This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
re/user_prop_race_thr.t: reduce timeout
[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
8203b3c3 334# define _DUMMY_COLLATE _DUMMY_CTYPE
948523db
KW
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
f8115d60 738 /* Here, we are switching locales. */
e9bc6d6b
KW
739
740# ifndef HAS_QUERYLOCALE
741
742 if (strEQ(locale, "")) {
743
744 /* For non-querylocale() systems, we do the setting of "" ourselves to
745 * be sure that we really know what's going on. We follow the Linux
746 * documented behavior (but if that differs from the actual behavior,
747 * this won't work exactly as the OS implements). We go out and
748 * examine the environment based on our understanding of how the system
749 * works, and use that to figure things out */
750
751 const char * const lc_all = PerlEnv_getenv("LC_ALL");
752
753 /* Use any "LC_ALL" environment variable, as it overrides everything
754 * else. */
755 if (lc_all && strNE(lc_all, "")) {
756 locale = lc_all;
757 }
758 else {
759
760 /* Otherwise, we need to dig deeper. Unless overridden, the
761 * default is the LANG environment variable; if it doesn't exist,
762 * then "C" */
763
764 const char * default_name;
765
766 /* To minimize other threads messing with the environment, we copy
767 * the variable, making it a temporary. But this doesn't work upon
768 * program initialization before any scopes are created, and at
769 * this time, there's nothing else going on that would interfere.
770 * So skip the copy in that case */
771 if (PL_scopestack_ix == 0) {
772 default_name = PerlEnv_getenv("LANG");
773 }
774 else {
775 default_name = savepv(PerlEnv_getenv("LANG"));
776 }
777
778 if (! default_name || strEQ(default_name, "")) {
929a7f8c 779 default_name = "C";
e9bc6d6b
KW
780 }
781 else if (PL_scopestack_ix != 0) {
782 SAVEFREEPV(default_name);
783 }
784
785 if (category != LC_ALL) {
786 const char * const name = PerlEnv_getenv(category_names[index]);
787
788 /* Here we are setting a single category. Assume will have the
789 * default name */
790 locale = default_name;
791
792 /* But then look for an overriding environment variable */
793 if (name && strNE(name, "")) {
794 locale = name;
795 }
796 }
797 else {
798 bool did_override = FALSE;
799 unsigned int i;
800
801 /* Here, we are getting LC_ALL. Any categories that don't have
802 * a corresponding environment variable set should be set to
803 * LANG, or to "C" if there is no LANG. If no individual
804 * categories differ from this, we can just set LC_ALL. This
805 * is buggy on systems that have extra categories that we don't
806 * know about. If there is an environment variable that sets
807 * that category, we won't know to look for it, and so our use
808 * of LANG or "C" improperly overrides it. On the other hand,
809 * if we don't do what is done here, and there is no
810 * environment variable, the category's locale should be set to
811 * LANG or "C". So there is no good solution. khw thinks the
812 * best is to look at systems to see what categories they have,
813 * and include them, and then to assume that we know the
814 * complete set */
815
816 for (i = 0; i < LC_ALL_INDEX; i++) {
817 const char * const env_override
818 = savepv(PerlEnv_getenv(category_names[i]));
819 const char * this_locale = ( env_override
820 && strNE(env_override, ""))
821 ? env_override
822 : default_name;
6435f98d
KW
823 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
824 {
825 Safefree(env_override);
826 return NULL;
827 }
e9bc6d6b
KW
828
829 if (strNE(this_locale, default_name)) {
830 did_override = TRUE;
831 }
832
833 Safefree(env_override);
834 }
835
836 /* If all the categories are the same, we can set LC_ALL to
837 * that */
838 if (! did_override) {
839 locale = default_name;
840 }
841 else {
842
843 /* Here, LC_ALL is no longer valid, as some individual
844 * categories don't match it. We call ourselves
845 * recursively, as that will execute the code that
846 * generates the proper locale string for this situation.
847 * We don't do the remainder of this function, as that is
848 * to update our records, and we've just done that for the
849 * individual categories in the loop above, and doing so
850 * would cause LC_ALL to be done as well */
851 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
852 }
853 }
854 }
855 }
856 else if (strchr(locale, ';')) {
857
858 /* LC_ALL may actually incude a conglomeration of various categories.
859 * Without querylocale, this code uses the glibc (as of this writing)
860 * syntax for representing that, but that is not a stable API, and
861 * other platforms do it differently, so we have to handle all cases
862 * ourselves */
863
f0023d47 864 unsigned int i;
e9bc6d6b
KW
865 const char * s = locale;
866 const char * e = locale + strlen(locale);
867 const char * p = s;
868 const char * category_end;
869 const char * name_start;
870 const char * name_end;
871
f0023d47
KW
872 /* If the string that gives what to set doesn't include all categories,
873 * the omitted ones get set to "C". To get this behavior, first set
874 * all the individual categories to "C", and override the furnished
875 * ones below */
876 for (i = 0; i < LC_ALL_INDEX; i++) {
877 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
878 return NULL;
879 }
880 }
881
e9bc6d6b 882 while (s < e) {
e9bc6d6b
KW
883
884 /* Parse through the category */
885 while (isWORDCHAR(*p)) {
886 p++;
887 }
888 category_end = p;
889
890 if (*p++ != '=') {
891 Perl_croak(aTHX_
892 "panic: %s: %d: Unexpected character in locale name '%02X",
893 __FILE__, __LINE__, *(p-1));
894 }
895
896 /* Parse through the locale name */
897 name_start = p;
bee74f4b
KW
898 while (p < e && *p != ';') {
899 if (! isGRAPH(*p)) {
900 Perl_croak(aTHX_
901 "panic: %s: %d: Unexpected character in locale name '%02X",
902 __FILE__, __LINE__, *(p-1));
903 }
e9bc6d6b
KW
904 p++;
905 }
906 name_end = p;
907
bee74f4b
KW
908 /* Space past the semi-colon */
909 if (p < e) {
910 p++;
e9bc6d6b
KW
911 }
912
913 /* Find the index of the category name in our lists */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 915 char * individ_locale;
e9bc6d6b
KW
916
917 /* Keep going if this isn't the index. The strnNE() avoids a
918 * Perl_form(), but would fail if ever a category name could be
919 * a substring of another one, like if there were a
920 * "LC_TIME_DATE" */
921 if strnNE(s, category_names[i], category_end - s) {
922 continue;
923 }
924
925 /* If this index is for the single category we're changing, we
926 * have found the locale to set it to. */
927 if (category == categories[i]) {
928 locale = Perl_form(aTHX_ "%.*s",
929 (int) (name_end - name_start),
930 name_start);
931 goto ready_to_set;
932 }
933
3c76a412 934 assert(category == LC_ALL);
bee74f4b
KW
935 individ_locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start), name_start);
6435f98d
KW
937 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
938 {
939 return NULL;
940 }
e9bc6d6b
KW
941 }
942
943 s = p;
944 }
945
946 /* Here we have set all the individual categories by recursive calls.
947 * These collectively should have fixed up LC_ALL, so can just query
948 * what that now is */
949 assert(category == LC_ALL);
950
951 return do_setlocale_c(LC_ALL, NULL);
952 }
953
954 ready_to_set: ;
955
f8115d60
KW
956 /* Here at the end of having to deal with the absence of querylocale().
957 * Some cases have already been fully handled by recursive calls to this
958 * function. But at this point, we haven't dealt with those, but are now
959 * prepared to, knowing what the locale name to set this category to is.
960 * This would have come for free if this system had had querylocale() */
961
e9bc6d6b
KW
962# endif /* end of ! querylocale */
963
f8115d60
KW
964 assert(PL_C_locale_obj);
965
966 /* Switching locales generally entails freeing the current one's space (at
967 * the C library's discretion). We need to stop using that locale before
968 * the switch. So switch to a known locale object that we don't otherwise
969 * mess with. This returns the locale object in effect at the time of the
970 * switch. */
971 old_obj = uselocale(PL_C_locale_obj);
972
973# ifdef DEBUGGING
974
975 if (DEBUG_Lv_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
977 }
978
979# endif
980
981 if (! old_obj) {
982
983# ifdef DEBUGGING
984
985 if (DEBUG_L_TEST || debug_initialization) {
986 dSAVE_ERRNO;
987 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
988 RESTORE_ERRNO;
989 }
990
991# endif
992
993 return NULL;
994 }
995
996# ifdef DEBUGGING
997
998 if (DEBUG_Lv_TEST || debug_initialization) {
999 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1000 }
1001
1002# endif
1003
1004 /* If we weren't in a thread safe locale, set so that newlocale() below
1005 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1006 The latter is defensive coding, so that, even if there is some bug, we
1007 will never end up trying to modify either of these, as if passed to
1008 newlocale(), they can be. */
1009 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1010 old_obj = (locale_t) 0;
1011 }
1012
e9bc6d6b
KW
1013 /* Ready to create a new locale by modification of the exising one */
1014 new_obj = newlocale(mask, locale, old_obj);
1015
1016 if (! new_obj) {
1017 dSAVE_ERRNO;
1018
1019# ifdef DEBUGGING
1020
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1023 }
1024
1025# endif
1026
1027 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1028
1029# ifdef DEBUGGING
1030
1031 if (DEBUG_L_TEST || debug_initialization) {
1032 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1033 }
1034
1035# endif
1036
1037 }
1038 RESTORE_ERRNO;
1039 return NULL;
1040 }
1041
1042# ifdef DEBUGGING
1043
1044 if (DEBUG_Lv_TEST || debug_initialization) {
d15b6e8b 1045 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p; should have freed %p\n", __FILE__, __LINE__, new_obj, old_obj);
e9bc6d6b
KW
1046 }
1047
1048# endif
1049
1050 /* And switch into it */
1051 if (! uselocale(new_obj)) {
1052 dSAVE_ERRNO;
1053
1054# ifdef DEBUGGING
1055
1056 if (DEBUG_L_TEST || debug_initialization) {
1057 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1058 }
1059
1060# endif
1061
1062 if (! uselocale(old_obj)) {
1063
1064# ifdef DEBUGGING
1065
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1068 }
1069
1070# endif
1071
1072 }
1073 freelocale(new_obj);
1074 RESTORE_ERRNO;
1075 return NULL;
1076 }
1077
1078# ifdef DEBUGGING
1079
1080 if (DEBUG_Lv_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1082 }
1083
1084# endif
1085
1086 /* We are done, except for updating our records (if the system doesn't keep
1087 * them) and in the case of locale "", we don't actually know what the
1088 * locale that got switched to is, as it came from the environment. So
1089 * have to find it */
1090
1091# ifdef HAS_QUERYLOCALE
1092
1093 if (strEQ(locale, "")) {
1094 locale = querylocale(mask, new_obj);
1095 }
1096
1097# else
1098
1099 /* Here, 'locale' is the return value */
1100
1101 /* Without querylocale(), we have to update our records */
1102
1103 if (category == LC_ALL) {
1104 unsigned int i;
1105
1106 /* For LC_ALL, we change all individual categories to correspond */
1107 /* PL_curlocales is a parallel array, so has same
1108 * length as 'categories' */
1109 for (i = 0; i <= LC_ALL_INDEX; i++) {
1110 Safefree(PL_curlocales[i]);
1111 PL_curlocales[i] = savepv(locale);
1112 }
1113 }
1114 else {
1115
1116 /* For a single category, if it's not the same as the one in LC_ALL, we
1117 * nullify LC_ALL */
1118
1119 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1120 Safefree(PL_curlocales[LC_ALL_INDEX]);
1121 PL_curlocales[LC_ALL_INDEX] = NULL;
1122 }
1123
1124 /* Then update the category's record */
1125 Safefree(PL_curlocales[index]);
1126 PL_curlocales[index] = savepv(locale);
1127 }
1128
1129# endif
1130
1131 return locale;
1132}
1133
1134#endif /* USE_POSIX_2008_LOCALE */
1135
1136#if 0 /* Code that was to emulate thread-safe locales on platforms that
1137 didn't natively support them */
1138
1139/* The way this would work is that we would keep a per-thread list of the
1140 * correct locale for that thread. Any operation that was locale-sensitive
1141 * would have to be changed so that it would look like this:
1142 *
1143 * LOCALE_LOCK;
1144 * setlocale to the correct locale for this operation
1145 * do operation
1146 * LOCALE_UNLOCK
1147 *
1148 * This leaves the global locale in the most recently used operation's, but it
1149 * was locked long enough to get the result. If that result is static, it
1150 * needs to be copied before the unlock.
1151 *
1152 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1153 * the setup, but are no-ops when not needed, and similarly,
1154 * END_LOCALE_DEPENDENT_OP for the tear-down
1155 *
1156 * But every call to a locale-sensitive function would have to be changed, and
1157 * if a module didn't cooperate by using the mutex, things would break.
1158 *
1159 * This code was abandoned before being completed or tested, and is left as-is
1160*/
1161
1162# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1163# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1164
1165STATIC char *
1166S_locking_setlocale(pTHX_
1167 const int category,
1168 const char * locale,
1169 int index,
1170 const bool is_index_valid
1171 )
1172{
1173 /* This function kind of performs a setlocale() on just the current thread;
1174 * thus it is kind of thread-safe. It does this by keeping a thread-level
1175 * array of the current locales for each category. Every time a locale is
1176 * switched to, it does the switch globally, but updates the thread's
1177 * array. A query as to what the current locale is just returns the
1178 * appropriate element from the array, and doesn't actually call the system
1179 * setlocale(). The saving into the array is done in an uninterruptible
1180 * section of code, so is unaffected by whatever any other threads might be
1181 * doing.
1182 *
1183 * All locale-sensitive operations must work by first starting a critical
1184 * section, then switching to the thread's locale as kept by this function,
1185 * and then doing the operation, then ending the critical section. Thus,
1186 * each gets done in the appropriate locale. simulating thread-safety.
1187 *
1188 * This function takes the same parameters, 'category' and 'locale', that
1189 * the regular setlocale() function does, but it also takes two additional
1190 * ones. This is because as described earlier. If we know on input the
1191 * index corresponding to the category into the array where we store the
1192 * current locales, we don't have to calculate it. If the caller knows at
1193 * compile time what the index is, it it can pass it, setting
1194 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1195 *
1196 */
1197
1198 /* If the input index might be incorrect, calculate the correct one */
1199 if (! is_index_valid) {
1200 unsigned int i;
1201
1202 if (DEBUG_Lv_TEST || debug_initialization) {
1203 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1204 }
1205
1206 for (i = 0; i <= LC_ALL_INDEX; i++) {
1207 if (category == categories[i]) {
1208 index = i;
1209 goto found_index;
1210 }
1211 }
1212
1213 /* Here, we don't know about this category, so can't handle it.
1214 * XXX best we can do is to unsafely set this
1215 * XXX warning */
1216
1217 return my_setlocale(category, locale);
1218
1219 found_index: ;
1220
1221 if (DEBUG_Lv_TEST || debug_initialization) {
1222 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1223 }
1224 }
1225
1226 /* For a query, just return what's in our records */
1227 if (new_locale == NULL) {
1228 return curlocales[index];
1229 }
1230
1231
1232 /* Otherwise, we need to do the switch, and save the result, all in a
1233 * critical section */
1234
1235 Safefree(curlocales[[index]]);
1236
1237 /* It might be that this is called from an already-locked section of code.
1238 * We would have to detect and skip the LOCK/UNLOCK if so */
1239 LOCALE_LOCK;
1240
1241 curlocales[index] = savepv(my_setlocale(category, new_locale));
1242
1243 if (strEQ(new_locale, "")) {
1244
1245#ifdef LC_ALL
1246
1247 /* The locale values come from the environment, and may not all be the
1248 * same, so for LC_ALL, we have to update all the others, while the
1249 * mutex is still locked */
1250
1251 if (category == LC_ALL) {
1252 unsigned int i;
1253 for (i = 0; i < LC_ALL_INDEX) {
1254 curlocales[i] = my_setlocale(categories[i], NULL);
1255 }
1256 }
1257 }
1258
1259#endif
1260
1261 LOCALE_UNLOCK;
1262
1263 return curlocales[index];
1264}
1265
1266#endif
d36adde0 1267#ifdef USE_LOCALE
837ce802 1268
a4f00dcc 1269STATIC void
86799d2d 1270S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1271{
86799d2d
KW
1272 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1273 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1274
86799d2d
KW
1275#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1276 || defined(HAS_NL_LANGINFO))
98994639 1277
87f8e8e7 1278 const char * radix = (use_locale)
4e6826bf 1279 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1280 /* FALSE => already in dest locale */
1281 : ".";
2213a3be 1282
87f8e8e7 1283 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1284
87f8e8e7
KW
1285 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1286 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1287 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1288 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1289 && _is_cur_LC_category_utf8(LC_NUMERIC))
1290 {
1291 SvUTF8_on(PL_numeric_radix_sv);
1292 }
86799d2d
KW
1293
1294# ifdef DEBUGGING
7d4bcc4a 1295
2fcc0ca9
KW
1296 if (DEBUG_L_TEST || debug_initialization) {
1297 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1298 SvPVX(PL_numeric_radix_sv),
1299 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1300 }
69014004 1301
86799d2d 1302# endif
d36adde0
KW
1303#else
1304
1305 PERL_UNUSED_ARG(use_locale);
1306
86799d2d 1307#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1308
98994639
HS
1309}
1310
398eeea9
KW
1311STATIC void
1312S_new_numeric(pTHX_ const char *newnum)
98994639 1313{
7d4bcc4a
KW
1314
1315#ifndef USE_LOCALE_NUMERIC
1316
1317 PERL_UNUSED_ARG(newnum);
1318
1319#else
0d071d52 1320
291a84fb 1321 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1322 * core Perl this and that 'newnum' is the name of the new locale.
1323 * It installs this locale as the current underlying default.
1324 *
1325 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1326 * set_numeric_underlying() and set_numeric_standard() functions, which
1327 * should probably not be called directly, but only via macros like
0d071d52
KW
1328 * SET_NUMERIC_STANDARD() in perl.h.
1329 *
1330 * The toggling is necessary mainly so that a non-dot radix decimal point
1331 * character can be output, while allowing internal calculations to use a
1332 * dot.
1333 *
1334 * This sets several interpreter-level variables:
bb304765 1335 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1336 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1337 * that the current locale is the program's underlying
1338 * locale
1339 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1340 * that the current locale is the C locale or
1341 * indistinguishable from the C locale. If non-zero, it
1342 * is in C; if > 1, it means it may not be toggled away
7738054c 1343 * from C.
4c68b815
KW
1344 * PL_numeric_underlying_is_standard A bool kept by this function
1345 * indicating that the underlying locale and the standard
1346 * C locale are indistinguishable for the purposes of
1347 * LC_NUMERIC. This happens when both of the above two
1348 * variables are true at the same time. (Toggling is a
1349 * no-op under these circumstances.) This variable is
1350 * used to avoid having to recalculate.
398eeea9 1351 */
0d071d52 1352
b03f34cf 1353 char *save_newnum;
98994639
HS
1354
1355 if (! newnum) {
43c5f42d
NC
1356 Safefree(PL_numeric_name);
1357 PL_numeric_name = NULL;
98994639 1358 PL_numeric_standard = TRUE;
892e6465 1359 PL_numeric_underlying = TRUE;
4c68b815 1360 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1361 return;
1362 }
1363
b03f34cf 1364 save_newnum = stdize_locale(savepv(newnum));
892e6465 1365 PL_numeric_underlying = TRUE;
4c68b815
KW
1366 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1367
69c5e0db
KW
1368#ifndef TS_W32_BROKEN_LOCALECONV
1369
4c68b815 1370 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1371 * them. But on broken Windows systems calling my_nl_langinfo() for
1372 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1373 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1374 if (! PL_numeric_standard) {
4e6826bf 1375 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1376 FALSE /* Don't toggle locale */ ))
4e6826bf 1377 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1378 }
abe1abcf 1379
69c5e0db
KW
1380#endif
1381
4c68b815 1382 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1383 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1384 Safefree(PL_numeric_name);
b03f34cf 1385 PL_numeric_name = save_newnum;
b03f34cf 1386 }
abe1abcf
KW
1387 else {
1388 Safefree(save_newnum);
1389 }
4c28b29c 1390
4c68b815
KW
1391 PL_numeric_underlying_is_standard = PL_numeric_standard;
1392
7e5377f7 1393# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1394
1395 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1396 PL_numeric_name,
1397 PL_underlying_numeric_obj);
1398
1399#endif
1400
4c68b815
KW
1401 if (DEBUG_L_TEST || debug_initialization) {
1402 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1403 }
1404
4c28b29c
KW
1405 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1406 * have to worry about the radix being a non-dot. (Core operations that
1407 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1408 if (PL_numeric_standard) {
1409 set_numeric_radix(0);
1410 }
1411 else {
1412 set_numeric_standard();
1413 }
4c28b29c 1414
98994639 1415#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1416
98994639
HS
1417}
1418
1419void
1420Perl_set_numeric_standard(pTHX)
1421{
7d4bcc4a 1422
98994639 1423#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1424
28c1bf33
KW
1425 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1426 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1427 * macro avoids calling this routine if toggling isn't necessary according
1428 * to our records (which could be wrong if some XS code has changed the
1429 * locale behind our back) */
0d071d52 1430
7d4bcc4a
KW
1431# ifdef DEBUGGING
1432
2fcc0ca9
KW
1433 if (DEBUG_L_TEST || debug_initialization) {
1434 PerlIO_printf(Perl_debug_log,
7bb8f702 1435 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1436 }
98994639 1437
7d4bcc4a 1438# endif
7bb8f702
KW
1439
1440 do_setlocale_c(LC_NUMERIC, "C");
1441 PL_numeric_standard = TRUE;
1442 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1443 set_numeric_radix(0);
1444
98994639 1445#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1446
98994639
HS
1447}
1448
1449void
5792c642 1450Perl_set_numeric_underlying(pTHX)
98994639 1451{
7d4bcc4a 1452
98994639 1453#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1454
28c1bf33 1455 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1456 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1457 * instead of calling this directly. The macro avoids calling this routine
1458 * if toggling isn't necessary according to our records (which could be
1459 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1460
7d4bcc4a
KW
1461# ifdef DEBUGGING
1462
2fcc0ca9
KW
1463 if (DEBUG_L_TEST || debug_initialization) {
1464 PerlIO_printf(Perl_debug_log,
7bb8f702 1465 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1466 PL_numeric_name);
1467 }
98994639 1468
7d4bcc4a 1469# endif
7bb8f702
KW
1470
1471 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1472 PL_numeric_standard = PL_numeric_underlying_is_standard;
1473 PL_numeric_underlying = TRUE;
1474 set_numeric_radix(! PL_numeric_standard);
1475
98994639 1476#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1477
98994639
HS
1478}
1479
1480/*
1481 * Set up for a new ctype locale.
1482 */
a4f00dcc
KW
1483STATIC void
1484S_new_ctype(pTHX_ const char *newctype)
98994639 1485{
7d4bcc4a
KW
1486
1487#ifndef USE_LOCALE_CTYPE
1488
7d4bcc4a
KW
1489 PERL_UNUSED_ARG(newctype);
1490 PERL_UNUSED_CONTEXT;
1491
1492#else
0d071d52 1493
291a84fb 1494 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1495 * core Perl this and that 'newctype' is the name of the new locale.
1496 *
1497 * This function sets up the folding arrays for all 256 bytes, assuming
1498 * that tofold() is tolc() since fold case is not a concept in POSIX,
1499 *
1500 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1501 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1502 * this function should be called directly only from this file and from
0d071d52
KW
1503 * POSIX::setlocale() */
1504
27da23d5 1505 dVAR;
013f4e03 1506 unsigned int i;
98994639 1507
8b7358b9
KW
1508 /* Don't check for problems if we are suppressing the warnings */
1509 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
30d8090d 1510 bool maybe_utf8_turkic = FALSE;
8b7358b9 1511
7918f24d
NC
1512 PERL_ARGS_ASSERT_NEW_CTYPE;
1513
215c5139
KW
1514 /* We will replace any bad locale warning with 1) nothing if the new one is
1515 * ok; or 2) a new warning for the bad new locale */
1516 if (PL_warn_locale) {
1517 SvREFCNT_dec_NN(PL_warn_locale);
1518 PL_warn_locale = NULL;
1519 }
1520
c1284011 1521 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1522
1523 /* A UTF-8 locale gets standard rules. But note that code still has to
1524 * handle this specially because of the three problematic code points */
1525 if (PL_in_utf8_CTYPE_locale) {
1526 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
30d8090d
KW
1527
1528 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1529 * Turkic. Make sure these two are the only anomalies. (We don't use
1530 * towupper and towlower because they aren't in C89.) */
1531 if (toupper('i') == 'i' && tolower('I') == 'I') {
1532 check_for_problems = TRUE;
1533 maybe_utf8_turkic = TRUE;
1534 }
31f05a37 1535 }
8b7358b9
KW
1536
1537 /* We don't populate the other lists if a UTF-8 locale, but do check that
1538 * everything works as expected, unless checking turned off */
1539 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1540 /* Assume enough space for every character being bad. 4 spaces each
1541 * for the 94 printable characters that are output like "'x' "; and 5
1542 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1543 * NUL */
8b7358b9 1544 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1545 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1546 to start */
1547 unsigned int bad_count = 0; /* Count of bad characters */
1548
baa60164 1549 for (i = 0; i < 256; i++) {
8b7358b9 1550 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1551 if (isupper(i))
1552 PL_fold_locale[i] = (U8) tolower(i);
1553 else if (islower(i))
1554 PL_fold_locale[i] = (U8) toupper(i);
1555 else
1556 PL_fold_locale[i] = (U8) i;
8b7358b9 1557 }
8c6180a9
KW
1558
1559 /* If checking for locale problems, see if the native ASCII-range
1560 * printables plus \n and \t are in their expected categories in
1561 * the new locale. If not, this could mean big trouble, upending
1562 * Perl's and most programs' assumptions, like having a
1563 * metacharacter with special meaning become a \w. Fortunately,
1564 * it's very rare to find locales that aren't supersets of ASCII
1565 * nowadays. It isn't a problem for most controls to be changed
1566 * into something else; we check only \n and \t, though perhaps \r
1567 * could be an issue as well. */
7d4bcc4a 1568 if ( check_for_problems
8c6180a9
KW
1569 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1570 {
8b7358b9 1571 bool is_bad = FALSE;
6726b4f4 1572 char name[4] = { '\0' };
8b7358b9
KW
1573
1574 /* Convert the name into a string */
6726b4f4 1575 if (isGRAPH_A(i)) {
8b7358b9
KW
1576 name[0] = i;
1577 name[1] = '\0';
1578 }
1579 else if (i == '\n') {
6726b4f4
KW
1580 my_strlcpy(name, "\\n", sizeof(name));
1581 }
1582 else if (i == '\t') {
1583 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1584 }
1585 else {
6726b4f4
KW
1586 assert(i == ' ');
1587 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1588 }
1589
1590 /* Check each possibe class */
1591 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1592 is_bad = TRUE;
1593 DEBUG_L(PerlIO_printf(Perl_debug_log,
1594 "isalnum('%s') unexpectedly is %d\n",
1595 name, cBOOL(isalnum(i))));
1596 }
1597 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1598 is_bad = TRUE;
1599 DEBUG_L(PerlIO_printf(Perl_debug_log,
1600 "isalpha('%s') unexpectedly is %d\n",
1601 name, cBOOL(isalpha(i))));
1602 }
1603 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1604 is_bad = TRUE;
1605 DEBUG_L(PerlIO_printf(Perl_debug_log,
1606 "isdigit('%s') unexpectedly is %d\n",
1607 name, cBOOL(isdigit(i))));
1608 }
1609 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1610 is_bad = TRUE;
1611 DEBUG_L(PerlIO_printf(Perl_debug_log,
1612 "isgraph('%s') unexpectedly is %d\n",
1613 name, cBOOL(isgraph(i))));
1614 }
1615 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1616 is_bad = TRUE;
1617 DEBUG_L(PerlIO_printf(Perl_debug_log,
1618 "islower('%s') unexpectedly is %d\n",
1619 name, cBOOL(islower(i))));
1620 }
1621 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1622 is_bad = TRUE;
1623 DEBUG_L(PerlIO_printf(Perl_debug_log,
1624 "isprint('%s') unexpectedly is %d\n",
1625 name, cBOOL(isprint(i))));
1626 }
1627 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1628 is_bad = TRUE;
1629 DEBUG_L(PerlIO_printf(Perl_debug_log,
1630 "ispunct('%s') unexpectedly is %d\n",
1631 name, cBOOL(ispunct(i))));
1632 }
1633 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1634 is_bad = TRUE;
1635 DEBUG_L(PerlIO_printf(Perl_debug_log,
1636 "isspace('%s') unexpectedly is %d\n",
1637 name, cBOOL(isspace(i))));
1638 }
1639 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1640 is_bad = TRUE;
1641 DEBUG_L(PerlIO_printf(Perl_debug_log,
1642 "isupper('%s') unexpectedly is %d\n",
1643 name, cBOOL(isupper(i))));
1644 }
1645 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1646 is_bad = TRUE;
1647 DEBUG_L(PerlIO_printf(Perl_debug_log,
1648 "isxdigit('%s') unexpectedly is %d\n",
1649 name, cBOOL(isxdigit(i))));
1650 }
1651 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1652 is_bad = TRUE;
1653 DEBUG_L(PerlIO_printf(Perl_debug_log,
1654 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1655 name, tolower(i), (int) toLOWER_A(i)));
1656 }
1657 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1658 is_bad = TRUE;
1659 DEBUG_L(PerlIO_printf(Perl_debug_log,
1660 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1661 name, toupper(i), (int) toUPPER_A(i)));
1662 }
1663 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1664 is_bad = TRUE;
1665 DEBUG_L(PerlIO_printf(Perl_debug_log,
1666 "'\\n' (=%02X) is not a control\n", (int) i));
1667 }
1668
1669 /* Add to the list; Separate multiple entries with a blank */
1670 if (is_bad) {
1671 if (bad_count) {
1672 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1673 }
8b7358b9
KW
1674 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1675 bad_count++;
8c6180a9
KW
1676 }
1677 }
1678 }
1679
30d8090d
KW
1680 if (bad_count == 2 && maybe_utf8_turkic) {
1681 bad_count = 0;
1682 *bad_chars_list = '\0';
1683 PL_fold_locale['I'] = 'I';
1684 PL_fold_locale['i'] = 'i';
1685 PL_in_utf8_turkic_locale = TRUE;
1686 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1687 __FILE__, __LINE__, newctype));
1688 }
1689 else {
b8df1494 1690 PL_in_utf8_turkic_locale = FALSE;
30d8090d 1691 }
b8df1494 1692
7d4bcc4a
KW
1693# ifdef MB_CUR_MAX
1694
8c6180a9 1695 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1696 * this locale requires more than one byte, there are going to be
8c6180a9 1697 * problems. */
9c8a6dc2
KW
1698 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1699 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1700 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1701
8b7358b9
KW
1702 if ( check_for_problems && MB_CUR_MAX > 1
1703 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1704
1705 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1706 * locale. Just assume that the implementation for them (plus
1707 * for POSIX) is correct and the > 1 value is spurious. (Since
1708 * these are specially handled to never be considered UTF-8
1709 * locales, as long as this is the only problem, everything
1710 * should work fine */
1711 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1712 {
8c6180a9
KW
1713 multi_byte_locale = TRUE;
1714 }
7d4bcc4a
KW
1715
1716# endif
8c6180a9 1717
30d8090d
KW
1718 /* If we found problems and we want them output, do so */
1719 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1720 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1721 {
8b7358b9
KW
1722 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1723 PL_warn_locale = Perl_newSVpvf(aTHX_
1724 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1725 " which have\nunexpected meanings: %s\nThe Perl program"
1726 " will use the expected meanings",
8b7358b9 1727 newctype, bad_chars_list);
8b7358b9
KW
1728 }
1729 else {
1730 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1731 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1732 newctype,
8c6180a9
KW
1733 (multi_byte_locale)
1734 ? " Some characters in it are not recognized by"
1735 " Perl."
1736 : "",
1737 (bad_count)
1738 ? "\nThe following characters (and maybe others)"
1739 " may not have the same meaning as the Perl"
1740 " program expects:\n"
1741 : "",
1742 (bad_count)
1743 ? bad_chars_list
1744 : ""
1745 );
8b7358b9
KW
1746 }
1747
b119c2be
KW
1748# ifdef HAS_NL_LANGINFO
1749
1750 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1751 /* parameter FALSE is a don't care here */
4e6826bf 1752 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1753
1754# endif
1755
8b7358b9
KW
1756 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1757
cc9eaeb0 1758 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1759 * output the message now. If not in that scope, we save the
1760 * message to be output at the first operation using this locale,
1761 * if that actually happens. Most programs don't use locales, so
1762 * they are immune to bad ones. */
cc9eaeb0 1763 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1764
780fcc9f
KW
1765 /* The '0' below suppresses a bogus gcc compiler warning */
1766 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1767
bddebb56
KW
1768 if (IN_LC(LC_CTYPE)) {
1769 SvREFCNT_dec_NN(PL_warn_locale);
1770 PL_warn_locale = NULL;
1771 }
780fcc9f 1772 }
baa60164 1773 }
31f05a37 1774 }
98994639
HS
1775
1776#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1777
98994639
HS
1778}
1779
98994639 1780void
2726666d
KW
1781Perl__warn_problematic_locale()
1782{
2726666d
KW
1783
1784#ifdef USE_LOCALE_CTYPE
1785
5f04a188
KW
1786 dTHX;
1787
1788 /* Internal-to-core function that outputs the message in PL_warn_locale,
1789 * and then NULLS it. Should be called only through the macro
1790 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1791
2726666d 1792 if (PL_warn_locale) {
2726666d
KW
1793 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1794 SvPVX(PL_warn_locale),
1795 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1796 SvREFCNT_dec_NN(PL_warn_locale);
1797 PL_warn_locale = NULL;
1798 }
1799
1800#endif
1801
1802}
1803
a4f00dcc
KW
1804STATIC void
1805S_new_collate(pTHX_ const char *newcoll)
98994639 1806{
7d4bcc4a
KW
1807
1808#ifndef USE_LOCALE_COLLATE
1809
1810 PERL_UNUSED_ARG(newcoll);
1811 PERL_UNUSED_CONTEXT;
1812
1813#else
0d071d52 1814
291a84fb 1815 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1816 * core Perl this and that 'newcoll' is the name of the new locale.
1817 *
d35fca5f
KW
1818 * The design of locale collation is that every locale change is given an
1819 * index 'PL_collation_ix'. The first time a string particpates in an
1820 * operation that requires collation while locale collation is active, it
1821 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1822 * magic includes the collation index, and the transformation of the string
1823 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1824 * instead of the string itself. If a string changes, the magic is
1825 * cleared. The next time the locale changes, the index is incremented,
1826 * and so we know during a comparison that the transformation is not
1827 * necessarily still valid, and so is recomputed. Note that if the locale
1828 * changes enough times, the index could wrap (a U32), and it is possible
1829 * that a transformation would improperly be considered valid, leading to
1830 * an unlikely bug */
0d071d52 1831
98994639
HS
1832 if (! newcoll) {
1833 if (PL_collation_name) {
1834 ++PL_collation_ix;
1835 Safefree(PL_collation_name);
1836 PL_collation_name = NULL;
1837 }
1838 PL_collation_standard = TRUE;
00bf60ca 1839 is_standard_collation:
98994639
HS
1840 PL_collxfrm_base = 0;
1841 PL_collxfrm_mult = 2;
165a1c52 1842 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1843 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1844 PL_strxfrm_max_cp = 0;
98994639
HS
1845 return;
1846 }
1847
d35fca5f 1848 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1849 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1850 ++PL_collation_ix;
1851 Safefree(PL_collation_name);
1852 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1853 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1854 if (PL_collation_standard) {
1855 goto is_standard_collation;
1856 }
98994639 1857
165a1c52 1858 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1859 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1860 PL_strxfrm_max_cp = 0;
165a1c52 1861
59c018b9
KW
1862 /* A locale collation definition includes primary, secondary, tertiary,
1863 * etc. weights for each character. To sort, the primary weights are
1864 * used, and only if they compare equal, then the secondary weights are
1865 * used, and only if they compare equal, then the tertiary, etc.
1866 *
1867 * strxfrm() works by taking the input string, say ABC, and creating an
1868 * output transformed string consisting of first the primary weights,
1869 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1870 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1871 * may not have weights at every level. In our example, let's say B
1872 * doesn't have a tertiary weight, and A doesn't have a secondary
1873 * weight. The constructed string is then going to be
1874 * A¹B¹C¹ B²C² A³C³ ....
1875 * This has the desired effect that strcmp() will look at the secondary
1876 * or tertiary weights only if the strings compare equal at all higher
1877 * priority weights. The spaces shown here, like in
c342d20e 1878 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1879 * are not just for readability. In the general case, these must
1880 * actually be bytes, which we will call here 'separator weights'; and
1881 * they must be smaller than any other weight value, but since these
1882 * are C strings, only the terminating one can be a NUL (some
1883 * implementations may include a non-NUL separator weight just before
1884 * the NUL). Implementations tend to reserve 01 for the separator
1885 * weights. They are needed so that a shorter string's secondary
1886 * weights won't be misconstrued as primary weights of a longer string,
1887 * etc. By making them smaller than any other weight, the shorter
1888 * string will sort first. (Actually, if all secondary weights are
1889 * smaller than all primary ones, there is no need for a separator
1890 * weight between those two levels, etc.)
1891 *
1892 * The length of the transformed string is roughly a linear function of
1893 * the input string. It's not exactly linear because some characters
1894 * don't have weights at all levels. When we call strxfrm() we have to
1895 * allocate some memory to hold the transformed string. The
1896 * calculations below try to find coefficients 'm' and 'b' for this
1897 * locale so that m*x + b equals how much space we need, given the size
1898 * of the input string in 'x'. If we calculate too small, we increase
1899 * the size as needed, and call strxfrm() again, but it is better to
1900 * get it right the first time to avoid wasted expensive string
1901 * transformations. */
1902
98994639 1903 {
79f120c8
KW
1904 /* We use the string below to find how long the tranformation of it
1905 * is. Almost all locales are supersets of ASCII, or at least the
1906 * ASCII letters. We use all of them, half upper half lower,
1907 * because if we used fewer, we might hit just the ones that are
1908 * outliers in a particular locale. Most of the strings being
1909 * collated will contain a preponderance of letters, and even if
1910 * they are above-ASCII, they are likely to have the same number of
1911 * weight levels as the ASCII ones. It turns out that digits tend
1912 * to have fewer levels, and some punctuation has more, but those
1913 * are relatively sparse in text, and khw believes this gives a
1914 * reasonable result, but it could be changed if experience so
1915 * dictates. */
1916 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1917 char * x_longer; /* Transformed 'longer' */
1918 Size_t x_len_longer; /* Length of 'x_longer' */
1919
1920 char * x_shorter; /* We also transform a substring of 'longer' */
1921 Size_t x_len_shorter;
1922
a4a439fb 1923 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1924 * are interested only in its length). It is used because it has
1925 * the intelligence to handle all cases, but to work, it needs some
1926 * values of 'm' and 'b' to get it started. For the purposes of
1927 * this calculation we use a very conservative estimate of 'm' and
1928 * 'b'. This assumes a weight can be multiple bytes, enough to
1929 * hold any UV on the platform, and there are 5 levels, 4 weight
1930 * bytes, and a trailing NUL. */
1931 PL_collxfrm_base = 5;
1932 PL_collxfrm_mult = 5 * sizeof(UV);
1933
1934 /* Find out how long the transformation really is */
a4a439fb
KW
1935 x_longer = _mem_collxfrm(longer,
1936 sizeof(longer) - 1,
1937 &x_len_longer,
1938
1939 /* We avoid converting to UTF-8 in the
1940 * called function by telling it the
1941 * string is in UTF-8 if the locale is a
1942 * UTF-8 one. Since the string passed
1943 * here is invariant under UTF-8, we can
1944 * claim it's UTF-8 even though it isn't.
1945 * */
1946 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1947 Safefree(x_longer);
1948
1949 /* Find out how long the transformation of a substring of 'longer'
1950 * is. Together the lengths of these transformations are
1951 * sufficient to calculate 'm' and 'b'. The substring is all of
1952 * 'longer' except the first character. This minimizes the chances
1953 * of being swayed by outliers */
a4a439fb 1954 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1955 sizeof(longer) - 2,
a4a439fb
KW
1956 &x_len_shorter,
1957 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1958 Safefree(x_shorter);
1959
1960 /* If the results are nonsensical for this simple test, the whole
1961 * locale definition is suspect. Mark it so that locale collation
1962 * is not active at all for it. XXX Should we warn? */
1963 if ( x_len_shorter == 0
1964 || x_len_longer == 0
1965 || x_len_shorter >= x_len_longer)
1966 {
1967 PL_collxfrm_mult = 0;
1968 PL_collxfrm_base = 0;
1969 }
1970 else {
1971 SSize_t base; /* Temporary */
1972
1973 /* We have both: m * strlen(longer) + b = x_len_longer
1974 * m * strlen(shorter) + b = x_len_shorter;
1975 * subtracting yields:
1976 * m * (strlen(longer) - strlen(shorter))
1977 * = x_len_longer - x_len_shorter
1978 * But we have set things up so that 'shorter' is 1 byte smaller
1979 * than 'longer'. Hence:
1980 * m = x_len_longer - x_len_shorter
1981 *
1982 * But if something went wrong, make sure the multiplier is at
1983 * least 1.
1984 */
1985 if (x_len_longer > x_len_shorter) {
1986 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1987 }
1988 else {
1989 PL_collxfrm_mult = 1;
1990 }
1991
1992 /* mx + b = len
1993 * so: b = len - mx
1994 * but in case something has gone wrong, make sure it is
1995 * non-negative */
1996 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1997 if (base < 0) {
1998 base = 0;
1999 }
2000
2001 /* Add 1 for the trailing NUL */
2002 PL_collxfrm_base = base + 1;
2003 }
58eebef2 2004
7d4bcc4a
KW
2005# ifdef DEBUGGING
2006
58eebef2
KW
2007 if (DEBUG_L_TEST || debug_initialization) {
2008 PerlIO_printf(Perl_debug_log,
b07929e4
KW
2009 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2010 "x_len_longer=%zu,"
2011 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
2012 __FILE__, __LINE__,
2013 PL_in_utf8_COLLATE_locale,
2014 x_len_shorter, x_len_longer,
2015 PL_collxfrm_mult, PL_collxfrm_base);
2016 }
7d4bcc4a
KW
2017# endif
2018
98994639
HS
2019 }
2020 }
2021
2022#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 2023
98994639
HS
2024}
2025
d36adde0
KW
2026#endif
2027
d2b24094 2028#ifdef WIN32
b8cc575c 2029
a4f00dcc 2030STATIC char *
b8cc575c 2031S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2032{
2033 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2034 * difference between the two unless the input locale is "", which normally
2035 * means on Windows to get the machine default, which is set via the
2036 * computer's "Regional and Language Options" (or its current equivalent).
2037 * In POSIX, it instead means to find the locale from the user's
2038 * environment. This routine changes the Windows behavior to first look in
2039 * the environment, and, if anything is found, use that instead of going to
2040 * the machine default. If there is no environment override, the machine
2041 * default is used, by calling the real setlocale() with "".
2042 *
2043 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2044 * use the particular category's variable if set; otherwise to use the LANG
2045 * variable. */
b385bb4d 2046
175c4cf9 2047 bool override_LC_ALL = FALSE;
89f7b9aa 2048 char * result;
e5f10d49 2049 unsigned int i;
89f7b9aa 2050
b385bb4d 2051 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2052
2053# ifdef LC_ALL
2054
b385bb4d
KW
2055 locale = PerlEnv_getenv("LC_ALL");
2056 if (! locale) {
e5f10d49
KW
2057 if (category == LC_ALL) {
2058 override_LC_ALL = TRUE;
2059 }
2060 else {
7d4bcc4a
KW
2061
2062# endif
7d4bcc4a 2063
e5f10d49
KW
2064 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2065 if (category == categories[i]) {
2066 locale = PerlEnv_getenv(category_names[i]);
2067 goto found_locale;
2068 }
2069 }
7d4bcc4a 2070
b385bb4d 2071 locale = PerlEnv_getenv("LANG");
481465ea 2072 if (! locale) {
b385bb4d
KW
2073 locale = "";
2074 }
e5f10d49
KW
2075
2076 found_locale: ;
7d4bcc4a
KW
2077
2078# ifdef LC_ALL
2079
e5f10d49 2080 }
b385bb4d 2081 }
7d4bcc4a
KW
2082
2083# endif
2084
b385bb4d
KW
2085 }
2086
89f7b9aa 2087 result = setlocale(category, locale);
48015184
KW
2088 DEBUG_L(STMT_START {
2089 dSAVE_ERRNO;
2090 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2091 setlocale_debug_string(category, locale, result));
2092 RESTORE_ERRNO;
2093 } STMT_END);
89f7b9aa 2094
481465ea 2095 if (! override_LC_ALL) {
89f7b9aa
KW
2096 return result;
2097 }
2098
dfd77d7a 2099 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2100 * LANG variable or the system default if there is no LANG. But these have
2101 * lower priority than the other LC_foo variables, so override it for each
2102 * one that is set. (If they are set to "", it means to use the same thing
2103 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2104
948523db 2105 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2106 result = PerlEnv_getenv(category_names[i]);
2107 if (result && strNE(result, "")) {
2108 setlocale(categories[i], result);
2109 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2110 __FILE__, __LINE__,
2111 setlocale_debug_string(categories[i], result, "not captured")));
2112 }
89f7b9aa 2113 }
7d4bcc4a 2114
bbc98134 2115 result = setlocale(LC_ALL, NULL);
48015184
KW
2116 DEBUG_L(STMT_START {
2117 dSAVE_ERRNO;
2118 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2119 __FILE__, __LINE__,
48015184
KW
2120 setlocale_debug_string(LC_ALL, NULL, result));
2121 RESTORE_ERRNO;
2122 } STMT_END);
89f7b9aa 2123
bbc98134 2124 return result;
b385bb4d
KW
2125}
2126
2127#endif
2128
9aac5db8
KW
2129/*
2130
2131=head1 Locale-related functions and macros
2132
2133=for apidoc Perl_setlocale
2134
2135This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2136taking the same parameters, and returning the same information, except that it
9487427b
KW
2137returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2138instead return C<C> if the underlying locale has a non-dot decimal point
2139character, or a non-empty thousands separator for displaying floating point
2140numbers. This is because perl keeps that locale category such that it has a
2141dot and empty separator, changing the locale briefly during the operations
2142where the underlying one is required. C<Perl_setlocale> knows about this, and
2143compensates; regular C<setlocale> doesn't.
9aac5db8 2144
e9bc6d6b 2145Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2146declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2147C<const> (presumably because its API was specified long ago, and can't be
2148updated; it is illegal to change the information C<setlocale> returns; doing
2149so leads to segfaults.)
9aac5db8 2150
e9bc6d6b
KW
2151Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2152C<setlocale> can be completely ineffective on some platforms under some
2153configurations.
2154
9aac5db8 2155C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2156where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2157the system C<setlocale()> is ineffective, returning the wrong information, and
2158failing to actually change the locale. C<Perl_setlocale>, however works
2159properly in all circumstances.
9aac5db8
KW
2160
2161The return points to a per-thread static buffer, which is overwritten the next
2162time C<Perl_setlocale> is called from the same thread.
2163
2164=cut
2165
2166*/
2167
2168const char *
2169Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2170{
2171 /* This wraps POSIX::setlocale() */
2172
d36adde0
KW
2173#ifdef NO_LOCALE
2174
2175 PERL_UNUSED_ARG(category);
2176 PERL_UNUSED_ARG(locale);
2177
2178 return "C";
2179
2180#else
2181
9aac5db8
KW
2182 const char * retval;
2183 const char * newlocale;
2184 dSAVEDERRNO;
a4f00dcc 2185 dTHX;
d36adde0 2186 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2187
a4f00dcc
KW
2188#ifdef USE_LOCALE_NUMERIC
2189
291a84fb
KW
2190 /* A NULL locale means only query what the current one is. We have the
2191 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2192 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2193 * the correct results. All other categories don't require special
2194 * handling */
a4f00dcc
KW
2195 if (locale == NULL) {
2196 if (category == LC_NUMERIC) {
9aac5db8
KW
2197
2198 /* We don't have to copy this return value, as it is a per-thread
2199 * variable, and won't change until a future setlocale */
2200 return PL_numeric_name;
a4f00dcc
KW
2201 }
2202
7d4bcc4a 2203# ifdef LC_ALL
a4f00dcc 2204
9aac5db8
KW
2205 else if (category == LC_ALL) {
2206 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2207 }
2208
7d4bcc4a 2209# endif
a4f00dcc
KW
2210
2211 }
2212
2213#endif
2214
33e5a354
KW
2215 retval = save_to_buffer(do_setlocale_r(category, locale),
2216 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2217 SAVE_ERRNO;
2218
2219#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2220
2221 if (locale == NULL && category == LC_ALL) {
2222 RESTORE_LC_NUMERIC();
2223 }
2224
2225#endif
a4f00dcc
KW
2226
2227 DEBUG_L(PerlIO_printf(Perl_debug_log,
2228 "%s:%d: %s\n", __FILE__, __LINE__,
2229 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2230
9aac5db8
KW
2231 RESTORE_ERRNO;
2232
2233 if (! retval) {
a4f00dcc
KW
2234 return NULL;
2235 }
2236
9aac5db8 2237 /* If locale == NULL, we are just querying the state */
a4f00dcc 2238 if (locale == NULL) {
a4f00dcc
KW
2239 return retval;
2240 }
a4f00dcc 2241
1159483a
KW
2242 /* Now that have switched locales, we have to update our records to
2243 * correspond. */
a4f00dcc 2244
1159483a 2245 switch (category) {
a4f00dcc 2246
1159483a 2247#ifdef USE_LOCALE_CTYPE
a4f00dcc 2248
1159483a
KW
2249 case LC_CTYPE:
2250 new_ctype(retval);
2251 break;
a4f00dcc 2252
1159483a 2253#endif
a4f00dcc
KW
2254#ifdef USE_LOCALE_COLLATE
2255
1159483a
KW
2256 case LC_COLLATE:
2257 new_collate(retval);
2258 break;
a4f00dcc 2259
1159483a 2260#endif
a4f00dcc
KW
2261#ifdef USE_LOCALE_NUMERIC
2262
1159483a
KW
2263 case LC_NUMERIC:
2264 new_numeric(retval);
2265 break;
a4f00dcc 2266
1159483a
KW
2267#endif
2268#ifdef LC_ALL
a4f00dcc 2269
1159483a 2270 case LC_ALL:
a4f00dcc 2271
1159483a
KW
2272 /* LC_ALL updates all the things we care about. The values may not
2273 * be the same as 'retval', as the locale "" may have set things
2274 * individually */
a4f00dcc 2275
1159483a 2276# ifdef USE_LOCALE_CTYPE
a4f00dcc 2277
b0bde642 2278 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2279 new_ctype(newlocale);
b0bde642 2280 Safefree(newlocale);
a4f00dcc 2281
1159483a
KW
2282# endif /* USE_LOCALE_CTYPE */
2283# ifdef USE_LOCALE_COLLATE
2284
b0bde642 2285 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2286 new_collate(newlocale);
b0bde642 2287 Safefree(newlocale);
a4f00dcc 2288
7d4bcc4a 2289# endif
1159483a 2290# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2291
b0bde642 2292 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2293 new_numeric(newlocale);
b0bde642 2294 Safefree(newlocale);
a4f00dcc 2295
1159483a
KW
2296# endif /* USE_LOCALE_NUMERIC */
2297#endif /* LC_ALL */
a4f00dcc 2298
1159483a
KW
2299 default:
2300 break;
a4f00dcc
KW
2301 }
2302
2303 return retval;
2304
d36adde0
KW
2305#endif
2306
f7416781
KW
2307}
2308
2309PERL_STATIC_INLINE const char *
2310S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2311{
2312 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2313 * growing it if necessary */
2314
0b6697c5 2315 Size_t string_size;
f7416781
KW
2316
2317 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2318
0b6697c5
KW
2319 if (! string) {
2320 return NULL;
2321 }
2322
2323 string_size = strlen(string) + offset + 1;
2324
f7416781
KW
2325 if (*buf_size == 0) {
2326 Newx(*buf, string_size, char);
2327 *buf_size = string_size;
2328 }
2329 else if (string_size > *buf_size) {
2330 Renew(*buf, string_size, char);
2331 *buf_size = string_size;
2332 }
2333
2334 Copy(string, *buf + offset, string_size - offset, char);
2335 return *buf;
2336}
2337
2338/*
2339
f7416781
KW
2340=for apidoc Perl_langinfo
2341
ce181bf6 2342This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2343taking the same C<item> parameter values, and returning the same information.
2344But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2345of Perl's locale handling from your code, and can be used on systems that lack
2346a native C<nl_langinfo>.
2347
2348Expanding on these:
2349
2350=over
2351
2352=item *
2353
ce181bf6
KW
2354The reason it isn't quite a drop-in replacement is actually an advantage. The
2355only difference is that it returns S<C<const char *>>, whereas plain
2356C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2357forbidden to write into the buffer. By declaring this C<const>, the compiler
2358enforces this restriction, so if it is violated, you know at compilation time,
2359rather than getting segfaults at runtime.
2360
2361=item *
2362
69e2275e 2363It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2364without you having to write extra code. The reason for the extra code would be
2365because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2366kept set by Perl so that the radix is a dot, and the separator is the empty
2367string, no matter what the underlying locale is supposed to be, and so to get
2368the expected results, you have to temporarily toggle into the underlying
2369locale, and later toggle back. (You could use plain C<nl_langinfo> and
2370C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2371the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2372(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2373(decimal point) character to be a dot.)
ce181bf6
KW
2374
2375=item *
2376
2377The system function it replaces can have its static return buffer trashed,
2378not only by a subesequent call to that function, but by a C<freelocale>,
2379C<setlocale>, or other locale change. The returned buffer of this function is
2380not changed until the next call to it, so the buffer is never in a trashed
2381state.
f7416781
KW
2382
2383=item *
2384
ce181bf6
KW
2385Its return buffer is per-thread, so it also is never overwritten by a call to
2386this function from another thread; unlike the function it replaces.
2387
2388=item *
2389
2390But most importantly, it works on systems that don't have C<nl_langinfo>, such
2391as Windows, hence makes your code more portable. Of the fifty-some possible
2392items specified by the POSIX 2008 standard,
f7416781 2393L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2394only one is completely unimplemented, though on non-Windows platforms, another
2395significant one is also not implemented). It uses various techniques to
2396recover the other items, including calling C<L<localeconv(3)>>, and
2397C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2398available. Later C<strftime()> versions have additional capabilities; C<""> is
2399returned for those not available on your system.
ce181bf6
KW
2400
2401It is important to note that when called with an item that is recovered by
2402using C<localeconv>, the buffer from any previous explicit call to
2403C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2404contents if you need to access them after a call to this function. (But note
2405that you might not want to be using C<localeconv()> directly anyway, because of
2406issues like the ones listed in the second item of this list (above) for
2407C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2408call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2409unpack).
9f0bc911 2410
6201c220
KW
2411The details for those items which may deviate from what this emulation returns
2412and what a native C<nl_langinfo()> would return are specified in
2413L<I18N::Langinfo>.
f7416781 2414
ce181bf6
KW
2415=back
2416
f7416781
KW
2417When using C<Perl_langinfo> on systems that don't have a native
2418C<nl_langinfo()>, you must
2419
2420 #include "perl_langinfo.h"
2421
2422before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2423C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2424C<langinfo.h> would try to import into the namespace for code that doesn't need
2425it.)
f7416781 2426
f7416781
KW
2427The original impetus for C<Perl_langinfo()> was so that code that needs to
2428find out the current currency symbol, floating point radix character, or digit
2429grouping separator can use, on all systems, the simpler and more
2430thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2431pain to make thread-friendly. For other fields returned by C<localeconv>, it
2432is better to use the methods given in L<perlcall> to call
2433L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2434
2435=cut
2436
2437*/
2438
2439const char *
2440#ifdef HAS_NL_LANGINFO
2441Perl_langinfo(const nl_item item)
2442#else
2443Perl_langinfo(const int item)
2444#endif
2445{
f61748ac
KW
2446 return my_nl_langinfo(item, TRUE);
2447}
2448
59a15af0 2449STATIC const char *
f61748ac
KW
2450#ifdef HAS_NL_LANGINFO
2451S_my_nl_langinfo(const nl_item item, bool toggle)
2452#else
2453S_my_nl_langinfo(const int item, bool toggle)
2454#endif
2455{
ae74815b 2456 dTHX;
0b6697c5 2457 const char * retval;
f7416781 2458
d36adde0
KW
2459#ifdef USE_LOCALE_NUMERIC
2460
5a854ab3
KW
2461 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2462 * two items, and only if not already there */
4e6826bf 2463 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2464 || PL_numeric_underlying))
d36adde0
KW
2465
2466#endif /* No toggling needed if not using LC_NUMERIC */
2467
5a854ab3 2468 toggle = FALSE;
5a854ab3 2469
ab340fff 2470#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2471# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2472 || ! defined(HAS_POSIX_2008_LOCALE) \
2473 || ! defined(DUPLOCALE)
f7416781 2474
ab340fff 2475 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2476 * for those items dependent on it. This must be copied to a buffer before
2477 * switching back, as some systems destroy the buffer when setlocale() is
2478 * called */
f7416781 2479
038d3702
KW
2480 {
2481 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2482
b2fee59c 2483 if (toggle) {
038d3702 2484 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2485 }
f7416781 2486
5acc4454
KW
2487 LOCALE_LOCK; /* Prevent interference from another thread executing
2488 this code section (the only call to nl_langinfo in
2489 the core) */
2490
ee90eb2d 2491
628ff75a
KW
2492 /* Copy to a per-thread buffer, which is also one that won't be
2493 * destroyed by a subsequent setlocale(), such as the
2494 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2495 retval = save_to_buffer(nl_langinfo(item),
2496 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2497
5acc4454
KW
2498 LOCALE_UNLOCK;
2499
b2fee59c 2500 if (toggle) {
038d3702 2501 RESTORE_LC_NUMERIC();
b2fee59c 2502 }
f7416781
KW
2503 }
2504
ab340fff
KW
2505# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2506
5a854ab3 2507 {
b2fee59c
KW
2508 bool do_free = FALSE;
2509 locale_t cur = uselocale((locale_t) 0);
ab340fff 2510
b2fee59c
KW
2511 if (cur == LC_GLOBAL_LOCALE) {
2512 cur = duplocale(LC_GLOBAL_LOCALE);
2513 do_free = TRUE;
2514 }
ab340fff 2515
d36adde0
KW
2516# ifdef USE_LOCALE_NUMERIC
2517
b2fee59c 2518 if (toggle) {
e1aa2579
KW
2519 if (PL_underlying_numeric_obj) {
2520 cur = PL_underlying_numeric_obj;
2521 }
2522 else {
2523 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2524 do_free = TRUE;
2525 }
b2fee59c 2526 }
ab340fff 2527
d36adde0
KW
2528# endif
2529
628ff75a
KW
2530 /* We have to save it to a buffer, because the freelocale() just below
2531 * can invalidate the internal one */
0b6697c5
KW
2532 retval = save_to_buffer(nl_langinfo_l(item, cur),
2533 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2534
b2fee59c
KW
2535 if (do_free) {
2536 freelocale(cur);
2537 }
5a854ab3 2538 }
ab340fff 2539
c1566110
KW
2540# endif
2541
0b6697c5 2542 if (strEQ(retval, "")) {
4e6826bf 2543 if (item == YESSTR) {
c1566110
KW
2544 return "yes";
2545 }
4e6826bf 2546 if (item == NOSTR) {
c1566110
KW
2547 return "no";
2548 }
2549 }
2550
0b6697c5 2551 return retval;
ab340fff 2552
f7416781 2553#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2554
2555 {
2556
f7416781
KW
2557# ifdef HAS_LOCALECONV
2558
43dd6b15 2559 const struct lconv* lc;
628ff75a 2560 const char * temp;
038d3702 2561 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2562
69c5e0db
KW
2563# ifdef TS_W32_BROKEN_LOCALECONV
2564
2565 const char * save_global;
2566 const char * save_thread;
2567 int needed_size;
2568 char * ptr;
2569 char * e;
2570 char * item_start;
2571
2572# endif
f7416781
KW
2573# endif
2574# ifdef HAS_STRFTIME
2575
43dd6b15
KW
2576 struct tm tm;
2577 bool return_format = FALSE; /* Return the %format, not the value */
2578 const char * format;
f7416781
KW
2579
2580# endif
2581
43dd6b15
KW
2582 /* We copy the results to a per-thread buffer, even if not
2583 * multi-threaded. This is in part to simplify this code, and partly
2584 * because we need a buffer anyway for strftime(), and partly because a
2585 * call of localeconv() could otherwise wipe out the buffer, and the
2586 * programmer would not be expecting this, as this is a nl_langinfo()
2587 * substitute after all, so s/he might be thinking their localeconv()
2588 * is safe until another localeconv() call. */
f7416781 2589
43dd6b15
KW
2590 switch (item) {
2591 Size_t len;
f7416781 2592
8d72e74e 2593 /* This is unimplemented */
4e6826bf 2594 case ERA: /* For use with strftime() %E modifier */
f7416781 2595
43dd6b15
KW
2596 default:
2597 return "";
f7416781 2598
43dd6b15 2599 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2600 case YESEXPR: return "^[+1yY]";
2601 case YESSTR: return "yes";
2602 case NOEXPR: return "^[-0nN]";
2603 case NOSTR: return "no";
2604
8d72e74e
KW
2605 case CODESET:
2606
2607# ifndef WIN32
2608
2609 /* On non-windows, this is unimplemented, in part because of
2610 * inconsistencies between vendors. The Darwin native
2611 * nl_langinfo() implementation simply looks at everything past
2612 * any dot in the name, but that doesn't work for other
2613 * vendors. Many Linux locales that don't have UTF-8 in their
2614 * names really are UTF-8, for example; z/OS locales that do
2615 * have UTF-8 in their names, aren't really UTF-8 */
2616 return "";
2617
2618# else
2619
2620 { /* But on Windows, the name does seem to be consistent, so
2621 use that. */
2622 const char * p;
2623 const char * first;
2624 Size_t offset = 0;
2625 const char * name = my_setlocale(LC_CTYPE, NULL);
2626
2627 if (isNAME_C_OR_POSIX(name)) {
2628 return "ANSI_X3.4-1968";
2629 }
2630
2631 /* Find the dot in the locale name */
2632 first = (const char *) strchr(name, '.');
2633 if (! first) {
2634 first = name;
2635 goto has_nondigit;
2636 }
2637
2638 /* Look at everything past the dot */
2639 first++;
2640 p = first;
2641
2642 while (*p) {
2643 if (! isDIGIT(*p)) {
2644 goto has_nondigit;
2645 }
2646
2647 p++;
2648 }
2649
2650 /* Here everything past the dot is a digit. Treat it as a
2651 * code page */
18503cbd 2652 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2653 &PL_langinfo_bufsize, 0);
8d72e74e 2654 offset = STRLENs("CP");
f7416781 2655
8d72e74e
KW
2656 has_nondigit:
2657
2658 retval = save_to_buffer(first, &PL_langinfo_buf,
2659 &PL_langinfo_bufsize, offset);
2660 }
2661
2662 break;
2663
2664# endif
f7416781
KW
2665# ifdef HAS_LOCALECONV
2666
4e6826bf 2667 case CRNCYSTR:
f7416781 2668
43dd6b15
KW
2669 /* We don't bother with localeconv_l() because any system that
2670 * has it is likely to also have nl_langinfo() */
291a84fb 2671
69c5e0db
KW
2672 LOCALE_LOCK_V; /* Prevent interference with other threads
2673 using localeconv() */
2674
2675# ifdef TS_W32_BROKEN_LOCALECONV
2676
2677 /* This is a workaround for a Windows bug prior to VS 15.
2678 * What we do here is, while locked, switch to the global
2679 * locale so localeconv() works; then switch back just before
2680 * the unlock. This can screw things up if some thread is
2681 * already using the global locale while assuming no other is.
2682 * A different workaround would be to call GetCurrencyFormat on
2683 * a known value, and parse it; patches welcome
2684 *
2685 * We have to use LC_ALL instead of LC_MONETARY because of
2686 * another bug in Windows */
2687
2688 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2689 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2690 save_global= savepv(my_setlocale(LC_ALL, NULL));
2691 my_setlocale(LC_ALL, save_thread);
2692
2693# endif
5acc4454 2694
43dd6b15
KW
2695 lc = localeconv();
2696 if ( ! lc
2697 || ! lc->currency_symbol
2698 || strEQ("", lc->currency_symbol))
2699 {
69c5e0db 2700 LOCALE_UNLOCK_V;
43dd6b15
KW
2701 return "";
2702 }
f7416781 2703
43dd6b15 2704 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2705 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2706 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2707 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2708 { /* khw couldn't figure out how the localedef specifications
2709 would show that the $ should replace the radix; this is
2710 just a guess as to how it might work.*/
a34edee3 2711 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2712 }
2713 else if (lc->p_cs_precedes) {
a34edee3 2714 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2715 }
2716 else {
a34edee3 2717 PL_langinfo_buf[0] = '+';
43dd6b15 2718 }
f7416781 2719
69c5e0db
KW
2720# ifdef TS_W32_BROKEN_LOCALECONV
2721
2722 my_setlocale(LC_ALL, save_global);
2723 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2724 my_setlocale(LC_ALL, save_thread);
2725 Safefree(save_global);
2726 Safefree(save_thread);
2727
2728# endif
2729
2730 LOCALE_UNLOCK_V;
43dd6b15 2731 break;
f7416781 2732
69c5e0db
KW
2733# ifdef TS_W32_BROKEN_LOCALECONV
2734
4e6826bf 2735 case RADIXCHAR:
69c5e0db
KW
2736
2737 /* For this, we output a known simple floating point number to
2738 * a buffer, and parse it, looking for the radix */
2739
2740 if (toggle) {
2741 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2742 }
2743
2744 if (PL_langinfo_bufsize < 10) {
2745 PL_langinfo_bufsize = 10;
2746 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2747 }
2748
2749 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2750 "%.1f", 1.5);
2751 if (needed_size >= (int) PL_langinfo_bufsize) {
2752 PL_langinfo_bufsize = needed_size + 1;
2753 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2754 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2755 "%.1f", 1.5);
2756 assert(needed_size < (int) PL_langinfo_bufsize);
2757 }
2758
2759 ptr = PL_langinfo_buf;
2760 e = PL_langinfo_buf + PL_langinfo_bufsize;
2761
2762 /* Find the '1' */
2763 while (ptr < e && *ptr != '1') {
2764 ptr++;
2765 }
2766 ptr++;
2767
2768 /* Find the '5' */
2769 item_start = ptr;
2770 while (ptr < e && *ptr != '5') {
2771 ptr++;
2772 }
2773
2774 /* Everything in between is the radix string */
2775 if (ptr >= e) {
2776 PL_langinfo_buf[0] = '?';
2777 PL_langinfo_buf[1] = '\0';
2778 }
2779 else {
2780 *ptr = '\0';
2781 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2782 }
2783
2784 if (toggle) {
2785 RESTORE_LC_NUMERIC();
2786 }
2787
2788 retval = PL_langinfo_buf;
2789 break;
2790
2791# else
2792
2793 case RADIXCHAR: /* No special handling needed */
2794
2795# endif
2796
4e6826bf 2797 case THOUSEP:
f7416781 2798
43dd6b15 2799 if (toggle) {
038d3702 2800 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2801 }
f7416781 2802
69c5e0db
KW
2803 LOCALE_LOCK_V; /* Prevent interference with other threads
2804 using localeconv() */
2805
2806# ifdef TS_W32_BROKEN_LOCALECONV
2807
2808 /* This should only be for the thousands separator. A
2809 * different work around would be to use GetNumberFormat on a
2810 * known value and parse the result to find the separator */
2811 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2812 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2813 save_global = savepv(my_setlocale(LC_ALL, NULL));
2814 my_setlocale(LC_ALL, save_thread);
2815# if 0
2816 /* This is the start of code that for broken Windows replaces
2817 * the above and below code, and instead calls
2818 * GetNumberFormat() and then would parse that to find the
2819 * thousands separator. It needs to handle UTF-16 vs -8
2820 * issues. */
2821
2822 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2823 DEBUG_L(PerlIO_printf(Perl_debug_log,
2824 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2825 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2826
2827# endif
2828# endif
5acc4454 2829
43dd6b15
KW
2830 lc = localeconv();
2831 if (! lc) {
628ff75a 2832 temp = "";
33394adc 2833 }
43dd6b15 2834 else {
4e6826bf 2835 temp = (item == RADIXCHAR)
43dd6b15
KW
2836 ? lc->decimal_point
2837 : lc->thousands_sep;
628ff75a
KW
2838 if (! temp) {
2839 temp = "";
43dd6b15
KW
2840 }
2841 }
f7416781 2842
0b6697c5
KW
2843 retval = save_to_buffer(temp, &PL_langinfo_buf,
2844 &PL_langinfo_bufsize, 0);
f7416781 2845
69c5e0db
KW
2846# ifdef TS_W32_BROKEN_LOCALECONV
2847
2848 my_setlocale(LC_ALL, save_global);
2849 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2850 my_setlocale(LC_ALL, save_thread);
2851 Safefree(save_global);
2852 Safefree(save_thread);
2853
2854# endif
2855
2856 LOCALE_UNLOCK_V;
5acc4454 2857
43dd6b15 2858 if (toggle) {
038d3702 2859 RESTORE_LC_NUMERIC();
43dd6b15 2860 }
f7416781 2861
43dd6b15 2862 break;
f7416781
KW
2863
2864# endif
2865# ifdef HAS_STRFTIME
2866
43dd6b15
KW
2867 /* These are defined by C89, so we assume that strftime supports
2868 * them, and so are returned unconditionally; they may not be what
2869 * the locale actually says, but should give good enough results
2870 * for someone using them as formats (as opposed to trying to parse
2871 * them to figure out what the locale says). The other format
2872 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2873 case D_FMT: return "%x";
2874 case T_FMT: return "%X";
2875 case D_T_FMT: return "%c";
43dd6b15
KW
2876
2877 /* These formats are only available in later strfmtime's */
4e6826bf 2878 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2879
2880 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2881 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2882 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2883 case ALT_DIGITS:
2884 case AM_STR: case PM_STR:
2885 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2886 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2887 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2888 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2889 case DAY_5: case DAY_6: case DAY_7:
2890 case MON_1: case MON_2: case MON_3: case MON_4:
2891 case MON_5: case MON_6: case MON_7: case MON_8:
2892 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2893
2894 LOCALE_LOCK;
2895
2896 init_tm(&tm); /* Precaution against core dumps */
2897 tm.tm_sec = 30;
2898 tm.tm_min = 30;
2899 tm.tm_hour = 6;
2900 tm.tm_year = 2017 - 1900;
2901 tm.tm_wday = 0;
2902 tm.tm_mon = 0;
2903 switch (item) {
2904 default:
2905 LOCALE_UNLOCK;
2906 Perl_croak(aTHX_
2907 "panic: %s: %d: switch case: %d problem",
2908 __FILE__, __LINE__, item);
2909 NOT_REACHED; /* NOTREACHED */
2910
4e6826bf
KW
2911 case PM_STR: tm.tm_hour = 18;
2912 case AM_STR:
43dd6b15
KW
2913 format = "%p";
2914 break;
2915
4e6826bf
KW
2916 case ABDAY_7: tm.tm_wday++;
2917 case ABDAY_6: tm.tm_wday++;
2918 case ABDAY_5: tm.tm_wday++;
2919 case ABDAY_4: tm.tm_wday++;
2920 case ABDAY_3: tm.tm_wday++;
2921 case ABDAY_2: tm.tm_wday++;
2922 case ABDAY_1:
43dd6b15
KW
2923 format = "%a";
2924 break;
2925
4e6826bf
KW
2926 case DAY_7: tm.tm_wday++;
2927 case DAY_6: tm.tm_wday++;
2928 case DAY_5: tm.tm_wday++;
2929 case DAY_4: tm.tm_wday++;
2930 case DAY_3: tm.tm_wday++;
2931 case DAY_2: tm.tm_wday++;
2932 case DAY_1:
43dd6b15
KW
2933 format = "%A";
2934 break;
2935
4e6826bf
KW
2936 case ABMON_12: tm.tm_mon++;
2937 case ABMON_11: tm.tm_mon++;
2938 case ABMON_10: tm.tm_mon++;
2939 case ABMON_9: tm.tm_mon++;
2940 case ABMON_8: tm.tm_mon++;
2941 case ABMON_7: tm.tm_mon++;
2942 case ABMON_6: tm.tm_mon++;
2943 case ABMON_5: tm.tm_mon++;
2944 case ABMON_4: tm.tm_mon++;
2945 case ABMON_3: tm.tm_mon++;
2946 case ABMON_2: tm.tm_mon++;
2947 case ABMON_1:
43dd6b15
KW
2948 format = "%b";
2949 break;
2950
4e6826bf
KW
2951 case MON_12: tm.tm_mon++;
2952 case MON_11: tm.tm_mon++;
2953 case MON_10: tm.tm_mon++;
2954 case MON_9: tm.tm_mon++;
2955 case MON_8: tm.tm_mon++;
2956 case MON_7: tm.tm_mon++;
2957 case MON_6: tm.tm_mon++;
2958 case MON_5: tm.tm_mon++;
2959 case MON_4: tm.tm_mon++;
2960 case MON_3: tm.tm_mon++;
2961 case MON_2: tm.tm_mon++;
2962 case MON_1:
43dd6b15
KW
2963 format = "%B";
2964 break;
2965
4e6826bf 2966 case T_FMT_AMPM:
43dd6b15
KW
2967 format = "%r";
2968 return_format = TRUE;
2969 break;
2970
4e6826bf 2971 case ERA_D_FMT:
43dd6b15
KW
2972 format = "%Ex";
2973 return_format = TRUE;
2974 break;
2975
4e6826bf 2976 case ERA_T_FMT:
43dd6b15
KW
2977 format = "%EX";
2978 return_format = TRUE;
2979 break;
2980
4e6826bf 2981 case ERA_D_T_FMT:
43dd6b15
KW
2982 format = "%Ec";
2983 return_format = TRUE;
2984 break;
2985
4e6826bf 2986 case ALT_DIGITS:
43dd6b15
KW
2987 tm.tm_wday = 0;
2988 format = "%Ow"; /* Find the alternate digit for 0 */
2989 break;
2990 }
f7416781 2991
43dd6b15
KW
2992 /* We can't use my_strftime() because it doesn't look at
2993 * tm_wday */
2994 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2995 format, &tm))
2996 {
2997 /* A zero return means one of:
2998 * a) there wasn't enough space in PL_langinfo_buf
2999 * b) the format, like a plain %p, returns empty
3000 * c) it was an illegal format, though some
3001 * implementations of strftime will just return the
3002 * illegal format as a plain character sequence.
3003 *
3004 * To quickly test for case 'b)', try again but precede
3005 * the format with a plain character. If that result is
3006 * still empty, the problem is either 'a)' or 'c)' */
3007
3008 Size_t format_size = strlen(format) + 1;
3009 Size_t mod_size = format_size + 1;
3010 char * mod_format;
3011 char * temp_result;
3012
3013 Newx(mod_format, mod_size, char);
3014 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 3015 *mod_format = ' ';
43dd6b15
KW
3016 my_strlcpy(mod_format + 1, format, mod_size);
3017 len = strftime(temp_result,
3018 PL_langinfo_bufsize,
3019 mod_format, &tm);
3020 Safefree(mod_format);
3021 Safefree(temp_result);
3022
3023 /* If 'len' is non-zero, it means that we had a case like
3024 * %p which means the current locale doesn't use a.m. or
3025 * p.m., and that is valid */
3026 if (len == 0) {
3027
3028 /* Here, still didn't work. If we get well beyond a
3029 * reasonable size, bail out to prevent an infinite
3030 * loop. */
3031
3032 if (PL_langinfo_bufsize > 100 * format_size) {
3033 *PL_langinfo_buf = '\0';
3034 }
3035 else {
3036 /* Double the buffer size to retry; Add 1 in case
3037 * original was 0, so we aren't stuck at 0. */
3038 PL_langinfo_bufsize *= 2;
3039 PL_langinfo_bufsize++;
3040 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3041 continue;
3042 }
3043 }
f7416781 3044
f7416781 3045 break;
43dd6b15 3046 }
f7416781 3047
43dd6b15
KW
3048 /* Here, we got a result.
3049 *
3050 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3051 * alternate format for wday 0. If the value is the same as
3052 * the normal 0, there isn't an alternate, so clear the buffer.
3053 * */
4e6826bf 3054 if ( item == ALT_DIGITS
43dd6b15
KW
3055 && strEQ(PL_langinfo_buf, "0"))
3056 {
3057 *PL_langinfo_buf = '\0';
3058 }
f7416781 3059
43dd6b15
KW
3060 /* ALT_DIGITS is problematic. Experiments on it showed that
3061 * strftime() did not always work properly when going from
3062 * alt-9 to alt-10. Only a few locales have this item defined,
3063 * and in all of them on Linux that khw was able to find,
3064 * nl_langinfo() merely returned the alt-0 character, possibly
3065 * doubled. Most Unicode digits are in blocks of 10
3066 * consecutive code points, so that is sufficient information
3067 * for those scripts, as we can infer alt-1, alt-2, .... But
3068 * for a Japanese locale, a CJK ideographic 0 is returned, and
3069 * the CJK digits are not in code point order, so you can't
3070 * really infer anything. The localedef for this locale did
3071 * specify the succeeding digits, so that strftime() works
3072 * properly on them, without needing to infer anything. But
3073 * the nl_langinfo() return did not give sufficient information
3074 * for the caller to understand what's going on. So until
3075 * there is evidence that it should work differently, this
3076 * returns the alt-0 string for ALT_DIGITS.
3077 *
3078 * wday was chosen because its range is all a single digit.
3079 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3080
43dd6b15 3081 LOCALE_UNLOCK;
f7416781 3082
d1b150d9
KW
3083 retval = PL_langinfo_buf;
3084
43dd6b15
KW
3085 /* If to return the format, not the value, overwrite the buffer
3086 * with it. But some strftime()s will keep the original format
3087 * if illegal, so change those to "" */
3088 if (return_format) {
3089 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3090 *PL_langinfo_buf = '\0';
3091 }
43dd6b15 3092 else {
0b6697c5
KW
3093 retval = save_to_buffer(format, &PL_langinfo_buf,
3094 &PL_langinfo_bufsize, 0);
f7416781
KW
3095 }
3096 }
3097
3098 break;
f7416781
KW
3099
3100# endif
3101
43dd6b15 3102 }
f7416781
KW
3103 }
3104
0b6697c5 3105 return retval;
f7416781
KW
3106
3107#endif
3108
a4f00dcc 3109}
b385bb4d 3110
98994639
HS
3111/*
3112 * Initialize locale awareness.
3113 */
3114int
3115Perl_init_i18nl10n(pTHX_ int printwarn)
3116{
0e92a118
KW
3117 /* printwarn is
3118 *
3119 * 0 if not to output warning when setup locale is bad
3120 * 1 if to output warning based on value of PERL_BADLANG
3121 * >1 if to output regardless of PERL_BADLANG
3122 *
3123 * returns
98994639 3124 * 1 = set ok or not applicable,
0e92a118
KW
3125 * 0 = fallback to a locale of lower priority
3126 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3127 *
3128 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3129 * set, debugging information is output.
3130 *
3131 * This looks more complicated than it is, mainly due to the #ifdefs.
3132 *
3133 * We try to set LC_ALL to the value determined by the environment. If
3134 * there is no LC_ALL on this platform, we try the individual categories we
3135 * know about. If this works, we are done.
3136 *
3137 * But if it doesn't work, we have to do something else. We search the
3138 * environment variables ourselves instead of relying on the system to do
3139 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3140 * think there is one), and the ultimate fallback "C". This is all done in
3141 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3142 * more complex. The 'trial_locales' array is initialized with just one
3143 * element; it causes the behavior described in the paragraph above this to
3144 * happen. If that fails, we add elements to 'trial_locales', and do extra
3145 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3146 *
3147 * On Ultrix, the locale MUST come from the environment, so there is
3148 * preliminary code to set it. I (khw) am not sure that it is necessary,
3149 * and that this couldn't be folded into the loop, but barring any real
3150 * platforms to test on, it's staying as-is
3151 *
3152 * A slight complication is that in embedded Perls, the locale may already
3153 * be set-up, and we don't want to get it from the normal environment
3154 * variables. This is handled by having a special environment variable
3155 * indicate we're in this situation. We simply set setlocale's 2nd
3156 * parameter to be a NULL instead of "". That indicates to setlocale that
3157 * it is not to change anything, but to return the current value,
3158 * effectively initializing perl's db to what the locale already is.
3159 *
3160 * We play the same trick with NULL if a LC_ALL succeeds. We call
3161 * setlocale() on the individual categores with NULL to get their existing
3162 * values for our db, instead of trying to change them.
3163 * */
98994639 3164
0e92a118
KW
3165 int ok = 1;
3166
7d4bcc4a
KW
3167#ifndef USE_LOCALE
3168
3169 PERL_UNUSED_ARG(printwarn);
3170
3171#else /* USE_LOCALE */
7d4bcc4a
KW
3172# ifdef __GLIBC__
3173
175c4cf9 3174 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3175
3176# endif
65ebb059 3177
ccd65d51
KW
3178 /* NULL uses the existing already set up locale */
3179 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3180 ? NULL
3181 : "";
c3fcd832
KW
3182 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3183 unsigned int trial_locales_count;
175c4cf9
KW
3184 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3185 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3186 bool setlocale_failure = FALSE;
65ebb059 3187 unsigned int i;
175c4cf9
KW
3188
3189 /* A later getenv() could zap this, so only use here */
3190 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3191
3192 const bool locwarn = (printwarn > 1
e5f10d49
KW
3193 || ( printwarn
3194 && ( ! bad_lang_use_once
22ff3130 3195 || (
e5f10d49
KW
3196 /* disallow with "" or "0" */
3197 *bad_lang_use_once
3198 && strNE("0", bad_lang_use_once)))));
ea92aad8 3199
291a84fb 3200 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3201 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3202
3203 /* current locale for given category; should have been copied so aren't
3204 * volatile */
8de4332b 3205 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3206
7d4bcc4a
KW
3207# ifdef WIN32
3208
6bce99ee
JH
3209 /* In some systems you can find out the system default locale
3210 * and use that as the fallback locale. */
7d4bcc4a
KW
3211# define SYSTEM_DEFAULT_LOCALE
3212# endif
3213# ifdef SYSTEM_DEFAULT_LOCALE
3214
65ebb059 3215 const char *system_default_locale = NULL;
98994639 3216
7d4bcc4a 3217# endif
948523db
KW
3218
3219# ifndef DEBUGGING
3220# define DEBUG_LOCALE_INIT(a,b,c)
3221# else
7d4bcc4a 3222
8298454c 3223 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3224
3225# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3226 STMT_START { \
3227 if (debug_initialization) { \
3228 PerlIO_printf(Perl_debug_log, \
3229 "%s:%d: %s\n", \
3230 __FILE__, __LINE__, \
a4f00dcc 3231 setlocale_debug_string(category, \
2fcc0ca9
KW
3232 locale, \
3233 result)); \
3234 } \
3235 } STMT_END
2fcc0ca9 3236
948523db
KW
3237/* Make sure the parallel arrays are properly set up */
3238# ifdef USE_LOCALE_NUMERIC
3239 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3240 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3241# ifdef USE_POSIX_2008_LOCALE
3242 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3243# endif
948523db
KW
3244# endif
3245# ifdef USE_LOCALE_CTYPE
3246 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3247 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3248# ifdef USE_POSIX_2008_LOCALE
3249 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3250# endif
948523db
KW
3251# endif
3252# ifdef USE_LOCALE_COLLATE
3253 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3254 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3255# ifdef USE_POSIX_2008_LOCALE
3256 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3257# endif
948523db
KW
3258# endif
3259# ifdef USE_LOCALE_TIME
3260 assert(categories[LC_TIME_INDEX] == LC_TIME);
3261 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3262# ifdef USE_POSIX_2008_LOCALE
3263 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3264# endif
948523db
KW
3265# endif
3266# ifdef USE_LOCALE_MESSAGES
3267 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3268 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3269# ifdef USE_POSIX_2008_LOCALE
3270 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3271# endif
948523db
KW
3272# endif
3273# ifdef USE_LOCALE_MONETARY
3274 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3275 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3276# ifdef USE_POSIX_2008_LOCALE
3277 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3278# endif
948523db 3279# endif
9821811f
KW
3280# ifdef USE_LOCALE_ADDRESS
3281 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3282 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3283# ifdef USE_POSIX_2008_LOCALE
3284 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3285# endif
9821811f
KW
3286# endif
3287# ifdef USE_LOCALE_IDENTIFICATION
3288 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3289 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3290# ifdef USE_POSIX_2008_LOCALE
3291 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3292# endif
9821811f
KW
3293# endif
3294# ifdef USE_LOCALE_MEASUREMENT
3295 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3296 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3297# ifdef USE_POSIX_2008_LOCALE
3298 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3299# endif
9821811f
KW
3300# endif
3301# ifdef USE_LOCALE_PAPER
3302 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3303 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3304# ifdef USE_POSIX_2008_LOCALE
3305 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3306# endif
9821811f
KW
3307# endif
3308# ifdef USE_LOCALE_TELEPHONE
3309 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3310 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3311# ifdef USE_POSIX_2008_LOCALE
3312 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3313# endif
9821811f 3314# endif
948523db
KW
3315# ifdef LC_ALL
3316 assert(categories[LC_ALL_INDEX] == LC_ALL);
3317 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3318 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3319# ifdef USE_POSIX_2008_LOCALE
3320 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3321# endif
948523db
KW
3322# endif
3323# endif /* DEBUGGING */
47280b20
KW
3324
3325 /* Initialize the cache of the program's UTF-8ness for the always known
3326 * locales C and POSIX */
3327 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3328 sizeof(PL_locale_utf8ness));
3329
e9bc6d6b
KW
3330# ifdef USE_THREAD_SAFE_LOCALE
3331# ifdef WIN32
3332
3333 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3334
3335# endif
3336# endif
39e69e77 3337# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3338
3339 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3340 if (! PL_C_locale_obj) {
3341 Perl_croak_nocontext(
3342 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3343 }
3344 if (DEBUG_Lv_TEST || debug_initialization) {
3345 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3346 }
3347
3348# endif
3349
78f7c09f
KW
3350# ifdef USE_LOCALE_NUMERIC
3351
3ca88433
KW
3352 PL_numeric_radix_sv = newSVpvs(".");
3353
78f7c09f
KW
3354# endif
3355
e9bc6d6b
KW
3356# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3357
3358 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3359 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3360
3361# endif
b56c4436 3362# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3363
3364 /*
3365 * Ultrix setlocale(..., "") fails if there are no environment
3366 * variables from which to get a locale name.
3367 */
3368
b56c4436
KW
3369# ifndef LC_ALL
3370# error Ultrix without LC_ALL not implemented
3371# else
7d4bcc4a 3372
b56c4436
KW
3373 {
3374 bool done = FALSE;
ea92aad8
KW
3375 if (lang) {
3376 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3377 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3378 if (sl_result[LC_ALL_INDEX])
3379 done = TRUE;
3380 else
e5f10d49 3381 setlocale_failure = TRUE;
ea92aad8
KW
3382 }
3383 if (! setlocale_failure) {
3384 const char * locale_param;
3385 for (i = 0; i < LC_ALL_INDEX; i++) {
3386 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3387 ? setlocale_init
3388 : NULL;
3389 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3390 if (! sl_result[i]) {
3391 setlocale_failure = TRUE;
3392 }
3393 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3394 }
c835d6be 3395 }
7d4bcc4a
KW
3396 }
3397
3398# endif /* LC_ALL */
e5f10d49 3399# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3400
65ebb059 3401 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3402 * the list. Normally the loop is executed just once. But if setting the
3403 * locale fails, inside the loop we add fallback trials to the array and so
3404 * will execute the loop multiple times */
c3fcd832
KW
3405 trial_locales[0] = setlocale_init;
3406 trial_locales_count = 1;
7d4bcc4a 3407
65ebb059
KW
3408 for (i= 0; i < trial_locales_count; i++) {
3409 const char * trial_locale = trial_locales[i];
3410
3411 if (i > 0) {
3412
3413 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3414 * when i==0, but I (khw) don't think that behavior makes much
3415 * sense */
3416 setlocale_failure = FALSE;
3417
7d4bcc4a 3418# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3419# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3420
65ebb059
KW
3421 /* On Windows machines, an entry of "" after the 0th means to use
3422 * the system default locale, which we now proceed to get. */
3423 if (strEQ(trial_locale, "")) {
3424 unsigned int j;
3425
3426 /* Note that this may change the locale, but we are going to do
3427 * that anyway just below */
837ce802 3428 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3429 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3430
7d4bcc4a 3431 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3432 * try */
3433 if (! system_default_locale) {
3434 goto next_iteration;
3435 }
3436 for (j = 0; j < trial_locales_count; j++) {
3437 if (strEQ(system_default_locale, trial_locales[j])) {
3438 goto next_iteration;
3439 }
3440 }
3441
3442 trial_locale = system_default_locale;
3443 }
ec0202b5
KW
3444# else
3445# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3446# endif
7d4bcc4a 3447# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3448
3449 } /* For i > 0 */
65ebb059 3450
7d4bcc4a
KW
3451# ifdef LC_ALL
3452
948523db
KW
3453 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3454 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3455 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3456 setlocale_failure = TRUE;
7cd8b568
KW
3457 }
3458 else {
3459 /* Since LC_ALL succeeded, it should have changed all the other
3460 * categories it can to its value; so we massage things so that the
3461 * setlocales below just return their category's current values.
3462 * This adequately handles the case in NetBSD where LC_COLLATE may
3463 * not be defined for a locale, and setting it individually will
7d4bcc4a 3464 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3465 * the POSIX locale. */
3466 trial_locale = NULL;
3467 }
7d4bcc4a
KW
3468
3469# endif /* LC_ALL */
98994639 3470
e5f10d49
KW
3471 if (! setlocale_failure) {
3472 unsigned int j;
3473 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3474 curlocales[j]
3475 = savepv(do_setlocale_r(categories[j], trial_locale));
3476 if (! curlocales[j]) {
3477 setlocale_failure = TRUE;
3478 }
3479 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3480 }
c835d6be 3481
e5f10d49
KW
3482 if (! setlocale_failure) { /* All succeeded */
3483 break; /* Exit trial_locales loop */
49c85077 3484 }
65ebb059 3485 }
98994639 3486
49c85077
KW
3487 /* Here, something failed; will need to try a fallback. */
3488 ok = 0;
65ebb059 3489
49c85077
KW
3490 if (i == 0) {
3491 unsigned int j;
98994639 3492
65ebb059 3493 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3494
3495# ifdef LC_ALL
98994639 3496
49c85077
KW
3497 PerlIO_printf(Perl_error_log,
3498 "perl: warning: Setting locale failed.\n");
98994639 3499
7d4bcc4a 3500# else /* !LC_ALL */
98994639 3501
49c85077
KW
3502 PerlIO_printf(Perl_error_log,
3503 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3504
e5f10d49
KW
3505 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3506 if (! curlocales[j]) {
3507 PerlIO_printf(Perl_error_log, category_names[j]);
3508 }
3509 else {
3510 Safefree(curlocales[j]);
3511 }
3512 }
7d4bcc4a 3513
7d4bcc4a 3514# endif /* LC_ALL */
98994639 3515
49c85077
KW
3516 PerlIO_printf(Perl_error_log,
3517 "perl: warning: Please check that your locale settings:\n");
98994639 3518
7d4bcc4a
KW
3519# ifdef __GLIBC__
3520
49c85077
KW
3521 PerlIO_printf(Perl_error_log,
3522 "\tLANGUAGE = %c%s%c,\n",
3523 language ? '"' : '(',
3524 language ? language : "unset",
3525 language ? '"' : ')');
7d4bcc4a 3526# endif
98994639 3527
49c85077
KW
3528 PerlIO_printf(Perl_error_log,
3529 "\tLC_ALL = %c%s%c,\n",
3530 lc_all ? '"' : '(',
3531 lc_all ? lc_all : "unset",
3532 lc_all ? '"' : ')');
98994639 3533
7d4bcc4a
KW
3534# if defined(USE_ENVIRON_ARRAY)
3535
49c85077 3536 {
cd999af9 3537 char **e;
d5e32b93
KW
3538
3539 /* Look through the environment for any variables of the
3540 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3541 * already handled above. These are assumed to be locale
3542 * settings. Output them and their values. */
cd999af9 3543 for (e = environ; *e; e++) {
d5e32b93
KW
3544 const STRLEN prefix_len = sizeof("LC_") - 1;
3545 STRLEN uppers_len;
3546
cd999af9 3547 if ( strBEGINs(*e, "LC_")
c8b388b0 3548 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3549 && (uppers_len = strspn(*e + prefix_len,
3550 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3551 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3552 {
3553 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3554 (int) (prefix_len + uppers_len), *e,
3555 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3556 }
3557 }
49c85077 3558 }
7d4bcc4a
KW
3559
3560# else
3561
49c85077
KW
3562 PerlIO_printf(Perl_error_log,
3563 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3564
3565# endif
98994639 3566
49c85077
KW
3567 PerlIO_printf(Perl_error_log,
3568 "\tLANG = %c%s%c\n",
3569 lang ? '"' : '(',
3570 lang ? lang : "unset",
3571 lang ? '"' : ')');
98994639 3572
49c85077
KW
3573 PerlIO_printf(Perl_error_log,
3574 " are supported and installed on your system.\n");
3575 }
98994639 3576
65ebb059 3577 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3578 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3579 * usually change the upper bound of the loop we are in.
3580 *
3581 * Since the system's default way of setting the locale has not
3582 * found one that works, We use Perl's defined ordering: LC_ALL,
3583 * LANG, and the C locale. We don't try the same locale twice, so
3584 * don't add to the list if already there. (On POSIX systems, the
3585 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3586 * but there's no harm done by doing it explicitly.
3587 *
3588 * Note that this tries the LC_ALL environment variable even on
3589 * systems which have no LC_ALL locale setting. This may or may
3590 * not have been originally intentional, but there's no real need
3591 * to change the behavior. */
65ebb059
KW
3592 if (lc_all) {
3593 for (j = 0; j < trial_locales_count; j++) {
3594 if (strEQ(lc_all, trial_locales[j])) {
3595 goto done_lc_all;
3596 }
3597 }
3598 trial_locales[trial_locales_count++] = lc_all;
3599 }
3600 done_lc_all:
98994639 3601
65ebb059
KW
3602 if (lang) {
3603 for (j = 0; j < trial_locales_count; j++) {
3604 if (strEQ(lang, trial_locales[j])) {
3605 goto done_lang;
3606 }
3607 }
3608 trial_locales[trial_locales_count++] = lang;
3609 }
3610 done_lang:
3611
7d4bcc4a
KW
3612# if defined(WIN32) && defined(LC_ALL)
3613
65ebb059
KW
3614 /* For Windows, we also try the system default locale before "C".
3615 * (If there exists a Windows without LC_ALL we skip this because
3616 * it gets too complicated. For those, the "C" is the next
3617 * fallback possibility). The "" is the same as the 0th element of
3618 * the array, but the code at the loop above knows to treat it
3619 * differently when not the 0th */
3620 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3621
3622# endif
65ebb059
KW
3623
3624 for (j = 0; j < trial_locales_count; j++) {
3625 if (strEQ("C", trial_locales[j])) {
3626 goto done_C;
3627 }
3628 }
3629 trial_locales[trial_locales_count++] = "C";
98994639 3630
65ebb059
KW
3631 done_C: ;
3632 } /* end of first time through the loop */
98994639 3633
7d4bcc4a
KW
3634# ifdef WIN32
3635
65ebb059 3636 next_iteration: ;
7d4bcc4a
KW
3637
3638# endif
65ebb059
KW
3639
3640 } /* end of looping through the trial locales */
3641
3642 if (ok < 1) { /* If we tried to fallback */
3643 const char* msg;
3644 if (! setlocale_failure) { /* fallback succeeded */
3645 msg = "Falling back to";
3646 }
3647 else { /* fallback failed */
e5f10d49 3648 unsigned int j;
98994639 3649
65ebb059
KW
3650 /* We dropped off the end of the loop, so have to decrement i to
3651 * get back to the value the last time through */
3652 i--;
98994639 3653
65ebb059
KW
3654 ok = -1;
3655 msg = "Failed to fall back to";
3656
3657 /* To continue, we should use whatever values we've got */
7d4bcc4a 3658
e5f10d49
KW
3659 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3660 Safefree(curlocales[j]);
3661 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3662 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3663 }
65ebb059
KW
3664 }
3665
3666 if (locwarn) {
3667 const char * description;
3668 const char * name = "";
3669 if (strEQ(trial_locales[i], "C")) {
3670 description = "the standard locale";
3671 name = "C";
3672 }
7d4bcc4a
KW
3673
3674# ifdef SYSTEM_DEFAULT_LOCALE
3675
65ebb059
KW
3676 else if (strEQ(trial_locales[i], "")) {
3677 description = "the system default locale";
3678 if (system_default_locale) {
3679 name = system_default_locale;
3680 }
3681 }
7d4bcc4a
KW
3682
3683# endif /* SYSTEM_DEFAULT_LOCALE */
3684
65ebb059
KW
3685 else {
3686 description = "a fallback locale";
3687 name = trial_locales[i];
3688 }
3689 if (name && strNE(name, "")) {
3690 PerlIO_printf(Perl_error_log,
3691 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3692 }
3693 else {
3694 PerlIO_printf(Perl_error_log,
3695 "perl: warning: %s %s.\n", msg, description);
3696 }
3697 }
3698 } /* End of tried to fallback */
98994639 3699
e5f10d49
KW
3700 /* Done with finding the locales; update our records */
3701
7d4bcc4a
KW
3702# ifdef USE_LOCALE_CTYPE
3703
948523db 3704 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3705
e5f10d49 3706# endif
7d4bcc4a
KW
3707# ifdef USE_LOCALE_COLLATE
3708
948523db 3709 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3710
e5f10d49 3711# endif
7d4bcc4a
KW
3712# ifdef USE_LOCALE_NUMERIC
3713
948523db 3714 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3715
3716# endif
3717
948523db 3718 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3719
e9bc6d6b 3720# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3721
3722 /* This caches whether each category's locale is UTF-8 or not. This
3723 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3724 * initialization time before any threads have started, but not later
3725 * unless thread-safe operations are used.
47280b20
KW
3726 * Caching means that if the program heeds our dictate not to change
3727 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3728 * it may get queried without having to change locales. If the
3729 * environment is such that all categories have the same locale, this
3730 * isn't needed, as the code will not change the locale; but this
3731 * handles the uncommon case where the environment has disparate
3732 * locales for the categories */
47280b20
KW
3733 (void) _is_cur_LC_category_utf8(categories[i]);
3734
3735# endif
3736
e5f10d49
KW
3737 Safefree(curlocales[i]);
3738 }
b310b053 3739
7d4bcc4a
KW
3740# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3741
49c85077 3742 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3743 * locale is UTF-8. The call to new_ctype() just above has already
3744 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3745 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3746 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3747 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3748 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3749
a05d7ebb 3750 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3751 This is an alternative to using the -C command line switch
3752 (the -C if present will override this). */
3753 {
dd374669 3754 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3755 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3756 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3757 PL_utf8cache = -1;
b310b053
JH
3758 }
3759
7d4bcc4a 3760# endif
7d4bcc4a
KW
3761# ifdef __GLIBC__
3762
175c4cf9 3763 Safefree(language);
7d4bcc4a
KW
3764
3765# endif
175c4cf9
KW
3766
3767 Safefree(lc_all);
3768 Safefree(lang);
3769
e3305790 3770#endif /* USE_LOCALE */
2fcc0ca9 3771#ifdef DEBUGGING
7d4bcc4a 3772
2fcc0ca9 3773 /* So won't continue to output stuff */
27cdc72e 3774 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3775
2fcc0ca9
KW
3776#endif
3777
98994639
HS
3778 return ok;
3779}
3780
98994639
HS
3781#ifdef USE_LOCALE_COLLATE
3782
a4a439fb 3783char *
a4a439fb
KW
3784Perl__mem_collxfrm(pTHX_ const char *input_string,
3785 STRLEN len, /* Length of 'input_string' */
3786 STRLEN *xlen, /* Set to length of returned string
3787 (not including the collation index
3788 prefix) */
3789 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3790 )
98994639 3791{
a4a439fb
KW
3792
3793 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3794 * differences. First, it handles embedded NULs. Second, it allocates a bit
3795 * more memory than needed for the transformed data itself. The real
55e5378d 3796 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3797 * the length of that, and doesn't include the collation index size.
3798 * Please see sv_collxfrm() to see how this is used. */
3799
55e5378d
KW
3800#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3801
6696cfa7
KW
3802 char * s = (char *) input_string;
3803 STRLEN s_strlen = strlen(input_string);
79f120c8 3804 char *xbuf = NULL;
55e5378d 3805 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3806 STRLEN length_in_chars;
c664130f 3807 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3808
a4a439fb
KW
3809 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3810
3811 /* Must be NUL-terminated */
3812 assert(*(input_string + len) == '\0');
7918f24d 3813
79f120c8
KW
3814 /* If this locale has defective collation, skip */
3815 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3816 DEBUG_L(PerlIO_printf(Perl_debug_log,
3817 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3818 goto bad;
3819 }
3820
6696cfa7
KW
3821 /* Replace any embedded NULs with the control that sorts before any others.
3822 * This will give as good as possible results on strings that don't
3823 * otherwise contain that character, but otherwise there may be
3824 * less-than-perfect results with that character and NUL. This is
fdc080f3 3825 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3826 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3827 NUL */
6696cfa7
KW
3828 char * e = s + len;
3829 char * sans_nuls;
fdc080f3 3830 STRLEN sans_nuls_len;
94762aa0 3831 int try_non_controls;
afc4976f
KW
3832 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3833 making sure 2nd byte is NUL.
3834 */
3835 STRLEN this_replacement_len;
3836
1e4c9676
KW
3837 /* If we don't know what non-NUL control character sorts lowest for
3838 * this locale, find it */
f28f4d2a 3839 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3840 int j;
afc4976f 3841 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3842 includes the collation index
3843 prefixed. */
3844
91c0e2e0 3845 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3846
3847 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3848 * NUL, in which case we instead look for any character. Controls
3849 * are preferred because collation order is, in general, context
3850 * sensitive, with adjoining characters affecting the order, and
3851 * controls are less likely to have such interactions, allowing the
3852 * NUL-replacement to stand on its own. (Another way to look at it
3853 * is to imagine what would happen if the NUL were replaced by a
3854 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3855 for (try_non_controls = 0;
3856 try_non_controls < 2;
3857 try_non_controls++)
3858 {
d4ff9586
KW
3859 /* Look through all legal code points (NUL isn't) */
3860 for (j = 1; j < 256; j++) {
3861 char * x; /* j's xfrm plus collation index */
3862 STRLEN x_len; /* length of 'x' */
3863 STRLEN trial_len = 1;
736a4fed