This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Don't use numeric unless LC_NUMERIC
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
e9bc6d6b
KW
35 *
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
166f8a29
DM
44 */
45
98994639
HS
46#include "EXTERN.h"
47#define PERL_IN_LOCALE_C
f7416781 48#include "perl_langinfo.h"
98994639
HS
49#include "perl.h"
50
a4af207c
JH
51#include "reentr.h"
52
0dec74cd
KW
53#ifdef I_WCHAR
54# include <wchar.h>
55#endif
56
2fcc0ca9
KW
57/* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
5a4b0634
KW
60#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61# define debug_initialization 0
62# define DEBUG_INITIALIZATION_set(v)
63#else
2fcc0ca9 64static bool debug_initialization = FALSE;
5a4b0634 65# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
66#endif
67
48015184
KW
68
69/* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
71 * done generally. */
72#define GET_ERRNO saved_errno
73
291a84fb
KW
74/* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
77#define STRLENs(s) (sizeof("" s "") - 1)
78
63e5b0d7
KW
79/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
85 * cntrl: 84-97 9B-9F
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89#define isNAME_C_OR_POSIX(name) \
90 ( (name) != NULL \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
93
8ef6e574
KW
94#ifdef USE_LOCALE
95
47280b20
KW
96/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
98
99#define UTF8NESS_SEP "\v"
100#define UTF8NESS_PREFIX "\f"
101
102/* So, the string looks like:
98994639 103 *
47280b20 104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 105 *
47280b20
KW
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
108
109STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
111
112#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
114
115/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
118
98994639
HS
119STATIC char *
120S_stdize_locale(pTHX_ char *locs)
121{
47280b20
KW
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
124 *
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
130 *
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
134 * */
135
7452cf6a 136 const char * const s = strchr(locs, '=');
98994639
HS
137 bool okay = TRUE;
138
7918f24d
NC
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
140
8772537c
AL
141 if (s) {
142 const char * const t = strchr(s, '.');
98994639 143 okay = FALSE;
8772537c
AL
144 if (t) {
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
149 locs[len] = 0;
150 okay = TRUE;
98994639
HS
151 }
152 }
153 }
154
155 if (!okay)
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
157
158 return locs;
159}
160
e5f10d49
KW
161/* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
163 * order. */
164
165const int categories[] = {
166
167# ifdef USE_LOCALE_NUMERIC
168 LC_NUMERIC,
169# endif
170# ifdef USE_LOCALE_CTYPE
171 LC_CTYPE,
172# endif
173# ifdef USE_LOCALE_COLLATE
174 LC_COLLATE,
175# endif
176# ifdef USE_LOCALE_TIME
177 LC_TIME,
178# endif
179# ifdef USE_LOCALE_MESSAGES
180 LC_MESSAGES,
181# endif
182# ifdef USE_LOCALE_MONETARY
183 LC_MONETARY,
184# endif
9821811f
KW
185# ifdef USE_LOCALE_ADDRESS
186 LC_ADDRESS,
187# endif
188# ifdef USE_LOCALE_IDENTIFICATION
189 LC_IDENTIFICATION,
190# endif
191# ifdef USE_LOCALE_MEASUREMENT
192 LC_MEASUREMENT,
193# endif
194# ifdef USE_LOCALE_PAPER
195 LC_PAPER,
196# endif
197# ifdef USE_LOCALE_TELEPHONE
198 LC_TELEPHONE,
199# endif
e5f10d49
KW
200# ifdef LC_ALL
201 LC_ALL,
202# endif
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
206};
207
208/* The top-most real element is LC_ALL */
209
210const char * category_names[] = {
211
212# ifdef USE_LOCALE_NUMERIC
213 "LC_NUMERIC",
214# endif
215# ifdef USE_LOCALE_CTYPE
216 "LC_CTYPE",
217# endif
218# ifdef USE_LOCALE_COLLATE
219 "LC_COLLATE",
220# endif
221# ifdef USE_LOCALE_TIME
222 "LC_TIME",
223# endif
224# ifdef USE_LOCALE_MESSAGES
225 "LC_MESSAGES",
226# endif
227# ifdef USE_LOCALE_MONETARY
228 "LC_MONETARY",
229# endif
9821811f
KW
230# ifdef USE_LOCALE_ADDRESS
231 "LC_ADDRESS",
232# endif
233# ifdef USE_LOCALE_IDENTIFICATION
234 "LC_IDENTIFICATION",
235# endif
236# ifdef USE_LOCALE_MEASUREMENT
237 "LC_MEASUREMENT",
238# endif
239# ifdef USE_LOCALE_PAPER
240 "LC_PAPER",
241# endif
242# ifdef USE_LOCALE_TELEPHONE
243 "LC_TELEPHONE",
244# endif
e5f10d49
KW
245# ifdef LC_ALL
246 "LC_ALL",
247# endif
248 NULL /* Placeholder */
249 };
250
251# ifdef LC_ALL
252
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
256
257# else
258
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
262
263# endif
264
265/* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
948523db
KW
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
e5f10d49 272
b09aaf40
KW
273STATIC const char *
274S_category_name(const int category)
275{
276 unsigned int i;
277
278#ifdef LC_ALL
279
280 if (category == LC_ALL) {
281 return "LC_ALL";
282 }
283
284#endif
285
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
289 }
290 }
291
292 {
293 const char suffix[] = " (unknown)";
294 int temp = category;
295 Size_t length = sizeof(suffix) + 1;
296 char * unknown;
297 dTHX;
298
299 if (temp < 0) {
300 length++;
301 temp = - temp;
302 }
303
304 /* Calculate the number of digits */
305 while (temp >= 10) {
306 temp /= 10;
307 length++;
308 }
309
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
312 SAVEFREEPV(unknown);
313 return unknown;
314 }
315}
316
948523db
KW
317/* Now create LC_foo_INDEX #defines for just those categories on this system */
318# ifdef USE_LOCALE_NUMERIC
319# define LC_NUMERIC_INDEX 0
320# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
321# else
322# define _DUMMY_NUMERIC -1
323# endif
324# ifdef USE_LOCALE_CTYPE
325# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326# define _DUMMY_CTYPE LC_CTYPE_INDEX
327# else
328# define _DUMMY_CTYPE _DUMMY_NUMERIC
329# endif
330# ifdef USE_LOCALE_COLLATE
331# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332# define _DUMMY_COLLATE LC_COLLATE_INDEX
333# else
334# define _DUMMY_COLLATE _DUMMY_COLLATE
335# endif
336# ifdef USE_LOCALE_TIME
337# define LC_TIME_INDEX _DUMMY_COLLATE + 1
338# define _DUMMY_TIME LC_TIME_INDEX
339# else
340# define _DUMMY_TIME _DUMMY_COLLATE
341# endif
342# ifdef USE_LOCALE_MESSAGES
343# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
345# else
346# define _DUMMY_MESSAGES _DUMMY_TIME
347# endif
348# ifdef USE_LOCALE_MONETARY
349# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350# define _DUMMY_MONETARY LC_MONETARY_INDEX
351# else
352# define _DUMMY_MONETARY _DUMMY_MESSAGES
353# endif
9821811f
KW
354# ifdef USE_LOCALE_ADDRESS
355# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
357# else
358# define _DUMMY_ADDRESS _DUMMY_MONETARY
359# endif
360# ifdef USE_LOCALE_IDENTIFICATION
361# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
363# else
364# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
365# endif
366# ifdef USE_LOCALE_MEASUREMENT
367# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
369# else
370# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
371# endif
372# ifdef USE_LOCALE_PAPER
373# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374# define _DUMMY_PAPER LC_PAPER_INDEX
375# else
376# define _DUMMY_PAPER _DUMMY_MEASUREMENT
377# endif
378# ifdef USE_LOCALE_TELEPHONE
379# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
381# else
382# define _DUMMY_TELEPHONE _DUMMY_PAPER
383# endif
948523db 384# ifdef LC_ALL
9821811f 385# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
386# endif
387#endif /* ifdef USE_LOCALE */
8ef6e574 388
d2b24094 389/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
390#ifdef WIN32
391# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
392#else
393# define my_setlocale(cat, locale) setlocale(cat, locale)
394#endif
395
396#ifndef USE_POSIX_2008_LOCALE
397
398/* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
402
403#else /* Below uses POSIX 2008 */
404
405/* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411# define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
414
415/* A third array, parallel to the ones above to map from category to its
416 * equivalent mask */
417const int category_masks[] = {
418# ifdef USE_LOCALE_NUMERIC
419 LC_NUMERIC_MASK,
420# endif
421# ifdef USE_LOCALE_CTYPE
422 LC_CTYPE_MASK,
423# endif
424# ifdef USE_LOCALE_COLLATE
425 LC_COLLATE_MASK,
426# endif
427# ifdef USE_LOCALE_TIME
428 LC_TIME_MASK,
429# endif
430# ifdef USE_LOCALE_MESSAGES
431 LC_MESSAGES_MASK,
432# endif
433# ifdef USE_LOCALE_MONETARY
434 LC_MONETARY_MASK,
435# endif
436# ifdef USE_LOCALE_ADDRESS
437 LC_ADDRESS_MASK,
438# endif
439# ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
441# endif
442# ifdef USE_LOCALE_MEASUREMENT
443 LC_MEASUREMENT_MASK,
444# endif
445# ifdef USE_LOCALE_PAPER
446 LC_PAPER_MASK,
447# endif
448# ifdef USE_LOCALE_TELEPHONE
449 LC_TELEPHONE_MASK,
450# endif
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
455 * time */
456 LC_ALL_MASK
457 };
458
459STATIC const char *
460S_emulate_setlocale(const int category,
461 const char * locale,
462 unsigned int index,
463 const bool is_index_valid
464 )
465{
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
473 *
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
476 *
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
486 *
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
501 */
502
503 int mask;
504 locale_t old_obj;
505 locale_t new_obj;
506 dTHX;
507
508# ifdef DEBUGGING
509
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
512 }
513
514# endif
515
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
518 unsigned int i;
519
520# ifdef DEBUGGING
521
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
524 }
525
526# endif
527
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
530 index = i;
531 goto found_index;
532 }
533 }
534
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
539 category, locale);
540 return NULL;
541
542 found_index: ;
543
544# ifdef DEBUGGING
545
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
548 }
549
550# endif
551
552 }
553
554 mask = category_masks[index];
555
556# ifdef DEBUGGING
557
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
560 }
561
562# endif
563
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
567
568# ifdef DEBUGGING
569
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
572 }
573
574# endif
575
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
578 }
579
580# ifdef HAS_QUERYLOCALE
581
582 return (char *) querylocale(mask, cur_obj);
583
d2b24094 584# else
ecdda939
KW
585
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
588
e9bc6d6b
KW
589# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
590
591 {
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
595
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
598 /*
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
602 */
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
607
608# ifdef USE_C_BACKTRACE
609
610 dump_c_backtrace(Perl_debug_log, 20, 1);
611
612# endif
613
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
618 }
619
620 return temp_name;
621 }
622 }
623
624# endif
625
626 /* Without querylocale(), we have to use our record-keeping we've
627 * done. */
628
629 if (category != LC_ALL) {
630
631# ifdef DEBUGGING
632
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 }
636
637# endif
638
639 return PL_curlocales[index];
640 }
641 else { /* For LC_ALL */
642 unsigned int i;
643 Size_t names_len = 0;
644 char * all_string;
7c93a471 645 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
646
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
649
650# ifdef DEBUGGING
651
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
654 }
655
656# endif
657
658 return PL_curlocales[LC_ALL_INDEX];
659 }
660
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
e9bc6d6b
KW
666 for (i = 0; i < LC_ALL_INDEX; i++) {
667
668# ifdef DEBUGGING
669
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
672 }
673
674# endif
675
676 names_len += strlen(category_names[i])
677 + 1 /* '=' */
678 + strlen(PL_curlocales[i])
679 + 1; /* ';' */
7c93a471
KW
680
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
683 }
684 }
685
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 692 }
7c93a471 693
e9bc6d6b
KW
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
696 *all_string = '\0';
697
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
700
701# ifdef DEBUGGING
702
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
705 }
706
707# endif
708
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
713 }
714
715# ifdef DEBUGGING
716
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
719 }
720
721 #endif
722
723 return all_string;
724 }
725
726# ifdef EINVAL
727
728 SETERRNO(EINVAL, LIB_INVARG);
729
730# endif
731
732 return NULL;
733
d2b24094
KW
734# endif
735
e9bc6d6b
KW
736 }
737
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);
1510
7918f24d
NC
1511 PERL_ARGS_ASSERT_NEW_CTYPE;
1512
215c5139
KW
1513 /* We will replace any bad locale warning with 1) nothing if the new one is
1514 * ok; or 2) a new warning for the bad new locale */
1515 if (PL_warn_locale) {
1516 SvREFCNT_dec_NN(PL_warn_locale);
1517 PL_warn_locale = NULL;
1518 }
1519
c1284011 1520 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1521
1522 /* A UTF-8 locale gets standard rules. But note that code still has to
1523 * handle this specially because of the three problematic code points */
1524 if (PL_in_utf8_CTYPE_locale) {
1525 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1526 }
8b7358b9
KW
1527
1528 /* We don't populate the other lists if a UTF-8 locale, but do check that
1529 * everything works as expected, unless checking turned off */
1530 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1531 /* Assume enough space for every character being bad. 4 spaces each
1532 * for the 94 printable characters that are output like "'x' "; and 5
1533 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1534 * NUL */
8b7358b9 1535 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1536 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1537 to start */
1538 unsigned int bad_count = 0; /* Count of bad characters */
1539
baa60164 1540 for (i = 0; i < 256; i++) {
8b7358b9 1541 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1542 if (isupper(i))
1543 PL_fold_locale[i] = (U8) tolower(i);
1544 else if (islower(i))
1545 PL_fold_locale[i] = (U8) toupper(i);
1546 else
1547 PL_fold_locale[i] = (U8) i;
8b7358b9 1548 }
8c6180a9
KW
1549
1550 /* If checking for locale problems, see if the native ASCII-range
1551 * printables plus \n and \t are in their expected categories in
1552 * the new locale. If not, this could mean big trouble, upending
1553 * Perl's and most programs' assumptions, like having a
1554 * metacharacter with special meaning become a \w. Fortunately,
1555 * it's very rare to find locales that aren't supersets of ASCII
1556 * nowadays. It isn't a problem for most controls to be changed
1557 * into something else; we check only \n and \t, though perhaps \r
1558 * could be an issue as well. */
7d4bcc4a 1559 if ( check_for_problems
8c6180a9
KW
1560 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1561 {
8b7358b9 1562 bool is_bad = FALSE;
6726b4f4 1563 char name[4] = { '\0' };
8b7358b9
KW
1564
1565 /* Convert the name into a string */
6726b4f4 1566 if (isGRAPH_A(i)) {
8b7358b9
KW
1567 name[0] = i;
1568 name[1] = '\0';
1569 }
1570 else if (i == '\n') {
6726b4f4
KW
1571 my_strlcpy(name, "\\n", sizeof(name));
1572 }
1573 else if (i == '\t') {
1574 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1575 }
1576 else {
6726b4f4
KW
1577 assert(i == ' ');
1578 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1579 }
1580
1581 /* Check each possibe class */
1582 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1583 is_bad = TRUE;
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isalnum('%s') unexpectedly is %d\n",
1586 name, cBOOL(isalnum(i))));
1587 }
1588 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1589 is_bad = TRUE;
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isalpha('%s') unexpectedly is %d\n",
1592 name, cBOOL(isalpha(i))));
1593 }
1594 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1595 is_bad = TRUE;
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isdigit('%s') unexpectedly is %d\n",
1598 name, cBOOL(isdigit(i))));
1599 }
1600 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1601 is_bad = TRUE;
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "isgraph('%s') unexpectedly is %d\n",
1604 name, cBOOL(isgraph(i))));
1605 }
1606 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1607 is_bad = TRUE;
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "islower('%s') unexpectedly is %d\n",
1610 name, cBOOL(islower(i))));
1611 }
1612 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1613 is_bad = TRUE;
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "isprint('%s') unexpectedly is %d\n",
1616 name, cBOOL(isprint(i))));
1617 }
1618 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1619 is_bad = TRUE;
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "ispunct('%s') unexpectedly is %d\n",
1622 name, cBOOL(ispunct(i))));
1623 }
1624 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1625 is_bad = TRUE;
1626 DEBUG_L(PerlIO_printf(Perl_debug_log,
1627 "isspace('%s') unexpectedly is %d\n",
1628 name, cBOOL(isspace(i))));
1629 }
1630 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1631 is_bad = TRUE;
1632 DEBUG_L(PerlIO_printf(Perl_debug_log,
1633 "isupper('%s') unexpectedly is %d\n",
1634 name, cBOOL(isupper(i))));
1635 }
1636 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1637 is_bad = TRUE;
1638 DEBUG_L(PerlIO_printf(Perl_debug_log,
1639 "isxdigit('%s') unexpectedly is %d\n",
1640 name, cBOOL(isxdigit(i))));
1641 }
1642 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1643 is_bad = TRUE;
1644 DEBUG_L(PerlIO_printf(Perl_debug_log,
1645 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1646 name, tolower(i), (int) toLOWER_A(i)));
1647 }
1648 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1649 is_bad = TRUE;
1650 DEBUG_L(PerlIO_printf(Perl_debug_log,
1651 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1652 name, toupper(i), (int) toUPPER_A(i)));
1653 }
1654 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1655 is_bad = TRUE;
1656 DEBUG_L(PerlIO_printf(Perl_debug_log,
1657 "'\\n' (=%02X) is not a control\n", (int) i));
1658 }
1659
1660 /* Add to the list; Separate multiple entries with a blank */
1661 if (is_bad) {
1662 if (bad_count) {
1663 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1664 }
8b7358b9
KW
1665 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1666 bad_count++;
8c6180a9
KW
1667 }
1668 }
1669 }
1670
7d4bcc4a
KW
1671# ifdef MB_CUR_MAX
1672
8c6180a9 1673 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1674 * this locale requires more than one byte, there are going to be
8c6180a9 1675 * problems. */
9c8a6dc2
KW
1676 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1677 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1678 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1679
8b7358b9
KW
1680 if ( check_for_problems && MB_CUR_MAX > 1
1681 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1682
1683 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1684 * locale. Just assume that the implementation for them (plus
1685 * for POSIX) is correct and the > 1 value is spurious. (Since
1686 * these are specially handled to never be considered UTF-8
1687 * locales, as long as this is the only problem, everything
1688 * should work fine */
1689 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1690 {
8c6180a9
KW
1691 multi_byte_locale = TRUE;
1692 }
7d4bcc4a
KW
1693
1694# endif
8c6180a9 1695
8b7358b9
KW
1696 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1697 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1698 PL_warn_locale = Perl_newSVpvf(aTHX_
1699 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1700 " which have\nunexpected meanings: %s\nThe Perl program"
1701 " will use the expected meanings",
8b7358b9 1702 newctype, bad_chars_list);
8b7358b9
KW
1703 }
1704 else {
1705 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1706 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1707 newctype,
8c6180a9
KW
1708 (multi_byte_locale)
1709 ? " Some characters in it are not recognized by"
1710 " Perl."
1711 : "",
1712 (bad_count)
1713 ? "\nThe following characters (and maybe others)"
1714 " may not have the same meaning as the Perl"
1715 " program expects:\n"
1716 : "",
1717 (bad_count)
1718 ? bad_chars_list
1719 : ""
1720 );
8b7358b9
KW
1721 }
1722
b119c2be
KW
1723# ifdef HAS_NL_LANGINFO
1724
1725 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1726 /* parameter FALSE is a don't care here */
4e6826bf 1727 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1728
1729# endif
1730
8b7358b9
KW
1731 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1732
cc9eaeb0 1733 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1734 * output the message now. If not in that scope, we save the
1735 * message to be output at the first operation using this locale,
1736 * if that actually happens. Most programs don't use locales, so
1737 * they are immune to bad ones. */
cc9eaeb0 1738 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1739
780fcc9f
KW
1740 /* The '0' below suppresses a bogus gcc compiler warning */
1741 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1742
bddebb56
KW
1743 if (IN_LC(LC_CTYPE)) {
1744 SvREFCNT_dec_NN(PL_warn_locale);
1745 PL_warn_locale = NULL;
1746 }
780fcc9f 1747 }
baa60164 1748 }
31f05a37 1749 }
98994639
HS
1750
1751#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1752
98994639
HS
1753}
1754
98994639 1755void
2726666d
KW
1756Perl__warn_problematic_locale()
1757{
2726666d
KW
1758
1759#ifdef USE_LOCALE_CTYPE
1760
5f04a188
KW
1761 dTHX;
1762
1763 /* Internal-to-core function that outputs the message in PL_warn_locale,
1764 * and then NULLS it. Should be called only through the macro
1765 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1766
2726666d 1767 if (PL_warn_locale) {
2726666d
KW
1768 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1769 SvPVX(PL_warn_locale),
1770 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1771 SvREFCNT_dec_NN(PL_warn_locale);
1772 PL_warn_locale = NULL;
1773 }
1774
1775#endif
1776
1777}
1778
a4f00dcc
KW
1779STATIC void
1780S_new_collate(pTHX_ const char *newcoll)
98994639 1781{
7d4bcc4a
KW
1782
1783#ifndef USE_LOCALE_COLLATE
1784
1785 PERL_UNUSED_ARG(newcoll);
1786 PERL_UNUSED_CONTEXT;
1787
1788#else
0d071d52 1789
291a84fb 1790 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1791 * core Perl this and that 'newcoll' is the name of the new locale.
1792 *
d35fca5f
KW
1793 * The design of locale collation is that every locale change is given an
1794 * index 'PL_collation_ix'. The first time a string particpates in an
1795 * operation that requires collation while locale collation is active, it
1796 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1797 * magic includes the collation index, and the transformation of the string
1798 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1799 * instead of the string itself. If a string changes, the magic is
1800 * cleared. The next time the locale changes, the index is incremented,
1801 * and so we know during a comparison that the transformation is not
1802 * necessarily still valid, and so is recomputed. Note that if the locale
1803 * changes enough times, the index could wrap (a U32), and it is possible
1804 * that a transformation would improperly be considered valid, leading to
1805 * an unlikely bug */
0d071d52 1806
98994639
HS
1807 if (! newcoll) {
1808 if (PL_collation_name) {
1809 ++PL_collation_ix;
1810 Safefree(PL_collation_name);
1811 PL_collation_name = NULL;
1812 }
1813 PL_collation_standard = TRUE;
00bf60ca 1814 is_standard_collation:
98994639
HS
1815 PL_collxfrm_base = 0;
1816 PL_collxfrm_mult = 2;
165a1c52 1817 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1818 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1819 PL_strxfrm_max_cp = 0;
98994639
HS
1820 return;
1821 }
1822
d35fca5f 1823 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1824 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1825 ++PL_collation_ix;
1826 Safefree(PL_collation_name);
1827 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1828 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1829 if (PL_collation_standard) {
1830 goto is_standard_collation;
1831 }
98994639 1832
165a1c52 1833 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1834 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1835 PL_strxfrm_max_cp = 0;
165a1c52 1836
59c018b9
KW
1837 /* A locale collation definition includes primary, secondary, tertiary,
1838 * etc. weights for each character. To sort, the primary weights are
1839 * used, and only if they compare equal, then the secondary weights are
1840 * used, and only if they compare equal, then the tertiary, etc.
1841 *
1842 * strxfrm() works by taking the input string, say ABC, and creating an
1843 * output transformed string consisting of first the primary weights,
1844 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1845 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1846 * may not have weights at every level. In our example, let's say B
1847 * doesn't have a tertiary weight, and A doesn't have a secondary
1848 * weight. The constructed string is then going to be
1849 * A¹B¹C¹ B²C² A³C³ ....
1850 * This has the desired effect that strcmp() will look at the secondary
1851 * or tertiary weights only if the strings compare equal at all higher
1852 * priority weights. The spaces shown here, like in
c342d20e 1853 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1854 * are not just for readability. In the general case, these must
1855 * actually be bytes, which we will call here 'separator weights'; and
1856 * they must be smaller than any other weight value, but since these
1857 * are C strings, only the terminating one can be a NUL (some
1858 * implementations may include a non-NUL separator weight just before
1859 * the NUL). Implementations tend to reserve 01 for the separator
1860 * weights. They are needed so that a shorter string's secondary
1861 * weights won't be misconstrued as primary weights of a longer string,
1862 * etc. By making them smaller than any other weight, the shorter
1863 * string will sort first. (Actually, if all secondary weights are
1864 * smaller than all primary ones, there is no need for a separator
1865 * weight between those two levels, etc.)
1866 *
1867 * The length of the transformed string is roughly a linear function of
1868 * the input string. It's not exactly linear because some characters
1869 * don't have weights at all levels. When we call strxfrm() we have to
1870 * allocate some memory to hold the transformed string. The
1871 * calculations below try to find coefficients 'm' and 'b' for this
1872 * locale so that m*x + b equals how much space we need, given the size
1873 * of the input string in 'x'. If we calculate too small, we increase
1874 * the size as needed, and call strxfrm() again, but it is better to
1875 * get it right the first time to avoid wasted expensive string
1876 * transformations. */
1877
98994639 1878 {
79f120c8
KW
1879 /* We use the string below to find how long the tranformation of it
1880 * is. Almost all locales are supersets of ASCII, or at least the
1881 * ASCII letters. We use all of them, half upper half lower,
1882 * because if we used fewer, we might hit just the ones that are
1883 * outliers in a particular locale. Most of the strings being
1884 * collated will contain a preponderance of letters, and even if
1885 * they are above-ASCII, they are likely to have the same number of
1886 * weight levels as the ASCII ones. It turns out that digits tend
1887 * to have fewer levels, and some punctuation has more, but those
1888 * are relatively sparse in text, and khw believes this gives a
1889 * reasonable result, but it could be changed if experience so
1890 * dictates. */
1891 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1892 char * x_longer; /* Transformed 'longer' */
1893 Size_t x_len_longer; /* Length of 'x_longer' */
1894
1895 char * x_shorter; /* We also transform a substring of 'longer' */
1896 Size_t x_len_shorter;
1897
a4a439fb 1898 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1899 * are interested only in its length). It is used because it has
1900 * the intelligence to handle all cases, but to work, it needs some
1901 * values of 'm' and 'b' to get it started. For the purposes of
1902 * this calculation we use a very conservative estimate of 'm' and
1903 * 'b'. This assumes a weight can be multiple bytes, enough to
1904 * hold any UV on the platform, and there are 5 levels, 4 weight
1905 * bytes, and a trailing NUL. */
1906 PL_collxfrm_base = 5;
1907 PL_collxfrm_mult = 5 * sizeof(UV);
1908
1909 /* Find out how long the transformation really is */
a4a439fb
KW
1910 x_longer = _mem_collxfrm(longer,
1911 sizeof(longer) - 1,
1912 &x_len_longer,
1913
1914 /* We avoid converting to UTF-8 in the
1915 * called function by telling it the
1916 * string is in UTF-8 if the locale is a
1917 * UTF-8 one. Since the string passed
1918 * here is invariant under UTF-8, we can
1919 * claim it's UTF-8 even though it isn't.
1920 * */
1921 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1922 Safefree(x_longer);
1923
1924 /* Find out how long the transformation of a substring of 'longer'
1925 * is. Together the lengths of these transformations are
1926 * sufficient to calculate 'm' and 'b'. The substring is all of
1927 * 'longer' except the first character. This minimizes the chances
1928 * of being swayed by outliers */
a4a439fb 1929 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1930 sizeof(longer) - 2,
a4a439fb
KW
1931 &x_len_shorter,
1932 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1933 Safefree(x_shorter);
1934
1935 /* If the results are nonsensical for this simple test, the whole
1936 * locale definition is suspect. Mark it so that locale collation
1937 * is not active at all for it. XXX Should we warn? */
1938 if ( x_len_shorter == 0
1939 || x_len_longer == 0
1940 || x_len_shorter >= x_len_longer)
1941 {
1942 PL_collxfrm_mult = 0;
1943 PL_collxfrm_base = 0;
1944 }
1945 else {
1946 SSize_t base; /* Temporary */
1947
1948 /* We have both: m * strlen(longer) + b = x_len_longer
1949 * m * strlen(shorter) + b = x_len_shorter;
1950 * subtracting yields:
1951 * m * (strlen(longer) - strlen(shorter))
1952 * = x_len_longer - x_len_shorter
1953 * But we have set things up so that 'shorter' is 1 byte smaller
1954 * than 'longer'. Hence:
1955 * m = x_len_longer - x_len_shorter
1956 *
1957 * But if something went wrong, make sure the multiplier is at
1958 * least 1.
1959 */
1960 if (x_len_longer > x_len_shorter) {
1961 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1962 }
1963 else {
1964 PL_collxfrm_mult = 1;
1965 }
1966
1967 /* mx + b = len
1968 * so: b = len - mx
1969 * but in case something has gone wrong, make sure it is
1970 * non-negative */
1971 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1972 if (base < 0) {
1973 base = 0;
1974 }
1975
1976 /* Add 1 for the trailing NUL */
1977 PL_collxfrm_base = base + 1;
1978 }
58eebef2 1979
7d4bcc4a
KW
1980# ifdef DEBUGGING
1981
58eebef2
KW
1982 if (DEBUG_L_TEST || debug_initialization) {
1983 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1984 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1985 "x_len_longer=%zu,"
1986 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1987 __FILE__, __LINE__,
1988 PL_in_utf8_COLLATE_locale,
1989 x_len_shorter, x_len_longer,
1990 PL_collxfrm_mult, PL_collxfrm_base);
1991 }
7d4bcc4a
KW
1992# endif
1993
98994639
HS
1994 }
1995 }
1996
1997#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1998
98994639
HS
1999}
2000
d36adde0
KW
2001#endif
2002
d2b24094 2003#ifdef WIN32
b8cc575c 2004
a4f00dcc 2005STATIC char *
b8cc575c 2006S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2007{
2008 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2009 * difference between the two unless the input locale is "", which normally
2010 * means on Windows to get the machine default, which is set via the
2011 * computer's "Regional and Language Options" (or its current equivalent).
2012 * In POSIX, it instead means to find the locale from the user's
2013 * environment. This routine changes the Windows behavior to first look in
2014 * the environment, and, if anything is found, use that instead of going to
2015 * the machine default. If there is no environment override, the machine
2016 * default is used, by calling the real setlocale() with "".
2017 *
2018 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2019 * use the particular category's variable if set; otherwise to use the LANG
2020 * variable. */
b385bb4d 2021
175c4cf9 2022 bool override_LC_ALL = FALSE;
89f7b9aa 2023 char * result;
e5f10d49 2024 unsigned int i;
89f7b9aa 2025
b385bb4d 2026 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2027
2028# ifdef LC_ALL
2029
b385bb4d
KW
2030 locale = PerlEnv_getenv("LC_ALL");
2031 if (! locale) {
e5f10d49
KW
2032 if (category == LC_ALL) {
2033 override_LC_ALL = TRUE;
2034 }
2035 else {
7d4bcc4a
KW
2036
2037# endif
7d4bcc4a 2038
e5f10d49
KW
2039 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2040 if (category == categories[i]) {
2041 locale = PerlEnv_getenv(category_names[i]);
2042 goto found_locale;
2043 }
2044 }
7d4bcc4a 2045
b385bb4d 2046 locale = PerlEnv_getenv("LANG");
481465ea 2047 if (! locale) {
b385bb4d
KW
2048 locale = "";
2049 }
e5f10d49
KW
2050
2051 found_locale: ;
7d4bcc4a
KW
2052
2053# ifdef LC_ALL
2054
e5f10d49 2055 }
b385bb4d 2056 }
7d4bcc4a
KW
2057
2058# endif
2059
b385bb4d
KW
2060 }
2061
89f7b9aa 2062 result = setlocale(category, locale);
48015184
KW
2063 DEBUG_L(STMT_START {
2064 dSAVE_ERRNO;
2065 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2066 setlocale_debug_string(category, locale, result));
2067 RESTORE_ERRNO;
2068 } STMT_END);
89f7b9aa 2069
481465ea 2070 if (! override_LC_ALL) {
89f7b9aa
KW
2071 return result;
2072 }
2073
dfd77d7a 2074 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2075 * LANG variable or the system default if there is no LANG. But these have
2076 * lower priority than the other LC_foo variables, so override it for each
2077 * one that is set. (If they are set to "", it means to use the same thing
2078 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2079
948523db 2080 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2081 result = PerlEnv_getenv(category_names[i]);
2082 if (result && strNE(result, "")) {
2083 setlocale(categories[i], result);
2084 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2085 __FILE__, __LINE__,
2086 setlocale_debug_string(categories[i], result, "not captured")));
2087 }
89f7b9aa 2088 }
7d4bcc4a 2089
bbc98134 2090 result = setlocale(LC_ALL, NULL);
48015184
KW
2091 DEBUG_L(STMT_START {
2092 dSAVE_ERRNO;
2093 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2094 __FILE__, __LINE__,
48015184
KW
2095 setlocale_debug_string(LC_ALL, NULL, result));
2096 RESTORE_ERRNO;
2097 } STMT_END);
89f7b9aa 2098
bbc98134 2099 return result;
b385bb4d
KW
2100}
2101
2102#endif
2103
9aac5db8
KW
2104/*
2105
2106=head1 Locale-related functions and macros
2107
2108=for apidoc Perl_setlocale
2109
2110This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2111taking the same parameters, and returning the same information, except that it
9487427b
KW
2112returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2113instead return C<C> if the underlying locale has a non-dot decimal point
2114character, or a non-empty thousands separator for displaying floating point
2115numbers. This is because perl keeps that locale category such that it has a
2116dot and empty separator, changing the locale briefly during the operations
2117where the underlying one is required. C<Perl_setlocale> knows about this, and
2118compensates; regular C<setlocale> doesn't.
9aac5db8 2119
e9bc6d6b 2120Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2121declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2122C<const> (presumably because its API was specified long ago, and can't be
2123updated; it is illegal to change the information C<setlocale> returns; doing
2124so leads to segfaults.)
9aac5db8 2125
e9bc6d6b
KW
2126Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2127C<setlocale> can be completely ineffective on some platforms under some
2128configurations.
2129
9aac5db8 2130C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2131where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2132the system C<setlocale()> is ineffective, returning the wrong information, and
2133failing to actually change the locale. C<Perl_setlocale>, however works
2134properly in all circumstances.
9aac5db8
KW
2135
2136The return points to a per-thread static buffer, which is overwritten the next
2137time C<Perl_setlocale> is called from the same thread.
2138
2139=cut
2140
2141*/
2142
2143const char *
2144Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2145{
2146 /* This wraps POSIX::setlocale() */
2147
d36adde0
KW
2148#ifdef NO_LOCALE
2149
2150 PERL_UNUSED_ARG(category);
2151 PERL_UNUSED_ARG(locale);
2152
2153 return "C";
2154
2155#else
2156
9aac5db8
KW
2157 const char * retval;
2158 const char * newlocale;
2159 dSAVEDERRNO;
a4f00dcc 2160 dTHX;
d36adde0 2161 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2162
a4f00dcc
KW
2163#ifdef USE_LOCALE_NUMERIC
2164
291a84fb
KW
2165 /* A NULL locale means only query what the current one is. We have the
2166 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2167 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2168 * the correct results. All other categories don't require special
2169 * handling */
a4f00dcc
KW
2170 if (locale == NULL) {
2171 if (category == LC_NUMERIC) {
9aac5db8
KW
2172
2173 /* We don't have to copy this return value, as it is a per-thread
2174 * variable, and won't change until a future setlocale */
2175 return PL_numeric_name;
a4f00dcc
KW
2176 }
2177
7d4bcc4a 2178# ifdef LC_ALL
a4f00dcc 2179
9aac5db8
KW
2180 else if (category == LC_ALL) {
2181 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2182 }
2183
7d4bcc4a 2184# endif
a4f00dcc
KW
2185
2186 }
2187
2188#endif
2189
33e5a354
KW
2190 retval = save_to_buffer(do_setlocale_r(category, locale),
2191 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2192 SAVE_ERRNO;
2193
2194#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2195
2196 if (locale == NULL && category == LC_ALL) {
2197 RESTORE_LC_NUMERIC();
2198 }
2199
2200#endif
a4f00dcc
KW
2201
2202 DEBUG_L(PerlIO_printf(Perl_debug_log,
2203 "%s:%d: %s\n", __FILE__, __LINE__,
2204 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2205
9aac5db8
KW
2206 RESTORE_ERRNO;
2207
2208 if (! retval) {
a4f00dcc
KW
2209 return NULL;
2210 }
2211
9aac5db8 2212 /* If locale == NULL, we are just querying the state */
a4f00dcc 2213 if (locale == NULL) {
a4f00dcc
KW
2214 return retval;
2215 }
a4f00dcc 2216
1159483a
KW
2217 /* Now that have switched locales, we have to update our records to
2218 * correspond. */
a4f00dcc 2219
1159483a 2220 switch (category) {
a4f00dcc 2221
1159483a 2222#ifdef USE_LOCALE_CTYPE
a4f00dcc 2223
1159483a
KW
2224 case LC_CTYPE:
2225 new_ctype(retval);
2226 break;
a4f00dcc 2227
1159483a 2228#endif
a4f00dcc
KW
2229#ifdef USE_LOCALE_COLLATE
2230
1159483a
KW
2231 case LC_COLLATE:
2232 new_collate(retval);
2233 break;
a4f00dcc 2234
1159483a 2235#endif
a4f00dcc
KW
2236#ifdef USE_LOCALE_NUMERIC
2237
1159483a
KW
2238 case LC_NUMERIC:
2239 new_numeric(retval);
2240 break;
a4f00dcc 2241
1159483a
KW
2242#endif
2243#ifdef LC_ALL
a4f00dcc 2244
1159483a 2245 case LC_ALL:
a4f00dcc 2246
1159483a
KW
2247 /* LC_ALL updates all the things we care about. The values may not
2248 * be the same as 'retval', as the locale "" may have set things
2249 * individually */
a4f00dcc 2250
1159483a 2251# ifdef USE_LOCALE_CTYPE
a4f00dcc 2252
b0bde642 2253 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2254 new_ctype(newlocale);
b0bde642 2255 Safefree(newlocale);
a4f00dcc 2256
1159483a
KW
2257# endif /* USE_LOCALE_CTYPE */
2258# ifdef USE_LOCALE_COLLATE
2259
b0bde642 2260 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2261 new_collate(newlocale);
b0bde642 2262 Safefree(newlocale);
a4f00dcc 2263
7d4bcc4a 2264# endif
1159483a 2265# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2266
b0bde642 2267 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2268 new_numeric(newlocale);
b0bde642 2269 Safefree(newlocale);
a4f00dcc 2270
1159483a
KW
2271# endif /* USE_LOCALE_NUMERIC */
2272#endif /* LC_ALL */
a4f00dcc 2273
1159483a
KW
2274 default:
2275 break;
a4f00dcc
KW
2276 }
2277
2278 return retval;
2279
d36adde0
KW
2280#endif
2281
f7416781
KW
2282}
2283
2284PERL_STATIC_INLINE const char *
2285S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2286{
2287 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2288 * growing it if necessary */
2289
0b6697c5 2290 Size_t string_size;
f7416781
KW
2291
2292 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2293
0b6697c5
KW
2294 if (! string) {
2295 return NULL;
2296 }
2297
2298 string_size = strlen(string) + offset + 1;
2299
f7416781
KW
2300 if (*buf_size == 0) {
2301 Newx(*buf, string_size, char);
2302 *buf_size = string_size;
2303 }
2304 else if (string_size > *buf_size) {
2305 Renew(*buf, string_size, char);
2306 *buf_size = string_size;
2307 }
2308
2309 Copy(string, *buf + offset, string_size - offset, char);
2310 return *buf;
2311}
2312
2313/*
2314
f7416781
KW
2315=for apidoc Perl_langinfo
2316
ce181bf6 2317This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2318taking the same C<item> parameter values, and returning the same information.
2319But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2320of Perl's locale handling from your code, and can be used on systems that lack
2321a native C<nl_langinfo>.
2322
2323Expanding on these:
2324
2325=over
2326
2327=item *
2328
ce181bf6
KW
2329The reason it isn't quite a drop-in replacement is actually an advantage. The
2330only difference is that it returns S<C<const char *>>, whereas plain
2331C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2332forbidden to write into the buffer. By declaring this C<const>, the compiler
2333enforces this restriction, so if it is violated, you know at compilation time,
2334rather than getting segfaults at runtime.
2335
2336=item *
2337
69e2275e 2338It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2339without you having to write extra code. The reason for the extra code would be
2340because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2341kept set by Perl so that the radix is a dot, and the separator is the empty
2342string, no matter what the underlying locale is supposed to be, and so to get
2343the expected results, you have to temporarily toggle into the underlying
2344locale, and later toggle back. (You could use plain C<nl_langinfo> and
2345C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2346the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2347(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2348(decimal point) character to be a dot.)
ce181bf6
KW
2349
2350=item *
2351
2352The system function it replaces can have its static return buffer trashed,
2353not only by a subesequent call to that function, but by a C<freelocale>,
2354C<setlocale>, or other locale change. The returned buffer of this function is
2355not changed until the next call to it, so the buffer is never in a trashed
2356state.
f7416781
KW
2357
2358=item *
2359
ce181bf6
KW
2360Its return buffer is per-thread, so it also is never overwritten by a call to
2361this function from another thread; unlike the function it replaces.
2362
2363=item *
2364
2365But most importantly, it works on systems that don't have C<nl_langinfo>, such
2366as Windows, hence makes your code more portable. Of the fifty-some possible
2367items specified by the POSIX 2008 standard,
f7416781 2368L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2369only one is completely unimplemented, though on non-Windows platforms, another
2370significant one is also not implemented). It uses various techniques to
2371recover the other items, including calling C<L<localeconv(3)>>, and
2372C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2373available. Later C<strftime()> versions have additional capabilities; C<""> is
2374returned for those not available on your system.
ce181bf6
KW
2375
2376It is important to note that when called with an item that is recovered by
2377using C<localeconv>, the buffer from any previous explicit call to
2378C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2379contents if you need to access them after a call to this function. (But note
2380that you might not want to be using C<localeconv()> directly anyway, because of
2381issues like the ones listed in the second item of this list (above) for
2382C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2383call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2384unpack).
9f0bc911 2385
6201c220
KW
2386The details for those items which may deviate from what this emulation returns
2387and what a native C<nl_langinfo()> would return are specified in
2388L<I18N::Langinfo>.
f7416781 2389
ce181bf6
KW
2390=back
2391
f7416781
KW
2392When using C<Perl_langinfo> on systems that don't have a native
2393C<nl_langinfo()>, you must
2394
2395 #include "perl_langinfo.h"
2396
2397before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2398C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2399C<langinfo.h> would try to import into the namespace for code that doesn't need
2400it.)
f7416781 2401
f7416781
KW
2402The original impetus for C<Perl_langinfo()> was so that code that needs to
2403find out the current currency symbol, floating point radix character, or digit
2404grouping separator can use, on all systems, the simpler and more
2405thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2406pain to make thread-friendly. For other fields returned by C<localeconv>, it
2407is better to use the methods given in L<perlcall> to call
2408L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2409
2410=cut
2411
2412*/
2413
2414const char *
2415#ifdef HAS_NL_LANGINFO
2416Perl_langinfo(const nl_item item)
2417#else
2418Perl_langinfo(const int item)
2419#endif
2420{
f61748ac
KW
2421 return my_nl_langinfo(item, TRUE);
2422}
2423
59a15af0 2424STATIC const char *
f61748ac
KW
2425#ifdef HAS_NL_LANGINFO
2426S_my_nl_langinfo(const nl_item item, bool toggle)
2427#else
2428S_my_nl_langinfo(const int item, bool toggle)
2429#endif
2430{
ae74815b 2431 dTHX;
0b6697c5 2432 const char * retval;
f7416781 2433
d36adde0
KW
2434#ifdef USE_LOCALE_NUMERIC
2435
5a854ab3
KW
2436 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2437 * two items, and only if not already there */
4e6826bf 2438 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2439 || PL_numeric_underlying))
d36adde0
KW
2440
2441#endif /* No toggling needed if not using LC_NUMERIC */
2442
5a854ab3 2443 toggle = FALSE;
5a854ab3 2444
ab340fff 2445#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2446# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2447 || ! defined(HAS_POSIX_2008_LOCALE) \
2448 || ! defined(DUPLOCALE)
f7416781 2449
ab340fff 2450 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2451 * for those items dependent on it. This must be copied to a buffer before
2452 * switching back, as some systems destroy the buffer when setlocale() is
2453 * called */
f7416781 2454
038d3702
KW
2455 {
2456 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2457
b2fee59c 2458 if (toggle) {
038d3702 2459 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2460 }
f7416781 2461
5acc4454
KW
2462 LOCALE_LOCK; /* Prevent interference from another thread executing
2463 this code section (the only call to nl_langinfo in
2464 the core) */
2465
ee90eb2d 2466
628ff75a
KW
2467 /* Copy to a per-thread buffer, which is also one that won't be
2468 * destroyed by a subsequent setlocale(), such as the
2469 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2470 retval = save_to_buffer(nl_langinfo(item),
2471 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2472
5acc4454
KW
2473 LOCALE_UNLOCK;
2474
b2fee59c 2475 if (toggle) {
038d3702 2476 RESTORE_LC_NUMERIC();
b2fee59c 2477 }
f7416781
KW
2478 }
2479
ab340fff
KW
2480# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2481
5a854ab3 2482 {
b2fee59c
KW
2483 bool do_free = FALSE;
2484 locale_t cur = uselocale((locale_t) 0);
ab340fff 2485
b2fee59c
KW
2486 if (cur == LC_GLOBAL_LOCALE) {
2487 cur = duplocale(LC_GLOBAL_LOCALE);
2488 do_free = TRUE;
2489 }
ab340fff 2490
d36adde0
KW
2491# ifdef USE_LOCALE_NUMERIC
2492
b2fee59c 2493 if (toggle) {
e1aa2579
KW
2494 if (PL_underlying_numeric_obj) {
2495 cur = PL_underlying_numeric_obj;
2496 }
2497 else {
2498 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2499 do_free = TRUE;
2500 }
b2fee59c 2501 }
ab340fff 2502
d36adde0
KW
2503# endif
2504
628ff75a
KW
2505 /* We have to save it to a buffer, because the freelocale() just below
2506 * can invalidate the internal one */
0b6697c5
KW
2507 retval = save_to_buffer(nl_langinfo_l(item, cur),
2508 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2509
b2fee59c
KW
2510 if (do_free) {
2511 freelocale(cur);
2512 }
5a854ab3 2513 }
ab340fff 2514
c1566110
KW
2515# endif
2516
0b6697c5 2517 if (strEQ(retval, "")) {
4e6826bf 2518 if (item == YESSTR) {
c1566110
KW
2519 return "yes";
2520 }
4e6826bf 2521 if (item == NOSTR) {
c1566110
KW
2522 return "no";
2523 }
2524 }
2525
0b6697c5 2526 return retval;
ab340fff 2527
f7416781 2528#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2529
2530 {
2531
f7416781
KW
2532# ifdef HAS_LOCALECONV
2533
43dd6b15 2534 const struct lconv* lc;
628ff75a 2535 const char * temp;
038d3702 2536 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2537
69c5e0db
KW
2538# ifdef TS_W32_BROKEN_LOCALECONV
2539
2540 const char * save_global;
2541 const char * save_thread;
2542 int needed_size;
2543 char * ptr;
2544 char * e;
2545 char * item_start;
2546
2547# endif
f7416781
KW
2548# endif
2549# ifdef HAS_STRFTIME
2550
43dd6b15
KW
2551 struct tm tm;
2552 bool return_format = FALSE; /* Return the %format, not the value */
2553 const char * format;
f7416781
KW
2554
2555# endif
2556
43dd6b15
KW
2557 /* We copy the results to a per-thread buffer, even if not
2558 * multi-threaded. This is in part to simplify this code, and partly
2559 * because we need a buffer anyway for strftime(), and partly because a
2560 * call of localeconv() could otherwise wipe out the buffer, and the
2561 * programmer would not be expecting this, as this is a nl_langinfo()
2562 * substitute after all, so s/he might be thinking their localeconv()
2563 * is safe until another localeconv() call. */
f7416781 2564
43dd6b15
KW
2565 switch (item) {
2566 Size_t len;
f7416781 2567
8d72e74e 2568 /* This is unimplemented */
4e6826bf 2569 case ERA: /* For use with strftime() %E modifier */
f7416781 2570
43dd6b15
KW
2571 default:
2572 return "";
f7416781 2573
43dd6b15 2574 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2575 case YESEXPR: return "^[+1yY]";
2576 case YESSTR: return "yes";
2577 case NOEXPR: return "^[-0nN]";
2578 case NOSTR: return "no";
2579
8d72e74e
KW
2580 case CODESET:
2581
2582# ifndef WIN32
2583
2584 /* On non-windows, this is unimplemented, in part because of
2585 * inconsistencies between vendors. The Darwin native
2586 * nl_langinfo() implementation simply looks at everything past
2587 * any dot in the name, but that doesn't work for other
2588 * vendors. Many Linux locales that don't have UTF-8 in their
2589 * names really are UTF-8, for example; z/OS locales that do
2590 * have UTF-8 in their names, aren't really UTF-8 */
2591 return "";
2592
2593# else
2594
2595 { /* But on Windows, the name does seem to be consistent, so
2596 use that. */
2597 const char * p;
2598 const char * first;
2599 Size_t offset = 0;
2600 const char * name = my_setlocale(LC_CTYPE, NULL);
2601
2602 if (isNAME_C_OR_POSIX(name)) {
2603 return "ANSI_X3.4-1968";
2604 }
2605
2606 /* Find the dot in the locale name */
2607 first = (const char *) strchr(name, '.');
2608 if (! first) {
2609 first = name;
2610 goto has_nondigit;
2611 }
2612
2613 /* Look at everything past the dot */
2614 first++;
2615 p = first;
2616
2617 while (*p) {
2618 if (! isDIGIT(*p)) {
2619 goto has_nondigit;
2620 }
2621
2622 p++;
2623 }
2624
2625 /* Here everything past the dot is a digit. Treat it as a
2626 * code page */
18503cbd 2627 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2628 &PL_langinfo_bufsize, 0);
8d72e74e 2629 offset = STRLENs("CP");
f7416781 2630
8d72e74e
KW
2631 has_nondigit:
2632
2633 retval = save_to_buffer(first, &PL_langinfo_buf,
2634 &PL_langinfo_bufsize, offset);
2635 }
2636
2637 break;
2638
2639# endif
f7416781
KW
2640# ifdef HAS_LOCALECONV
2641
4e6826bf 2642 case CRNCYSTR:
f7416781 2643
43dd6b15
KW
2644 /* We don't bother with localeconv_l() because any system that
2645 * has it is likely to also have nl_langinfo() */
291a84fb 2646
69c5e0db
KW
2647 LOCALE_LOCK_V; /* Prevent interference with other threads
2648 using localeconv() */
2649
2650# ifdef TS_W32_BROKEN_LOCALECONV
2651
2652 /* This is a workaround for a Windows bug prior to VS 15.
2653 * What we do here is, while locked, switch to the global
2654 * locale so localeconv() works; then switch back just before
2655 * the unlock. This can screw things up if some thread is
2656 * already using the global locale while assuming no other is.
2657 * A different workaround would be to call GetCurrencyFormat on
2658 * a known value, and parse it; patches welcome
2659 *
2660 * We have to use LC_ALL instead of LC_MONETARY because of
2661 * another bug in Windows */
2662
2663 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2664 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2665 save_global= savepv(my_setlocale(LC_ALL, NULL));
2666 my_setlocale(LC_ALL, save_thread);
2667
2668# endif
5acc4454 2669
43dd6b15
KW
2670 lc = localeconv();
2671 if ( ! lc
2672 || ! lc->currency_symbol
2673 || strEQ("", lc->currency_symbol))
2674 {
69c5e0db 2675 LOCALE_UNLOCK_V;
43dd6b15
KW
2676 return "";
2677 }
f7416781 2678
43dd6b15 2679 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2680 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2681 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2682 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2683 { /* khw couldn't figure out how the localedef specifications
2684 would show that the $ should replace the radix; this is
2685 just a guess as to how it might work.*/
a34edee3 2686 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2687 }
2688 else if (lc->p_cs_precedes) {
a34edee3 2689 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2690 }
2691 else {
a34edee3 2692 PL_langinfo_buf[0] = '+';
43dd6b15 2693 }
f7416781 2694
69c5e0db
KW
2695# ifdef TS_W32_BROKEN_LOCALECONV
2696
2697 my_setlocale(LC_ALL, save_global);
2698 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2699 my_setlocale(LC_ALL, save_thread);
2700 Safefree(save_global);
2701 Safefree(save_thread);
2702
2703# endif
2704
2705 LOCALE_UNLOCK_V;
43dd6b15 2706 break;
f7416781 2707
69c5e0db
KW
2708# ifdef TS_W32_BROKEN_LOCALECONV
2709
4e6826bf 2710 case RADIXCHAR:
69c5e0db
KW
2711
2712 /* For this, we output a known simple floating point number to
2713 * a buffer, and parse it, looking for the radix */
2714
2715 if (toggle) {
2716 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2717 }
2718
2719 if (PL_langinfo_bufsize < 10) {
2720 PL_langinfo_bufsize = 10;
2721 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2722 }
2723
2724 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2725 "%.1f", 1.5);
2726 if (needed_size >= (int) PL_langinfo_bufsize) {
2727 PL_langinfo_bufsize = needed_size + 1;
2728 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2729 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2730 "%.1f", 1.5);
2731 assert(needed_size < (int) PL_langinfo_bufsize);
2732 }
2733
2734 ptr = PL_langinfo_buf;
2735 e = PL_langinfo_buf + PL_langinfo_bufsize;
2736
2737 /* Find the '1' */
2738 while (ptr < e && *ptr != '1') {
2739 ptr++;
2740 }
2741 ptr++;
2742
2743 /* Find the '5' */
2744 item_start = ptr;
2745 while (ptr < e && *ptr != '5') {
2746 ptr++;
2747 }
2748
2749 /* Everything in between is the radix string */
2750 if (ptr >= e) {
2751 PL_langinfo_buf[0] = '?';
2752 PL_langinfo_buf[1] = '\0';
2753 }
2754 else {
2755 *ptr = '\0';
2756 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2757 }
2758
2759 if (toggle) {
2760 RESTORE_LC_NUMERIC();
2761 }
2762
2763 retval = PL_langinfo_buf;
2764 break;
2765
2766# else
2767
2768 case RADIXCHAR: /* No special handling needed */
2769
2770# endif
2771
4e6826bf 2772 case THOUSEP:
f7416781 2773
43dd6b15 2774 if (toggle) {
038d3702 2775 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2776 }
f7416781 2777
69c5e0db
KW
2778 LOCALE_LOCK_V; /* Prevent interference with other threads
2779 using localeconv() */
2780
2781# ifdef TS_W32_BROKEN_LOCALECONV
2782
2783 /* This should only be for the thousands separator. A
2784 * different work around would be to use GetNumberFormat on a
2785 * known value and parse the result to find the separator */
2786 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2787 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2788 save_global = savepv(my_setlocale(LC_ALL, NULL));
2789 my_setlocale(LC_ALL, save_thread);
2790# if 0
2791 /* This is the start of code that for broken Windows replaces
2792 * the above and below code, and instead calls
2793 * GetNumberFormat() and then would parse that to find the
2794 * thousands separator. It needs to handle UTF-16 vs -8
2795 * issues. */
2796
2797 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2798 DEBUG_L(PerlIO_printf(Perl_debug_log,
2799 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2800 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2801
2802# endif
2803# endif
5acc4454 2804
43dd6b15
KW
2805 lc = localeconv();
2806 if (! lc) {
628ff75a 2807 temp = "";
33394adc 2808 }
43dd6b15 2809 else {
4e6826bf 2810 temp = (item == RADIXCHAR)
43dd6b15
KW
2811 ? lc->decimal_point
2812 : lc->thousands_sep;
628ff75a
KW
2813 if (! temp) {
2814 temp = "";
43dd6b15
KW
2815 }
2816 }
f7416781 2817
0b6697c5
KW
2818 retval = save_to_buffer(temp, &PL_langinfo_buf,
2819 &PL_langinfo_bufsize, 0);
f7416781 2820
69c5e0db
KW
2821# ifdef TS_W32_BROKEN_LOCALECONV
2822
2823 my_setlocale(LC_ALL, save_global);
2824 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2825 my_setlocale(LC_ALL, save_thread);
2826 Safefree(save_global);
2827 Safefree(save_thread);
2828
2829# endif
2830
2831 LOCALE_UNLOCK_V;
5acc4454 2832
43dd6b15 2833 if (toggle) {
038d3702 2834 RESTORE_LC_NUMERIC();
43dd6b15 2835 }
f7416781 2836
43dd6b15 2837 break;
f7416781
KW
2838
2839# endif
2840# ifdef HAS_STRFTIME
2841
43dd6b15
KW
2842 /* These are defined by C89, so we assume that strftime supports
2843 * them, and so are returned unconditionally; they may not be what
2844 * the locale actually says, but should give good enough results
2845 * for someone using them as formats (as opposed to trying to parse
2846 * them to figure out what the locale says). The other format
2847 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2848 case D_FMT: return "%x";
2849 case T_FMT: return "%X";
2850 case D_T_FMT: return "%c";
43dd6b15
KW
2851
2852 /* These formats are only available in later strfmtime's */
4e6826bf 2853 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2854
2855 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2856 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2857 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2858 case ALT_DIGITS:
2859 case AM_STR: case PM_STR:
2860 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2861 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2862 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2863 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2864 case DAY_5: case DAY_6: case DAY_7:
2865 case MON_1: case MON_2: case MON_3: case MON_4:
2866 case MON_5: case MON_6: case MON_7: case MON_8:
2867 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2868
2869 LOCALE_LOCK;
2870
2871 init_tm(&tm); /* Precaution against core dumps */
2872 tm.tm_sec = 30;
2873 tm.tm_min = 30;
2874 tm.tm_hour = 6;
2875 tm.tm_year = 2017 - 1900;
2876 tm.tm_wday = 0;
2877 tm.tm_mon = 0;
2878 switch (item) {
2879 default:
2880 LOCALE_UNLOCK;
2881 Perl_croak(aTHX_
2882 "panic: %s: %d: switch case: %d problem",
2883 __FILE__, __LINE__, item);
2884 NOT_REACHED; /* NOTREACHED */
2885
4e6826bf
KW
2886 case PM_STR: tm.tm_hour = 18;
2887 case AM_STR:
43dd6b15
KW
2888 format = "%p";
2889 break;
2890
4e6826bf
KW
2891 case ABDAY_7: tm.tm_wday++;
2892 case ABDAY_6: tm.tm_wday++;
2893 case ABDAY_5: tm.tm_wday++;
2894 case ABDAY_4: tm.tm_wday++;
2895 case ABDAY_3: tm.tm_wday++;
2896 case ABDAY_2: tm.tm_wday++;
2897 case ABDAY_1:
43dd6b15
KW
2898 format = "%a";
2899 break;
2900
4e6826bf
KW
2901 case DAY_7: tm.tm_wday++;
2902 case DAY_6: tm.tm_wday++;
2903 case DAY_5: tm.tm_wday++;
2904 case DAY_4: tm.tm_wday++;
2905 case DAY_3: tm.tm_wday++;
2906 case DAY_2: tm.tm_wday++;
2907 case DAY_1:
43dd6b15
KW
2908 format = "%A";
2909 break;
2910
4e6826bf
KW
2911 case ABMON_12: tm.tm_mon++;
2912 case ABMON_11: tm.tm_mon++;
2913 case ABMON_10: tm.tm_mon++;
2914 case ABMON_9: tm.tm_mon++;
2915 case ABMON_8: tm.tm_mon++;
2916 case ABMON_7: tm.tm_mon++;
2917 case ABMON_6: tm.tm_mon++;
2918 case ABMON_5: tm.tm_mon++;
2919 case ABMON_4: tm.tm_mon++;
2920 case ABMON_3: tm.tm_mon++;
2921 case ABMON_2: tm.tm_mon++;
2922 case ABMON_1:
43dd6b15
KW
2923 format = "%b";
2924 break;
2925
4e6826bf
KW
2926 case MON_12: tm.tm_mon++;
2927 case MON_11: tm.tm_mon++;
2928 case MON_10: tm.tm_mon++;
2929 case MON_9: tm.tm_mon++;
2930 case MON_8: tm.tm_mon++;
2931 case MON_7: tm.tm_mon++;
2932 case MON_6: tm.tm_mon++;
2933 case MON_5: tm.tm_mon++;
2934 case MON_4: tm.tm_mon++;
2935 case MON_3: tm.tm_mon++;
2936 case MON_2: tm.tm_mon++;
2937 case MON_1:
43dd6b15
KW
2938 format = "%B";
2939 break;
2940
4e6826bf 2941 case T_FMT_AMPM:
43dd6b15
KW
2942 format = "%r";
2943 return_format = TRUE;
2944 break;
2945
4e6826bf 2946 case ERA_D_FMT:
43dd6b15
KW
2947 format = "%Ex";
2948 return_format = TRUE;
2949 break;
2950
4e6826bf 2951 case ERA_T_FMT:
43dd6b15
KW
2952 format = "%EX";
2953 return_format = TRUE;
2954 break;
2955
4e6826bf 2956 case ERA_D_T_FMT:
43dd6b15
KW
2957 format = "%Ec";
2958 return_format = TRUE;
2959 break;
2960
4e6826bf 2961 case ALT_DIGITS:
43dd6b15
KW
2962 tm.tm_wday = 0;
2963 format = "%Ow"; /* Find the alternate digit for 0 */
2964 break;
2965 }
f7416781 2966
43dd6b15
KW
2967 /* We can't use my_strftime() because it doesn't look at
2968 * tm_wday */
2969 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2970 format, &tm))
2971 {
2972 /* A zero return means one of:
2973 * a) there wasn't enough space in PL_langinfo_buf
2974 * b) the format, like a plain %p, returns empty
2975 * c) it was an illegal format, though some
2976 * implementations of strftime will just return the
2977 * illegal format as a plain character sequence.
2978 *
2979 * To quickly test for case 'b)', try again but precede
2980 * the format with a plain character. If that result is
2981 * still empty, the problem is either 'a)' or 'c)' */
2982
2983 Size_t format_size = strlen(format) + 1;
2984 Size_t mod_size = format_size + 1;
2985 char * mod_format;
2986 char * temp_result;
2987
2988 Newx(mod_format, mod_size, char);
2989 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2990 *mod_format = ' ';
43dd6b15
KW
2991 my_strlcpy(mod_format + 1, format, mod_size);
2992 len = strftime(temp_result,
2993 PL_langinfo_bufsize,
2994 mod_format, &tm);
2995 Safefree(mod_format);
2996 Safefree(temp_result);
2997
2998 /* If 'len' is non-zero, it means that we had a case like
2999 * %p which means the current locale doesn't use a.m. or
3000 * p.m., and that is valid */
3001 if (len == 0) {
3002
3003 /* Here, still didn't work. If we get well beyond a
3004 * reasonable size, bail out to prevent an infinite
3005 * loop. */
3006
3007 if (PL_langinfo_bufsize > 100 * format_size) {
3008 *PL_langinfo_buf = '\0';
3009 }
3010 else {
3011 /* Double the buffer size to retry; Add 1 in case
3012 * original was 0, so we aren't stuck at 0. */
3013 PL_langinfo_bufsize *= 2;
3014 PL_langinfo_bufsize++;
3015 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3016 continue;
3017 }
3018 }
f7416781 3019
f7416781 3020 break;
43dd6b15 3021 }
f7416781 3022
43dd6b15
KW
3023 /* Here, we got a result.
3024 *
3025 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3026 * alternate format for wday 0. If the value is the same as
3027 * the normal 0, there isn't an alternate, so clear the buffer.
3028 * */
4e6826bf 3029 if ( item == ALT_DIGITS
43dd6b15
KW
3030 && strEQ(PL_langinfo_buf, "0"))
3031 {
3032 *PL_langinfo_buf = '\0';
3033 }
f7416781 3034
43dd6b15
KW
3035 /* ALT_DIGITS is problematic. Experiments on it showed that
3036 * strftime() did not always work properly when going from
3037 * alt-9 to alt-10. Only a few locales have this item defined,
3038 * and in all of them on Linux that khw was able to find,
3039 * nl_langinfo() merely returned the alt-0 character, possibly
3040 * doubled. Most Unicode digits are in blocks of 10
3041 * consecutive code points, so that is sufficient information
3042 * for those scripts, as we can infer alt-1, alt-2, .... But
3043 * for a Japanese locale, a CJK ideographic 0 is returned, and
3044 * the CJK digits are not in code point order, so you can't
3045 * really infer anything. The localedef for this locale did
3046 * specify the succeeding digits, so that strftime() works
3047 * properly on them, without needing to infer anything. But
3048 * the nl_langinfo() return did not give sufficient information
3049 * for the caller to understand what's going on. So until
3050 * there is evidence that it should work differently, this
3051 * returns the alt-0 string for ALT_DIGITS.
3052 *
3053 * wday was chosen because its range is all a single digit.
3054 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3055
43dd6b15 3056 LOCALE_UNLOCK;
f7416781 3057
d1b150d9
KW
3058 retval = PL_langinfo_buf;
3059
43dd6b15
KW
3060 /* If to return the format, not the value, overwrite the buffer
3061 * with it. But some strftime()s will keep the original format
3062 * if illegal, so change those to "" */
3063 if (return_format) {
3064 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3065 *PL_langinfo_buf = '\0';
3066 }
43dd6b15 3067 else {
0b6697c5
KW
3068 retval = save_to_buffer(format, &PL_langinfo_buf,
3069 &PL_langinfo_bufsize, 0);
f7416781
KW
3070 }
3071 }
3072
3073 break;
f7416781
KW
3074
3075# endif
3076
43dd6b15 3077 }
f7416781
KW
3078 }
3079
0b6697c5 3080 return retval;
f7416781
KW
3081
3082#endif
3083
a4f00dcc 3084}
b385bb4d 3085
98994639
HS
3086/*
3087 * Initialize locale awareness.
3088 */
3089int
3090Perl_init_i18nl10n(pTHX_ int printwarn)
3091{
0e92a118
KW
3092 /* printwarn is
3093 *
3094 * 0 if not to output warning when setup locale is bad
3095 * 1 if to output warning based on value of PERL_BADLANG
3096 * >1 if to output regardless of PERL_BADLANG
3097 *
3098 * returns
98994639 3099 * 1 = set ok or not applicable,
0e92a118
KW
3100 * 0 = fallback to a locale of lower priority
3101 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3102 *
3103 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3104 * set, debugging information is output.
3105 *
3106 * This looks more complicated than it is, mainly due to the #ifdefs.
3107 *
3108 * We try to set LC_ALL to the value determined by the environment. If
3109 * there is no LC_ALL on this platform, we try the individual categories we
3110 * know about. If this works, we are done.
3111 *
3112 * But if it doesn't work, we have to do something else. We search the
3113 * environment variables ourselves instead of relying on the system to do
3114 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3115 * think there is one), and the ultimate fallback "C". This is all done in
3116 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3117 * more complex. The 'trial_locales' array is initialized with just one
3118 * element; it causes the behavior described in the paragraph above this to
3119 * happen. If that fails, we add elements to 'trial_locales', and do extra
3120 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3121 *
3122 * On Ultrix, the locale MUST come from the environment, so there is
3123 * preliminary code to set it. I (khw) am not sure that it is necessary,
3124 * and that this couldn't be folded into the loop, but barring any real
3125 * platforms to test on, it's staying as-is
3126 *
3127 * A slight complication is that in embedded Perls, the locale may already
3128 * be set-up, and we don't want to get it from the normal environment
3129 * variables. This is handled by having a special environment variable
3130 * indicate we're in this situation. We simply set setlocale's 2nd
3131 * parameter to be a NULL instead of "". That indicates to setlocale that
3132 * it is not to change anything, but to return the current value,
3133 * effectively initializing perl's db to what the locale already is.
3134 *
3135 * We play the same trick with NULL if a LC_ALL succeeds. We call
3136 * setlocale() on the individual categores with NULL to get their existing
3137 * values for our db, instead of trying to change them.
3138 * */
98994639 3139
0e92a118
KW
3140 int ok = 1;
3141
7d4bcc4a
KW
3142#ifndef USE_LOCALE
3143
3144 PERL_UNUSED_ARG(printwarn);
3145
3146#else /* USE_LOCALE */
7d4bcc4a
KW
3147# ifdef __GLIBC__
3148
175c4cf9 3149 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3150
3151# endif
65ebb059 3152
ccd65d51
KW
3153 /* NULL uses the existing already set up locale */
3154 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3155 ? NULL
3156 : "";
c3fcd832
KW
3157 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3158 unsigned int trial_locales_count;
175c4cf9
KW
3159 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3160 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3161 bool setlocale_failure = FALSE;
65ebb059 3162 unsigned int i;
175c4cf9
KW
3163
3164 /* A later getenv() could zap this, so only use here */
3165 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3166
3167 const bool locwarn = (printwarn > 1
e5f10d49
KW
3168 || ( printwarn
3169 && ( ! bad_lang_use_once
22ff3130 3170 || (
e5f10d49
KW
3171 /* disallow with "" or "0" */
3172 *bad_lang_use_once
3173 && strNE("0", bad_lang_use_once)))));
ea92aad8 3174
291a84fb 3175 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3176 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3177
3178 /* current locale for given category; should have been copied so aren't
3179 * volatile */
8de4332b 3180 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3181
7d4bcc4a
KW
3182# ifdef WIN32
3183
6bce99ee
JH
3184 /* In some systems you can find out the system default locale
3185 * and use that as the fallback locale. */
7d4bcc4a
KW
3186# define SYSTEM_DEFAULT_LOCALE
3187# endif
3188# ifdef SYSTEM_DEFAULT_LOCALE
3189
65ebb059 3190 const char *system_default_locale = NULL;
98994639 3191
7d4bcc4a 3192# endif
948523db
KW
3193
3194# ifndef DEBUGGING
3195# define DEBUG_LOCALE_INIT(a,b,c)
3196# else
7d4bcc4a 3197
8298454c 3198 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3199
3200# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3201 STMT_START { \
3202 if (debug_initialization) { \
3203 PerlIO_printf(Perl_debug_log, \
3204 "%s:%d: %s\n", \
3205 __FILE__, __LINE__, \
a4f00dcc 3206 setlocale_debug_string(category, \
2fcc0ca9
KW
3207 locale, \
3208 result)); \
3209 } \
3210 } STMT_END
2fcc0ca9 3211
948523db
KW
3212/* Make sure the parallel arrays are properly set up */
3213# ifdef USE_LOCALE_NUMERIC
3214 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3215 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3216# ifdef USE_POSIX_2008_LOCALE
3217 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3218# endif
948523db
KW
3219# endif
3220# ifdef USE_LOCALE_CTYPE
3221 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3222 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3223# ifdef USE_POSIX_2008_LOCALE
3224 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3225# endif
948523db
KW
3226# endif
3227# ifdef USE_LOCALE_COLLATE
3228 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3229 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3230# ifdef USE_POSIX_2008_LOCALE
3231 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3232# endif
948523db
KW
3233# endif
3234# ifdef USE_LOCALE_TIME
3235 assert(categories[LC_TIME_INDEX] == LC_TIME);
3236 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3237# ifdef USE_POSIX_2008_LOCALE
3238 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3239# endif
948523db
KW
3240# endif
3241# ifdef USE_LOCALE_MESSAGES
3242 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3243 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3244# ifdef USE_POSIX_2008_LOCALE
3245 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3246# endif
948523db
KW
3247# endif
3248# ifdef USE_LOCALE_MONETARY
3249 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3250 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3251# ifdef USE_POSIX_2008_LOCALE
3252 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3253# endif
948523db 3254# endif
9821811f
KW
3255# ifdef USE_LOCALE_ADDRESS
3256 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3257 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3258# ifdef USE_POSIX_2008_LOCALE
3259 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3260# endif
9821811f
KW
3261# endif
3262# ifdef USE_LOCALE_IDENTIFICATION
3263 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3264 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3265# ifdef USE_POSIX_2008_LOCALE
3266 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3267# endif
9821811f
KW
3268# endif
3269# ifdef USE_LOCALE_MEASUREMENT
3270 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3271 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3272# ifdef USE_POSIX_2008_LOCALE
3273 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3274# endif
9821811f
KW
3275# endif
3276# ifdef USE_LOCALE_PAPER
3277 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3278 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3279# ifdef USE_POSIX_2008_LOCALE
3280 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3281# endif
9821811f
KW
3282# endif
3283# ifdef USE_LOCALE_TELEPHONE
3284 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3285 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3286# ifdef USE_POSIX_2008_LOCALE
3287 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3288# endif
9821811f 3289# endif
948523db
KW
3290# ifdef LC_ALL
3291 assert(categories[LC_ALL_INDEX] == LC_ALL);
3292 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3293 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3294# ifdef USE_POSIX_2008_LOCALE
3295 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3296# endif
948523db
KW
3297# endif
3298# endif /* DEBUGGING */
47280b20
KW
3299
3300 /* Initialize the cache of the program's UTF-8ness for the always known
3301 * locales C and POSIX */
3302 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3303 sizeof(PL_locale_utf8ness));
3304
e9bc6d6b
KW
3305# ifdef USE_THREAD_SAFE_LOCALE
3306# ifdef WIN32
3307
3308 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3309
3310# endif
3311# endif
39e69e77 3312# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3313
3314 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3315 if (! PL_C_locale_obj) {
3316 Perl_croak_nocontext(
3317 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3318 }
3319 if (DEBUG_Lv_TEST || debug_initialization) {
3320 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3321 }
3322
3323# endif
3324
78f7c09f
KW
3325# ifdef USE_LOCALE_NUMERIC
3326
3ca88433
KW
3327 PL_numeric_radix_sv = newSVpvs(".");
3328
78f7c09f
KW
3329# endif
3330
e9bc6d6b
KW
3331# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3332
3333 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3334 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3335
3336# endif
b56c4436 3337# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3338
3339 /*
3340 * Ultrix setlocale(..., "") fails if there are no environment
3341 * variables from which to get a locale name.
3342 */
3343
b56c4436
KW
3344# ifndef LC_ALL
3345# error Ultrix without LC_ALL not implemented
3346# else
7d4bcc4a 3347
b56c4436
KW
3348 {
3349 bool done = FALSE;
ea92aad8
KW
3350 if (lang) {
3351 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3352 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3353 if (sl_result[LC_ALL_INDEX])
3354 done = TRUE;
3355 else
e5f10d49 3356 setlocale_failure = TRUE;
ea92aad8
KW
3357 }
3358 if (! setlocale_failure) {
3359 const char * locale_param;
3360 for (i = 0; i < LC_ALL_INDEX; i++) {
3361 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3362 ? setlocale_init
3363 : NULL;
3364 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3365 if (! sl_result[i]) {
3366 setlocale_failure = TRUE;
3367 }
3368 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3369 }
c835d6be 3370 }
7d4bcc4a
KW
3371 }
3372
3373# endif /* LC_ALL */
e5f10d49 3374# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3375
65ebb059 3376 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3377 * the list. Normally the loop is executed just once. But if setting the
3378 * locale fails, inside the loop we add fallback trials to the array and so
3379 * will execute the loop multiple times */
c3fcd832
KW
3380 trial_locales[0] = setlocale_init;
3381 trial_locales_count = 1;
7d4bcc4a 3382
65ebb059
KW
3383 for (i= 0; i < trial_locales_count; i++) {
3384 const char * trial_locale = trial_locales[i];
3385
3386 if (i > 0) {
3387
3388 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3389 * when i==0, but I (khw) don't think that behavior makes much
3390 * sense */
3391 setlocale_failure = FALSE;
3392
7d4bcc4a 3393# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3394# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3395
65ebb059
KW
3396 /* On Windows machines, an entry of "" after the 0th means to use
3397 * the system default locale, which we now proceed to get. */
3398 if (strEQ(trial_locale, "")) {
3399 unsigned int j;
3400
3401 /* Note that this may change the locale, but we are going to do
3402 * that anyway just below */
837ce802 3403 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3404 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3405
7d4bcc4a 3406 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3407 * try */
3408 if (! system_default_locale) {
3409 goto next_iteration;
3410 }
3411 for (j = 0; j < trial_locales_count; j++) {
3412 if (strEQ(system_default_locale, trial_locales[j])) {
3413 goto next_iteration;
3414 }
3415 }
3416
3417 trial_locale = system_default_locale;
3418 }
ec0202b5
KW
3419# else
3420# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3421# endif
7d4bcc4a 3422# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3423
3424 } /* For i > 0 */
65ebb059 3425
7d4bcc4a
KW
3426# ifdef LC_ALL
3427
948523db
KW
3428 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3429 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3430 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3431 setlocale_failure = TRUE;
7cd8b568
KW
3432 }
3433 else {
3434 /* Since LC_ALL succeeded, it should have changed all the other
3435 * categories it can to its value; so we massage things so that the
3436 * setlocales below just return their category's current values.
3437 * This adequately handles the case in NetBSD where LC_COLLATE may
3438 * not be defined for a locale, and setting it individually will
7d4bcc4a 3439 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3440 * the POSIX locale. */
3441 trial_locale = NULL;
3442 }
7d4bcc4a
KW
3443
3444# endif /* LC_ALL */
98994639 3445
e5f10d49
KW
3446 if (! setlocale_failure) {
3447 unsigned int j;
3448 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3449 curlocales[j]
3450 = savepv(do_setlocale_r(categories[j], trial_locale));
3451 if (! curlocales[j]) {
3452 setlocale_failure = TRUE;
3453 }
3454 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3455 }
c835d6be 3456
e5f10d49
KW
3457 if (! setlocale_failure) { /* All succeeded */
3458 break; /* Exit trial_locales loop */
49c85077 3459 }
65ebb059 3460 }
98994639 3461
49c85077
KW
3462 /* Here, something failed; will need to try a fallback. */
3463 ok = 0;
65ebb059 3464
49c85077
KW
3465 if (i == 0) {
3466 unsigned int j;
98994639 3467
65ebb059 3468 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3469
3470# ifdef LC_ALL
98994639 3471
49c85077
KW
3472 PerlIO_printf(Perl_error_log,
3473 "perl: warning: Setting locale failed.\n");
98994639 3474
7d4bcc4a 3475# else /* !LC_ALL */
98994639 3476
49c85077
KW
3477 PerlIO_printf(Perl_error_log,
3478 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3479
e5f10d49
KW
3480 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3481 if (! curlocales[j]) {
3482 PerlIO_printf(Perl_error_log, category_names[j]);
3483 }
3484 else {
3485 Safefree(curlocales[j]);
3486 }
3487 }
7d4bcc4a 3488
7d4bcc4a 3489# endif /* LC_ALL */
98994639 3490
49c85077
KW
3491 PerlIO_printf(Perl_error_log,
3492 "perl: warning: Please check that your locale settings:\n");
98994639 3493
7d4bcc4a
KW
3494# ifdef __GLIBC__
3495
49c85077
KW
3496 PerlIO_printf(Perl_error_log,
3497 "\tLANGUAGE = %c%s%c,\n",
3498 language ? '"' : '(',
3499 language ? language : "unset",
3500 language ? '"' : ')');
7d4bcc4a 3501# endif
98994639 3502
49c85077
KW
3503 PerlIO_printf(Perl_error_log,
3504 "\tLC_ALL = %c%s%c,\n",
3505 lc_all ? '"' : '(',
3506 lc_all ? lc_all : "unset",
3507 lc_all ? '"' : ')');
98994639 3508
7d4bcc4a
KW
3509# if defined(USE_ENVIRON_ARRAY)
3510
49c85077 3511 {
cd999af9 3512 char **e;
d5e32b93
KW
3513
3514 /* Look through the environment for any variables of the
3515 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3516 * already handled above. These are assumed to be locale
3517 * settings. Output them and their values. */
cd999af9 3518 for (e = environ; *e; e++) {
d5e32b93
KW
3519 const STRLEN prefix_len = sizeof("LC_") - 1;
3520 STRLEN uppers_len;
3521
cd999af9 3522 if ( strBEGINs(*e, "LC_")
c8b388b0 3523 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3524 && (uppers_len = strspn(*e + prefix_len,
3525 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3526 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3527 {
3528 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3529 (int) (prefix_len + uppers_len), *e,
3530 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3531 }
3532 }
49c85077 3533 }
7d4bcc4a
KW
3534
3535# else
3536
49c85077
KW
3537 PerlIO_printf(Perl_error_log,
3538 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3539
3540# endif
98994639 3541
49c85077
KW
3542 PerlIO_printf(Perl_error_log,
3543 "\tLANG = %c%s%c\n",
3544 lang ? '"' : '(',
3545 lang ? lang : "unset",
3546 lang ? '"' : ')');
98994639 3547
49c85077
KW
3548 PerlIO_printf(Perl_error_log,
3549 " are supported and installed on your system.\n");
3550 }
98994639 3551
65ebb059 3552 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3553 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3554 * usually change the upper bound of the loop we are in.
3555 *
3556 * Since the system's default way of setting the locale has not
3557 * found one that works, We use Perl's defined ordering: LC_ALL,
3558 * LANG, and the C locale. We don't try the same locale twice, so
3559 * don't add to the list if already there. (On POSIX systems, the
3560 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3561 * but there's no harm done by doing it explicitly.
3562 *
3563 * Note that this tries the LC_ALL environment variable even on
3564 * systems which have no LC_ALL locale setting. This may or may
3565 * not have been originally intentional, but there's no real need
3566 * to change the behavior. */
65ebb059
KW
3567 if (lc_all) {
3568 for (j = 0; j < trial_locales_count; j++) {
3569 if (strEQ(lc_all, trial_locales[j])) {
3570 goto done_lc_all;
3571 }
3572 }
3573 trial_locales[trial_locales_count++] = lc_all;
3574 }
3575 done_lc_all:
98994639 3576
65ebb059
KW
3577 if (lang) {
3578 for (j = 0; j < trial_locales_count; j++) {
3579 if (strEQ(lang, trial_locales[j])) {
3580 goto done_lang;
3581 }
3582 }
3583 trial_locales[trial_locales_count++] = lang;
3584 }
3585 done_lang:
3586
7d4bcc4a
KW
3587# if defined(WIN32) && defined(LC_ALL)
3588
65ebb059
KW
3589 /* For Windows, we also try the system default locale before "C".
3590 * (If there exists a Windows without LC_ALL we skip this because
3591 * it gets too complicated. For those, the "C" is the next
3592 * fallback possibility). The "" is the same as the 0th element of
3593 * the array, but the code at the loop above knows to treat it
3594 * differently when not the 0th */
3595 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3596
3597# endif
65ebb059
KW
3598
3599 for (j = 0; j < trial_locales_count; j++) {
3600 if (strEQ("C", trial_locales[j])) {
3601 goto done_C;
3602 }
3603 }
3604 trial_locales[trial_locales_count++] = "C";
98994639 3605
65ebb059
KW
3606 done_C: ;
3607 } /* end of first time through the loop */
98994639 3608
7d4bcc4a
KW
3609# ifdef WIN32
3610
65ebb059 3611 next_iteration: ;
7d4bcc4a
KW
3612
3613# endif
65ebb059
KW
3614
3615 } /* end of looping through the trial locales */
3616
3617 if (ok < 1) { /* If we tried to fallback */
3618 const char* msg;
3619 if (! setlocale_failure) { /* fallback succeeded */
3620 msg = "Falling back to";
3621 }
3622 else { /* fallback failed */
e5f10d49 3623 unsigned int j;
98994639 3624
65ebb059
KW
3625 /* We dropped off the end of the loop, so have to decrement i to
3626 * get back to the value the last time through */
3627 i--;
98994639 3628
65ebb059
KW
3629 ok = -1;
3630 msg = "Failed to fall back to";
3631
3632 /* To continue, we should use whatever values we've got */
7d4bcc4a 3633
e5f10d49
KW
3634 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3635 Safefree(curlocales[j]);
3636 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3637 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3638 }
65ebb059
KW
3639 }
3640
3641 if (locwarn) {
3642 const char * description;
3643 const char * name = "";
3644 if (strEQ(trial_locales[i], "C")) {
3645 description = "the standard locale";
3646 name = "C";
3647 }
7d4bcc4a
KW
3648
3649# ifdef SYSTEM_DEFAULT_LOCALE
3650
65ebb059
KW
3651 else if (strEQ(trial_locales[i], "")) {
3652 description = "the system default locale";
3653 if (system_default_locale) {
3654 name = system_default_locale;
3655 }
3656 }
7d4bcc4a
KW
3657
3658# endif /* SYSTEM_DEFAULT_LOCALE */
3659
65ebb059
KW
3660 else {
3661 description = "a fallback locale";
3662 name = trial_locales[i];
3663 }
3664 if (name && strNE(name, "")) {
3665 PerlIO_printf(Perl_error_log,
3666 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3667 }
3668 else {
3669 PerlIO_printf(Perl_error_log,
3670 "perl: warning: %s %s.\n", msg, description);
3671 }
3672 }
3673 } /* End of tried to fallback */
98994639 3674
e5f10d49
KW
3675 /* Done with finding the locales; update our records */
3676
7d4bcc4a
KW
3677# ifdef USE_LOCALE_CTYPE
3678
948523db 3679 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3680
e5f10d49 3681# endif
7d4bcc4a
KW
3682# ifdef USE_LOCALE_COLLATE
3683
948523db 3684 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3685
e5f10d49 3686# endif
7d4bcc4a
KW
3687# ifdef USE_LOCALE_NUMERIC
3688
948523db 3689 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3690
3691# endif
3692
948523db 3693 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3694
e9bc6d6b 3695# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3696
3697 /* This caches whether each category's locale is UTF-8 or not. This
3698 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3699 * initialization time before any threads have started, but not later
3700 * unless thread-safe operations are used.
47280b20
KW
3701 * Caching means that if the program heeds our dictate not to change
3702 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3703 * it may get queried without having to change locales. If the
3704 * environment is such that all categories have the same locale, this
3705 * isn't needed, as the code will not change the locale; but this
3706 * handles the uncommon case where the environment has disparate
3707 * locales for the categories */
47280b20
KW
3708 (void) _is_cur_LC_category_utf8(categories[i]);
3709
3710# endif
3711
e5f10d49
KW
3712 Safefree(curlocales[i]);
3713 }
b310b053 3714
7d4bcc4a
KW
3715# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3716
49c85077 3717 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3718 * locale is UTF-8. The call to new_ctype() just above has already
3719 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3720 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3721 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3722 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3723 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3724
a05d7ebb 3725 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3726 This is an alternative to using the -C command line switch
3727 (the -C if present will override this). */
3728 {
dd374669 3729 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3730 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3731 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3732 PL_utf8cache = -1;
b310b053
JH
3733 }
3734
7d4bcc4a 3735# endif
7d4bcc4a
KW
3736# ifdef __GLIBC__
3737
175c4cf9 3738 Safefree(language);
7d4bcc4a
KW
3739
3740# endif
175c4cf9
KW
3741
3742 Safefree(lc_all);
3743 Safefree(lang);
3744
e3305790 3745#endif /* USE_LOCALE */
2fcc0ca9 3746#ifdef DEBUGGING
7d4bcc4a 3747
2fcc0ca9 3748 /* So won't continue to output stuff */
27cdc72e 3749 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3750
2fcc0ca9
KW
3751#endif
3752
98994639
HS
3753 return ok;
3754}
3755
98994639
HS
3756#ifdef USE_LOCALE_COLLATE
3757
a4a439fb 3758char *
a4a439fb
KW
3759Perl__mem_collxfrm(pTHX_ const char *input_string,
3760 STRLEN len, /* Length of 'input_string' */
3761 STRLEN *xlen, /* Set to length of returned string
3762 (not including the collation index
3763 prefix) */
3764 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3765 )
98994639 3766{
a4a439fb
KW
3767
3768 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3769 * differences. First, it handles embedded NULs. Second, it allocates a bit
3770 * more memory than needed for the transformed data itself. The real
55e5378d 3771 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3772 * the length of that, and doesn't include the collation index size.
3773 * Please see sv_collxfrm() to see how this is used. */
3774
55e5378d
KW
3775#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3776
6696cfa7
KW
3777 char * s = (char *) input_string;
3778 STRLEN s_strlen = strlen(input_string);
79f120c8 3779 char *xbuf = NULL;
55e5378d 3780 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3781 STRLEN length_in_chars;
c664130f 3782 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3783
a4a439fb
KW
3784 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3785
3786 /* Must be NUL-terminated */
3787 assert(*(input_string + len) == '\0');
7918f24d 3788
79f120c8
KW
3789 /* If this locale has defective collation, skip */
3790 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3791 DEBUG_L(PerlIO_printf(Perl_debug_log,
3792 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3793 goto bad;
3794 }
3795
6696cfa7
KW
3796 /* Replace any embedded NULs with the control that sorts before any others.
3797 * This will give as good as possible results on strings that don't
3798 * otherwise contain that character, but otherwise there may be
3799 * less-than-perfect results with that character and NUL. This is
fdc080f3 3800 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3801 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3802 NUL */
6696cfa7
KW
3803 char * e = s + len;
3804 char * sans_nuls;
fdc080f3 3805 STRLEN sans_nuls_len;
94762aa0 3806 int try_non_controls;
afc4976f
KW
3807 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3808 making sure 2nd byte is NUL.
3809 */
3810 STRLEN this_replacement_len;
3811
1e4c9676
KW
3812 /* If we don't know what non-NUL control character sorts lowest for
3813 * this locale, find it */
f28f4d2a 3814 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3815 int j;
afc4976f 3816 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3817 includes the collation index
3818 prefixed. */
3819
91c0e2e0 3820 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3821
3822 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3823 * NUL, in which case we instead look for any character. Controls
3824 * are preferred because collation order is, in general, context
3825 * sensitive, with adjoining characters affecting the order, and
3826 * controls are less likely to have such interactions, allowing the
3827 * NUL-replacement to stand on its own. (Another way to look at it
3828 * is to imagine what would happen if the NUL were replaced by a
3829 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3830 for (try_non_controls = 0;
3831 try_non_controls < 2;
3832 try_non_controls++)
3833 {
d4ff9586
KW
3834 /* Look through all legal code points (NUL isn't) */
3835 for (j = 1; j < 256; j++) {
3836 char * x; /* j's xfrm plus collation index */
3837 STRLEN x_len; /* length of 'x' */
3838 STRLEN trial_len = 1;
736a4fed 3839 char cur_source[] = { '\0', '\0' };
d4ff9586 3840
736a4fed
KW
3841 /* Skip non-controls the first time through the loop. The
3842 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3843 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3844 ? ! isCNTRL_L1(j)
3845 : ! isCNTRL_LC(j))
3846 {
d4ff9586 3847 continue;
6696cfa7 3848 }
6696cfa7 3849
736a4fed
KW
3850 /* Create a 1-char string of the current code point */
3851 cur_source[0] = (char) j;
3852
d4ff9586
KW
3853 /* Then transform it */
3854 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3855 0 /* The string is not in UTF-8 */);
6696cfa7 3856
1e4c9676 3857 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3858 * */
3859 if (! x) {
3860 continue;
3861 }