 goto found_index;
532 }
533 }
534
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
539 category, locale);
540 return NULL;
541
542 found_index: ;
543
544# ifdef DEBUGGING
545
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
548 }
549
550# endif
551
552 }
553
554 mask = category_masks[index];
555
556# ifdef DEBUGGING
557
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
560 }
561
562# endif
563
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
567
568# ifdef DEBUGGING
569
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
572 }
573
574# endif
575
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
578 }
579
580# ifdef HAS_QUERYLOCALE
581
582 return (char *) querylocale(mask, cur_obj);
583
d2b24094 584# else
ecdda939
KW
585
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
588
e9bc6d6b
KW
589# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
590
591 {
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
595
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
598 /*
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
602 */
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
607
608# ifdef USE_C_BACKTRACE
609
610 dump_c_backtrace(Perl_debug_log, 20, 1);
611
612# endif
613
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
618 }
619
620 return temp_name;
621 }
622 }
623
624# endif
625
626 /* Without querylocale(), we have to use our record-keeping we've
627 * done. */
628
629 if (category != LC_ALL) {
630
631# ifdef DEBUGGING
632
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 }
636
637# endif
638
639 return PL_curlocales[index];
640 }
641 else { /* For LC_ALL */
642 unsigned int i;
643 Size_t names_len = 0;
644 char * all_string;
7c93a471 645 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
646
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
649
650# ifdef DEBUGGING
651
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
654 }
655
656# endif
657
658 return PL_curlocales[LC_ALL_INDEX];
659 }
660
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
e9bc6d6b
KW
666 for (i = 0; i < LC_ALL_INDEX; i++) {
667
668# ifdef DEBUGGING
669
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
672 }
673
674# endif
675
676 names_len += strlen(category_names[i])
677 + 1 /* '=' */
678 + strlen(PL_curlocales[i])
679 + 1; /* ';' */
7c93a471
KW
680
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
683 }
684 }
685
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 692 }
7c93a471 693
e9bc6d6b
KW
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
696 *all_string = '\0';
697
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
700
701# ifdef DEBUGGING
702
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
705 }
706
707# endif
708
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
713 }
714
715# ifdef DEBUGGING
716
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
719 }
720
721 #endif
722
723 return all_string;
724 }
725
726# ifdef EINVAL
727
728 SETERRNO(EINVAL, LIB_INVARG);
729
730# endif
731
732 return NULL;
733
d2b24094
KW
734# endif
735
e9bc6d6b
KW
736 }
737
738 assert(PL_C_locale_obj);
739
740 /* Otherwise, we are switching locales. This will generally entail freeing
741 * the current one's space (at the C library's discretion). We need to
742 * stop using that locale before the switch. So switch to a known locale
743 * object that we don't otherwise mess with. This returns the locale
744 * object in effect at the time of the switch. */
745 old_obj = uselocale(PL_C_locale_obj);
746
747# ifdef DEBUGGING
748
749 if (DEBUG_Lv_TEST || debug_initialization) {
750 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
751 }
752
753# endif
754
755 if (! old_obj) {
756
757# ifdef DEBUGGING
758
759 if (DEBUG_L_TEST || debug_initialization) {
760 dSAVE_ERRNO;
761 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
762 RESTORE_ERRNO;
763 }
764
765# endif
766
767 return NULL;
768 }
769
770# ifdef DEBUGGING
771
772 if (DEBUG_Lv_TEST || debug_initialization) {
773 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
774 }
775
776# endif
777
778 /* If we weren't in a thread safe locale, set so that newlocale() below
779 which uses 'old_obj', uses an empty one. Same for our reserved C object.
780 The latter is defensive coding, so that, even if there is some bug, we
781 will never end up trying to modify either of these, as if passed to
782 newlocale(), they can be. */
783 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
784 old_obj = (locale_t) 0;
785 }
786
787 /* Create the new locale (it may actually modify the current one). */
788
789# ifndef HAS_QUERYLOCALE
790
791 if (strEQ(locale, "")) {
792
793 /* For non-querylocale() systems, we do the setting of "" ourselves to
794 * be sure that we really know what's going on. We follow the Linux
795 * documented behavior (but if that differs from the actual behavior,
796 * this won't work exactly as the OS implements). We go out and
797 * examine the environment based on our understanding of how the system
798 * works, and use that to figure things out */
799
800 const char * const lc_all = PerlEnv_getenv("LC_ALL");
801
802 /* Use any "LC_ALL" environment variable, as it overrides everything
803 * else. */
804 if (lc_all && strNE(lc_all, "")) {
805 locale = lc_all;
806 }
807 else {
808
809 /* Otherwise, we need to dig deeper. Unless overridden, the
810 * default is the LANG environment variable; if it doesn't exist,
811 * then "C" */
812
813 const char * default_name;
814
815 /* To minimize other threads messing with the environment, we copy
816 * the variable, making it a temporary. But this doesn't work upon
817 * program initialization before any scopes are created, and at
818 * this time, there's nothing else going on that would interfere.
819 * So skip the copy in that case */
820 if (PL_scopestack_ix == 0) {
821 default_name = PerlEnv_getenv("LANG");
822 }
823 else {
824 default_name = savepv(PerlEnv_getenv("LANG"));
825 }
826
827 if (! default_name || strEQ(default_name, "")) {
929a7f8c 828 default_name = "C";
e9bc6d6b
KW
829 }
830 else if (PL_scopestack_ix != 0) {
831 SAVEFREEPV(default_name);
832 }
833
834 if (category != LC_ALL) {
835 const char * const name = PerlEnv_getenv(category_names[index]);
836
837 /* Here we are setting a single category. Assume will have the
838 * default name */
839 locale = default_name;
840
841 /* But then look for an overriding environment variable */
842 if (name && strNE(name, "")) {
843 locale = name;
844 }
845 }
846 else {
847 bool did_override = FALSE;
848 unsigned int i;
849
850 /* Here, we are getting LC_ALL. Any categories that don't have
851 * a corresponding environment variable set should be set to
852 * LANG, or to "C" if there is no LANG. If no individual
853 * categories differ from this, we can just set LC_ALL. This
854 * is buggy on systems that have extra categories that we don't
855 * know about. If there is an environment variable that sets
856 * that category, we won't know to look for it, and so our use
857 * of LANG or "C" improperly overrides it. On the other hand,
858 * if we don't do what is done here, and there is no
859 * environment variable, the category's locale should be set to
860 * LANG or "C". So there is no good solution. khw thinks the
861 * best is to look at systems to see what categories they have,
862 * and include them, and then to assume that we know the
863 * complete set */
864
865 for (i = 0; i < LC_ALL_INDEX; i++) {
866 const char * const env_override
867 = savepv(PerlEnv_getenv(category_names[i]));
868 const char * this_locale = ( env_override
869 && strNE(env_override, ""))
870 ? env_override
871 : default_name;
6435f98d
KW
872 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
873 {
874 Safefree(env_override);
875 return NULL;
876 }
e9bc6d6b
KW
877
878 if (strNE(this_locale, default_name)) {
879 did_override = TRUE;
880 }
881
882 Safefree(env_override);
883 }
884
885 /* If all the categories are the same, we can set LC_ALL to
886 * that */
887 if (! did_override) {
888 locale = default_name;
889 }
890 else {
891
892 /* Here, LC_ALL is no longer valid, as some individual
893 * categories don't match it. We call ourselves
894 * recursively, as that will execute the code that
895 * generates the proper locale string for this situation.
896 * We don't do the remainder of this function, as that is
897 * to update our records, and we've just done that for the
898 * individual categories in the loop above, and doing so
899 * would cause LC_ALL to be done as well */
900 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
901 }
902 }
903 }
904 }
905 else if (strchr(locale, ';')) {
906
907 /* LC_ALL may actually incude a conglomeration of various categories.
908 * Without querylocale, this code uses the glibc (as of this writing)
909 * syntax for representing that, but that is not a stable API, and
910 * other platforms do it differently, so we have to handle all cases
911 * ourselves */
912
f0023d47 913 unsigned int i;
e9bc6d6b
KW
914 const char * s = locale;
915 const char * e = locale + strlen(locale);
916 const char * p = s;
917 const char * category_end;
918 const char * name_start;
919 const char * name_end;
920
f0023d47
KW
921 /* If the string that gives what to set doesn't include all categories,
922 * the omitted ones get set to "C". To get this behavior, first set
923 * all the individual categories to "C", and override the furnished
924 * ones below */
925 for (i = 0; i < LC_ALL_INDEX; i++) {
926 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
927 return NULL;
928 }
929 }
930
e9bc6d6b 931 while (s < e) {
e9bc6d6b
KW
932
933 /* Parse through the category */
934 while (isWORDCHAR(*p)) {
935 p++;
936 }
937 category_end = p;
938
939 if (*p++ != '=') {
940 Perl_croak(aTHX_
941 "panic: %s: %d: Unexpected character in locale name '%02X",
942 __FILE__, __LINE__, *(p-1));
943 }
944
945 /* Parse through the locale name */
946 name_start = p;
bee74f4b
KW
947 while (p < e && *p != ';') {
948 if (! isGRAPH(*p)) {
949 Perl_croak(aTHX_
950 "panic: %s: %d: Unexpected character in locale name '%02X",
951 __FILE__, __LINE__, *(p-1));
952 }
e9bc6d6b
KW
953 p++;
954 }
955 name_end = p;
956
bee74f4b
KW
957 /* Space past the semi-colon */
958 if (p < e) {
959 p++;
e9bc6d6b
KW
960 }
961
962 /* Find the index of the category name in our lists */
963 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 964 char * individ_locale;
e9bc6d6b
KW
965
966 /* Keep going if this isn't the index. The strnNE() avoids a
967 * Perl_form(), but would fail if ever a category name could be
968 * a substring of another one, like if there were a
969 * "LC_TIME_DATE" */
970 if strnNE(s, category_names[i], category_end - s) {
971 continue;
972 }
973
974 /* If this index is for the single category we're changing, we
975 * have found the locale to set it to. */
976 if (category == categories[i]) {
977 locale = Perl_form(aTHX_ "%.*s",
978 (int) (name_end - name_start),
979 name_start);
980 goto ready_to_set;
981 }
982
3c76a412 983 assert(category == LC_ALL);
bee74f4b
KW
984 individ_locale = Perl_form(aTHX_ "%.*s",
985 (int) (name_end - name_start), name_start);
6435f98d
KW
986 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
987 {
988 return NULL;
989 }
e9bc6d6b
KW
990 }
991
992 s = p;
993 }
994
995 /* Here we have set all the individual categories by recursive calls.
996 * These collectively should have fixed up LC_ALL, so can just query
997 * what that now is */
998 assert(category == LC_ALL);
999
1000 return do_setlocale_c(LC_ALL, NULL);
1001 }
1002
1003 ready_to_set: ;
1004
1005# endif /* end of ! querylocale */
1006
1007 /* Ready to create a new locale by modification of the exising one */
1008 new_obj = newlocale(mask, locale, old_obj);
1009
1010 if (! new_obj) {
1011 dSAVE_ERRNO;
1012
1013# ifdef DEBUGGING
1014
1015 if (DEBUG_L_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1017 }
1018
1019# endif
1020
1021 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1022
1023# ifdef DEBUGGING
1024
1025 if (DEBUG_L_TEST || debug_initialization) {
1026 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1027 }
1028
1029# endif
1030
1031 }
1032 RESTORE_ERRNO;
1033 return NULL;
1034 }
1035
1036# ifdef DEBUGGING
1037
1038 if (DEBUG_Lv_TEST || debug_initialization) {
1039 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1040 }
1041
1042# endif
1043
1044 /* And switch into it */
1045 if (! uselocale(new_obj)) {
1046 dSAVE_ERRNO;
1047
1048# ifdef DEBUGGING
1049
1050 if (DEBUG_L_TEST || debug_initialization) {
1051 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1052 }
1053
1054# endif
1055
1056 if (! uselocale(old_obj)) {
1057
1058# ifdef DEBUGGING
1059
1060 if (DEBUG_L_TEST || debug_initialization) {
1061 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1062 }
1063
1064# endif
1065
1066 }
1067 freelocale(new_obj);
1068 RESTORE_ERRNO;
1069 return NULL;
1070 }
1071
1072# ifdef DEBUGGING
1073
1074 if (DEBUG_Lv_TEST || debug_initialization) {
1075 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1076 }
1077
1078# endif
1079
1080 /* We are done, except for updating our records (if the system doesn't keep
1081 * them) and in the case of locale "", we don't actually know what the
1082 * locale that got switched to is, as it came from the environment. So
1083 * have to find it */
1084
1085# ifdef HAS_QUERYLOCALE
1086
1087 if (strEQ(locale, "")) {
1088 locale = querylocale(mask, new_obj);
1089 }
1090
1091# else
1092
1093 /* Here, 'locale' is the return value */
1094
1095 /* Without querylocale(), we have to update our records */
1096
1097 if (category == LC_ALL) {
1098 unsigned int i;
1099
1100 /* For LC_ALL, we change all individual categories to correspond */
1101 /* PL_curlocales is a parallel array, so has same
1102 * length as 'categories' */
1103 for (i = 0; i <= LC_ALL_INDEX; i++) {
1104 Safefree(PL_curlocales[i]);
1105 PL_curlocales[i] = savepv(locale);
1106 }
1107 }
1108 else {
1109
1110 /* For a single category, if it's not the same as the one in LC_ALL, we
1111 * nullify LC_ALL */
1112
1113 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1114 Safefree(PL_curlocales[LC_ALL_INDEX]);
1115 PL_curlocales[LC_ALL_INDEX] = NULL;
1116 }
1117
1118 /* Then update the category's record */
1119 Safefree(PL_curlocales[index]);
1120 PL_curlocales[index] = savepv(locale);
1121 }
1122
1123# endif
1124
1125 return locale;
1126}
1127
1128#endif /* USE_POSIX_2008_LOCALE */
1129
1130#if 0 /* Code that was to emulate thread-safe locales on platforms that
1131 didn't natively support them */
1132
1133/* The way this would work is that we would keep a per-thread list of the
1134 * correct locale for that thread. Any operation that was locale-sensitive
1135 * would have to be changed so that it would look like this:
1136 *
1137 * LOCALE_LOCK;
1138 * setlocale to the correct locale for this operation
1139 * do operation
1140 * LOCALE_UNLOCK
1141 *
1142 * This leaves the global locale in the most recently used operation's, but it
1143 * was locked long enough to get the result. If that result is static, it
1144 * needs to be copied before the unlock.
1145 *
1146 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1147 * the setup, but are no-ops when not needed, and similarly,
1148 * END_LOCALE_DEPENDENT_OP for the tear-down
1149 *
1150 * But every call to a locale-sensitive function would have to be changed, and
1151 * if a module didn't cooperate by using the mutex, things would break.
1152 *
1153 * This code was abandoned before being completed or tested, and is left as-is
1154*/
1155
1156# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1157# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1158
1159STATIC char *
1160S_locking_setlocale(pTHX_
1161 const int category,
1162 const char * locale,
1163 int index,
1164 const bool is_index_valid
1165 )
1166{
1167 /* This function kind of performs a setlocale() on just the current thread;
1168 * thus it is kind of thread-safe. It does this by keeping a thread-level
1169 * array of the current locales for each category. Every time a locale is
1170 * switched to, it does the switch globally, but updates the thread's
1171 * array. A query as to what the current locale is just returns the
1172 * appropriate element from the array, and doesn't actually call the system
1173 * setlocale(). The saving into the array is done in an uninterruptible
1174 * section of code, so is unaffected by whatever any other threads might be
1175 * doing.
1176 *
1177 * All locale-sensitive operations must work by first starting a critical
1178 * section, then switching to the thread's locale as kept by this function,
1179 * and then doing the operation, then ending the critical section. Thus,
1180 * each gets done in the appropriate locale. simulating thread-safety.
1181 *
1182 * This function takes the same parameters, 'category' and 'locale', that
1183 * the regular setlocale() function does, but it also takes two additional
1184 * ones. This is because as described earlier. If we know on input the
1185 * index corresponding to the category into the array where we store the
1186 * current locales, we don't have to calculate it. If the caller knows at
1187 * compile time what the index is, it it can pass it, setting
1188 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1189 *
1190 */
1191
1192 /* If the input index might be incorrect, calculate the correct one */
1193 if (! is_index_valid) {
1194 unsigned int i;
1195
1196 if (DEBUG_Lv_TEST || debug_initialization) {
1197 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1198 }
1199
1200 for (i = 0; i <= LC_ALL_INDEX; i++) {
1201 if (category == categories[i]) {
1202 index = i;
1203 goto found_index;
1204 }
1205 }
1206
1207 /* Here, we don't know about this category, so can't handle it.
1208 * XXX best we can do is to unsafely set this
1209 * XXX warning */
1210
1211 return my_setlocale(category, locale);
1212
1213 found_index: ;
1214
1215 if (DEBUG_Lv_TEST || debug_initialization) {
1216 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1217 }
1218 }
1219
1220 /* For a query, just return what's in our records */
1221 if (new_locale == NULL) {
1222 return curlocales[index];
1223 }
1224
1225
1226 /* Otherwise, we need to do the switch, and save the result, all in a
1227 * critical section */
1228
1229 Safefree(curlocales[[index]]);
1230
1231 /* It might be that this is called from an already-locked section of code.
1232 * We would have to detect and skip the LOCK/UNLOCK if so */
1233 LOCALE_LOCK;
1234
1235 curlocales[index] = savepv(my_setlocale(category, new_locale));
1236
1237 if (strEQ(new_locale, "")) {
1238
1239#ifdef LC_ALL
1240
1241 /* The locale values come from the environment, and may not all be the
1242 * same, so for LC_ALL, we have to update all the others, while the
1243 * mutex is still locked */
1244
1245 if (category == LC_ALL) {
1246 unsigned int i;
1247 for (i = 0; i < LC_ALL_INDEX) {
1248 curlocales[i] = my_setlocale(categories[i], NULL);
1249 }
1250 }
1251 }
1252
1253#endif
1254
1255 LOCALE_UNLOCK;
1256
1257 return curlocales[index];
1258}
1259
1260#endif
837ce802 1261
a4f00dcc 1262STATIC void
86799d2d 1263S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1264{
86799d2d
KW
1265 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1266 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1267
86799d2d
KW
1268#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1269 || defined(HAS_NL_LANGINFO))
98994639 1270
87f8e8e7 1271 const char * radix = (use_locale)
4e6826bf 1272 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1273 /* FALSE => already in dest locale */
1274 : ".";
2213a3be 1275
87f8e8e7 1276 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1277
87f8e8e7
KW
1278 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1279 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1280 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1281 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1282 && _is_cur_LC_category_utf8(LC_NUMERIC))
1283 {
1284 SvUTF8_on(PL_numeric_radix_sv);
1285 }
86799d2d
KW
1286
1287# ifdef DEBUGGING
7d4bcc4a 1288
2fcc0ca9
KW
1289 if (DEBUG_L_TEST || debug_initialization) {
1290 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1291 SvPVX(PL_numeric_radix_sv),
1292 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1293 }
69014004 1294
86799d2d
KW
1295# endif
1296#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1297
98994639
HS
1298}
1299
398eeea9
KW
1300STATIC void
1301S_new_numeric(pTHX_ const char *newnum)
98994639 1302{
7d4bcc4a
KW
1303
1304#ifndef USE_LOCALE_NUMERIC
1305
1306 PERL_UNUSED_ARG(newnum);
1307
1308#else
0d071d52 1309
291a84fb 1310 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1311 * core Perl this and that 'newnum' is the name of the new locale.
1312 * It installs this locale as the current underlying default.
1313 *
1314 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1315 * set_numeric_underlying() and set_numeric_standard() functions, which
1316 * should probably not be called directly, but only via macros like
0d071d52
KW
1317 * SET_NUMERIC_STANDARD() in perl.h.
1318 *
1319 * The toggling is necessary mainly so that a non-dot radix decimal point
1320 * character can be output, while allowing internal calculations to use a
1321 * dot.
1322 *
1323 * This sets several interpreter-level variables:
bb304765 1324 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1325 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1326 * that the current locale is the program's underlying
1327 * locale
1328 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1329 * that the current locale is the C locale or
1330 * indistinguishable from the C locale. If non-zero, it
1331 * is in C; if > 1, it means it may not be toggled away
7738054c 1332 * from C.
4c68b815
KW
1333 * PL_numeric_underlying_is_standard A bool kept by this function
1334 * indicating that the underlying locale and the standard
1335 * C locale are indistinguishable for the purposes of
1336 * LC_NUMERIC. This happens when both of the above two
1337 * variables are true at the same time. (Toggling is a
1338 * no-op under these circumstances.) This variable is
1339 * used to avoid having to recalculate.
398eeea9 1340 */
0d071d52 1341
b03f34cf 1342 char *save_newnum;
98994639
HS
1343
1344 if (! newnum) {
43c5f42d
NC
1345 Safefree(PL_numeric_name);
1346 PL_numeric_name = NULL;
98994639 1347 PL_numeric_standard = TRUE;
892e6465 1348 PL_numeric_underlying = TRUE;
4c68b815 1349 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1350 return;
1351 }
1352
b03f34cf 1353 save_newnum = stdize_locale(savepv(newnum));
892e6465 1354 PL_numeric_underlying = TRUE;
4c68b815
KW
1355 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1356
69c5e0db
KW
1357#ifndef TS_W32_BROKEN_LOCALECONV
1358
4c68b815 1359 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1360 * them. But on broken Windows systems calling my_nl_langinfo() for
1361 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1362 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1363 if (! PL_numeric_standard) {
4e6826bf 1364 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1365 FALSE /* Don't toggle locale */ ))
4e6826bf 1366 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1367 }
abe1abcf 1368
69c5e0db
KW
1369#endif
1370
4c68b815 1371 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1372 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1373 Safefree(PL_numeric_name);
b03f34cf 1374 PL_numeric_name = save_newnum;
b03f34cf 1375 }
abe1abcf
KW
1376 else {
1377 Safefree(save_newnum);
1378 }
4c28b29c 1379
4c68b815
KW
1380 PL_numeric_underlying_is_standard = PL_numeric_standard;
1381
7e5377f7 1382# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1383
1384 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1385 PL_numeric_name,
1386 PL_underlying_numeric_obj);
1387
1388#endif
1389
4c68b815
KW
1390 if (DEBUG_L_TEST || debug_initialization) {
1391 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1392 }
1393
4c28b29c
KW
1394 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1395 * have to worry about the radix being a non-dot. (Core operations that
1396 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1397 if (PL_numeric_standard) {
1398 set_numeric_radix(0);
1399 }
1400 else {
1401 set_numeric_standard();
1402 }
4c28b29c 1403
98994639 1404#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1405
98994639
HS
1406}
1407
1408void
1409Perl_set_numeric_standard(pTHX)
1410{
7d4bcc4a 1411
98994639 1412#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1413
28c1bf33
KW
1414 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1415 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1416 * macro avoids calling this routine if toggling isn't necessary according
1417 * to our records (which could be wrong if some XS code has changed the
1418 * locale behind our back) */
0d071d52 1419
837ce802 1420 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1421 PL_numeric_standard = TRUE;
4c68b815 1422 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1423 set_numeric_radix(0);
7d4bcc4a
KW
1424
1425# ifdef DEBUGGING
1426
2fcc0ca9
KW
1427 if (DEBUG_L_TEST || debug_initialization) {
1428 PerlIO_printf(Perl_debug_log,
58e4a467 1429 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1430 }
98994639 1431
7d4bcc4a 1432# endif
98994639 1433#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1434
98994639
HS
1435}
1436
1437void
5792c642 1438Perl_set_numeric_underlying(pTHX)
98994639 1439{
7d4bcc4a 1440
98994639 1441#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1442
28c1bf33 1443 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1444 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1445 * instead of calling this directly. The macro avoids calling this routine
1446 * if toggling isn't necessary according to our records (which could be
1447 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1448
837ce802 1449 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1450 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1451 PL_numeric_underlying = TRUE;
3c62fd46 1452 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1453
1454# ifdef DEBUGGING
1455
2fcc0ca9
KW
1456 if (DEBUG_L_TEST || debug_initialization) {
1457 PerlIO_printf(Perl_debug_log,
58e4a467 1458 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1459 PL_numeric_name);
1460 }
98994639 1461
7d4bcc4a 1462# endif
98994639 1463#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1464
98994639
HS
1465}
1466
1467/*
1468 * Set up for a new ctype locale.
1469 */
a4f00dcc
KW
1470STATIC void
1471S_new_ctype(pTHX_ const char *newctype)
98994639 1472{
7d4bcc4a
KW
1473
1474#ifndef USE_LOCALE_CTYPE
1475
1476 PERL_ARGS_ASSERT_NEW_CTYPE;
1477 PERL_UNUSED_ARG(newctype);
1478 PERL_UNUSED_CONTEXT;
1479
1480#else
0d071d52 1481
291a84fb 1482 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1483 * core Perl this and that 'newctype' is the name of the new locale.
1484 *
1485 * This function sets up the folding arrays for all 256 bytes, assuming
1486 * that tofold() is tolc() since fold case is not a concept in POSIX,
1487 *
1488 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1489 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1490 * this function should be called directly only from this file and from
0d071d52
KW
1491 * POSIX::setlocale() */
1492
27da23d5 1493 dVAR;
013f4e03 1494 unsigned int i;
98994639 1495
8b7358b9
KW
1496 /* Don't check for problems if we are suppressing the warnings */
1497 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1498
7918f24d
NC
1499 PERL_ARGS_ASSERT_NEW_CTYPE;
1500
215c5139
KW
1501 /* We will replace any bad locale warning with 1) nothing if the new one is
1502 * ok; or 2) a new warning for the bad new locale */
1503 if (PL_warn_locale) {
1504 SvREFCNT_dec_NN(PL_warn_locale);
1505 PL_warn_locale = NULL;
1506 }
1507
c1284011 1508 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1509
1510 /* A UTF-8 locale gets standard rules. But note that code still has to
1511 * handle this specially because of the three problematic code points */
1512 if (PL_in_utf8_CTYPE_locale) {
1513 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1514 }
8b7358b9
KW
1515
1516 /* We don't populate the other lists if a UTF-8 locale, but do check that
1517 * everything works as expected, unless checking turned off */
1518 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1519 /* Assume enough space for every character being bad. 4 spaces each
1520 * for the 94 printable characters that are output like "'x' "; and 5
1521 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1522 * NUL */
8b7358b9 1523 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1524 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1525 to start */
1526 unsigned int bad_count = 0; /* Count of bad characters */
1527
baa60164 1528 for (i = 0; i < 256; i++) {
8b7358b9 1529 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1530 if (isupper(i))
1531 PL_fold_locale[i] = (U8) tolower(i);
1532 else if (islower(i))
1533 PL_fold_locale[i] = (U8) toupper(i);
1534 else
1535 PL_fold_locale[i] = (U8) i;
8b7358b9 1536 }
8c6180a9
KW
1537
1538 /* If checking for locale problems, see if the native ASCII-range
1539 * printables plus \n and \t are in their expected categories in
1540 * the new locale. If not, this could mean big trouble, upending
1541 * Perl's and most programs' assumptions, like having a
1542 * metacharacter with special meaning become a \w. Fortunately,
1543 * it's very rare to find locales that aren't supersets of ASCII
1544 * nowadays. It isn't a problem for most controls to be changed
1545 * into something else; we check only \n and \t, though perhaps \r
1546 * could be an issue as well. */
7d4bcc4a 1547 if ( check_for_problems
8c6180a9
KW
1548 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1549 {
8b7358b9 1550 bool is_bad = FALSE;
6726b4f4 1551 char name[4] = { '\0' };
8b7358b9
KW
1552
1553 /* Convert the name into a string */
6726b4f4 1554 if (isGRAPH_A(i)) {
8b7358b9
KW
1555 name[0] = i;
1556 name[1] = '\0';
1557 }
1558 else if (i == '\n') {
6726b4f4
KW
1559 my_strlcpy(name, "\\n", sizeof(name));
1560 }
1561 else if (i == '\t') {
1562 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1563 }
1564 else {
6726b4f4
KW
1565 assert(i == ' ');
1566 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1567 }
1568
1569 /* Check each possibe class */
1570 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1571 is_bad = TRUE;
1572 DEBUG_L(PerlIO_printf(Perl_debug_log,
1573 "isalnum('%s') unexpectedly is %d\n",
1574 name, cBOOL(isalnum(i))));
1575 }
1576 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1577 is_bad = TRUE;
1578 DEBUG_L(PerlIO_printf(Perl_debug_log,
1579 "isalpha('%s') unexpectedly is %d\n",
1580 name, cBOOL(isalpha(i))));
1581 }
1582 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1583 is_bad = TRUE;
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isdigit('%s') unexpectedly is %d\n",
1586 name, cBOOL(isdigit(i))));
1587 }
1588 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1589 is_bad = TRUE;
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isgraph('%s') unexpectedly is %d\n",
1592 name, cBOOL(isgraph(i))));
1593 }
1594 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1595 is_bad = TRUE;
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "islower('%s') unexpectedly is %d\n",
1598 name, cBOOL(islower(i))));
1599 }
1600 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1601 is_bad = TRUE;
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "isprint('%s') unexpectedly is %d\n",
1604 name, cBOOL(isprint(i))));
1605 }
1606 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1607 is_bad = TRUE;
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "ispunct('%s') unexpectedly is %d\n",
1610 name, cBOOL(ispunct(i))));
1611 }
1612 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1613 is_bad = TRUE;
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "isspace('%s') unexpectedly is %d\n",
1616 name, cBOOL(isspace(i))));
1617 }
1618 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1619 is_bad = TRUE;
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "isupper('%s') unexpectedly is %d\n",
1622 name, cBOOL(isupper(i))));
1623 }
1624 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1625 is_bad = TRUE;
1626 DEBUG_L(PerlIO_printf(Perl_debug_log,
1627 "isxdigit('%s') unexpectedly is %d\n",
1628 name, cBOOL(isxdigit(i))));
1629 }
1630 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1631 is_bad = TRUE;
1632 DEBUG_L(PerlIO_printf(Perl_debug_log,
1633 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1634 name, tolower(i), (int) toLOWER_A(i)));
1635 }
1636 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1637 is_bad = TRUE;
1638 DEBUG_L(PerlIO_printf(Perl_debug_log,
1639 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1640 name, toupper(i), (int) toUPPER_A(i)));
1641 }
1642 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1643 is_bad = TRUE;
1644 DEBUG_L(PerlIO_printf(Perl_debug_log,
1645 "'\\n' (=%02X) is not a control\n", (int) i));
1646 }
1647
1648 /* Add to the list; Separate multiple entries with a blank */
1649 if (is_bad) {
1650 if (bad_count) {
1651 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1652 }
8b7358b9
KW
1653 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1654 bad_count++;
8c6180a9
KW
1655 }
1656 }
1657 }
1658
7d4bcc4a
KW
1659# ifdef MB_CUR_MAX
1660
8c6180a9 1661 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1662 * this locale requires more than one byte, there are going to be
8c6180a9 1663 * problems. */
9c8a6dc2
KW
1664 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1665 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1666 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1667
8b7358b9
KW
1668 if ( check_for_problems && MB_CUR_MAX > 1
1669 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1670
1671 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1672 * locale. Just assume that the implementation for them (plus
1673 * for POSIX) is correct and the > 1 value is spurious. (Since
1674 * these are specially handled to never be considered UTF-8
1675 * locales, as long as this is the only problem, everything
1676 * should work fine */
1677 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1678 {
8c6180a9
KW
1679 multi_byte_locale = TRUE;
1680 }
7d4bcc4a
KW
1681
1682# endif
8c6180a9 1683
8b7358b9
KW
1684 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1685 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1686 PL_warn_locale = Perl_newSVpvf(aTHX_
1687 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1688 " which have\nunexpected meanings: %s\nThe Perl program"
1689 " will use the expected meanings",
8b7358b9 1690 newctype, bad_chars_list);
8b7358b9
KW
1691 }
1692 else {
1693 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1694 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1695 newctype,
8c6180a9
KW
1696 (multi_byte_locale)
1697 ? " Some characters in it are not recognized by"
1698 " Perl."
1699 : "",
1700 (bad_count)
1701 ? "\nThe following characters (and maybe others)"
1702 " may not have the same meaning as the Perl"
1703 " program expects:\n"
1704 : "",
1705 (bad_count)
1706 ? bad_chars_list
1707 : ""
1708 );
8b7358b9
KW
1709 }
1710
b119c2be
KW
1711# ifdef HAS_NL_LANGINFO
1712
1713 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1714 /* parameter FALSE is a don't care here */
4e6826bf 1715 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1716
1717# endif
1718
8b7358b9
KW
1719 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1720
cc9eaeb0 1721 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1722 * output the message now. If not in that scope, we save the
1723 * message to be output at the first operation using this locale,
1724 * if that actually happens. Most programs don't use locales, so
1725 * they are immune to bad ones. */
cc9eaeb0 1726 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1727
780fcc9f
KW
1728 /* The '0' below suppresses a bogus gcc compiler warning */
1729 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1730
bddebb56
KW
1731 if (IN_LC(LC_CTYPE)) {
1732 SvREFCNT_dec_NN(PL_warn_locale);
1733 PL_warn_locale = NULL;
1734 }
780fcc9f 1735 }
baa60164 1736 }
31f05a37 1737 }
98994639
HS
1738
1739#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1740
98994639
HS
1741}
1742
98994639 1743void
2726666d
KW
1744Perl__warn_problematic_locale()
1745{
2726666d
KW
1746
1747#ifdef USE_LOCALE_CTYPE
1748
5f04a188
KW
1749 dTHX;
1750
1751 /* Internal-to-core function that outputs the message in PL_warn_locale,
1752 * and then NULLS it. Should be called only through the macro
1753 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1754
2726666d 1755 if (PL_warn_locale) {
2726666d
KW
1756 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1757 SvPVX(PL_warn_locale),
1758 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1759 SvREFCNT_dec_NN(PL_warn_locale);
1760 PL_warn_locale = NULL;
1761 }
1762
1763#endif
1764
1765}
1766
a4f00dcc
KW
1767STATIC void
1768S_new_collate(pTHX_ const char *newcoll)
98994639 1769{
7d4bcc4a
KW
1770
1771#ifndef USE_LOCALE_COLLATE
1772
1773 PERL_UNUSED_ARG(newcoll);
1774 PERL_UNUSED_CONTEXT;
1775
1776#else
0d071d52 1777
291a84fb 1778 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1779 * core Perl this and that 'newcoll' is the name of the new locale.
1780 *
d35fca5f
KW
1781 * The design of locale collation is that every locale change is given an
1782 * index 'PL_collation_ix'. The first time a string particpates in an
1783 * operation that requires collation while locale collation is active, it
1784 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1785 * magic includes the collation index, and the transformation of the string
1786 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1787 * instead of the string itself. If a string changes, the magic is
1788 * cleared. The next time the locale changes, the index is incremented,
1789 * and so we know during a comparison that the transformation is not
1790 * necessarily still valid, and so is recomputed. Note that if the locale
1791 * changes enough times, the index could wrap (a U32), and it is possible
1792 * that a transformation would improperly be considered valid, leading to
1793 * an unlikely bug */
0d071d52 1794
98994639
HS
1795 if (! newcoll) {
1796 if (PL_collation_name) {
1797 ++PL_collation_ix;
1798 Safefree(PL_collation_name);
1799 PL_collation_name = NULL;
1800 }
1801 PL_collation_standard = TRUE;
00bf60ca 1802 is_standard_collation:
98994639
HS
1803 PL_collxfrm_base = 0;
1804 PL_collxfrm_mult = 2;
165a1c52 1805 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1806 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1807 PL_strxfrm_max_cp = 0;
98994639
HS
1808 return;
1809 }
1810
d35fca5f 1811 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1812 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1813 ++PL_collation_ix;
1814 Safefree(PL_collation_name);
1815 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1816 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1817 if (PL_collation_standard) {
1818 goto is_standard_collation;
1819 }
98994639 1820
165a1c52 1821 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1822 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1823 PL_strxfrm_max_cp = 0;
165a1c52 1824
59c018b9
KW
1825 /* A locale collation definition includes primary, secondary, tertiary,
1826 * etc. weights for each character. To sort, the primary weights are
1827 * used, and only if they compare equal, then the secondary weights are
1828 * used, and only if they compare equal, then the tertiary, etc.
1829 *
1830 * strxfrm() works by taking the input string, say ABC, and creating an
1831 * output transformed string consisting of first the primary weights,
1832 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1833 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1834 * may not have weights at every level. In our example, let's say B
1835 * doesn't have a tertiary weight, and A doesn't have a secondary
1836 * weight. The constructed string is then going to be
1837 * A¹B¹C¹ B²C² A³C³ ....
1838 * This has the desired effect that strcmp() will look at the secondary
1839 * or tertiary weights only if the strings compare equal at all higher
1840 * priority weights. The spaces shown here, like in
c342d20e 1841 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1842 * are not just for readability. In the general case, these must
1843 * actually be bytes, which we will call here 'separator weights'; and
1844 * they must be smaller than any other weight value, but since these
1845 * are C strings, only the terminating one can be a NUL (some
1846 * implementations may include a non-NUL separator weight just before
1847 * the NUL). Implementations tend to reserve 01 for the separator
1848 * weights. They are needed so that a shorter string's secondary
1849 * weights won't be misconstrued as primary weights of a longer string,
1850 * etc. By making them smaller than any other weight, the shorter
1851 * string will sort first. (Actually, if all secondary weights are
1852 * smaller than all primary ones, there is no need for a separator
1853 * weight between those two levels, etc.)
1854 *
1855 * The length of the transformed string is roughly a linear function of
1856 * the input string. It's not exactly linear because some characters
1857 * don't have weights at all levels. When we call strxfrm() we have to
1858 * allocate some memory to hold the transformed string. The
1859 * calculations below try to find coefficients 'm' and 'b' for this
1860 * locale so that m*x + b equals how much space we need, given the size
1861 * of the input string in 'x'. If we calculate too small, we increase
1862 * the size as needed, and call strxfrm() again, but it is better to
1863 * get it right the first time to avoid wasted expensive string
1864 * transformations. */
1865
98994639 1866 {
79f120c8
KW
1867 /* We use the string below to find how long the tranformation of it
1868 * is. Almost all locales are supersets of ASCII, or at least the
1869 * ASCII letters. We use all of them, half upper half lower,
1870 * because if we used fewer, we might hit just the ones that are
1871 * outliers in a particular locale. Most of the strings being
1872 * collated will contain a preponderance of letters, and even if
1873 * they are above-ASCII, they are likely to have the same number of
1874 * weight levels as the ASCII ones. It turns out that digits tend
1875 * to have fewer levels, and some punctuation has more, but those
1876 * are relatively sparse in text, and khw believes this gives a
1877 * reasonable result, but it could be changed if experience so
1878 * dictates. */
1879 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1880 char * x_longer; /* Transformed 'longer' */
1881 Size_t x_len_longer; /* Length of 'x_longer' */
1882
1883 char * x_shorter; /* We also transform a substring of 'longer' */
1884 Size_t x_len_shorter;
1885
a4a439fb 1886 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1887 * are interested only in its length). It is used because it has
1888 * the intelligence to handle all cases, but to work, it needs some
1889 * values of 'm' and 'b' to get it started. For the purposes of
1890 * this calculation we use a very conservative estimate of 'm' and
1891 * 'b'. This assumes a weight can be multiple bytes, enough to
1892 * hold any UV on the platform, and there are 5 levels, 4 weight
1893 * bytes, and a trailing NUL. */
1894 PL_collxfrm_base = 5;
1895 PL_collxfrm_mult = 5 * sizeof(UV);
1896
1897 /* Find out how long the transformation really is */
a4a439fb
KW
1898 x_longer = _mem_collxfrm(longer,
1899 sizeof(longer) - 1,
1900 &x_len_longer,
1901
1902 /* We avoid converting to UTF-8 in the
1903 * called function by telling it the
1904 * string is in UTF-8 if the locale is a
1905 * UTF-8 one. Since the string passed
1906 * here is invariant under UTF-8, we can
1907 * claim it's UTF-8 even though it isn't.
1908 * */
1909 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1910 Safefree(x_longer);
1911
1912 /* Find out how long the transformation of a substring of 'longer'
1913 * is. Together the lengths of these transformations are
1914 * sufficient to calculate 'm' and 'b'. The substring is all of
1915 * 'longer' except the first character. This minimizes the chances
1916 * of being swayed by outliers */
a4a439fb 1917 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1918 sizeof(longer) - 2,
a4a439fb
KW
1919 &x_len_shorter,
1920 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1921 Safefree(x_shorter);
1922
1923 /* If the results are nonsensical for this simple test, the whole
1924 * locale definition is suspect. Mark it so that locale collation
1925 * is not active at all for it. XXX Should we warn? */
1926 if ( x_len_shorter == 0
1927 || x_len_longer == 0
1928 || x_len_shorter >= x_len_longer)
1929 {
1930 PL_collxfrm_mult = 0;
1931 PL_collxfrm_base = 0;
1932 }
1933 else {
1934 SSize_t base; /* Temporary */
1935
1936 /* We have both: m * strlen(longer) + b = x_len_longer
1937 * m * strlen(shorter) + b = x_len_shorter;
1938 * subtracting yields:
1939 * m * (strlen(longer) - strlen(shorter))
1940 * = x_len_longer - x_len_shorter
1941 * But we have set things up so that 'shorter' is 1 byte smaller
1942 * than 'longer'. Hence:
1943 * m = x_len_longer - x_len_shorter
1944 *
1945 * But if something went wrong, make sure the multiplier is at
1946 * least 1.
1947 */
1948 if (x_len_longer > x_len_shorter) {
1949 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1950 }
1951 else {
1952 PL_collxfrm_mult = 1;
1953 }
1954
1955 /* mx + b = len
1956 * so: b = len - mx
1957 * but in case something has gone wrong, make sure it is
1958 * non-negative */
1959 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1960 if (base < 0) {
1961 base = 0;
1962 }
1963
1964 /* Add 1 for the trailing NUL */
1965 PL_collxfrm_base = base + 1;
1966 }
58eebef2 1967
7d4bcc4a
KW
1968# ifdef DEBUGGING
1969
58eebef2
KW
1970 if (DEBUG_L_TEST || debug_initialization) {
1971 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1972 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1973 "x_len_longer=%zu,"
1974 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1975 __FILE__, __LINE__,
1976 PL_in_utf8_COLLATE_locale,
1977 x_len_shorter, x_len_longer,
1978 PL_collxfrm_mult, PL_collxfrm_base);
1979 }
7d4bcc4a
KW
1980# endif
1981
98994639
HS
1982 }
1983 }
1984
1985#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1986
98994639
HS
1987}
1988
d2b24094 1989#ifdef WIN32
b8cc575c 1990
a4f00dcc 1991STATIC char *
b8cc575c 1992S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1993{
1994 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1995 * difference between the two unless the input locale is "", which normally
1996 * means on Windows to get the machine default, which is set via the
1997 * computer's "Regional and Language Options" (or its current equivalent).
1998 * In POSIX, it instead means to find the locale from the user's
1999 * environment. This routine changes the Windows behavior to first look in
2000 * the environment, and, if anything is found, use that instead of going to
2001 * the machine default. If there is no environment override, the machine
2002 * default is used, by calling the real setlocale() with "".
2003 *
2004 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2005 * use the particular category's variable if set; otherwise to use the LANG
2006 * variable. */
b385bb4d 2007
175c4cf9 2008 bool override_LC_ALL = FALSE;
89f7b9aa 2009 char * result;
e5f10d49 2010 unsigned int i;
89f7b9aa 2011
b385bb4d 2012 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2013
2014# ifdef LC_ALL
2015
b385bb4d
KW
2016 locale = PerlEnv_getenv("LC_ALL");
2017 if (! locale) {
e5f10d49
KW
2018 if (category == LC_ALL) {
2019 override_LC_ALL = TRUE;
2020 }
2021 else {
7d4bcc4a
KW
2022
2023# endif
7d4bcc4a 2024
e5f10d49
KW
2025 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2026 if (category == categories[i]) {
2027 locale = PerlEnv_getenv(category_names[i]);
2028 goto found_locale;
2029 }
2030 }
7d4bcc4a 2031
b385bb4d 2032 locale = PerlEnv_getenv("LANG");
481465ea 2033 if (! locale) {
b385bb4d
KW
2034 locale = "";
2035 }
e5f10d49
KW
2036
2037 found_locale: ;
7d4bcc4a
KW
2038
2039# ifdef LC_ALL
2040
e5f10d49 2041 }
b385bb4d 2042 }
7d4bcc4a
KW
2043
2044# endif
2045
b385bb4d
KW
2046 }
2047
89f7b9aa 2048 result = setlocale(category, locale);
48015184
KW
2049 DEBUG_L(STMT_START {
2050 dSAVE_ERRNO;
2051 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2052 setlocale_debug_string(category, locale, result));
2053 RESTORE_ERRNO;
2054 } STMT_END);
89f7b9aa 2055
481465ea 2056 if (! override_LC_ALL) {
89f7b9aa
KW
2057 return result;
2058 }
2059
dfd77d7a 2060 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2061 * LANG variable or the system default if there is no LANG. But these have
2062 * lower priority than the other LC_foo variables, so override it for each
2063 * one that is set. (If they are set to "", it means to use the same thing
2064 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2065
948523db 2066 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2067 result = PerlEnv_getenv(category_names[i]);
2068 if (result && strNE(result, "")) {
2069 setlocale(categories[i], result);
2070 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2071 __FILE__, __LINE__,
2072 setlocale_debug_string(categories[i], result, "not captured")));
2073 }
89f7b9aa 2074 }
7d4bcc4a 2075
bbc98134 2076 result = setlocale(LC_ALL, NULL);
48015184
KW
2077 DEBUG_L(STMT_START {
2078 dSAVE_ERRNO;
2079 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2080 __FILE__, __LINE__,
48015184
KW
2081 setlocale_debug_string(LC_ALL, NULL, result));
2082 RESTORE_ERRNO;
2083 } STMT_END);
89f7b9aa 2084
bbc98134 2085 return result;
b385bb4d
KW
2086}
2087
2088#endif
2089
9aac5db8
KW
2090/*
2091
2092=head1 Locale-related functions and macros
2093
2094=for apidoc Perl_setlocale
2095
2096This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2097taking the same parameters, and returning the same information, except that it
9487427b
KW
2098returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2099instead return C<C> if the underlying locale has a non-dot decimal point
2100character, or a non-empty thousands separator for displaying floating point
2101numbers. This is because perl keeps that locale category such that it has a
2102dot and empty separator, changing the locale briefly during the operations
2103where the underlying one is required. C<Perl_setlocale> knows about this, and
2104compensates; regular C<setlocale> doesn't.
9aac5db8 2105
e9bc6d6b 2106Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2107declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2108C<const> (presumably because its API was specified long ago, and can't be
2109updated; it is illegal to change the information C<setlocale> returns; doing
2110so leads to segfaults.)
9aac5db8 2111
e9bc6d6b
KW
2112Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2113C<setlocale> can be completely ineffective on some platforms under some
2114configurations.
2115
9aac5db8 2116C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2117where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2118the system C<setlocale()> is ineffective, returning the wrong information, and
2119failing to actually change the locale. C<Perl_setlocale>, however works
2120properly in all circumstances.
9aac5db8
KW
2121
2122The return points to a per-thread static buffer, which is overwritten the next
2123time C<Perl_setlocale> is called from the same thread.
2124
2125=cut
2126
2127*/
2128
2129const char *
2130Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2131{
2132 /* This wraps POSIX::setlocale() */
2133
9aac5db8
KW
2134 const char * retval;
2135 const char * newlocale;
2136 dSAVEDERRNO;
2137 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2138 dTHX;
2139
a4f00dcc
KW
2140#ifdef USE_LOCALE_NUMERIC
2141
291a84fb
KW
2142 /* A NULL locale means only query what the current one is. We have the
2143 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2144 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2145 * the correct results. All other categories don't require special
2146 * handling */
a4f00dcc
KW
2147 if (locale == NULL) {
2148 if (category == LC_NUMERIC) {
9aac5db8
KW
2149
2150 /* We don't have to copy this return value, as it is a per-thread
2151 * variable, and won't change until a future setlocale */
2152 return PL_numeric_name;
a4f00dcc
KW
2153 }
2154
7d4bcc4a 2155# ifdef LC_ALL
a4f00dcc 2156
9aac5db8
KW
2157 else if (category == LC_ALL) {
2158 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2159 }
2160
7d4bcc4a 2161# endif
a4f00dcc
KW
2162
2163 }
2164
2165#endif
2166
33e5a354
KW
2167 retval = save_to_buffer(do_setlocale_r(category, locale),
2168 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2169 SAVE_ERRNO;
2170
2171#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2172
2173 if (locale == NULL && category == LC_ALL) {
2174 RESTORE_LC_NUMERIC();
2175 }
2176
2177#endif
a4f00dcc
KW
2178
2179 DEBUG_L(PerlIO_printf(Perl_debug_log,
2180 "%s:%d: %s\n", __FILE__, __LINE__,
2181 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2182
9aac5db8
KW
2183 RESTORE_ERRNO;
2184
2185 if (! retval) {
a4f00dcc
KW
2186 return NULL;
2187 }
2188
9aac5db8 2189 /* If locale == NULL, we are just querying the state */
a4f00dcc 2190 if (locale == NULL) {
a4f00dcc
KW
2191 return retval;
2192 }
a4f00dcc 2193
1159483a
KW
2194 /* Now that have switched locales, we have to update our records to
2195 * correspond. */
a4f00dcc 2196
1159483a 2197 switch (category) {
a4f00dcc 2198
1159483a 2199#ifdef USE_LOCALE_CTYPE
a4f00dcc 2200
1159483a
KW
2201 case LC_CTYPE:
2202 new_ctype(retval);
2203 break;
a4f00dcc 2204
1159483a 2205#endif
a4f00dcc
KW
2206#ifdef USE_LOCALE_COLLATE
2207
1159483a
KW
2208 case LC_COLLATE:
2209 new_collate(retval);
2210 break;
a4f00dcc 2211
1159483a 2212#endif
a4f00dcc
KW
2213#ifdef USE_LOCALE_NUMERIC
2214
1159483a
KW
2215 case LC_NUMERIC:
2216 new_numeric(retval);
2217 break;
a4f00dcc 2218
1159483a
KW
2219#endif
2220#ifdef LC_ALL
a4f00dcc 2221
1159483a 2222 case LC_ALL:
a4f00dcc 2223
1159483a
KW
2224 /* LC_ALL updates all the things we care about. The values may not
2225 * be the same as 'retval', as the locale "" may have set things
2226 * individually */
a4f00dcc 2227
1159483a 2228# ifdef USE_LOCALE_CTYPE
a4f00dcc 2229
b0bde642 2230 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2231 new_ctype(newlocale);
b0bde642 2232 Safefree(newlocale);
a4f00dcc 2233
1159483a
KW
2234# endif /* USE_LOCALE_CTYPE */
2235# ifdef USE_LOCALE_COLLATE
2236
b0bde642 2237 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2238 new_collate(newlocale);
b0bde642 2239 Safefree(newlocale);
a4f00dcc 2240
7d4bcc4a 2241# endif
1159483a 2242# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2243
b0bde642 2244 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2245 new_numeric(newlocale);
b0bde642 2246 Safefree(newlocale);
a4f00dcc 2247
1159483a
KW
2248# endif /* USE_LOCALE_NUMERIC */
2249#endif /* LC_ALL */
a4f00dcc 2250
1159483a
KW
2251 default:
2252 break;
a4f00dcc
KW
2253 }
2254
2255 return retval;
2256
f7416781
KW
2257}
2258
2259PERL_STATIC_INLINE const char *
2260S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2261{
2262 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2263 * growing it if necessary */
2264
0b6697c5 2265 Size_t string_size;
f7416781
KW
2266
2267 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2268
0b6697c5
KW
2269 if (! string) {
2270 return NULL;
2271 }
2272
2273 string_size = strlen(string) + offset + 1;
2274
f7416781
KW
2275 if (*buf_size == 0) {
2276 Newx(*buf, string_size, char);
2277 *buf_size = string_size;
2278 }
2279 else if (string_size > *buf_size) {
2280 Renew(*buf, string_size, char);
2281 *buf_size = string_size;
2282 }
2283
2284 Copy(string, *buf + offset, string_size - offset, char);
2285 return *buf;
2286}
2287
2288/*
2289
f7416781
KW
2290=for apidoc Perl_langinfo
2291
ce181bf6 2292This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2293taking the same C<item> parameter values, and returning the same information.
2294But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2295of Perl's locale handling from your code, and can be used on systems that lack
2296a native C<nl_langinfo>.
2297
2298Expanding on these:
2299
2300=over
2301
2302=item *
2303
ce181bf6
KW
2304The reason it isn't quite a drop-in replacement is actually an advantage. The
2305only difference is that it returns S<C<const char *>>, whereas plain
2306C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2307forbidden to write into the buffer. By declaring this C<const>, the compiler
2308enforces this restriction, so if it is violated, you know at compilation time,
2309rather than getting segfaults at runtime.
2310
2311=item *
2312
69e2275e 2313It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2314without you having to write extra code. The reason for the extra code would be
2315because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2316kept set by Perl so that the radix is a dot, and the separator is the empty
2317string, no matter what the underlying locale is supposed to be, and so to get
2318the expected results, you have to temporarily toggle into the underlying
2319locale, and later toggle back. (You could use plain C<nl_langinfo> and
2320C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2321the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2322(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2323(decimal point) character to be a dot.)
ce181bf6
KW
2324
2325=item *
2326
2327The system function it replaces can have its static return buffer trashed,
2328not only by a subesequent call to that function, but by a C<freelocale>,
2329C<setlocale>, or other locale change. The returned buffer of this function is
2330not changed until the next call to it, so the buffer is never in a trashed
2331state.
f7416781
KW
2332
2333=item *
2334
ce181bf6
KW
2335Its return buffer is per-thread, so it also is never overwritten by a call to
2336this function from another thread; unlike the function it replaces.
2337
2338=item *
2339
2340But most importantly, it works on systems that don't have C<nl_langinfo>, such
2341as Windows, hence makes your code more portable. Of the fifty-some possible
2342items specified by the POSIX 2008 standard,
f7416781 2343L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2344only one is completely unimplemented, though on non-Windows platforms, another
2345significant one is also not implemented). It uses various techniques to
2346recover the other items, including calling C<L<localeconv(3)>>, and
2347C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2348available. Later C<strftime()> versions have additional capabilities; C<""> is
2349returned for those not available on your system.
ce181bf6
KW
2350
2351It is important to note that when called with an item that is recovered by
2352using C<localeconv>, the buffer from any previous explicit call to
2353C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2354contents if you need to access them after a call to this function. (But note
2355that you might not want to be using C<localeconv()> directly anyway, because of
2356issues like the ones listed in the second item of this list (above) for
2357C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2358call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2359unpack).
9f0bc911 2360
6201c220
KW
2361The details for those items which may deviate from what this emulation returns
2362and what a native C<nl_langinfo()> would return are specified in
2363L<I18N::Langinfo>.
f7416781 2364
ce181bf6
KW
2365=back
2366
f7416781
KW
2367When using C<Perl_langinfo> on systems that don't have a native
2368C<nl_langinfo()>, you must
2369
2370 #include "perl_langinfo.h"
2371
2372before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2373C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2374C<langinfo.h> would try to import into the namespace for code that doesn't need
2375it.)
f7416781 2376
f7416781
KW
2377The original impetus for C<Perl_langinfo()> was so that code that needs to
2378find out the current currency symbol, floating point radix character, or digit
2379grouping separator can use, on all systems, the simpler and more
2380thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2381pain to make thread-friendly. For other fields returned by C<localeconv>, it
2382is better to use the methods given in L<perlcall> to call
2383L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2384
2385=cut
2386
2387*/
2388
2389const char *
2390#ifdef HAS_NL_LANGINFO
2391Perl_langinfo(const nl_item item)
2392#else
2393Perl_langinfo(const int item)
2394#endif
2395{
f61748ac
KW
2396 return my_nl_langinfo(item, TRUE);
2397}
2398
59a15af0 2399STATIC const char *
f61748ac
KW
2400#ifdef HAS_NL_LANGINFO
2401S_my_nl_langinfo(const nl_item item, bool toggle)
2402#else
2403S_my_nl_langinfo(const int item, bool toggle)
2404#endif
2405{
ae74815b 2406 dTHX;
0b6697c5 2407 const char * retval;
f7416781 2408
5a854ab3
KW
2409 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2410 * two items, and only if not already there */
4e6826bf 2411 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3
KW
2412 || PL_numeric_underlying))
2413 {
2414 toggle = FALSE;
2415 }
2416
ab340fff 2417#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2418# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2419 || ! defined(HAS_POSIX_2008_LOCALE) \
2420 || ! defined(DUPLOCALE)
f7416781 2421
ab340fff 2422 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2423 * for those items dependent on it. This must be copied to a buffer before
2424 * switching back, as some systems destroy the buffer when setlocale() is
2425 * called */
f7416781 2426
038d3702
KW
2427 {
2428 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2429
b2fee59c 2430 if (toggle) {
038d3702 2431 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2432 }
f7416781 2433
5acc4454
KW
2434 LOCALE_LOCK; /* Prevent interference from another thread executing
2435 this code section (the only call to nl_langinfo in
2436 the core) */
2437
ee90eb2d 2438
628ff75a
KW
2439 /* Copy to a per-thread buffer, which is also one that won't be
2440 * destroyed by a subsequent setlocale(), such as the
2441 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2442 retval = save_to_buffer(nl_langinfo(item),
2443 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2444
5acc4454
KW
2445 LOCALE_UNLOCK;
2446
b2fee59c 2447 if (toggle) {
038d3702 2448 RESTORE_LC_NUMERIC();
b2fee59c 2449 }
f7416781
KW
2450 }
2451
ab340fff
KW
2452# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2453
5a854ab3 2454 {
b2fee59c
KW
2455 bool do_free = FALSE;
2456 locale_t cur = uselocale((locale_t) 0);
ab340fff 2457
b2fee59c
KW
2458 if (cur == LC_GLOBAL_LOCALE) {
2459 cur = duplocale(LC_GLOBAL_LOCALE);
2460 do_free = TRUE;
2461 }
ab340fff 2462
b2fee59c 2463 if (toggle) {
e1aa2579
KW
2464 if (PL_underlying_numeric_obj) {
2465 cur = PL_underlying_numeric_obj;
2466 }
2467 else {
2468 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2469 do_free = TRUE;
2470 }
b2fee59c 2471 }
ab340fff 2472
628ff75a
KW
2473 /* We have to save it to a buffer, because the freelocale() just below
2474 * can invalidate the internal one */
0b6697c5
KW
2475 retval = save_to_buffer(nl_langinfo_l(item, cur),
2476 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2477
b2fee59c
KW
2478 if (do_free) {
2479 freelocale(cur);
2480 }
5a854ab3 2481 }
ab340fff 2482
c1566110
KW
2483# endif
2484
0b6697c5 2485 if (strEQ(retval, "")) {
4e6826bf 2486 if (item == YESSTR) {
c1566110
KW
2487 return "yes";
2488 }
4e6826bf 2489 if (item == NOSTR) {
c1566110
KW
2490 return "no";
2491 }
2492 }
2493
0b6697c5 2494 return retval;
ab340fff 2495
f7416781 2496#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2497
2498 {
2499
f7416781
KW
2500# ifdef HAS_LOCALECONV
2501
43dd6b15 2502 const struct lconv* lc;
628ff75a 2503 const char * temp;
038d3702 2504 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2505
69c5e0db
KW
2506# ifdef TS_W32_BROKEN_LOCALECONV
2507
2508 const char * save_global;
2509 const char * save_thread;
2510 int needed_size;
2511 char * ptr;
2512 char * e;
2513 char * item_start;
2514
2515# endif
f7416781
KW
2516# endif
2517# ifdef HAS_STRFTIME
2518
43dd6b15
KW
2519 struct tm tm;
2520 bool return_format = FALSE; /* Return the %format, not the value */
2521 const char * format;
f7416781
KW
2522
2523# endif
2524
43dd6b15
KW
2525 /* We copy the results to a per-thread buffer, even if not
2526 * multi-threaded. This is in part to simplify this code, and partly
2527 * because we need a buffer anyway for strftime(), and partly because a
2528 * call of localeconv() could otherwise wipe out the buffer, and the
2529 * programmer would not be expecting this, as this is a nl_langinfo()
2530 * substitute after all, so s/he might be thinking their localeconv()
2531 * is safe until another localeconv() call. */
f7416781 2532
43dd6b15
KW
2533 switch (item) {
2534 Size_t len;
f7416781 2535
8d72e74e 2536 /* This is unimplemented */
4e6826bf 2537 case ERA: /* For use with strftime() %E modifier */
f7416781 2538
43dd6b15
KW
2539 default:
2540 return "";
f7416781 2541
43dd6b15 2542 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2543 case YESEXPR: return "^[+1yY]";
2544 case YESSTR: return "yes";
2545 case NOEXPR: return "^[-0nN]";
2546 case NOSTR: return "no";
2547
8d72e74e
KW
2548 case CODESET:
2549
2550# ifndef WIN32
2551
2552 /* On non-windows, this is unimplemented, in part because of
2553 * inconsistencies between vendors. The Darwin native
2554 * nl_langinfo() implementation simply looks at everything past
2555 * any dot in the name, but that doesn't work for other
2556 * vendors. Many Linux locales that don't have UTF-8 in their
2557 * names really are UTF-8, for example; z/OS locales that do
2558 * have UTF-8 in their names, aren't really UTF-8 */
2559 return "";
2560
2561# else
2562
2563 { /* But on Windows, the name does seem to be consistent, so
2564 use that. */
2565 const char * p;
2566 const char * first;
2567 Size_t offset = 0;
2568 const char * name = my_setlocale(LC_CTYPE, NULL);
2569
2570 if (isNAME_C_OR_POSIX(name)) {
2571 return "ANSI_X3.4-1968";
2572 }
2573
2574 /* Find the dot in the locale name */
2575 first = (const char *) strchr(name, '.');
2576 if (! first) {
2577 first = name;
2578 goto has_nondigit;
2579 }
2580
2581 /* Look at everything past the dot */
2582 first++;
2583 p = first;
2584
2585 while (*p) {
2586 if (! isDIGIT(*p)) {
2587 goto has_nondigit;
2588 }
2589
2590 p++;
2591 }
2592
2593 /* Here everything past the dot is a digit. Treat it as a
2594 * code page */
32a62865
KW
2595 (void) save_to_buffer("CP", &PL_langinfo_buf,
2596 &PL_langinfo_bufsize, 0);
8d72e74e 2597 offset = STRLENs("CP");
f7416781 2598
8d72e74e
KW
2599 has_nondigit:
2600
2601 retval = save_to_buffer(first, &PL_langinfo_buf,
2602 &PL_langinfo_bufsize, offset);
2603 }
2604
2605 break;
2606
2607# endif
f7416781
KW
2608# ifdef HAS_LOCALECONV
2609
4e6826bf 2610 case CRNCYSTR:
f7416781 2611
43dd6b15
KW
2612 /* We don't bother with localeconv_l() because any system that
2613 * has it is likely to also have nl_langinfo() */
291a84fb 2614
69c5e0db
KW
2615 LOCALE_LOCK_V; /* Prevent interference with other threads
2616 using localeconv() */
2617
2618# ifdef TS_W32_BROKEN_LOCALECONV
2619
2620 /* This is a workaround for a Windows bug prior to VS 15.
2621 * What we do here is, while locked, switch to the global
2622 * locale so localeconv() works; then switch back just before
2623 * the unlock. This can screw things up if some thread is
2624 * already using the global locale while assuming no other is.
2625 * A different workaround would be to call GetCurrencyFormat on
2626 * a known value, and parse it; patches welcome
2627 *
2628 * We have to use LC_ALL instead of LC_MONETARY because of
2629 * another bug in Windows */
2630
2631 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2632 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2633 save_global= savepv(my_setlocale(LC_ALL, NULL));
2634 my_setlocale(LC_ALL, save_thread);
2635
2636# endif
5acc4454 2637
43dd6b15
KW
2638 lc = localeconv();
2639 if ( ! lc
2640 || ! lc->currency_symbol
2641 || strEQ("", lc->currency_symbol))
2642 {
69c5e0db 2643 LOCALE_UNLOCK_V;
43dd6b15
KW
2644 return "";
2645 }
f7416781 2646
43dd6b15 2647 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2648 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2649 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2650 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2651 { /* khw couldn't figure out how the localedef specifications
2652 would show that the $ should replace the radix; this is
2653 just a guess as to how it might work.*/
a34edee3 2654 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2655 }
2656 else if (lc->p_cs_precedes) {
a34edee3 2657 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2658 }
2659 else {
a34edee3 2660 PL_langinfo_buf[0] = '+';
43dd6b15 2661 }
f7416781 2662
69c5e0db
KW
2663# ifdef TS_W32_BROKEN_LOCALECONV
2664
2665 my_setlocale(LC_ALL, save_global);
2666 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2667 my_setlocale(LC_ALL, save_thread);
2668 Safefree(save_global);
2669 Safefree(save_thread);
2670
2671# endif
2672
2673 LOCALE_UNLOCK_V;
43dd6b15 2674 break;
f7416781 2675
69c5e0db
KW
2676# ifdef TS_W32_BROKEN_LOCALECONV
2677
4e6826bf 2678 case RADIXCHAR:
69c5e0db
KW
2679
2680 /* For this, we output a known simple floating point number to
2681 * a buffer, and parse it, looking for the radix */
2682
2683 if (toggle) {
2684 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2685 }
2686
2687 if (PL_langinfo_bufsize < 10) {
2688 PL_langinfo_bufsize = 10;
2689 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2690 }
2691
2692 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2693 "%.1f", 1.5);
2694 if (needed_size >= (int) PL_langinfo_bufsize) {
2695 PL_langinfo_bufsize = needed_size + 1;
2696 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2697 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2698 "%.1f", 1.5);
2699 assert(needed_size < (int) PL_langinfo_bufsize);
2700 }
2701
2702 ptr = PL_langinfo_buf;
2703 e = PL_langinfo_buf + PL_langinfo_bufsize;
2704
2705 /* Find the '1' */
2706 while (ptr < e && *ptr != '1') {
2707 ptr++;
2708 }
2709 ptr++;
2710
2711 /* Find the '5' */
2712 item_start = ptr;
2713 while (ptr < e && *ptr != '5') {
2714 ptr++;
2715 }
2716
2717 /* Everything in between is the radix string */
2718 if (ptr >= e) {
2719 PL_langinfo_buf[0] = '?';
2720 PL_langinfo_buf[1] = '\0';
2721 }
2722 else {
2723 *ptr = '\0';
2724 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2725 }
2726
2727 if (toggle) {
2728 RESTORE_LC_NUMERIC();
2729 }
2730
2731 retval = PL_langinfo_buf;
2732 break;
2733
2734# else
2735
2736 case RADIXCHAR: /* No special handling needed */
2737
2738# endif
2739
4e6826bf 2740 case THOUSEP:
f7416781 2741
43dd6b15 2742 if (toggle) {
038d3702 2743 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2744 }
f7416781 2745
69c5e0db
KW
2746 LOCALE_LOCK_V; /* Prevent interference with other threads
2747 using localeconv() */
2748
2749# ifdef TS_W32_BROKEN_LOCALECONV
2750
2751 /* This should only be for the thousands separator. A
2752 * different work around would be to use GetNumberFormat on a
2753 * known value and parse the result to find the separator */
2754 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2755 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2756 save_global = savepv(my_setlocale(LC_ALL, NULL));
2757 my_setlocale(LC_ALL, save_thread);
2758# if 0
2759 /* This is the start of code that for broken Windows replaces
2760 * the above and below code, and instead calls
2761 * GetNumberFormat() and then would parse that to find the
2762 * thousands separator. It needs to handle UTF-16 vs -8
2763 * issues. */
2764
2765 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2766 DEBUG_L(PerlIO_printf(Perl_debug_log,
2767 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2768 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2769
2770# endif
2771# endif
5acc4454 2772
43dd6b15
KW
2773 lc = localeconv();
2774 if (! lc) {
628ff75a 2775 temp = "";
33394adc 2776 }
43dd6b15 2777 else {
4e6826bf 2778 temp = (item == RADIXCHAR)
43dd6b15
KW
2779 ? lc->decimal_point
2780 : lc->thousands_sep;
628ff75a
KW
2781 if (! temp) {
2782 temp = "";
43dd6b15
KW
2783 }
2784 }
f7416781 2785
0b6697c5
KW
2786 retval = save_to_buffer(temp, &PL_langinfo_buf,
2787 &PL_langinfo_bufsize, 0);
f7416781 2788
69c5e0db
KW
2789# ifdef TS_W32_BROKEN_LOCALECONV
2790
2791 my_setlocale(LC_ALL, save_global);
2792 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2793 my_setlocale(LC_ALL, save_thread);
2794 Safefree(save_global);
2795 Safefree(save_thread);
2796
2797# endif
2798
2799 LOCALE_UNLOCK_V;
5acc4454 2800
43dd6b15 2801 if (toggle) {
038d3702 2802 RESTORE_LC_NUMERIC();
43dd6b15 2803 }
f7416781 2804
43dd6b15 2805 break;
f7416781
KW
2806
2807# endif
2808# ifdef HAS_STRFTIME
2809
43dd6b15
KW
2810 /* These are defined by C89, so we assume that strftime supports
2811 * them, and so are returned unconditionally; they may not be what
2812 * the locale actually says, but should give good enough results
2813 * for someone using them as formats (as opposed to trying to parse
2814 * them to figure out what the locale says). The other format
2815 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2816 case D_FMT: return "%x";
2817 case T_FMT: return "%X";
2818 case D_T_FMT: return "%c";
43dd6b15
KW
2819
2820 /* These formats are only available in later strfmtime's */
4e6826bf 2821 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2822
2823 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2824 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2825 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2826 case ALT_DIGITS:
2827 case AM_STR: case PM_STR:
2828 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2829 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2830 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2831 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2832 case DAY_5: case DAY_6: case DAY_7:
2833 case MON_1: case MON_2: case MON_3: case MON_4:
2834 case MON_5: case MON_6: case MON_7: case MON_8:
2835 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2836
2837 LOCALE_LOCK;
2838
2839 init_tm(&tm); /* Precaution against core dumps */
2840 tm.tm_sec = 30;
2841 tm.tm_min = 30;
2842 tm.tm_hour = 6;
2843 tm.tm_year = 2017 - 1900;
2844 tm.tm_wday = 0;
2845 tm.tm_mon = 0;
2846 switch (item) {
2847 default:
2848 LOCALE_UNLOCK;
2849 Perl_croak(aTHX_
2850 "panic: %s: %d: switch case: %d problem",
2851 __FILE__, __LINE__, item);
2852 NOT_REACHED; /* NOTREACHED */
2853
4e6826bf
KW
2854 case PM_STR: tm.tm_hour = 18;
2855 case AM_STR:
43dd6b15
KW
2856 format = "%p";
2857 break;
2858
4e6826bf
KW
2859 case ABDAY_7: tm.tm_wday++;
2860 case ABDAY_6: tm.tm_wday++;
2861 case ABDAY_5: tm.tm_wday++;
2862 case ABDAY_4: tm.tm_wday++;
2863 case ABDAY_3: tm.tm_wday++;
2864 case ABDAY_2: tm.tm_wday++;
2865 case ABDAY_1:
43dd6b15
KW
2866 format = "%a";
2867 break;
2868
4e6826bf
KW
2869 case DAY_7: tm.tm_wday++;
2870 case DAY_6: tm.tm_wday++;
2871 case DAY_5: tm.tm_wday++;
2872 case DAY_4: tm.tm_wday++;
2873 case DAY_3: tm.tm_wday++;
2874 case DAY_2: tm.tm_wday++;
2875 case DAY_1:
43dd6b15
KW
2876 format = "%A";
2877 break;
2878
4e6826bf
KW
2879 case ABMON_12: tm.tm_mon++;
2880 case ABMON_11: tm.tm_mon++;
2881 case ABMON_10: tm.tm_mon++;
2882 case ABMON_9: tm.tm_mon++;
2883 case ABMON_8: tm.tm_mon++;
2884 case ABMON_7: tm.tm_mon++;
2885 case ABMON_6: tm.tm_mon++;
2886 case ABMON_5: tm.tm_mon++;
2887 case ABMON_4: tm.tm_mon++;
2888 case ABMON_3: tm.tm_mon++;
2889 case ABMON_2: tm.tm_mon++;
2890 case ABMON_1:
43dd6b15
KW
2891 format = "%b";
2892 break;
2893
4e6826bf
KW
2894 case MON_12: tm.tm_mon++;
2895 case MON_11: tm.tm_mon++;
2896 case MON_10: tm.tm_mon++;
2897 case MON_9: tm.tm_mon++;
2898 case MON_8: tm.tm_mon++;
2899 case MON_7: tm.tm_mon++;
2900 case MON_6: tm.tm_mon++;
2901 case MON_5: tm.tm_mon++;
2902 case MON_4: tm.tm_mon++;
2903 case MON_3: tm.tm_mon++;
2904 case MON_2: tm.tm_mon++;
2905 case MON_1:
43dd6b15
KW
2906 format = "%B";
2907 break;
2908
4e6826bf 2909 case T_FMT_AMPM:
43dd6b15
KW
2910 format = "%r";
2911 return_format = TRUE;
2912 break;
2913
4e6826bf 2914 case ERA_D_FMT:
43dd6b15
KW
2915 format = "%Ex";
2916 return_format = TRUE;
2917 break;
2918
4e6826bf 2919 case ERA_T_FMT:
43dd6b15
KW
2920 format = "%EX";
2921 return_format = TRUE;
2922 break;
2923
4e6826bf 2924 case ERA_D_T_FMT:
43dd6b15
KW
2925 format = "%Ec";
2926 return_format = TRUE;
2927 break;
2928
4e6826bf 2929 case ALT_DIGITS:
43dd6b15
KW
2930 tm.tm_wday = 0;
2931 format = "%Ow"; /* Find the alternate digit for 0 */
2932 break;
2933 }
f7416781 2934
43dd6b15
KW
2935 /* We can't use my_strftime() because it doesn't look at
2936 * tm_wday */
2937 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2938 format, &tm))
2939 {
2940 /* A zero return means one of:
2941 * a) there wasn't enough space in PL_langinfo_buf
2942 * b) the format, like a plain %p, returns empty
2943 * c) it was an illegal format, though some
2944 * implementations of strftime will just return the
2945 * illegal format as a plain character sequence.
2946 *
2947 * To quickly test for case 'b)', try again but precede
2948 * the format with a plain character. If that result is
2949 * still empty, the problem is either 'a)' or 'c)' */
2950
2951 Size_t format_size = strlen(format) + 1;
2952 Size_t mod_size = format_size + 1;
2953 char * mod_format;
2954 char * temp_result;
2955
2956 Newx(mod_format, mod_size, char);
2957 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2958 *mod_format = ' ';
43dd6b15
KW
2959 my_strlcpy(mod_format + 1, format, mod_size);
2960 len = strftime(temp_result,
2961 PL_langinfo_bufsize,
2962 mod_format, &tm);
2963 Safefree(mod_format);
2964 Safefree(temp_result);
2965
2966 /* If 'len' is non-zero, it means that we had a case like
2967 * %p which means the current locale doesn't use a.m. or
2968 * p.m., and that is valid */
2969 if (len == 0) {
2970
2971 /* Here, still didn't work. If we get well beyond a
2972 * reasonable size, bail out to prevent an infinite
2973 * loop. */
2974
2975 if (PL_langinfo_bufsize > 100 * format_size) {
2976 *PL_langinfo_buf = '\0';
2977 }
2978 else {
2979 /* Double the buffer size to retry; Add 1 in case
2980 * original was 0, so we aren't stuck at 0. */
2981 PL_langinfo_bufsize *= 2;
2982 PL_langinfo_bufsize++;
2983 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2984 continue;
2985 }
2986 }
f7416781 2987
f7416781 2988 break;
43dd6b15 2989 }
f7416781 2990
43dd6b15
KW
2991 /* Here, we got a result.
2992 *
2993 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2994 * alternate format for wday 0. If the value is the same as
2995 * the normal 0, there isn't an alternate, so clear the buffer.
2996 * */
4e6826bf 2997 if ( item == ALT_DIGITS
43dd6b15
KW
2998 && strEQ(PL_langinfo_buf, "0"))
2999 {
3000 *PL_langinfo_buf = '\0';
3001 }
f7416781 3002
43dd6b15
KW
3003 /* ALT_DIGITS is problematic. Experiments on it showed that
3004 * strftime() did not always work properly when going from
3005 * alt-9 to alt-10. Only a few locales have this item defined,
3006 * and in all of them on Linux that khw was able to find,
3007 * nl_langinfo() merely returned the alt-0 character, possibly
3008 * doubled. Most Unicode digits are in blocks of 10
3009 * consecutive code points, so that is sufficient information
3010 * for those scripts, as we can infer alt-1, alt-2, .... But
3011 * for a Japanese locale, a CJK ideographic 0 is returned, and
3012 * the CJK digits are not in code point order, so you can't
3013 * really infer anything. The localedef for this locale did
3014 * specify the succeeding digits, so that strftime() works
3015 * properly on them, without needing to infer anything. But
3016 * the nl_langinfo() return did not give sufficient information
3017 * for the caller to understand what's going on. So until
3018 * there is evidence that it should work differently, this
3019 * returns the alt-0 string for ALT_DIGITS.
3020 *
3021 * wday was chosen because its range is all a single digit.
3022 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3023
43dd6b15 3024 LOCALE_UNLOCK;
f7416781 3025
d1b150d9
KW
3026 retval = PL_langinfo_buf;
3027
43dd6b15
KW
3028 /* If to return the format, not the value, overwrite the buffer
3029 * with it. But some strftime()s will keep the original format
3030 * if illegal, so change those to "" */
3031 if (return_format) {
3032 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3033 *PL_langinfo_buf = '\0';
3034 }
43dd6b15 3035 else {
0b6697c5
KW
3036 retval = save_to_buffer(format, &PL_langinfo_buf,
3037 &PL_langinfo_bufsize, 0);
f7416781
KW
3038 }
3039 }
3040
3041 break;
f7416781
KW
3042
3043# endif
3044
43dd6b15 3045 }
f7416781
KW
3046 }
3047
0b6697c5 3048 return retval;
f7416781
KW
3049
3050#endif
3051
a4f00dcc 3052}
b385bb4d 3053
98994639
HS
3054/*
3055 * Initialize locale awareness.
3056 */
3057int
3058Perl_init_i18nl10n(pTHX_ int printwarn)
3059{
0e92a118
KW
3060 /* printwarn is
3061 *
3062 * 0 if not to output warning when setup locale is bad
3063 * 1 if to output warning based on value of PERL_BADLANG
3064 * >1 if to output regardless of PERL_BADLANG
3065 *
3066 * returns
98994639 3067 * 1 = set ok or not applicable,
0e92a118
KW
3068 * 0 = fallback to a locale of lower priority
3069 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3070 *
3071 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3072 * set, debugging information is output.
3073 *
3074 * This looks more complicated than it is, mainly due to the #ifdefs.
3075 *
3076 * We try to set LC_ALL to the value determined by the environment. If
3077 * there is no LC_ALL on this platform, we try the individual categories we
3078 * know about. If this works, we are done.
3079 *
3080 * But if it doesn't work, we have to do something else. We search the
3081 * environment variables ourselves instead of relying on the system to do
3082 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3083 * think there is one), and the ultimate fallback "C". This is all done in
3084 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3085 * more complex. The 'trial_locales' array is initialized with just one
3086 * element; it causes the behavior described in the paragraph above this to
3087 * happen. If that fails, we add elements to 'trial_locales', and do extra
3088 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3089 *
3090 * On Ultrix, the locale MUST come from the environment, so there is
3091 * preliminary code to set it. I (khw) am not sure that it is necessary,
3092 * and that this couldn't be folded into the loop, but barring any real
3093 * platforms to test on, it's staying as-is
3094 *
3095 * A slight complication is that in embedded Perls, the locale may already
3096 * be set-up, and we don't want to get it from the normal environment
3097 * variables. This is handled by having a special environment variable
3098 * indicate we're in this situation. We simply set setlocale's 2nd
3099 * parameter to be a NULL instead of "". That indicates to setlocale that
3100 * it is not to change anything, but to return the current value,
3101 * effectively initializing perl's db to what the locale already is.
3102 *
3103 * We play the same trick with NULL if a LC_ALL succeeds. We call
3104 * setlocale() on the individual categores with NULL to get their existing
3105 * values for our db, instead of trying to change them.
3106 * */
98994639 3107
0e92a118
KW
3108 int ok = 1;
3109
7d4bcc4a
KW
3110#ifndef USE_LOCALE
3111
3112 PERL_UNUSED_ARG(printwarn);
3113
3114#else /* USE_LOCALE */
7d4bcc4a
KW
3115# ifdef __GLIBC__
3116
175c4cf9 3117 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3118
3119# endif
65ebb059 3120
ccd65d51
KW
3121 /* NULL uses the existing already set up locale */
3122 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3123 ? NULL
3124 : "";
c3fcd832
KW
3125 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3126 unsigned int trial_locales_count;
175c4cf9
KW
3127 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3128 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3129 bool setlocale_failure = FALSE;
65ebb059 3130 unsigned int i;
175c4cf9
KW
3131
3132 /* A later getenv() could zap this, so only use here */
3133 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3134
3135 const bool locwarn = (printwarn > 1
e5f10d49
KW
3136 || ( printwarn
3137 && ( ! bad_lang_use_once
22ff3130 3138 || (
e5f10d49
KW
3139 /* disallow with "" or "0" */
3140 *bad_lang_use_once
3141 && strNE("0", bad_lang_use_once)))));
ea92aad8 3142
291a84fb 3143 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3144 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3145
3146 /* current locale for given category; should have been copied so aren't
3147 * volatile */
8de4332b 3148 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3149
7d4bcc4a
KW
3150# ifdef WIN32
3151
6bce99ee
JH
3152 /* In some systems you can find out the system default locale
3153 * and use that as the fallback locale. */
7d4bcc4a
KW
3154# define SYSTEM_DEFAULT_LOCALE
3155# endif
3156# ifdef SYSTEM_DEFAULT_LOCALE
3157
65ebb059 3158 const char *system_default_locale = NULL;
98994639 3159
7d4bcc4a 3160# endif
948523db
KW
3161
3162# ifndef DEBUGGING
3163# define DEBUG_LOCALE_INIT(a,b,c)
3164# else
7d4bcc4a 3165
8298454c 3166 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3167
3168# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3169 STMT_START { \
3170 if (debug_initialization) { \
3171 PerlIO_printf(Perl_debug_log, \
3172 "%s:%d: %s\n", \
3173 __FILE__, __LINE__, \
a4f00dcc 3174 setlocale_debug_string(category, \
2fcc0ca9
KW
3175 locale, \
3176 result)); \
3177 } \
3178 } STMT_END
2fcc0ca9 3179
948523db
KW
3180/* Make sure the parallel arrays are properly set up */
3181# ifdef USE_LOCALE_NUMERIC
3182 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3183 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3184# ifdef USE_POSIX_2008_LOCALE
3185 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3186# endif
948523db
KW
3187# endif
3188# ifdef USE_LOCALE_CTYPE
3189 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3190 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3191# ifdef USE_POSIX_2008_LOCALE
3192 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3193# endif
948523db
KW
3194# endif
3195# ifdef USE_LOCALE_COLLATE
3196 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3197 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3198# ifdef USE_POSIX_2008_LOCALE
3199 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3200# endif
948523db
KW
3201# endif
3202# ifdef USE_LOCALE_TIME
3203 assert(categories[LC_TIME_INDEX] == LC_TIME);
3204 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3205# ifdef USE_POSIX_2008_LOCALE
3206 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3207# endif
948523db
KW
3208# endif
3209# ifdef USE_LOCALE_MESSAGES
3210 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3211 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3212# ifdef USE_POSIX_2008_LOCALE
3213 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3214# endif
948523db
KW
3215# endif
3216# ifdef USE_LOCALE_MONETARY
3217 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3218 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3219# ifdef USE_POSIX_2008_LOCALE
3220 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3221# endif
948523db 3222# endif
9821811f
KW
3223# ifdef USE_LOCALE_ADDRESS
3224 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3225 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3226# ifdef USE_POSIX_2008_LOCALE
3227 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3228# endif
9821811f
KW
3229# endif
3230# ifdef USE_LOCALE_IDENTIFICATION
3231 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3232 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3233# ifdef USE_POSIX_2008_LOCALE
3234 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3235# endif
9821811f
KW
3236# endif
3237# ifdef USE_LOCALE_MEASUREMENT
3238 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3239 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3240# ifdef USE_POSIX_2008_LOCALE
3241 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3242# endif
9821811f
KW
3243# endif
3244# ifdef USE_LOCALE_PAPER
3245 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3246 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3247# ifdef USE_POSIX_2008_LOCALE
3248 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3249# endif
9821811f
KW
3250# endif
3251# ifdef USE_LOCALE_TELEPHONE
3252 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3253 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3254# ifdef USE_POSIX_2008_LOCALE
3255 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3256# endif
9821811f 3257# endif
948523db
KW
3258# ifdef LC_ALL
3259 assert(categories[LC_ALL_INDEX] == LC_ALL);
3260 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3261 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3262# ifdef USE_POSIX_2008_LOCALE
3263 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3264# endif
948523db
KW
3265# endif
3266# endif /* DEBUGGING */
47280b20
KW
3267
3268 /* Initialize the cache of the program's UTF-8ness for the always known
3269 * locales C and POSIX */
3270 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3271 sizeof(PL_locale_utf8ness));
3272
e9bc6d6b
KW
3273# ifdef USE_THREAD_SAFE_LOCALE
3274# ifdef WIN32
3275
3276 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3277
3278# endif
3279# endif
39e69e77 3280# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3281
3282 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3283 if (! PL_C_locale_obj) {
3284 Perl_croak_nocontext(
3285 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3286 }
3287 if (DEBUG_Lv_TEST || debug_initialization) {
3288 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3289 }
3290
3291# endif
3292
3ca88433
KW
3293 PL_numeric_radix_sv = newSVpvs(".");
3294
e9bc6d6b
KW
3295# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3296
3297 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3298 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3299
3300# endif
b56c4436 3301# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3302
3303 /*
3304 * Ultrix setlocale(..., "") fails if there are no environment
3305 * variables from which to get a locale name.
3306 */
3307
b56c4436
KW
3308# ifndef LC_ALL
3309# error Ultrix without LC_ALL not implemented
3310# else
7d4bcc4a 3311
b56c4436
KW
3312 {
3313 bool done = FALSE;
ea92aad8
KW
3314 if (lang) {
3315 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3316 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3317 if (sl_result[LC_ALL_INDEX])
3318 done = TRUE;
3319 else
e5f10d49 3320 setlocale_failure = TRUE;
ea92aad8
KW
3321 }
3322 if (! setlocale_failure) {
3323 const char * locale_param;
3324 for (i = 0; i < LC_ALL_INDEX; i++) {
3325 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3326 ? setlocale_init
3327 : NULL;
3328 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3329 if (! sl_result[i]) {
3330 setlocale_failure = TRUE;
3331 }
3332 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3333 }
c835d6be 3334 }
7d4bcc4a
KW
3335 }
3336
3337# endif /* LC_ALL */
e5f10d49 3338# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3339
65ebb059 3340 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3341 * the list. Normally the loop is executed just once. But if setting the
3342 * locale fails, inside the loop we add fallback trials to the array and so
3343 * will execute the loop multiple times */
c3fcd832
KW
3344 trial_locales[0] = setlocale_init;
3345 trial_locales_count = 1;
7d4bcc4a 3346
65ebb059
KW
3347 for (i= 0; i < trial_locales_count; i++) {
3348 const char * trial_locale = trial_locales[i];
3349
3350 if (i > 0) {
3351
3352 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3353 * when i==0, but I (khw) don't think that behavior makes much
3354 * sense */
3355 setlocale_failure = FALSE;
3356
7d4bcc4a 3357# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3358# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3359
65ebb059
KW
3360 /* On Windows machines, an entry of "" after the 0th means to use
3361 * the system default locale, which we now proceed to get. */
3362 if (strEQ(trial_locale, "")) {
3363 unsigned int j;
3364
3365 /* Note that this may change the locale, but we are going to do
3366 * that anyway just below */
837ce802 3367 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3368 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3369
7d4bcc4a 3370 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3371 * try */
3372 if (! system_default_locale) {
3373 goto next_iteration;
3374 }
3375 for (j = 0; j < trial_locales_count; j++) {
3376 if (strEQ(system_default_locale, trial_locales[j])) {
3377 goto next_iteration;
3378 }
3379 }
3380
3381 trial_locale = system_default_locale;
3382 }
ec0202b5
KW
3383# else
3384# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3385# endif
7d4bcc4a 3386# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3387
3388 } /* For i > 0 */
65ebb059 3389
7d4bcc4a
KW
3390# ifdef LC_ALL
3391
948523db
KW
3392 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3393 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3394 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3395 setlocale_failure = TRUE;
7cd8b568
KW
3396 }
3397 else {
3398 /* Since LC_ALL succeeded, it should have changed all the other
3399 * categories it can to its value; so we massage things so that the
3400 * setlocales below just return their category's current values.
3401 * This adequately handles the case in NetBSD where LC_COLLATE may
3402 * not be defined for a locale, and setting it individually will
7d4bcc4a 3403 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3404 * the POSIX locale. */
3405 trial_locale = NULL;
3406 }
7d4bcc4a
KW
3407
3408# endif /* LC_ALL */
98994639 3409
e5f10d49
KW
3410 if (! setlocale_failure) {
3411 unsigned int j;
3412 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3413 curlocales[j]
3414 = savepv(do_setlocale_r(categories[j], trial_locale));
3415 if (! curlocales[j]) {
3416 setlocale_failure = TRUE;
3417 }
3418 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3419 }
c835d6be 3420
e5f10d49
KW
3421 if (! setlocale_failure) { /* All succeeded */
3422 break; /* Exit trial_locales loop */
49c85077 3423 }
65ebb059 3424 }
98994639 3425
49c85077
KW
3426 /* Here, something failed; will need to try a fallback. */
3427 ok = 0;
65ebb059 3428
49c85077
KW
3429 if (i == 0) {
3430 unsigned int j;
98994639 3431
65ebb059 3432 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3433
3434# ifdef LC_ALL
98994639 3435
49c85077
KW
3436 PerlIO_printf(Perl_error_log,
3437 "perl: warning: Setting locale failed.\n");
98994639 3438
7d4bcc4a 3439# else /* !LC_ALL */
98994639 3440
49c85077
KW
3441 PerlIO_printf(Perl_error_log,
3442 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3443
e5f10d49
KW
3444 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3445 if (! curlocales[j]) {
3446 PerlIO_printf(Perl_error_log, category_names[j]);
3447 }
3448 else {
3449 Safefree(curlocales[j]);
3450 }
3451 }
7d4bcc4a 3452
7d4bcc4a 3453# endif /* LC_ALL */
98994639 3454
49c85077
KW
3455 PerlIO_printf(Perl_error_log,
3456 "perl: warning: Please check that your locale settings:\n");
98994639 3457
7d4bcc4a
KW
3458# ifdef __GLIBC__
3459
49c85077
KW
3460 PerlIO_printf(Perl_error_log,
3461 "\tLANGUAGE = %c%s%c,\n",
3462 language ? '"' : '(',
3463 language ? language : "unset",
3464 language ? '"' : ')');
7d4bcc4a 3465# endif
98994639 3466
49c85077
KW
3467 PerlIO_printf(Perl_error_log,
3468 "\tLC_ALL = %c%s%c,\n",
3469 lc_all ? '"' : '(',
3470 lc_all ? lc_all : "unset",
3471 lc_all ? '"' : ')');
98994639 3472
7d4bcc4a
KW
3473# if defined(USE_ENVIRON_ARRAY)
3474
49c85077 3475 {
cd999af9 3476 char **e;
d5e32b93
KW
3477
3478 /* Look through the environment for any variables of the
3479 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3480 * already handled above. These are assumed to be locale
3481 * settings. Output them and their values. */
cd999af9 3482 for (e = environ; *e; e++) {
d5e32b93
KW
3483 const STRLEN prefix_len = sizeof("LC_") - 1;
3484 STRLEN uppers_len;
3485
cd999af9 3486 if ( strBEGINs(*e, "LC_")
c8b388b0 3487 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3488 && (uppers_len = strspn(*e + prefix_len,
3489 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3490 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3491 {
3492 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3493 (int) (prefix_len + uppers_len), *e,
3494 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3495 }
3496 }
49c85077 3497 }
7d4bcc4a
KW
3498
3499# else
3500
49c85077
KW
3501 PerlIO_printf(Perl_error_log,
3502 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3503
3504# endif
98994639 3505
49c85077
KW
3506 PerlIO_printf(Perl_error_log,
3507 "\tLANG = %c%s%c\n",
3508 lang ? '"' : '(',
3509 lang ? lang : "unset",
3510 lang ? '"' : ')');
98994639 3511
49c85077
KW
3512 PerlIO_printf(Perl_error_log,
3513 " are supported and installed on your system.\n");
3514 }
98994639 3515
65ebb059 3516 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3517 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3518 * usually change the upper bound of the loop we are in.
3519 *
3520 * Since the system's default way of setting the locale has not
3521 * found one that works, We use Perl's defined ordering: LC_ALL,
3522 * LANG, and the C locale. We don't try the same locale twice, so
3523 * don't add to the list if already there. (On POSIX systems, the
3524 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3525 * but there's no harm done by doing it explicitly.
3526 *
3527 * Note that this tries the LC_ALL environment variable even on
3528 * systems which have no LC_ALL locale setting. This may or may
3529 * not have been originally intentional, but there's no real need
3530 * to change the behavior. */
65ebb059
KW
3531 if (lc_all) {
3532 for (j = 0; j < trial_locales_count; j++) {
3533 if (strEQ(lc_all, trial_locales[j])) {
3534 goto done_lc_all;
3535 }
3536 }
3537 trial_locales[trial_locales_count++] = lc_all;
3538 }
3539 done_lc_all:
98994639 3540
65ebb059
KW
3541 if (lang) {
3542 for (j = 0; j < trial_locales_count; j++) {
3543 if (strEQ(lang, trial_locales[j])) {
3544 goto done_lang;
3545 }
3546 }
3547 trial_locales[trial_locales_count++] = lang;
3548 }
3549 done_lang:
3550
7d4bcc4a
KW
3551# if defined(WIN32) && defined(LC_ALL)
3552
65ebb059
KW
3553 /* For Windows, we also try the system default locale before "C".
3554 * (If there exists a Windows without LC_ALL we skip this because
3555 * it gets too complicated. For those, the "C" is the next
3556 * fallback possibility). The "" is the same as the 0th element of
3557 * the array, but the code at the loop above knows to treat it
3558 * differently when not the 0th */
3559 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3560
3561# endif
65ebb059
KW
3562
3563 for (j = 0; j < trial_locales_count; j++) {
3564 if (strEQ("C", trial_locales[j])) {
3565 goto done_C;
3566 }
3567 }
3568 trial_locales[trial_locales_count++] = "C";
98994639 3569
65ebb059
KW
3570 done_C: ;
3571 } /* end of first time through the loop */
98994639 3572
7d4bcc4a
KW
3573# ifdef WIN32
3574
65ebb059 3575 next_iteration: ;
7d4bcc4a
KW
3576
3577# endif
65ebb059
KW
3578
3579 } /* end of looping through the trial locales */
3580
3581 if (ok < 1) { /* If we tried to fallback */
3582 const char* msg;
3583 if (! setlocale_failure) { /* fallback succeeded */
3584 msg = "Falling back to";
3585 }
3586 else { /* fallback failed */
e5f10d49 3587 unsigned int j;
98994639 3588
65ebb059
KW
3589 /* We dropped off the end of the loop, so have to decrement i to
3590 * get back to the value the last time through */
3591 i--;
98994639 3592
65ebb059
KW
3593 ok = -1;
3594 msg = "Failed to fall back to";
3595
3596 /* To continue, we should use whatever values we've got */
7d4bcc4a 3597
e5f10d49
KW
3598 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3599 Safefree(curlocales[j]);
3600 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3601 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3602 }
65ebb059
KW
3603 }
3604
3605 if (locwarn) {
3606 const char * description;
3607 const char * name = "";
3608 if (strEQ(trial_locales[i], "C")) {
3609 description = "the standard locale";
3610 name = "C";
3611 }
7d4bcc4a
KW
3612
3613# ifdef SYSTEM_DEFAULT_LOCALE
3614
65ebb059
KW
3615 else if (strEQ(trial_locales[i], "")) {
3616 description = "the system default locale";
3617 if (system_default_locale) {
3618 name = system_default_locale;
3619 }
3620 }
7d4bcc4a
KW
3621
3622# endif /* SYSTEM_DEFAULT_LOCALE */
3623
65ebb059
KW
3624 else {
3625 description = "a fallback locale";
3626 name = trial_locales[i];
3627 }
3628 if (name && strNE(name, "")) {
3629 PerlIO_printf(Perl_error_log,
3630 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3631 }
3632 else {
3633 PerlIO_printf(Perl_error_log,
3634 "perl: warning: %s %s.\n", msg, description);
3635 }
3636 }
3637 } /* End of tried to fallback */
98994639 3638
e5f10d49
KW
3639 /* Done with finding the locales; update our records */
3640
7d4bcc4a
KW
3641# ifdef USE_LOCALE_CTYPE
3642
948523db 3643 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3644
e5f10d49 3645# endif
7d4bcc4a
KW
3646# ifdef USE_LOCALE_COLLATE
3647
948523db 3648 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3649
e5f10d49 3650# endif
7d4bcc4a
KW
3651# ifdef USE_LOCALE_NUMERIC
3652
948523db 3653 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3654
3655# endif
3656
948523db 3657 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3658
e9bc6d6b 3659# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3660
3661 /* This caches whether each category's locale is UTF-8 or not. This
3662 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3663 * initialization time before any threads have started, but not later
3664 * unless thread-safe operations are used.
47280b20
KW
3665 * Caching means that if the program heeds our dictate not to change
3666 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3667 * it may get queried without having to change locales. If the
3668 * environment is such that all categories have the same locale, this
3669 * isn't needed, as the code will not change the locale; but this
3670 * handles the uncommon case where the environment has disparate
3671 * locales for the categories */
47280b20
KW
3672 (void) _is_cur_LC_category_utf8(categories[i]);
3673
3674# endif
3675
e5f10d49
KW
3676 Safefree(curlocales[i]);
3677 }
b310b053 3678
7d4bcc4a
KW
3679# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3680
49c85077 3681 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3682 * locale is UTF-8. The call to new_ctype() just above has already
3683 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3684 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3685 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3686 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3687 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3688
a05d7ebb 3689 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3690 This is an alternative to using the -C command line switch
3691 (the -C if present will override this). */
3692 {
dd374669 3693 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3694 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3695 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3696 PL_utf8cache = -1;
b310b053
JH
3697 }
3698
7d4bcc4a 3699# endif
7d4bcc4a
KW
3700# ifdef __GLIBC__
3701
175c4cf9 3702 Safefree(language);
7d4bcc4a
KW
3703
3704# endif
175c4cf9
KW
3705
3706 Safefree(lc_all);
3707 Safefree(lang);
3708
e3305790 3709#endif /* USE_LOCALE */
2fcc0ca9 3710#ifdef DEBUGGING
7d4bcc4a 3711
2fcc0ca9 3712 /* So won't continue to output stuff */
27cdc72e 3713 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3714
2fcc0ca9
KW
3715#endif
3716
98994639
HS
3717 return ok;
3718}
3719
98994639
HS
3720#ifdef USE_LOCALE_COLLATE
3721
a4a439fb 3722char *
a4a439fb
KW
3723Perl__mem_collxfrm(pTHX_ const char *input_string,
3724 STRLEN len, /* Length of 'input_string' */
3725 STRLEN *xlen, /* Set to length of returned string
3726 (not including the collation index
3727 prefix) */
3728 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3729 )
98994639 3730{
a4a439fb
KW
3731
3732 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3733 * differences. First, it handles embedded NULs. Second, it allocates a bit
3734 * more memory than needed for the transformed data itself. The real
55e5378d 3735 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3736 * the length of that, and doesn't include the collation index size.
3737 * Please see sv_collxfrm() to see how this is used. */
3738
55e5378d
KW
3739#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3740
6696cfa7
KW
3741 char * s = (char *) input_string;
3742 STRLEN s_strlen = strlen(input_string);
79f120c8 3743 char *xbuf = NULL;
55e5378d 3744 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3745 STRLEN length_in_chars;
c664130f 3746 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3747
a4a439fb
KW
3748 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3749
3750 /* Must be NUL-terminated */
3751 assert(*(input_string + len) == '\0');
7918f24d 3752
79f120c8
KW
3753 /* If this locale has defective collation, skip */
3754 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3755 DEBUG_L(PerlIO_printf(Perl_debug_log,
3756 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3757 goto bad;
3758 }
3759
6696cfa7
KW
3760 /* Replace any embedded NULs with the control that sorts before any others.
3761 * This will give as good as possible results on strings that don't
3762 * otherwise contain that character, but otherwise there may be
3763 * less-than-perfect results with that character and NUL. This is
fdc080f3 3764 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3765 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3766 NUL */
6696cfa7
KW
3767 char * e = s + len;
3768 char * sans_nuls;
fdc080f3 3769 STRLEN sans_nuls_len;
94762aa0 3770 int try_non_controls;
afc4976f
KW
3771 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3772 making sure 2nd byte is NUL.
3773 */
3774 STRLEN this_replacement_len;
3775
1e4c9676
KW
3776 /* If we don't know what non-NUL control character sorts lowest for
3777 * this locale, find it */
f28f4d2a 3778 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3779 int j;
afc4976f 3780 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3781 includes the collation index
3782 prefixed. */
3783
91c0e2e0 3784 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3785
3786 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3787 * NUL, in which case we instead look for any character. Controls
3788 * are preferred because collation order is, in general, context
3789 * sensitive, with adjoining characters affecting the order, and
3790 * controls are less likely to have such interactions, allowing the
3791 * NUL-replacement to stand on its own. (Another way to look at it
3792 * is to imagine what would happen if the NUL were replaced by a
3793 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3794 for (try_non_controls = 0;
3795 try_non_controls < 2;
3796 try_non_controls++)
3797 {
d4ff9586
KW
3798 /* Look through all legal code points (NUL isn't) */
3799 for (j = 1; j < 256; j++) {
3800 char * x; /* j's xfrm plus collation index */
3801 STRLEN x_len; /* length of 'x' */
3802 STRLEN trial_len = 1;
736a4fed 3803 char cur_source[] = { '\0', '\0' };
d4ff9586 3804
736a4fed
KW
3805 /* Skip non-controls the first time through the loop. The
3806 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3807 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3808 ? ! isCNTRL_L1(j)
3809 : ! isCNTRL_LC(j))
3810 {
d4ff9586 3811 continue;
6696cfa7 3812 }
6696cfa7 3813
736a4fed
KW
3814 /* Create a 1-char string of the current code point */
3815 cur_source[0] = (char) j;
3816
d4ff9586
KW
3817 /* Then transform it */
3818 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3819 0 /* The string is not in UTF-8 */);
6696cfa7 3820
1e4c9676 3821 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3822 * */
3823 if (! x) {
3824 continue;
3825 }
6696cfa7 3826
d4ff9586
KW
3827 /* If this character's transformation is lower than
3828 * the current lowest, this one becomes the lowest */
3829 if ( cur_min_x == NULL
3830 || strLT(x + COLLXFRM_HDR_LEN,
3831 cur_min_x + COLLXFRM_HDR_LEN))
3832 {
f28f4d2a 3833 PL_strxfrm_NUL_replacement = j;
d4ff9586 3834 cur_min_x = x;
d4ff9586
KW
3835 }
3836 else {
3837 Safefree(x);
3838 }
1e4c9676 3839 } /* end of loop through all 255 characters */
6696cfa7 3840
1e4c9676 3841 /* Stop looking if found */
94762aa0
KW
3842 if (cur_min_x) {
3843 break;
3844 }
3845
3846 /* Unlikely, but possible, if there aren't any controls that
3847 * work in the locale, repeat the loop, looking for any
3848 * character that works */
3849 DEBUG_L(PerlIO_printf(Perl_debug_log,
3850 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3851 } /* End of loop to try first the controls, then any char */
6696cfa7 3852
94762aa0
KW
3853 if (! cur_min_x) {
3854 DEBUG_L(PerlIO_printf(Perl_debug_log,
3855 "_mem_collxfrm: Couldn't find any character to replace"
3856 " embedded NULs in locale %s with", PL_collation_name));
3857 goto bad;
58eebef2
KW
3858 }
3859
94762aa0
KW
3860 DEBUG_L(PerlIO_printf(Perl_debug_log,
3861 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3862 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));