This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: White-space only
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
e9bc6d6b
KW
35 *
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
166f8a29
DM
44 */
45
98994639
HS
46#include "EXTERN.h"
47#define PERL_IN_LOCALE_C
f7416781 48#include "perl_langinfo.h"
98994639
HS
49#include "perl.h"
50
a4af207c
JH
51#include "reentr.h"
52
0dec74cd
KW
53#ifdef I_WCHAR
54# include <wchar.h>
55#endif
56
2fcc0ca9
KW
57/* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
5a4b0634
KW
60#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61# define debug_initialization 0
62# define DEBUG_INITIALIZATION_set(v)
63#else
2fcc0ca9 64static bool debug_initialization = FALSE;
5a4b0634 65# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
66#endif
67
48015184
KW
68
69/* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
71 * done generally. */
72#define GET_ERRNO saved_errno
73
291a84fb
KW
74/* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
77#define STRLENs(s) (sizeof("" s "") - 1)
78
63e5b0d7
KW
79/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
85 * cntrl: 84-97 9B-9F
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89#define isNAME_C_OR_POSIX(name) \
90 ( (name) != NULL \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
93
8ef6e574
KW
94#ifdef USE_LOCALE
95
47280b20
KW
96/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
98
99#define UTF8NESS_SEP "\v"
100#define UTF8NESS_PREFIX "\f"
101
102/* So, the string looks like:
98994639 103 *
47280b20 104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 105 *
47280b20
KW
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
108
109STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
111
112#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
114
115/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
118
98994639
HS
119STATIC char *
120S_stdize_locale(pTHX_ char *locs)
121{
47280b20
KW
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
124 *
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
130 *
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
134 * */
135
7452cf6a 136 const char * const s = strchr(locs, '=');
98994639
HS
137 bool okay = TRUE;
138
7918f24d
NC
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
140
8772537c
AL
141 if (s) {
142 const char * const t = strchr(s, '.');
98994639 143 okay = FALSE;
8772537c
AL
144 if (t) {
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
149 locs[len] = 0;
150 okay = TRUE;
98994639
HS
151 }
152 }
153 }
154
155 if (!okay)
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
157
158 return locs;
159}
160
e5f10d49
KW
161/* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
163 * order. */
164
165const int categories[] = {
166
167# ifdef USE_LOCALE_NUMERIC
168 LC_NUMERIC,
169# endif
170# ifdef USE_LOCALE_CTYPE
171 LC_CTYPE,
172# endif
173# ifdef USE_LOCALE_COLLATE
174 LC_COLLATE,
175# endif
176# ifdef USE_LOCALE_TIME
177 LC_TIME,
178# endif
179# ifdef USE_LOCALE_MESSAGES
180 LC_MESSAGES,
181# endif
182# ifdef USE_LOCALE_MONETARY
183 LC_MONETARY,
184# endif
9821811f
KW
185# ifdef USE_LOCALE_ADDRESS
186 LC_ADDRESS,
187# endif
188# ifdef USE_LOCALE_IDENTIFICATION
189 LC_IDENTIFICATION,
190# endif
191# ifdef USE_LOCALE_MEASUREMENT
192 LC_MEASUREMENT,
193# endif
194# ifdef USE_LOCALE_PAPER
195 LC_PAPER,
196# endif
197# ifdef USE_LOCALE_TELEPHONE
198 LC_TELEPHONE,
199# endif
e5f10d49
KW
200# ifdef LC_ALL
201 LC_ALL,
202# endif
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
206};
207
208/* The top-most real element is LC_ALL */
209
210const char * category_names[] = {
211
212# ifdef USE_LOCALE_NUMERIC
213 "LC_NUMERIC",
214# endif
215# ifdef USE_LOCALE_CTYPE
216 "LC_CTYPE",
217# endif
218# ifdef USE_LOCALE_COLLATE
219 "LC_COLLATE",
220# endif
221# ifdef USE_LOCALE_TIME
222 "LC_TIME",
223# endif
224# ifdef USE_LOCALE_MESSAGES
225 "LC_MESSAGES",
226# endif
227# ifdef USE_LOCALE_MONETARY
228 "LC_MONETARY",
229# endif
9821811f
KW
230# ifdef USE_LOCALE_ADDRESS
231 "LC_ADDRESS",
232# endif
233# ifdef USE_LOCALE_IDENTIFICATION
234 "LC_IDENTIFICATION",
235# endif
236# ifdef USE_LOCALE_MEASUREMENT
237 "LC_MEASUREMENT",
238# endif
239# ifdef USE_LOCALE_PAPER
240 "LC_PAPER",
241# endif
242# ifdef USE_LOCALE_TELEPHONE
243 "LC_TELEPHONE",
244# endif
e5f10d49
KW
245# ifdef LC_ALL
246 "LC_ALL",
247# endif
248 NULL /* Placeholder */
249 };
250
251# ifdef LC_ALL
252
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
256
257# else
258
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
262
263# endif
264
265/* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
948523db
KW
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
e5f10d49 272
b09aaf40
KW
273STATIC const char *
274S_category_name(const int category)
275{
276 unsigned int i;
277
278#ifdef LC_ALL
279
280 if (category == LC_ALL) {
281 return "LC_ALL";
282 }
283
284#endif
285
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
289 }
290 }
291
292 {
293 const char suffix[] = " (unknown)";
294 int temp = category;
295 Size_t length = sizeof(suffix) + 1;
296 char * unknown;
297 dTHX;
298
299 if (temp < 0) {
300 length++;
301 temp = - temp;
302 }
303
304 /* Calculate the number of digits */
305 while (temp >= 10) {
306 temp /= 10;
307 length++;
308 }
309
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
312 SAVEFREEPV(unknown);
313 return unknown;
314 }
315}
316
948523db
KW
317/* Now create LC_foo_INDEX #defines for just those categories on this system */
318# ifdef USE_LOCALE_NUMERIC
319# define LC_NUMERIC_INDEX 0
320# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
321# else
322# define _DUMMY_NUMERIC -1
323# endif
324# ifdef USE_LOCALE_CTYPE
325# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326# define _DUMMY_CTYPE LC_CTYPE_INDEX
327# else
328# define _DUMMY_CTYPE _DUMMY_NUMERIC
329# endif
330# ifdef USE_LOCALE_COLLATE
331# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332# define _DUMMY_COLLATE LC_COLLATE_INDEX
333# else
8203b3c3 334# define _DUMMY_COLLATE _DUMMY_CTYPE
948523db
KW
335# endif
336# ifdef USE_LOCALE_TIME
337# define LC_TIME_INDEX _DUMMY_COLLATE + 1
338# define _DUMMY_TIME LC_TIME_INDEX
339# else
340# define _DUMMY_TIME _DUMMY_COLLATE
341# endif
342# ifdef USE_LOCALE_MESSAGES
343# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
345# else
346# define _DUMMY_MESSAGES _DUMMY_TIME
347# endif
348# ifdef USE_LOCALE_MONETARY
349# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350# define _DUMMY_MONETARY LC_MONETARY_INDEX
351# else
352# define _DUMMY_MONETARY _DUMMY_MESSAGES
353# endif
9821811f
KW
354# ifdef USE_LOCALE_ADDRESS
355# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
357# else
358# define _DUMMY_ADDRESS _DUMMY_MONETARY
359# endif
360# ifdef USE_LOCALE_IDENTIFICATION
361# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
363# else
364# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
365# endif
366# ifdef USE_LOCALE_MEASUREMENT
367# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
369# else
370# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
371# endif
372# ifdef USE_LOCALE_PAPER
373# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374# define _DUMMY_PAPER LC_PAPER_INDEX
375# else
376# define _DUMMY_PAPER _DUMMY_MEASUREMENT
377# endif
378# ifdef USE_LOCALE_TELEPHONE
379# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
381# else
382# define _DUMMY_TELEPHONE _DUMMY_PAPER
383# endif
948523db 384# ifdef LC_ALL
9821811f 385# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
386# endif
387#endif /* ifdef USE_LOCALE */
8ef6e574 388
d2b24094 389/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
390#ifdef WIN32
391# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
392#else
393# define my_setlocale(cat, locale) setlocale(cat, locale)
394#endif
395
396#ifndef USE_POSIX_2008_LOCALE
397
398/* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
402
403#else /* Below uses POSIX 2008 */
404
405/* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411# define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
414
415/* A third array, parallel to the ones above to map from category to its
416 * equivalent mask */
417const int category_masks[] = {
418# ifdef USE_LOCALE_NUMERIC
419 LC_NUMERIC_MASK,
420# endif
421# ifdef USE_LOCALE_CTYPE
422 LC_CTYPE_MASK,
423# endif
424# ifdef USE_LOCALE_COLLATE
425 LC_COLLATE_MASK,
426# endif
427# ifdef USE_LOCALE_TIME
428 LC_TIME_MASK,
429# endif
430# ifdef USE_LOCALE_MESSAGES
431 LC_MESSAGES_MASK,
432# endif
433# ifdef USE_LOCALE_MONETARY
434 LC_MONETARY_MASK,
435# endif
436# ifdef USE_LOCALE_ADDRESS
437 LC_ADDRESS_MASK,
438# endif
439# ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
441# endif
442# ifdef USE_LOCALE_MEASUREMENT
443 LC_MEASUREMENT_MASK,
444# endif
445# ifdef USE_LOCALE_PAPER
446 LC_PAPER_MASK,
447# endif
448# ifdef USE_LOCALE_TELEPHONE
449 LC_TELEPHONE_MASK,
450# endif
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
455 * time */
456 LC_ALL_MASK
457 };
458
459STATIC const char *
460S_emulate_setlocale(const int category,
461 const char * locale,
462 unsigned int index,
463 const bool is_index_valid
464 )
465{
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
473 *
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
476 *
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
486 *
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
501 */
502
503 int mask;
504 locale_t old_obj;
505 locale_t new_obj;
506 dTHX;
507
508# ifdef DEBUGGING
509
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
512 }
513
514# endif
515
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
518 unsigned int i;
519
520# ifdef DEBUGGING
521
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
524 }
525
526# endif
527
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
530 index = i;
531 goto found_index;
532 }
533 }
534
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
539 category, locale);
540 return NULL;
541
542 found_index: ;
543
544# ifdef DEBUGGING
545
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
548 }
549
550# endif
551
552 }
553
554 mask = category_masks[index];
555
556# ifdef DEBUGGING
557
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
560 }
561
562# endif
563
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
567
568# ifdef DEBUGGING
569
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
572 }
573
574# endif
575
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
578 }
579
580# ifdef HAS_QUERYLOCALE
581
582 return (char *) querylocale(mask, cur_obj);
583
d2b24094 584# else
ecdda939
KW
585
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
588
e9bc6d6b
KW
589# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
590
591 {
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
595
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
598 /*
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
602 */
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
607
608# ifdef USE_C_BACKTRACE
609
610 dump_c_backtrace(Perl_debug_log, 20, 1);
611
612# endif
613
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
618 }
619
620 return temp_name;
621 }
622 }
623
624# endif
625
626 /* Without querylocale(), we have to use our record-keeping we've
627 * done. */
628
629 if (category != LC_ALL) {
630
631# ifdef DEBUGGING
632
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 }
636
637# endif
638
639 return PL_curlocales[index];
640 }
641 else { /* For LC_ALL */
642 unsigned int i;
643 Size_t names_len = 0;
644 char * all_string;
7c93a471 645 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
646
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
649
650# ifdef DEBUGGING
651
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
654 }
655
656# endif
657
658 return PL_curlocales[LC_ALL_INDEX];
659 }
660
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
e9bc6d6b
KW
666 for (i = 0; i < LC_ALL_INDEX; i++) {
667
668# ifdef DEBUGGING
669
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
672 }
673
674# endif
675
676 names_len += strlen(category_names[i])
677 + 1 /* '=' */
678 + strlen(PL_curlocales[i])
679 + 1; /* ';' */
7c93a471
KW
680
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
683 }
684 }
685
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 692 }
7c93a471 693
e9bc6d6b
KW
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
696 *all_string = '\0';
697
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
700
701# ifdef DEBUGGING
702
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
705 }
706
707# endif
708
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
713 }
714
715# ifdef DEBUGGING
716
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
719 }
720
721 #endif
722
723 return all_string;
724 }
725
726# ifdef EINVAL
727
728 SETERRNO(EINVAL, LIB_INVARG);
729
730# endif
731
732 return NULL;
733
d2b24094
KW
734# endif
735
e9bc6d6b
KW
736 }
737
f8115d60 738 /* Here, we are switching locales. */
e9bc6d6b
KW
739
740# ifndef HAS_QUERYLOCALE
741
742 if (strEQ(locale, "")) {
743
744 /* For non-querylocale() systems, we do the setting of "" ourselves to
745 * be sure that we really know what's going on. We follow the Linux
746 * documented behavior (but if that differs from the actual behavior,
747 * this won't work exactly as the OS implements). We go out and
748 * examine the environment based on our understanding of how the system
749 * works, and use that to figure things out */
750
751 const char * const lc_all = PerlEnv_getenv("LC_ALL");
752
753 /* Use any "LC_ALL" environment variable, as it overrides everything
754 * else. */
755 if (lc_all && strNE(lc_all, "")) {
756 locale = lc_all;
757 }
758 else {
759
760 /* Otherwise, we need to dig deeper. Unless overridden, the
761 * default is the LANG environment variable; if it doesn't exist,
762 * then "C" */
763
764 const char * default_name;
765
766 /* To minimize other threads messing with the environment, we copy
767 * the variable, making it a temporary. But this doesn't work upon
768 * program initialization before any scopes are created, and at
769 * this time, there's nothing else going on that would interfere.
770 * So skip the copy in that case */
771 if (PL_scopestack_ix == 0) {
772 default_name = PerlEnv_getenv("LANG");
773 }
774 else {
775 default_name = savepv(PerlEnv_getenv("LANG"));
776 }
777
778 if (! default_name || strEQ(default_name, "")) {
929a7f8c 779 default_name = "C";
e9bc6d6b
KW
780 }
781 else if (PL_scopestack_ix != 0) {
782 SAVEFREEPV(default_name);
783 }
784
785 if (category != LC_ALL) {
786 const char * const name = PerlEnv_getenv(category_names[index]);
787
788 /* Here we are setting a single category. Assume will have the
789 * default name */
790 locale = default_name;
791
792 /* But then look for an overriding environment variable */
793 if (name && strNE(name, "")) {
794 locale = name;
795 }
796 }
797 else {
798 bool did_override = FALSE;
799 unsigned int i;
800
801 /* Here, we are getting LC_ALL. Any categories that don't have
802 * a corresponding environment variable set should be set to
803 * LANG, or to "C" if there is no LANG. If no individual
804 * categories differ from this, we can just set LC_ALL. This
805 * is buggy on systems that have extra categories that we don't
806 * know about. If there is an environment variable that sets
807 * that category, we won't know to look for it, and so our use
808 * of LANG or "C" improperly overrides it. On the other hand,
809 * if we don't do what is done here, and there is no
810 * environment variable, the category's locale should be set to
811 * LANG or "C". So there is no good solution. khw thinks the
812 * best is to look at systems to see what categories they have,
813 * and include them, and then to assume that we know the
814 * complete set */
815
816 for (i = 0; i < LC_ALL_INDEX; i++) {
817 const char * const env_override
818 = savepv(PerlEnv_getenv(category_names[i]));
819 const char * this_locale = ( env_override
820 && strNE(env_override, ""))
821 ? env_override
822 : default_name;
6435f98d
KW
823 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
824 {
825 Safefree(env_override);
826 return NULL;
827 }
e9bc6d6b
KW
828
829 if (strNE(this_locale, default_name)) {
830 did_override = TRUE;
831 }
832
833 Safefree(env_override);
834 }
835
836 /* If all the categories are the same, we can set LC_ALL to
837 * that */
838 if (! did_override) {
839 locale = default_name;
840 }
841 else {
842
843 /* Here, LC_ALL is no longer valid, as some individual
844 * categories don't match it. We call ourselves
845 * recursively, as that will execute the code that
846 * generates the proper locale string for this situation.
847 * We don't do the remainder of this function, as that is
848 * to update our records, and we've just done that for the
849 * individual categories in the loop above, and doing so
850 * would cause LC_ALL to be done as well */
851 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
852 }
853 }
854 }
855 }
856 else if (strchr(locale, ';')) {
857
858 /* LC_ALL may actually incude a conglomeration of various categories.
859 * Without querylocale, this code uses the glibc (as of this writing)
860 * syntax for representing that, but that is not a stable API, and
861 * other platforms do it differently, so we have to handle all cases
862 * ourselves */
863
f0023d47 864 unsigned int i;
e9bc6d6b
KW
865 const char * s = locale;
866 const char * e = locale + strlen(locale);
867 const char * p = s;
868 const char * category_end;
869 const char * name_start;
870 const char * name_end;
871
f0023d47
KW
872 /* If the string that gives what to set doesn't include all categories,
873 * the omitted ones get set to "C". To get this behavior, first set
874 * all the individual categories to "C", and override the furnished
875 * ones below */
876 for (i = 0; i < LC_ALL_INDEX; i++) {
877 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
878 return NULL;
879 }
880 }
881
e9bc6d6b 882 while (s < e) {
e9bc6d6b
KW
883
884 /* Parse through the category */
885 while (isWORDCHAR(*p)) {
886 p++;
887 }
888 category_end = p;
889
890 if (*p++ != '=') {
891 Perl_croak(aTHX_
892 "panic: %s: %d: Unexpected character in locale name '%02X",
893 __FILE__, __LINE__, *(p-1));
894 }
895
896 /* Parse through the locale name */
897 name_start = p;
bee74f4b
KW
898 while (p < e && *p != ';') {
899 if (! isGRAPH(*p)) {
900 Perl_croak(aTHX_
901 "panic: %s: %d: Unexpected character in locale name '%02X",
902 __FILE__, __LINE__, *(p-1));
903 }
e9bc6d6b
KW
904 p++;
905 }
906 name_end = p;
907
bee74f4b
KW
908 /* Space past the semi-colon */
909 if (p < e) {
910 p++;
e9bc6d6b
KW
911 }
912
913 /* Find the index of the category name in our lists */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 915 char * individ_locale;
e9bc6d6b
KW
916
917 /* Keep going if this isn't the index. The strnNE() avoids a
918 * Perl_form(), but would fail if ever a category name could be
919 * a substring of another one, like if there were a
920 * "LC_TIME_DATE" */
921 if strnNE(s, category_names[i], category_end - s) {
922 continue;
923 }
924
925 /* If this index is for the single category we're changing, we
926 * have found the locale to set it to. */
927 if (category == categories[i]) {
928 locale = Perl_form(aTHX_ "%.*s",
929 (int) (name_end - name_start),
930 name_start);
931 goto ready_to_set;
932 }
933
3c76a412 934 assert(category == LC_ALL);
bee74f4b
KW
935 individ_locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start), name_start);
6435f98d
KW
937 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
938 {
939 return NULL;
940 }
e9bc6d6b
KW
941 }
942
943 s = p;
944 }
945
946 /* Here we have set all the individual categories by recursive calls.
947 * These collectively should have fixed up LC_ALL, so can just query
948 * what that now is */
949 assert(category == LC_ALL);
950
951 return do_setlocale_c(LC_ALL, NULL);
952 }
953
954 ready_to_set: ;
955
f8115d60
KW
956 /* Here at the end of having to deal with the absence of querylocale().
957 * Some cases have already been fully handled by recursive calls to this
958 * function. But at this point, we haven't dealt with those, but are now
959 * prepared to, knowing what the locale name to set this category to is.
960 * This would have come for free if this system had had querylocale() */
961
e9bc6d6b
KW
962# endif /* end of ! querylocale */
963
f8115d60
KW
964 assert(PL_C_locale_obj);
965
966 /* Switching locales generally entails freeing the current one's space (at
967 * the C library's discretion). We need to stop using that locale before
968 * the switch. So switch to a known locale object that we don't otherwise
969 * mess with. This returns the locale object in effect at the time of the
970 * switch. */
971 old_obj = uselocale(PL_C_locale_obj);
972
973# ifdef DEBUGGING
974
975 if (DEBUG_Lv_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
977 }
978
979# endif
980
981 if (! old_obj) {
982
983# ifdef DEBUGGING
984
985 if (DEBUG_L_TEST || debug_initialization) {
986 dSAVE_ERRNO;
987 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
988 RESTORE_ERRNO;
989 }
990
991# endif
992
993 return NULL;
994 }
995
996# ifdef DEBUGGING
997
998 if (DEBUG_Lv_TEST || debug_initialization) {
999 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1000 }
1001
1002# endif
1003
1004 /* If we weren't in a thread safe locale, set so that newlocale() below
1005 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1006 The latter is defensive coding, so that, even if there is some bug, we
1007 will never end up trying to modify either of these, as if passed to
1008 newlocale(), they can be. */
1009 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1010 old_obj = (locale_t) 0;
1011 }
1012
e9bc6d6b
KW
1013 /* Ready to create a new locale by modification of the exising one */
1014 new_obj = newlocale(mask, locale, old_obj);
1015
1016 if (! new_obj) {
1017 dSAVE_ERRNO;
1018
1019# ifdef DEBUGGING
1020
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1023 }
1024
1025# endif
1026
1027 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1028
1029# ifdef DEBUGGING
1030
1031 if (DEBUG_L_TEST || debug_initialization) {
1032 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1033 }
1034
1035# endif
1036
1037 }
1038 RESTORE_ERRNO;
1039 return NULL;
1040 }
1041
1042# ifdef DEBUGGING
1043
1044 if (DEBUG_Lv_TEST || debug_initialization) {
d15b6e8b 1045 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p; should have freed %p\n", __FILE__, __LINE__, new_obj, old_obj);
e9bc6d6b
KW
1046 }
1047
1048# endif
1049
1050 /* And switch into it */
1051 if (! uselocale(new_obj)) {
1052 dSAVE_ERRNO;
1053
1054# ifdef DEBUGGING
1055
1056 if (DEBUG_L_TEST || debug_initialization) {
1057 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1058 }
1059
1060# endif
1061
1062 if (! uselocale(old_obj)) {
1063
1064# ifdef DEBUGGING
1065
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1068 }
1069
1070# endif
1071
1072 }
1073 freelocale(new_obj);
1074 RESTORE_ERRNO;
1075 return NULL;
1076 }
1077
1078# ifdef DEBUGGING
1079
1080 if (DEBUG_Lv_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1082 }
1083
1084# endif
1085
1086 /* We are done, except for updating our records (if the system doesn't keep
1087 * them) and in the case of locale "", we don't actually know what the
1088 * locale that got switched to is, as it came from the environment. So
1089 * have to find it */
1090
1091# ifdef HAS_QUERYLOCALE
1092
1093 if (strEQ(locale, "")) {
1094 locale = querylocale(mask, new_obj);
1095 }
1096
1097# else
1098
1099 /* Here, 'locale' is the return value */
1100
1101 /* Without querylocale(), we have to update our records */
1102
1103 if (category == LC_ALL) {
1104 unsigned int i;
1105
1106 /* For LC_ALL, we change all individual categories to correspond */
1107 /* PL_curlocales is a parallel array, so has same
1108 * length as 'categories' */
1109 for (i = 0; i <= LC_ALL_INDEX; i++) {
1110 Safefree(PL_curlocales[i]);
1111 PL_curlocales[i] = savepv(locale);
1112 }
1113 }
1114 else {
1115
1116 /* For a single category, if it's not the same as the one in LC_ALL, we
1117 * nullify LC_ALL */
1118
1119 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1120 Safefree(PL_curlocales[LC_ALL_INDEX]);
1121 PL_curlocales[LC_ALL_INDEX] = NULL;
1122 }
1123
1124 /* Then update the category's record */
1125 Safefree(PL_curlocales[index]);
1126 PL_curlocales[index] = savepv(locale);
1127 }
1128
1129# endif
1130
1131 return locale;
1132}
1133
1134#endif /* USE_POSIX_2008_LOCALE */
1135
1136#if 0 /* Code that was to emulate thread-safe locales on platforms that
1137 didn't natively support them */
1138
1139/* The way this would work is that we would keep a per-thread list of the
1140 * correct locale for that thread. Any operation that was locale-sensitive
1141 * would have to be changed so that it would look like this:
1142 *
1143 * LOCALE_LOCK;
1144 * setlocale to the correct locale for this operation
1145 * do operation
1146 * LOCALE_UNLOCK
1147 *
1148 * This leaves the global locale in the most recently used operation's, but it
1149 * was locked long enough to get the result. If that result is static, it
1150 * needs to be copied before the unlock.
1151 *
1152 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1153 * the setup, but are no-ops when not needed, and similarly,
1154 * END_LOCALE_DEPENDENT_OP for the tear-down
1155 *
1156 * But every call to a locale-sensitive function would have to be changed, and
1157 * if a module didn't cooperate by using the mutex, things would break.
1158 *
1159 * This code was abandoned before being completed or tested, and is left as-is
1160*/
1161
1162# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1163# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1164
1165STATIC char *
1166S_locking_setlocale(pTHX_
1167 const int category,
1168 const char * locale,
1169 int index,
1170 const bool is_index_valid
1171 )
1172{
1173 /* This function kind of performs a setlocale() on just the current thread;
1174 * thus it is kind of thread-safe. It does this by keeping a thread-level
1175 * array of the current locales for each category. Every time a locale is
1176 * switched to, it does the switch globally, but updates the thread's
1177 * array. A query as to what the current locale is just returns the
1178 * appropriate element from the array, and doesn't actually call the system
1179 * setlocale(). The saving into the array is done in an uninterruptible
1180 * section of code, so is unaffected by whatever any other threads might be
1181 * doing.
1182 *
1183 * All locale-sensitive operations must work by first starting a critical
1184 * section, then switching to the thread's locale as kept by this function,
1185 * and then doing the operation, then ending the critical section. Thus,
1186 * each gets done in the appropriate locale. simulating thread-safety.
1187 *
1188 * This function takes the same parameters, 'category' and 'locale', that
1189 * the regular setlocale() function does, but it also takes two additional
1190 * ones. This is because as described earlier. If we know on input the
1191 * index corresponding to the category into the array where we store the
1192 * current locales, we don't have to calculate it. If the caller knows at
1193 * compile time what the index is, it it can pass it, setting
1194 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1195 *
1196 */
1197
1198 /* If the input index might be incorrect, calculate the correct one */
1199 if (! is_index_valid) {
1200 unsigned int i;
1201
1202 if (DEBUG_Lv_TEST || debug_initialization) {
1203 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1204 }
1205
1206 for (i = 0; i <= LC_ALL_INDEX; i++) {
1207 if (category == categories[i]) {
1208 index = i;
1209 goto found_index;
1210 }
1211 }
1212
1213 /* Here, we don't know about this category, so can't handle it.
1214 * XXX best we can do is to unsafely set this
1215 * XXX warning */
1216
1217 return my_setlocale(category, locale);
1218
1219 found_index: ;
1220
1221 if (DEBUG_Lv_TEST || debug_initialization) {
1222 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1223 }
1224 }
1225
1226 /* For a query, just return what's in our records */
1227 if (new_locale == NULL) {
1228 return curlocales[index];
1229 }
1230
1231
1232 /* Otherwise, we need to do the switch, and save the result, all in a
1233 * critical section */
1234
1235 Safefree(curlocales[[index]]);
1236
1237 /* It might be that this is called from an already-locked section of code.
1238 * We would have to detect and skip the LOCK/UNLOCK if so */
1239 LOCALE_LOCK;
1240
1241 curlocales[index] = savepv(my_setlocale(category, new_locale));
1242
1243 if (strEQ(new_locale, "")) {
1244
1245#ifdef LC_ALL
1246
1247 /* The locale values come from the environment, and may not all be the
1248 * same, so for LC_ALL, we have to update all the others, while the
1249 * mutex is still locked */
1250
1251 if (category == LC_ALL) {
1252 unsigned int i;
1253 for (i = 0; i < LC_ALL_INDEX) {
1254 curlocales[i] = my_setlocale(categories[i], NULL);
1255 }
1256 }
1257 }
1258
1259#endif
1260
1261 LOCALE_UNLOCK;
1262
1263 return curlocales[index];
1264}
1265
1266#endif
d36adde0 1267#ifdef USE_LOCALE
837ce802 1268
a4f00dcc 1269STATIC void
86799d2d 1270S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1271{
86799d2d
KW
1272 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1273 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1274
86799d2d
KW
1275#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1276 || defined(HAS_NL_LANGINFO))
98994639 1277
87f8e8e7 1278 const char * radix = (use_locale)
4e6826bf 1279 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1280 /* FALSE => already in dest locale */
1281 : ".";
2213a3be 1282
87f8e8e7 1283 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1284
87f8e8e7
KW
1285 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1286 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1287 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1288 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1289 && _is_cur_LC_category_utf8(LC_NUMERIC))
1290 {
1291 SvUTF8_on(PL_numeric_radix_sv);
1292 }
86799d2d
KW
1293
1294# ifdef DEBUGGING
7d4bcc4a 1295
2fcc0ca9
KW
1296 if (DEBUG_L_TEST || debug_initialization) {
1297 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1298 SvPVX(PL_numeric_radix_sv),
1299 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1300 }
69014004 1301
86799d2d 1302# endif
d36adde0
KW
1303#else
1304
1305 PERL_UNUSED_ARG(use_locale);
1306
86799d2d 1307#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1308
98994639
HS
1309}
1310
398eeea9
KW
1311STATIC void
1312S_new_numeric(pTHX_ const char *newnum)
98994639 1313{
7d4bcc4a
KW
1314
1315#ifndef USE_LOCALE_NUMERIC
1316
1317 PERL_UNUSED_ARG(newnum);
1318
1319#else
0d071d52 1320
291a84fb 1321 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1322 * core Perl this and that 'newnum' is the name of the new locale.
1323 * It installs this locale as the current underlying default.
1324 *
1325 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1326 * set_numeric_underlying() and set_numeric_standard() functions, which
1327 * should probably not be called directly, but only via macros like
0d071d52
KW
1328 * SET_NUMERIC_STANDARD() in perl.h.
1329 *
1330 * The toggling is necessary mainly so that a non-dot radix decimal point
1331 * character can be output, while allowing internal calculations to use a
1332 * dot.
1333 *
1334 * This sets several interpreter-level variables:
bb304765 1335 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1336 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1337 * that the current locale is the program's underlying
1338 * locale
1339 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1340 * that the current locale is the C locale or
1341 * indistinguishable from the C locale. If non-zero, it
1342 * is in C; if > 1, it means it may not be toggled away
7738054c 1343 * from C.
4c68b815
KW
1344 * PL_numeric_underlying_is_standard A bool kept by this function
1345 * indicating that the underlying locale and the standard
1346 * C locale are indistinguishable for the purposes of
1347 * LC_NUMERIC. This happens when both of the above two
1348 * variables are true at the same time. (Toggling is a
1349 * no-op under these circumstances.) This variable is
1350 * used to avoid having to recalculate.
398eeea9 1351 */
0d071d52 1352
b03f34cf 1353 char *save_newnum;
98994639
HS
1354
1355 if (! newnum) {
43c5f42d
NC
1356 Safefree(PL_numeric_name);
1357 PL_numeric_name = NULL;
98994639 1358 PL_numeric_standard = TRUE;
892e6465 1359 PL_numeric_underlying = TRUE;
4c68b815 1360 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1361 return;
1362 }
1363
b03f34cf 1364 save_newnum = stdize_locale(savepv(newnum));
892e6465 1365 PL_numeric_underlying = TRUE;
4c68b815
KW
1366 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1367
69c5e0db
KW
1368#ifndef TS_W32_BROKEN_LOCALECONV
1369
4c68b815 1370 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1371 * them. But on broken Windows systems calling my_nl_langinfo() for
1372 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1373 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1374 if (! PL_numeric_standard) {
4e6826bf 1375 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1376 FALSE /* Don't toggle locale */ ))
4e6826bf 1377 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1378 }
abe1abcf 1379
69c5e0db
KW
1380#endif
1381
4c68b815 1382 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1383 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1384 Safefree(PL_numeric_name);
b03f34cf 1385 PL_numeric_name = save_newnum;
b03f34cf 1386 }
abe1abcf
KW
1387 else {
1388 Safefree(save_newnum);
1389 }
4c28b29c 1390
4c68b815
KW
1391 PL_numeric_underlying_is_standard = PL_numeric_standard;
1392
7e5377f7 1393# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1394
1395 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1396 PL_numeric_name,
1397 PL_underlying_numeric_obj);
1398
1399#endif
1400
4c68b815
KW
1401 if (DEBUG_L_TEST || debug_initialization) {
1402 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1403 }
1404
4c28b29c
KW
1405 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1406 * have to worry about the radix being a non-dot. (Core operations that
1407 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1408 if (PL_numeric_standard) {
1409 set_numeric_radix(0);
1410 }
1411 else {
1412 set_numeric_standard();
1413 }
4c28b29c 1414
98994639 1415#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1416
98994639
HS
1417}
1418
1419void
1420Perl_set_numeric_standard(pTHX)
1421{
7d4bcc4a 1422
98994639 1423#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1424
28c1bf33
KW
1425 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1426 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1427 * macro avoids calling this routine if toggling isn't necessary according
1428 * to our records (which could be wrong if some XS code has changed the
1429 * locale behind our back) */
0d071d52 1430
7d4bcc4a
KW
1431# ifdef DEBUGGING
1432
2fcc0ca9
KW
1433 if (DEBUG_L_TEST || debug_initialization) {
1434 PerlIO_printf(Perl_debug_log,
7bb8f702 1435 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1436 }
98994639 1437
7d4bcc4a 1438# endif
7bb8f702
KW
1439
1440 do_setlocale_c(LC_NUMERIC, "C");
1441 PL_numeric_standard = TRUE;
1442 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1443 set_numeric_radix(0);
1444
98994639 1445#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1446
98994639
HS
1447}
1448
1449void
5792c642 1450Perl_set_numeric_underlying(pTHX)
98994639 1451{
7d4bcc4a 1452
98994639 1453#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1454
28c1bf33 1455 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1456 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1457 * instead of calling this directly. The macro avoids calling this routine
1458 * if toggling isn't necessary according to our records (which could be
1459 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1460
7d4bcc4a
KW
1461# ifdef DEBUGGING
1462
2fcc0ca9
KW
1463 if (DEBUG_L_TEST || debug_initialization) {
1464 PerlIO_printf(Perl_debug_log,
7bb8f702 1465 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1466 PL_numeric_name);
1467 }
98994639 1468
7d4bcc4a 1469# endif
7bb8f702
KW
1470
1471 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1472 PL_numeric_standard = PL_numeric_underlying_is_standard;
1473 PL_numeric_underlying = TRUE;
1474 set_numeric_radix(! PL_numeric_standard);
1475
98994639 1476#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1477
98994639
HS
1478}
1479
1480/*
1481 * Set up for a new ctype locale.
1482 */
a4f00dcc
KW
1483STATIC void
1484S_new_ctype(pTHX_ const char *newctype)
98994639 1485{
7d4bcc4a
KW
1486
1487#ifndef USE_LOCALE_CTYPE
1488
7d4bcc4a
KW
1489 PERL_UNUSED_ARG(newctype);
1490 PERL_UNUSED_CONTEXT;
1491
1492#else
0d071d52 1493
291a84fb 1494 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1495 * core Perl this and that 'newctype' is the name of the new locale.
1496 *
1497 * This function sets up the folding arrays for all 256 bytes, assuming
1498 * that tofold() is tolc() since fold case is not a concept in POSIX,
1499 *
1500 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1501 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1502 * this function should be called directly only from this file and from
0d071d52
KW
1503 * POSIX::setlocale() */
1504
27da23d5 1505 dVAR;
013f4e03 1506 unsigned int i;
98994639 1507
8b7358b9
KW
1508 /* Don't check for problems if we are suppressing the warnings */
1509 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
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
b8df1494
KW
1671 PL_in_utf8_turkic_locale = FALSE;
1672
7d4bcc4a
KW
1673# ifdef MB_CUR_MAX
1674
8c6180a9 1675 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1676 * this locale requires more than one byte, there are going to be
8c6180a9 1677 * problems. */
9c8a6dc2
KW
1678 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1679 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1680 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1681
8b7358b9
KW
1682 if ( check_for_problems && MB_CUR_MAX > 1
1683 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1684
1685 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1686 * locale. Just assume that the implementation for them (plus
1687 * for POSIX) is correct and the > 1 value is spurious. (Since
1688 * these are specially handled to never be considered UTF-8
1689 * locales, as long as this is the only problem, everything
1690 * should work fine */
1691 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1692 {
8c6180a9
KW
1693 multi_byte_locale = TRUE;
1694 }
7d4bcc4a
KW
1695
1696# endif
8c6180a9 1697
8b7358b9
KW
1698 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1699 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1700 PL_warn_locale = Perl_newSVpvf(aTHX_
1701 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1702 " which have\nunexpected meanings: %s\nThe Perl program"
1703 " will use the expected meanings",
8b7358b9 1704 newctype, bad_chars_list);
8b7358b9
KW
1705 }
1706 else {
1707 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1708 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1709 newctype,
8c6180a9
KW
1710 (multi_byte_locale)
1711 ? " Some characters in it are not recognized by"
1712 " Perl."
1713 : "",
1714 (bad_count)
1715 ? "\nThe following characters (and maybe others)"
1716 " may not have the same meaning as the Perl"
1717 " program expects:\n"
1718 : "",
1719 (bad_count)
1720 ? bad_chars_list
1721 : ""
1722 );
8b7358b9
KW
1723 }
1724
b119c2be
KW
1725# ifdef HAS_NL_LANGINFO
1726
1727 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1728 /* parameter FALSE is a don't care here */
4e6826bf 1729 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1730
1731# endif
1732
8b7358b9
KW
1733 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1734
cc9eaeb0 1735 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1736 * output the message now. If not in that scope, we save the
1737 * message to be output at the first operation using this locale,
1738 * if that actually happens. Most programs don't use locales, so
1739 * they are immune to bad ones. */
cc9eaeb0 1740 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1741
780fcc9f
KW
1742 /* The '0' below suppresses a bogus gcc compiler warning */
1743 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1744
bddebb56
KW
1745 if (IN_LC(LC_CTYPE)) {
1746 SvREFCNT_dec_NN(PL_warn_locale);
1747 PL_warn_locale = NULL;
1748 }
780fcc9f 1749 }
baa60164 1750 }
31f05a37 1751 }
98994639
HS
1752
1753#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1754
98994639
HS
1755}
1756
98994639 1757void
2726666d
KW
1758Perl__warn_problematic_locale()
1759{
2726666d
KW
1760
1761#ifdef USE_LOCALE_CTYPE
1762
5f04a188
KW
1763 dTHX;
1764
1765 /* Internal-to-core function that outputs the message in PL_warn_locale,
1766 * and then NULLS it. Should be called only through the macro
1767 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1768
2726666d 1769 if (PL_warn_locale) {
2726666d
KW
1770 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1771 SvPVX(PL_warn_locale),
1772 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1773 SvREFCNT_dec_NN(PL_warn_locale);
1774 PL_warn_locale = NULL;
1775 }
1776
1777#endif
1778
1779}
1780
a4f00dcc
KW
1781STATIC void
1782S_new_collate(pTHX_ const char *newcoll)
98994639 1783{
7d4bcc4a
KW
1784
1785#ifndef USE_LOCALE_COLLATE
1786
1787 PERL_UNUSED_ARG(newcoll);
1788 PERL_UNUSED_CONTEXT;
1789
1790#else
0d071d52 1791
291a84fb 1792 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1793 * core Perl this and that 'newcoll' is the name of the new locale.
1794 *
d35fca5f
KW
1795 * The design of locale collation is that every locale change is given an
1796 * index 'PL_collation_ix'. The first time a string particpates in an
1797 * operation that requires collation while locale collation is active, it
1798 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1799 * magic includes the collation index, and the transformation of the string
1800 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1801 * instead of the string itself. If a string changes, the magic is
1802 * cleared. The next time the locale changes, the index is incremented,
1803 * and so we know during a comparison that the transformation is not
1804 * necessarily still valid, and so is recomputed. Note that if the locale
1805 * changes enough times, the index could wrap (a U32), and it is possible
1806 * that a transformation would improperly be considered valid, leading to
1807 * an unlikely bug */
0d071d52 1808
98994639
HS
1809 if (! newcoll) {
1810 if (PL_collation_name) {
1811 ++PL_collation_ix;
1812 Safefree(PL_collation_name);
1813 PL_collation_name = NULL;
1814 }
1815 PL_collation_standard = TRUE;
00bf60ca 1816 is_standard_collation:
98994639
HS
1817 PL_collxfrm_base = 0;
1818 PL_collxfrm_mult = 2;
165a1c52 1819 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1820 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1821 PL_strxfrm_max_cp = 0;
98994639
HS
1822 return;
1823 }
1824
d35fca5f 1825 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1826 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1827 ++PL_collation_ix;
1828 Safefree(PL_collation_name);
1829 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1830 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1831 if (PL_collation_standard) {
1832 goto is_standard_collation;
1833 }
98994639 1834
165a1c52 1835 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1836 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1837 PL_strxfrm_max_cp = 0;
165a1c52 1838
59c018b9
KW
1839 /* A locale collation definition includes primary, secondary, tertiary,
1840 * etc. weights for each character. To sort, the primary weights are
1841 * used, and only if they compare equal, then the secondary weights are
1842 * used, and only if they compare equal, then the tertiary, etc.
1843 *
1844 * strxfrm() works by taking the input string, say ABC, and creating an
1845 * output transformed string consisting of first the primary weights,
1846 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1847 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1848 * may not have weights at every level. In our example, let's say B
1849 * doesn't have a tertiary weight, and A doesn't have a secondary
1850 * weight. The constructed string is then going to be
1851 * A¹B¹C¹ B²C² A³C³ ....
1852 * This has the desired effect that strcmp() will look at the secondary
1853 * or tertiary weights only if the strings compare equal at all higher
1854 * priority weights. The spaces shown here, like in
c342d20e 1855 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1856 * are not just for readability. In the general case, these must
1857 * actually be bytes, which we will call here 'separator weights'; and
1858 * they must be smaller than any other weight value, but since these
1859 * are C strings, only the terminating one can be a NUL (some
1860 * implementations may include a non-NUL separator weight just before
1861 * the NUL). Implementations tend to reserve 01 for the separator
1862 * weights. They are needed so that a shorter string's secondary
1863 * weights won't be misconstrued as primary weights of a longer string,
1864 * etc. By making them smaller than any other weight, the shorter
1865 * string will sort first. (Actually, if all secondary weights are
1866 * smaller than all primary ones, there is no need for a separator
1867 * weight between those two levels, etc.)
1868 *
1869 * The length of the transformed string is roughly a linear function of
1870 * the input string. It's not exactly linear because some characters
1871 * don't have weights at all levels. When we call strxfrm() we have to
1872 * allocate some memory to hold the transformed string. The
1873 * calculations below try to find coefficients 'm' and 'b' for this
1874 * locale so that m*x + b equals how much space we need, given the size
1875 * of the input string in 'x'. If we calculate too small, we increase
1876 * the size as needed, and call strxfrm() again, but it is better to
1877 * get it right the first time to avoid wasted expensive string
1878 * transformations. */
1879
98994639 1880 {
79f120c8
KW
1881 /* We use the string below to find how long the tranformation of it
1882 * is. Almost all locales are supersets of ASCII, or at least the
1883 * ASCII letters. We use all of them, half upper half lower,
1884 * because if we used fewer, we might hit just the ones that are
1885 * outliers in a particular locale. Most of the strings being
1886 * collated will contain a preponderance of letters, and even if
1887 * they are above-ASCII, they are likely to have the same number of
1888 * weight levels as the ASCII ones. It turns out that digits tend
1889 * to have fewer levels, and some punctuation has more, but those
1890 * are relatively sparse in text, and khw believes this gives a
1891 * reasonable result, but it could be changed if experience so
1892 * dictates. */
1893 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1894 char * x_longer; /* Transformed 'longer' */
1895 Size_t x_len_longer; /* Length of 'x_longer' */
1896
1897 char * x_shorter; /* We also transform a substring of 'longer' */
1898 Size_t x_len_shorter;
1899
a4a439fb 1900 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1901 * are interested only in its length). It is used because it has
1902 * the intelligence to handle all cases, but to work, it needs some
1903 * values of 'm' and 'b' to get it started. For the purposes of
1904 * this calculation we use a very conservative estimate of 'm' and
1905 * 'b'. This assumes a weight can be multiple bytes, enough to
1906 * hold any UV on the platform, and there are 5 levels, 4 weight
1907 * bytes, and a trailing NUL. */
1908 PL_collxfrm_base = 5;
1909 PL_collxfrm_mult = 5 * sizeof(UV);
1910
1911 /* Find out how long the transformation really is */
a4a439fb
KW
1912 x_longer = _mem_collxfrm(longer,
1913 sizeof(longer) - 1,
1914 &x_len_longer,
1915
1916 /* We avoid converting to UTF-8 in the
1917 * called function by telling it the
1918 * string is in UTF-8 if the locale is a
1919 * UTF-8 one. Since the string passed
1920 * here is invariant under UTF-8, we can
1921 * claim it's UTF-8 even though it isn't.
1922 * */
1923 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1924 Safefree(x_longer);
1925
1926 /* Find out how long the transformation of a substring of 'longer'
1927 * is. Together the lengths of these transformations are
1928 * sufficient to calculate 'm' and 'b'. The substring is all of
1929 * 'longer' except the first character. This minimizes the chances
1930 * of being swayed by outliers */
a4a439fb 1931 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1932 sizeof(longer) - 2,
a4a439fb
KW
1933 &x_len_shorter,
1934 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1935 Safefree(x_shorter);
1936
1937 /* If the results are nonsensical for this simple test, the whole
1938 * locale definition is suspect. Mark it so that locale collation
1939 * is not active at all for it. XXX Should we warn? */
1940 if ( x_len_shorter == 0
1941 || x_len_longer == 0
1942 || x_len_shorter >= x_len_longer)
1943 {
1944 PL_collxfrm_mult = 0;
1945 PL_collxfrm_base = 0;
1946 }
1947 else {
1948 SSize_t base; /* Temporary */
1949
1950 /* We have both: m * strlen(longer) + b = x_len_longer
1951 * m * strlen(shorter) + b = x_len_shorter;
1952 * subtracting yields:
1953 * m * (strlen(longer) - strlen(shorter))
1954 * = x_len_longer - x_len_shorter
1955 * But we have set things up so that 'shorter' is 1 byte smaller
1956 * than 'longer'. Hence:
1957 * m = x_len_longer - x_len_shorter
1958 *
1959 * But if something went wrong, make sure the multiplier is at
1960 * least 1.
1961 */
1962 if (x_len_longer > x_len_shorter) {
1963 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1964 }
1965 else {
1966 PL_collxfrm_mult = 1;
1967 }
1968
1969 /* mx + b = len
1970 * so: b = len - mx
1971 * but in case something has gone wrong, make sure it is
1972 * non-negative */
1973 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1974 if (base < 0) {
1975 base = 0;
1976 }
1977
1978 /* Add 1 for the trailing NUL */
1979 PL_collxfrm_base = base + 1;
1980 }
58eebef2 1981
7d4bcc4a
KW
1982# ifdef DEBUGGING
1983
58eebef2
KW
1984 if (DEBUG_L_TEST || debug_initialization) {
1985 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1986 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1987 "x_len_longer=%zu,"
1988 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1989 __FILE__, __LINE__,
1990 PL_in_utf8_COLLATE_locale,
1991 x_len_shorter, x_len_longer,
1992 PL_collxfrm_mult, PL_collxfrm_base);
1993 }
7d4bcc4a
KW
1994# endif
1995
98994639
HS
1996 }
1997 }
1998
1999#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 2000
98994639
HS
2001}
2002
d36adde0
KW
2003#endif
2004
d2b24094 2005#ifdef WIN32
b8cc575c 2006
a4f00dcc 2007STATIC char *
b8cc575c 2008S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2009{
2010 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2011 * difference between the two unless the input locale is "", which normally
2012 * means on Windows to get the machine default, which is set via the
2013 * computer's "Regional and Language Options" (or its current equivalent).
2014 * In POSIX, it instead means to find the locale from the user's
2015 * environment. This routine changes the Windows behavior to first look in
2016 * the environment, and, if anything is found, use that instead of going to
2017 * the machine default. If there is no environment override, the machine
2018 * default is used, by calling the real setlocale() with "".
2019 *
2020 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2021 * use the particular category's variable if set; otherwise to use the LANG
2022 * variable. */
b385bb4d 2023
175c4cf9 2024 bool override_LC_ALL = FALSE;
89f7b9aa 2025 char * result;
e5f10d49 2026 unsigned int i;
89f7b9aa 2027
b385bb4d 2028 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2029
2030# ifdef LC_ALL
2031
b385bb4d
KW
2032 locale = PerlEnv_getenv("LC_ALL");
2033 if (! locale) {
e5f10d49
KW
2034 if (category == LC_ALL) {
2035 override_LC_ALL = TRUE;
2036 }
2037 else {
7d4bcc4a
KW
2038
2039# endif
7d4bcc4a 2040
e5f10d49
KW
2041 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2042 if (category == categories[i]) {
2043 locale = PerlEnv_getenv(category_names[i]);
2044 goto found_locale;
2045 }
2046 }
7d4bcc4a 2047
b385bb4d 2048 locale = PerlEnv_getenv("LANG");
481465ea 2049 if (! locale) {
b385bb4d
KW
2050 locale = "";
2051 }
e5f10d49
KW
2052
2053 found_locale: ;
7d4bcc4a
KW
2054
2055# ifdef LC_ALL
2056
e5f10d49 2057 }
b385bb4d 2058 }
7d4bcc4a
KW
2059
2060# endif
2061
b385bb4d
KW
2062 }
2063
89f7b9aa 2064 result = setlocale(category, locale);
48015184
KW
2065 DEBUG_L(STMT_START {
2066 dSAVE_ERRNO;
2067 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2068 setlocale_debug_string(category, locale, result));
2069 RESTORE_ERRNO;
2070 } STMT_END);
89f7b9aa 2071
481465ea 2072 if (! override_LC_ALL) {
89f7b9aa
KW
2073 return result;
2074 }
2075
dfd77d7a 2076 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2077 * LANG variable or the system default if there is no LANG. But these have
2078 * lower priority than the other LC_foo variables, so override it for each
2079 * one that is set. (If they are set to "", it means to use the same thing
2080 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2081
948523db 2082 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2083 result = PerlEnv_getenv(category_names[i]);
2084 if (result && strNE(result, "")) {
2085 setlocale(categories[i], result);
2086 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2087 __FILE__, __LINE__,
2088 setlocale_debug_string(categories[i], result, "not captured")));
2089 }
89f7b9aa 2090 }
7d4bcc4a 2091
bbc98134 2092 result = setlocale(LC_ALL, NULL);
48015184
KW
2093 DEBUG_L(STMT_START {
2094 dSAVE_ERRNO;
2095 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2096 __FILE__, __LINE__,
48015184
KW
2097 setlocale_debug_string(LC_ALL, NULL, result));
2098 RESTORE_ERRNO;
2099 } STMT_END);
89f7b9aa 2100
bbc98134 2101 return result;
b385bb4d
KW
2102}
2103
2104#endif
2105
9aac5db8
KW
2106/*
2107
2108=head1 Locale-related functions and macros
2109
2110=for apidoc Perl_setlocale
2111
2112This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2113taking the same parameters, and returning the same information, except that it
9487427b
KW
2114returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2115instead return C<C> if the underlying locale has a non-dot decimal point
2116character, or a non-empty thousands separator for displaying floating point
2117numbers. This is because perl keeps that locale category such that it has a
2118dot and empty separator, changing the locale briefly during the operations
2119where the underlying one is required. C<Perl_setlocale> knows about this, and
2120compensates; regular C<setlocale> doesn't.
9aac5db8 2121
e9bc6d6b 2122Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2123declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2124C<const> (presumably because its API was specified long ago, and can't be
2125updated; it is illegal to change the information C<setlocale> returns; doing
2126so leads to segfaults.)
9aac5db8 2127
e9bc6d6b
KW
2128Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2129C<setlocale> can be completely ineffective on some platforms under some
2130configurations.
2131
9aac5db8 2132C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2133where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2134the system C<setlocale()> is ineffective, returning the wrong information, and
2135failing to actually change the locale. C<Perl_setlocale>, however works
2136properly in all circumstances.
9aac5db8
KW
2137
2138The return points to a per-thread static buffer, which is overwritten the next
2139time C<Perl_setlocale> is called from the same thread.
2140
2141=cut
2142
2143*/
2144
2145const char *
2146Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2147{
2148 /* This wraps POSIX::setlocale() */
2149
d36adde0
KW
2150#ifdef NO_LOCALE
2151
2152 PERL_UNUSED_ARG(category);
2153 PERL_UNUSED_ARG(locale);
2154
2155 return "C";
2156
2157#else
2158
9aac5db8
KW
2159 const char * retval;
2160 const char * newlocale;
2161 dSAVEDERRNO;
a4f00dcc 2162 dTHX;
d36adde0 2163 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2164
a4f00dcc
KW
2165#ifdef USE_LOCALE_NUMERIC
2166
291a84fb
KW
2167 /* A NULL locale means only query what the current one is. We have the
2168 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2169 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2170 * the correct results. All other categories don't require special
2171 * handling */
a4f00dcc
KW
2172 if (locale == NULL) {
2173 if (category == LC_NUMERIC) {
9aac5db8
KW
2174
2175 /* We don't have to copy this return value, as it is a per-thread
2176 * variable, and won't change until a future setlocale */
2177 return PL_numeric_name;
a4f00dcc
KW
2178 }
2179
7d4bcc4a 2180# ifdef LC_ALL
a4f00dcc 2181
9aac5db8
KW
2182 else if (category == LC_ALL) {
2183 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2184 }
2185
7d4bcc4a 2186# endif
a4f00dcc
KW
2187
2188 }
2189
2190#endif
2191
33e5a354
KW
2192 retval = save_to_buffer(do_setlocale_r(category, locale),
2193 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2194 SAVE_ERRNO;
2195
2196#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2197
2198 if (locale == NULL && category == LC_ALL) {
2199 RESTORE_LC_NUMERIC();
2200 }
2201
2202#endif
a4f00dcc
KW
2203
2204 DEBUG_L(PerlIO_printf(Perl_debug_log,
2205 "%s:%d: %s\n", __FILE__, __LINE__,
2206 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2207
9aac5db8
KW
2208 RESTORE_ERRNO;
2209
2210 if (! retval) {
a4f00dcc
KW
2211 return NULL;
2212 }
2213
9aac5db8 2214 /* If locale == NULL, we are just querying the state */
a4f00dcc 2215 if (locale == NULL) {
a4f00dcc
KW
2216 return retval;
2217 }
a4f00dcc 2218
1159483a
KW
2219 /* Now that have switched locales, we have to update our records to
2220 * correspond. */
a4f00dcc 2221
1159483a 2222 switch (category) {
a4f00dcc 2223
1159483a 2224#ifdef USE_LOCALE_CTYPE
a4f00dcc 2225
1159483a
KW
2226 case LC_CTYPE:
2227 new_ctype(retval);
2228 break;
a4f00dcc 2229
1159483a 2230#endif
a4f00dcc
KW
2231#ifdef USE_LOCALE_COLLATE
2232
1159483a
KW
2233 case LC_COLLATE:
2234 new_collate(retval);
2235 break;
a4f00dcc 2236
1159483a 2237#endif
a4f00dcc
KW
2238#ifdef USE_LOCALE_NUMERIC
2239
1159483a
KW
2240 case LC_NUMERIC:
2241 new_numeric(retval);
2242 break;
a4f00dcc 2243
1159483a
KW
2244#endif
2245#ifdef LC_ALL
a4f00dcc 2246
1159483a 2247 case LC_ALL:
a4f00dcc 2248
1159483a
KW
2249 /* LC_ALL updates all the things we care about. The values may not
2250 * be the same as 'retval', as the locale "" may have set things
2251 * individually */
a4f00dcc 2252
1159483a 2253# ifdef USE_LOCALE_CTYPE
a4f00dcc 2254
b0bde642 2255 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2256 new_ctype(newlocale);
b0bde642 2257 Safefree(newlocale);
a4f00dcc 2258
1159483a
KW
2259# endif /* USE_LOCALE_CTYPE */
2260# ifdef USE_LOCALE_COLLATE
2261
b0bde642 2262 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2263 new_collate(newlocale);
b0bde642 2264 Safefree(newlocale);
a4f00dcc 2265
7d4bcc4a 2266# endif
1159483a 2267# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2268
b0bde642 2269 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2270 new_numeric(newlocale);
b0bde642 2271 Safefree(newlocale);
a4f00dcc 2272
1159483a
KW
2273# endif /* USE_LOCALE_NUMERIC */
2274#endif /* LC_ALL */
a4f00dcc 2275
1159483a
KW
2276 default:
2277 break;
a4f00dcc
KW
2278 }
2279
2280 return retval;
2281
d36adde0
KW
2282#endif
2283
f7416781
KW
2284}
2285
2286PERL_STATIC_INLINE const char *
2287S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2288{
2289 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2290 * growing it if necessary */
2291
0b6697c5 2292 Size_t string_size;
f7416781
KW
2293
2294 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2295
0b6697c5
KW
2296 if (! string) {
2297 return NULL;
2298 }
2299
2300 string_size = strlen(string) + offset + 1;
2301
f7416781
KW
2302 if (*buf_size == 0) {
2303 Newx(*buf, string_size, char);
2304 *buf_size = string_size;
2305 }
2306 else if (string_size > *buf_size) {
2307 Renew(*buf, string_size, char);
2308 *buf_size = string_size;
2309 }
2310
2311 Copy(string, *buf + offset, string_size - offset, char);
2312 return *buf;
2313}
2314
2315/*
2316
f7416781
KW
2317=for apidoc Perl_langinfo
2318
ce181bf6 2319This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2320taking the same C<item> parameter values, and returning the same information.
2321But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2322of Perl's locale handling from your code, and can be used on systems that lack
2323a native C<nl_langinfo>.
2324
2325Expanding on these:
2326
2327=over
2328
2329=item *
2330
ce181bf6
KW
2331The reason it isn't quite a drop-in replacement is actually an advantage. The
2332only difference is that it returns S<C<const char *>>, whereas plain
2333C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2334forbidden to write into the buffer. By declaring this C<const>, the compiler
2335enforces this restriction, so if it is violated, you know at compilation time,
2336rather than getting segfaults at runtime.
2337
2338=item *
2339
69e2275e 2340It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2341without you having to write extra code. The reason for the extra code would be
2342because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2343kept set by Perl so that the radix is a dot, and the separator is the empty
2344string, no matter what the underlying locale is supposed to be, and so to get
2345the expected results, you have to temporarily toggle into the underlying
2346locale, and later toggle back. (You could use plain C<nl_langinfo> and
2347C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2348the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2349(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2350(decimal point) character to be a dot.)
ce181bf6
KW
2351
2352=item *
2353
2354The system function it replaces can have its static return buffer trashed,
2355not only by a subesequent call to that function, but by a C<freelocale>,
2356C<setlocale>, or other locale change. The returned buffer of this function is
2357not changed until the next call to it, so the buffer is never in a trashed
2358state.
f7416781
KW
2359
2360=item *
2361
ce181bf6
KW
2362Its return buffer is per-thread, so it also is never overwritten by a call to
2363this function from another thread; unlike the function it replaces.
2364
2365=item *
2366
2367But most importantly, it works on systems that don't have C<nl_langinfo>, such
2368as Windows, hence makes your code more portable. Of the fifty-some possible
2369items specified by the POSIX 2008 standard,
f7416781 2370L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2371only one is completely unimplemented, though on non-Windows platforms, another
2372significant one is also not implemented). It uses various techniques to
2373recover the other items, including calling C<L<localeconv(3)>>, and
2374C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2375available. Later C<strftime()> versions have additional capabilities; C<""> is
2376returned for those not available on your system.
ce181bf6
KW
2377
2378It is important to note that when called with an item that is recovered by
2379using C<localeconv>, the buffer from any previous explicit call to
2380C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2381contents if you need to access them after a call to this function. (But note
2382that you might not want to be using C<localeconv()> directly anyway, because of
2383issues like the ones listed in the second item of this list (above) for
2384C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2385call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2386unpack).
9f0bc911 2387
6201c220
KW
2388The details for those items which may deviate from what this emulation returns
2389and what a native C<nl_langinfo()> would return are specified in
2390L<I18N::Langinfo>.
f7416781 2391
ce181bf6
KW
2392=back
2393
f7416781
KW
2394When using C<Perl_langinfo> on systems that don't have a native
2395C<nl_langinfo()>, you must
2396
2397 #include "perl_langinfo.h"
2398
2399before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2400C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2401C<langinfo.h> would try to import into the namespace for code that doesn't need
2402it.)
f7416781 2403
f7416781
KW
2404The original impetus for C<Perl_langinfo()> was so that code that needs to
2405find out the current currency symbol, floating point radix character, or digit
2406grouping separator can use, on all systems, the simpler and more
2407thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2408pain to make thread-friendly. For other fields returned by C<localeconv>, it
2409is better to use the methods given in L<perlcall> to call
2410L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2411
2412=cut
2413
2414*/
2415
2416const char *
2417#ifdef HAS_NL_LANGINFO
2418Perl_langinfo(const nl_item item)
2419#else
2420Perl_langinfo(const int item)
2421#endif
2422{
f61748ac
KW
2423 return my_nl_langinfo(item, TRUE);
2424}
2425
59a15af0 2426STATIC const char *
f61748ac
KW
2427#ifdef HAS_NL_LANGINFO
2428S_my_nl_langinfo(const nl_item item, bool toggle)
2429#else
2430S_my_nl_langinfo(const int item, bool toggle)
2431#endif
2432{
ae74815b 2433 dTHX;
0b6697c5 2434 const char * retval;
f7416781 2435
d36adde0
KW
2436#ifdef USE_LOCALE_NUMERIC
2437
5a854ab3
KW
2438 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2439 * two items, and only if not already there */
4e6826bf 2440 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2441 || PL_numeric_underlying))
d36adde0
KW
2442
2443#endif /* No toggling needed if not using LC_NUMERIC */
2444
5a854ab3 2445 toggle = FALSE;
5a854ab3 2446
ab340fff 2447#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2448# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2449 || ! defined(HAS_POSIX_2008_LOCALE) \
2450 || ! defined(DUPLOCALE)
f7416781 2451
ab340fff 2452 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2453 * for those items dependent on it. This must be copied to a buffer before
2454 * switching back, as some systems destroy the buffer when setlocale() is
2455 * called */
f7416781 2456
038d3702
KW
2457 {
2458 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2459
b2fee59c 2460 if (toggle) {
038d3702 2461 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2462 }
f7416781 2463
5acc4454
KW
2464 LOCALE_LOCK; /* Prevent interference from another thread executing
2465 this code section (the only call to nl_langinfo in
2466 the core) */
2467
ee90eb2d 2468
628ff75a
KW
2469 /* Copy to a per-thread buffer, which is also one that won't be
2470 * destroyed by a subsequent setlocale(), such as the
2471 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2472 retval = save_to_buffer(nl_langinfo(item),
2473 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2474
5acc4454
KW
2475 LOCALE_UNLOCK;
2476
b2fee59c 2477 if (toggle) {
038d3702 2478 RESTORE_LC_NUMERIC();
b2fee59c 2479 }
f7416781
KW
2480 }
2481
ab340fff
KW
2482# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2483
5a854ab3 2484 {
b2fee59c
KW
2485 bool do_free = FALSE;
2486 locale_t cur = uselocale((locale_t) 0);
ab340fff 2487
b2fee59c
KW
2488 if (cur == LC_GLOBAL_LOCALE) {
2489 cur = duplocale(LC_GLOBAL_LOCALE);
2490 do_free = TRUE;
2491 }
ab340fff 2492
d36adde0
KW
2493# ifdef USE_LOCALE_NUMERIC
2494
b2fee59c 2495 if (toggle) {
e1aa2579
KW
2496 if (PL_underlying_numeric_obj) {
2497 cur = PL_underlying_numeric_obj;
2498 }
2499 else {
2500 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2501 do_free = TRUE;
2502 }
b2fee59c 2503 }
ab340fff 2504
d36adde0
KW
2505# endif
2506
628ff75a
KW
2507 /* We have to save it to a buffer, because the freelocale() just below
2508 * can invalidate the internal one */
0b6697c5
KW
2509 retval = save_to_buffer(nl_langinfo_l(item, cur),
2510 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2511
b2fee59c
KW
2512 if (do_free) {
2513 freelocale(cur);
2514 }
5a854ab3 2515 }
ab340fff 2516
c1566110
KW
2517# endif
2518
0b6697c5 2519 if (strEQ(retval, "")) {
4e6826bf 2520 if (item == YESSTR) {
c1566110
KW
2521 return "yes";
2522 }
4e6826bf 2523 if (item == NOSTR) {
c1566110
KW
2524 return "no";
2525 }
2526 }
2527
0b6697c5 2528 return retval;
ab340fff 2529
f7416781 2530#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2531
2532 {
2533
f7416781
KW
2534# ifdef HAS_LOCALECONV
2535
43dd6b15 2536 const struct lconv* lc;
628ff75a 2537 const char * temp;
038d3702 2538 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2539
69c5e0db
KW
2540# ifdef TS_W32_BROKEN_LOCALECONV
2541
2542 const char * save_global;
2543 const char * save_thread;
2544 int needed_size;
2545 char * ptr;
2546 char * e;
2547 char * item_start;
2548
2549# endif
f7416781
KW
2550# endif
2551# ifdef HAS_STRFTIME
2552
43dd6b15
KW
2553 struct tm tm;
2554 bool return_format = FALSE; /* Return the %format, not the value */
2555 const char * format;
f7416781
KW
2556
2557# endif
2558
43dd6b15
KW
2559 /* We copy the results to a per-thread buffer, even if not
2560 * multi-threaded. This is in part to simplify this code, and partly
2561 * because we need a buffer anyway for strftime(), and partly because a
2562 * call of localeconv() could otherwise wipe out the buffer, and the
2563 * programmer would not be expecting this, as this is a nl_langinfo()
2564 * substitute after all, so s/he might be thinking their localeconv()
2565 * is safe until another localeconv() call. */
f7416781 2566
43dd6b15
KW
2567 switch (item) {
2568 Size_t len;
f7416781 2569
8d72e74e 2570 /* This is unimplemented */
4e6826bf 2571 case ERA: /* For use with strftime() %E modifier */
f7416781 2572
43dd6b15
KW
2573 default:
2574 return "";
f7416781 2575
43dd6b15 2576 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2577 case YESEXPR: return "^[+1yY]";
2578 case YESSTR: return "yes";
2579 case NOEXPR: return "^[-0nN]";
2580 case NOSTR: return "no";
2581
8d72e74e
KW
2582 case CODESET:
2583
2584# ifndef WIN32
2585
2586 /* On non-windows, this is unimplemented, in part because of
2587 * inconsistencies between vendors. The Darwin native
2588 * nl_langinfo() implementation simply looks at everything past
2589 * any dot in the name, but that doesn't work for other
2590 * vendors. Many Linux locales that don't have UTF-8 in their
2591 * names really are UTF-8, for example; z/OS locales that do
2592 * have UTF-8 in their names, aren't really UTF-8 */
2593 return "";
2594
2595# else
2596
2597 { /* But on Windows, the name does seem to be consistent, so
2598 use that. */
2599 const char * p;
2600 const char * first;
2601 Size_t offset = 0;
2602 const char * name = my_setlocale(LC_CTYPE, NULL);
2603
2604 if (isNAME_C_OR_POSIX(name)) {
2605 return "ANSI_X3.4-1968";
2606 }
2607
2608 /* Find the dot in the locale name */
2609 first = (const char *) strchr(name, '.');
2610 if (! first) {
2611 first = name;
2612 goto has_nondigit;
2613 }
2614
2615 /* Look at everything past the dot */
2616 first++;
2617 p = first;
2618
2619 while (*p) {
2620 if (! isDIGIT(*p)) {
2621 goto has_nondigit;
2622 }
2623
2624 p++;
2625 }
2626
2627 /* Here everything past the dot is a digit. Treat it as a
2628 * code page */
18503cbd 2629 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2630 &PL_langinfo_bufsize, 0);
8d72e74e 2631 offset = STRLENs("CP");
f7416781 2632
8d72e74e
KW
2633 has_nondigit:
2634
2635 retval = save_to_buffer(first, &PL_langinfo_buf,
2636 &PL_langinfo_bufsize, offset);
2637 }
2638
2639 break;
2640
2641# endif
f7416781
KW
2642# ifdef HAS_LOCALECONV
2643
4e6826bf 2644 case CRNCYSTR:
f7416781 2645
43dd6b15
KW
2646 /* We don't bother with localeconv_l() because any system that
2647 * has it is likely to also have nl_langinfo() */
291a84fb 2648
69c5e0db
KW
2649 LOCALE_LOCK_V; /* Prevent interference with other threads
2650 using localeconv() */
2651
2652# ifdef TS_W32_BROKEN_LOCALECONV
2653
2654 /* This is a workaround for a Windows bug prior to VS 15.
2655 * What we do here is, while locked, switch to the global
2656 * locale so localeconv() works; then switch back just before
2657 * the unlock. This can screw things up if some thread is
2658 * already using the global locale while assuming no other is.
2659 * A different workaround would be to call GetCurrencyFormat on
2660 * a known value, and parse it; patches welcome
2661 *
2662 * We have to use LC_ALL instead of LC_MONETARY because of
2663 * another bug in Windows */
2664
2665 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2666 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2667 save_global= savepv(my_setlocale(LC_ALL, NULL));
2668 my_setlocale(LC_ALL, save_thread);
2669
2670# endif
5acc4454 2671
43dd6b15
KW
2672 lc = localeconv();
2673 if ( ! lc
2674 || ! lc->currency_symbol
2675 || strEQ("", lc->currency_symbol))
2676 {
69c5e0db 2677 LOCALE_UNLOCK_V;
43dd6b15
KW
2678 return "";
2679 }
f7416781 2680
43dd6b15 2681 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2682 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2683 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2684 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2685 { /* khw couldn't figure out how the localedef specifications
2686 would show that the $ should replace the radix; this is
2687 just a guess as to how it might work.*/
a34edee3 2688 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2689 }
2690 else if (lc->p_cs_precedes) {
a34edee3 2691 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2692 }
2693 else {
a34edee3 2694 PL_langinfo_buf[0] = '+';
43dd6b15 2695 }
f7416781 2696
69c5e0db
KW
2697# ifdef TS_W32_BROKEN_LOCALECONV
2698
2699 my_setlocale(LC_ALL, save_global);
2700 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2701 my_setlocale(LC_ALL, save_thread);
2702 Safefree(save_global);
2703 Safefree(save_thread);
2704
2705# endif
2706
2707 LOCALE_UNLOCK_V;
43dd6b15 2708 break;
f7416781 2709
69c5e0db
KW
2710# ifdef TS_W32_BROKEN_LOCALECONV
2711
4e6826bf 2712 case RADIXCHAR:
69c5e0db
KW
2713
2714 /* For this, we output a known simple floating point number to
2715 * a buffer, and parse it, looking for the radix */
2716
2717 if (toggle) {
2718 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2719 }
2720
2721 if (PL_langinfo_bufsize < 10) {
2722 PL_langinfo_bufsize = 10;
2723 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2724 }
2725
2726 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2727 "%.1f", 1.5);
2728 if (needed_size >= (int) PL_langinfo_bufsize) {
2729 PL_langinfo_bufsize = needed_size + 1;
2730 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2731 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2732 "%.1f", 1.5);
2733 assert(needed_size < (int) PL_langinfo_bufsize);
2734 }
2735
2736 ptr = PL_langinfo_buf;
2737 e = PL_langinfo_buf + PL_langinfo_bufsize;
2738
2739 /* Find the '1' */
2740 while (ptr < e && *ptr != '1') {
2741 ptr++;
2742 }
2743 ptr++;
2744
2745 /* Find the '5' */
2746 item_start = ptr;
2747 while (ptr < e && *ptr != '5') {
2748 ptr++;
2749 }
2750
2751 /* Everything in between is the radix string */
2752 if (ptr >= e) {
2753 PL_langinfo_buf[0] = '?';
2754 PL_langinfo_buf[1] = '\0';
2755 }
2756 else {
2757 *ptr = '\0';
2758 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2759 }
2760
2761 if (toggle) {
2762 RESTORE_LC_NUMERIC();
2763 }
2764
2765 retval = PL_langinfo_buf;
2766 break;
2767
2768# else
2769
2770 case RADIXCHAR: /* No special handling needed */
2771
2772# endif
2773
4e6826bf 2774 case THOUSEP:
f7416781 2775
43dd6b15 2776 if (toggle) {
038d3702 2777 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2778 }
f7416781 2779
69c5e0db
KW
2780 LOCALE_LOCK_V; /* Prevent interference with other threads
2781 using localeconv() */
2782
2783# ifdef TS_W32_BROKEN_LOCALECONV
2784
2785 /* This should only be for the thousands separator. A
2786 * different work around would be to use GetNumberFormat on a
2787 * known value and parse the result to find the separator */
2788 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2789 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2790 save_global = savepv(my_setlocale(LC_ALL, NULL));
2791 my_setlocale(LC_ALL, save_thread);
2792# if 0
2793 /* This is the start of code that for broken Windows replaces
2794 * the above and below code, and instead calls
2795 * GetNumberFormat() and then would parse that to find the
2796 * thousands separator. It needs to handle UTF-16 vs -8
2797 * issues. */
2798
2799 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2800 DEBUG_L(PerlIO_printf(Perl_debug_log,
2801 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2802 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2803
2804# endif
2805# endif
5acc4454 2806
43dd6b15
KW
2807 lc = localeconv();
2808 if (! lc) {
628ff75a 2809 temp = "";
33394adc 2810 }
43dd6b15 2811 else {
4e6826bf 2812 temp = (item == RADIXCHAR)
43dd6b15
KW
2813 ? lc->decimal_point
2814 : lc->thousands_sep;
628ff75a
KW
2815 if (! temp) {
2816 temp = "";
43dd6b15
KW
2817 }
2818 }
f7416781 2819
0b6697c5
KW
2820 retval = save_to_buffer(temp, &PL_langinfo_buf,
2821 &PL_langinfo_bufsize, 0);
f7416781 2822
69c5e0db
KW
2823# ifdef TS_W32_BROKEN_LOCALECONV
2824
2825 my_setlocale(LC_ALL, save_global);
2826 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2827 my_setlocale(LC_ALL, save_thread);
2828 Safefree(save_global);
2829 Safefree(save_thread);
2830
2831# endif
2832
2833 LOCALE_UNLOCK_V;
5acc4454 2834
43dd6b15 2835 if (toggle) {
038d3702 2836 RESTORE_LC_NUMERIC();
43dd6b15 2837 }
f7416781 2838
43dd6b15 2839 break;
f7416781
KW
2840
2841# endif
2842# ifdef HAS_STRFTIME
2843
43dd6b15
KW
2844 /* These are defined by C89, so we assume that strftime supports
2845 * them, and so are returned unconditionally; they may not be what
2846 * the locale actually says, but should give good enough results
2847 * for someone using them as formats (as opposed to trying to parse
2848 * them to figure out what the locale says). The other format
2849 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2850 case D_FMT: return "%x";
2851 case T_FMT: return "%X";
2852 case D_T_FMT: return "%c";
43dd6b15
KW
2853
2854 /* These formats are only available in later strfmtime's */
4e6826bf 2855 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2856
2857 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2858 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2859 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2860 case ALT_DIGITS:
2861 case AM_STR: case PM_STR:
2862 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2863 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2864 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2865 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2866 case DAY_5: case DAY_6: case DAY_7:
2867 case MON_1: case MON_2: case MON_3: case MON_4:
2868 case MON_5: case MON_6: case MON_7: case MON_8:
2869 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2870
2871 LOCALE_LOCK;
2872
2873 init_tm(&tm); /* Precaution against core dumps */
2874 tm.tm_sec = 30;
2875 tm.tm_min = 30;
2876 tm.tm_hour = 6;
2877 tm.tm_year = 2017 - 1900;
2878 tm.tm_wday = 0;
2879 tm.tm_mon = 0;
2880 switch (item) {
2881 default:
2882 LOCALE_UNLOCK;
2883 Perl_croak(aTHX_
2884 "panic: %s: %d: switch case: %d problem",
2885 __FILE__, __LINE__, item);
2886 NOT_REACHED; /* NOTREACHED */
2887
4e6826bf
KW
2888 case PM_STR: tm.tm_hour = 18;
2889 case AM_STR:
43dd6b15
KW
2890 format = "%p";
2891 break;
2892
4e6826bf
KW
2893 case ABDAY_7: tm.tm_wday++;
2894 case ABDAY_6: tm.tm_wday++;
2895 case ABDAY_5: tm.tm_wday++;
2896 case ABDAY_4: tm.tm_wday++;
2897 case ABDAY_3: tm.tm_wday++;
2898 case ABDAY_2: tm.tm_wday++;
2899 case ABDAY_1:
43dd6b15
KW
2900 format = "%a";
2901 break;
2902
4e6826bf
KW
2903 case DAY_7: tm.tm_wday++;
2904 case DAY_6: tm.tm_wday++;
2905 case DAY_5: tm.tm_wday++;
2906 case DAY_4: tm.tm_wday++;
2907 case DAY_3: tm.tm_wday++;
2908 case DAY_2: tm.tm_wday++;
2909 case DAY_1:
43dd6b15
KW
2910 format = "%A";
2911 break;
2912
4e6826bf
KW
2913 case ABMON_12: tm.tm_mon++;
2914 case ABMON_11: tm.tm_mon++;
2915 case ABMON_10: tm.tm_mon++;
2916 case ABMON_9: tm.tm_mon++;
2917 case ABMON_8: tm.tm_mon++;
2918 case ABMON_7: tm.tm_mon++;
2919 case ABMON_6: tm.tm_mon++;
2920 case ABMON_5: tm.tm_mon++;
2921 case ABMON_4: tm.tm_mon++;
2922 case ABMON_3: tm.tm_mon++;
2923 case ABMON_2: tm.tm_mon++;
2924 case ABMON_1:
43dd6b15
KW
2925 format = "%b";
2926 break;
2927
4e6826bf
KW
2928 case MON_12: tm.tm_mon++;
2929 case MON_11: tm.tm_mon++;
2930 case MON_10: tm.tm_mon++;
2931 case MON_9: tm.tm_mon++;
2932 case MON_8: tm.tm_mon++;
2933 case MON_7: tm.tm_mon++;
2934 case MON_6: tm.tm_mon++;
2935 case MON_5: tm.tm_mon++;
2936 case MON_4: tm.tm_mon++;
2937 case MON_3: tm.tm_mon++;
2938 case MON_2: tm.tm_mon++;
2939 case MON_1:
43dd6b15
KW
2940 format = "%B";
2941 break;
2942
4e6826bf 2943 case T_FMT_AMPM:
43dd6b15
KW
2944 format = "%r";
2945 return_format = TRUE;
2946 break;
2947
4e6826bf 2948 case ERA_D_FMT:
43dd6b15
KW
2949 format = "%Ex";
2950 return_format = TRUE;
2951 break;
2952
4e6826bf 2953 case ERA_T_FMT:
43dd6b15
KW
2954 format = "%EX";
2955 return_format = TRUE;
2956 break;
2957
4e6826bf 2958 case ERA_D_T_FMT:
43dd6b15
KW
2959 format = "%Ec";
2960 return_format = TRUE;
2961 break;
2962
4e6826bf 2963 case ALT_DIGITS:
43dd6b15
KW
2964 tm.tm_wday = 0;
2965 format = "%Ow"; /* Find the alternate digit for 0 */
2966 break;
2967 }
f7416781 2968
43dd6b15
KW
2969 /* We can't use my_strftime() because it doesn't look at
2970 * tm_wday */
2971 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2972 format, &tm))
2973 {
2974 /* A zero return means one of:
2975 * a) there wasn't enough space in PL_langinfo_buf
2976 * b) the format, like a plain %p, returns empty
2977 * c) it was an illegal format, though some
2978 * implementations of strftime will just return the
2979 * illegal format as a plain character sequence.
2980 *
2981 * To quickly test for case 'b)', try again but precede
2982 * the format with a plain character. If that result is
2983 * still empty, the problem is either 'a)' or 'c)' */
2984
2985 Size_t format_size = strlen(format) + 1;
2986 Size_t mod_size = format_size + 1;
2987 char * mod_format;
2988 char * temp_result;
2989
2990 Newx(mod_format, mod_size, char);
2991 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2992 *mod_format = ' ';
43dd6b15
KW
2993 my_strlcpy(mod_format + 1, format, mod_size);
2994 len = strftime(temp_result,
2995 PL_langinfo_bufsize,
2996 mod_format, &tm);
2997 Safefree(mod_format);
2998 Safefree(temp_result);
2999
3000 /* If 'len' is non-zero, it means that we had a case like
3001 * %p which means the current locale doesn't use a.m. or
3002 * p.m., and that is valid */
3003 if (len == 0) {
3004
3005 /* Here, still didn't work. If we get well beyond a
3006 * reasonable size, bail out to prevent an infinite
3007 * loop. */
3008
3009 if (PL_langinfo_bufsize > 100 * format_size) {
3010 *PL_langinfo_buf = '\0';
3011 }
3012 else {
3013 /* Double the buffer size to retry; Add 1 in case
3014 * original was 0, so we aren't stuck at 0. */
3015 PL_langinfo_bufsize *= 2;
3016 PL_langinfo_bufsize++;
3017 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3018 continue;
3019 }
3020 }
f7416781 3021
f7416781 3022 break;
43dd6b15 3023 }
f7416781 3024
43dd6b15
KW
3025 /* Here, we got a result.
3026 *
3027 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3028 * alternate format for wday 0. If the value is the same as
3029 * the normal 0, there isn't an alternate, so clear the buffer.
3030 * */
4e6826bf 3031 if ( item == ALT_DIGITS
43dd6b15
KW
3032 && strEQ(PL_langinfo_buf, "0"))
3033 {
3034 *PL_langinfo_buf = '\0';
3035 }
f7416781 3036
43dd6b15
KW
3037 /* ALT_DIGITS is problematic. Experiments on it showed that
3038 * strftime() did not always work properly when going from
3039 * alt-9 to alt-10. Only a few locales have this item defined,
3040 * and in all of them on Linux that khw was able to find,
3041 * nl_langinfo() merely returned the alt-0 character, possibly
3042 * doubled. Most Unicode digits are in blocks of 10
3043 * consecutive code points, so that is sufficient information
3044 * for those scripts, as we can infer alt-1, alt-2, .... But
3045 * for a Japanese locale, a CJK ideographic 0 is returned, and
3046 * the CJK digits are not in code point order, so you can't
3047 * really infer anything. The localedef for this locale did
3048 * specify the succeeding digits, so that strftime() works
3049 * properly on them, without needing to infer anything. But
3050 * the nl_langinfo() return did not give sufficient information
3051 * for the caller to understand what's going on. So until
3052 * there is evidence that it should work differently, this
3053 * returns the alt-0 string for ALT_DIGITS.
3054 *
3055 * wday was chosen because its range is all a single digit.
3056 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3057
43dd6b15 3058 LOCALE_UNLOCK;
f7416781 3059
d1b150d9
KW
3060 retval = PL_langinfo_buf;
3061
43dd6b15
KW
3062 /* If to return the format, not the value, overwrite the buffer
3063 * with it. But some strftime()s will keep the original format
3064 * if illegal, so change those to "" */
3065 if (return_format) {
3066 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3067 *PL_langinfo_buf = '\0';
3068 }
43dd6b15 3069 else {
0b6697c5
KW
3070 retval = save_to_buffer(format, &PL_langinfo_buf,
3071 &PL_langinfo_bufsize, 0);
f7416781
KW
3072 }
3073 }
3074
3075 break;
f7416781
KW
3076
3077# endif
3078
43dd6b15 3079 }
f7416781
KW
3080 }
3081
0b6697c5 3082 return retval;
f7416781
KW
3083
3084#endif
3085
a4f00dcc 3086}
b385bb4d 3087
98994639
HS
3088/*
3089 * Initialize locale awareness.
3090 */
3091int
3092Perl_init_i18nl10n(pTHX_ int printwarn)
3093{
0e92a118
KW
3094 /* printwarn is
3095 *
3096 * 0 if not to output warning when setup locale is bad
3097 * 1 if to output warning based on value of PERL_BADLANG
3098 * >1 if to output regardless of PERL_BADLANG
3099 *
3100 * returns
98994639 3101 * 1 = set ok or not applicable,
0e92a118
KW
3102 * 0 = fallback to a locale of lower priority
3103 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3104 *
3105 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3106 * set, debugging information is output.
3107 *
3108 * This looks more complicated than it is, mainly due to the #ifdefs.
3109 *
3110 * We try to set LC_ALL to the value determined by the environment. If
3111 * there is no LC_ALL on this platform, we try the individual categories we
3112 * know about. If this works, we are done.
3113 *
3114 * But if it doesn't work, we have to do something else. We search the
3115 * environment variables ourselves instead of relying on the system to do
3116 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3117 * think there is one), and the ultimate fallback "C". This is all done in
3118 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3119 * more complex. The 'trial_locales' array is initialized with just one
3120 * element; it causes the behavior described in the paragraph above this to
3121 * happen. If that fails, we add elements to 'trial_locales', and do extra
3122 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3123 *
3124 * On Ultrix, the locale MUST come from the environment, so there is
3125 * preliminary code to set it. I (khw) am not sure that it is necessary,
3126 * and that this couldn't be folded into the loop, but barring any real
3127 * platforms to test on, it's staying as-is
3128 *
3129 * A slight complication is that in embedded Perls, the locale may already
3130 * be set-up, and we don't want to get it from the normal environment
3131 * variables. This is handled by having a special environment variable
3132 * indicate we're in this situation. We simply set setlocale's 2nd
3133 * parameter to be a NULL instead of "". That indicates to setlocale that
3134 * it is not to change anything, but to return the current value,
3135 * effectively initializing perl's db to what the locale already is.
3136 *
3137 * We play the same trick with NULL if a LC_ALL succeeds. We call
3138 * setlocale() on the individual categores with NULL to get their existing
3139 * values for our db, instead of trying to change them.
3140 * */
98994639 3141
0e92a118
KW
3142 int ok = 1;
3143
7d4bcc4a
KW
3144#ifndef USE_LOCALE
3145
3146 PERL_UNUSED_ARG(printwarn);
3147
3148#else /* USE_LOCALE */
7d4bcc4a
KW
3149# ifdef __GLIBC__
3150
175c4cf9 3151 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3152
3153# endif
65ebb059 3154
ccd65d51
KW
3155 /* NULL uses the existing already set up locale */
3156 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3157 ? NULL
3158 : "";
c3fcd832
KW
3159 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3160 unsigned int trial_locales_count;
175c4cf9
KW
3161 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3162 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3163 bool setlocale_failure = FALSE;
65ebb059 3164 unsigned int i;
175c4cf9
KW
3165
3166 /* A later getenv() could zap this, so only use here */
3167 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3168
3169 const bool locwarn = (printwarn > 1
e5f10d49
KW
3170 || ( printwarn
3171 && ( ! bad_lang_use_once
22ff3130 3172 || (
e5f10d49
KW
3173 /* disallow with "" or "0" */
3174 *bad_lang_use_once
3175 && strNE("0", bad_lang_use_once)))));
ea92aad8 3176
291a84fb 3177 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3178 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3179
3180 /* current locale for given category; should have been copied so aren't
3181 * volatile */
8de4332b 3182 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3183
7d4bcc4a
KW
3184# ifdef WIN32
3185
6bce99ee
JH
3186 /* In some systems you can find out the system default locale
3187 * and use that as the fallback locale. */
7d4bcc4a
KW
3188# define SYSTEM_DEFAULT_LOCALE
3189# endif
3190# ifdef SYSTEM_DEFAULT_LOCALE
3191
65ebb059 3192 const char *system_default_locale = NULL;
98994639 3193
7d4bcc4a 3194# endif
948523db
KW
3195
3196# ifndef DEBUGGING
3197# define DEBUG_LOCALE_INIT(a,b,c)
3198# else
7d4bcc4a 3199
8298454c 3200 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3201
3202# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3203 STMT_START { \
3204 if (debug_initialization) { \
3205 PerlIO_printf(Perl_debug_log, \
3206 "%s:%d: %s\n", \
3207 __FILE__, __LINE__, \
a4f00dcc 3208 setlocale_debug_string(category, \
2fcc0ca9
KW
3209 locale, \
3210 result)); \
3211 } \
3212 } STMT_END
2fcc0ca9 3213
948523db
KW
3214/* Make sure the parallel arrays are properly set up */
3215# ifdef USE_LOCALE_NUMERIC
3216 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3217 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3218# ifdef USE_POSIX_2008_LOCALE
3219 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3220# endif
948523db
KW
3221# endif
3222# ifdef USE_LOCALE_CTYPE
3223 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3224 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3225# ifdef USE_POSIX_2008_LOCALE
3226 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3227# endif
948523db
KW
3228# endif
3229# ifdef USE_LOCALE_COLLATE
3230 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3231 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3232# ifdef USE_POSIX_2008_LOCALE
3233 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3234# endif
948523db
KW
3235# endif
3236# ifdef USE_LOCALE_TIME
3237 assert(categories[LC_TIME_INDEX] == LC_TIME);
3238 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3239# ifdef USE_POSIX_2008_LOCALE
3240 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3241# endif
948523db
KW
3242# endif
3243# ifdef USE_LOCALE_MESSAGES
3244 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3245 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3246# ifdef USE_POSIX_2008_LOCALE
3247 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3248# endif
948523db
KW
3249# endif
3250# ifdef USE_LOCALE_MONETARY
3251 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3252 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3253# ifdef USE_POSIX_2008_LOCALE
3254 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3255# endif
948523db 3256# endif
9821811f
KW
3257# ifdef USE_LOCALE_ADDRESS
3258 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3259 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3260# ifdef USE_POSIX_2008_LOCALE
3261 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3262# endif
9821811f
KW
3263# endif
3264# ifdef USE_LOCALE_IDENTIFICATION
3265 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3266 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3267# ifdef USE_POSIX_2008_LOCALE
3268 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3269# endif
9821811f
KW
3270# endif
3271# ifdef USE_LOCALE_MEASUREMENT
3272 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3273 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3274# ifdef USE_POSIX_2008_LOCALE
3275 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3276# endif
9821811f
KW
3277# endif
3278# ifdef USE_LOCALE_PAPER
3279 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3280 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3281# ifdef USE_POSIX_2008_LOCALE
3282 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3283# endif
9821811f
KW
3284# endif
3285# ifdef USE_LOCALE_TELEPHONE
3286 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3287 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3288# ifdef USE_POSIX_2008_LOCALE
3289 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3290# endif
9821811f 3291# endif
948523db
KW
3292# ifdef LC_ALL
3293 assert(categories[LC_ALL_INDEX] == LC_ALL);
3294 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3295 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3296# ifdef USE_POSIX_2008_LOCALE
3297 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3298# endif
948523db
KW
3299# endif
3300# endif /* DEBUGGING */
47280b20
KW
3301
3302 /* Initialize the cache of the program's UTF-8ness for the always known
3303 * locales C and POSIX */
3304 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3305 sizeof(PL_locale_utf8ness));
3306
e9bc6d6b
KW
3307# ifdef USE_THREAD_SAFE_LOCALE
3308# ifdef WIN32
3309
3310 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3311
3312# endif
3313# endif
39e69e77 3314# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3315
3316 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3317 if (! PL_C_locale_obj) {
3318 Perl_croak_nocontext(
3319 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3320 }
3321 if (DEBUG_Lv_TEST || debug_initialization) {
3322 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3323 }
3324
3325# endif
3326
78f7c09f
KW
3327# ifdef USE_LOCALE_NUMERIC
3328
3ca88433
KW
3329 PL_numeric_radix_sv = newSVpvs(".");
3330
78f7c09f
KW
3331# endif
3332
e9bc6d6b
KW
3333# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3334
3335 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3336 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3337
3338# endif
b56c4436 3339# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3340
3341 /*
3342 * Ultrix setlocale(..., "") fails if there are no environment
3343 * variables from which to get a locale name.
3344 */
3345
b56c4436
KW
3346# ifndef LC_ALL
3347# error Ultrix without LC_ALL not implemented
3348# else
7d4bcc4a 3349
b56c4436
KW
3350 {
3351 bool done = FALSE;
ea92aad8
KW
3352 if (lang) {
3353 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3354 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3355 if (sl_result[LC_ALL_INDEX])
3356 done = TRUE;
3357 else
e5f10d49 3358 setlocale_failure = TRUE;
ea92aad8
KW
3359 }
3360 if (! setlocale_failure) {
3361 const char * locale_param;
3362 for (i = 0; i < LC_ALL_INDEX; i++) {
3363 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3364 ? setlocale_init
3365 : NULL;
3366 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3367 if (! sl_result[i]) {
3368 setlocale_failure = TRUE;
3369 }
3370 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3371 }
c835d6be 3372 }
7d4bcc4a
KW
3373 }
3374
3375# endif /* LC_ALL */
e5f10d49 3376# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3377
65ebb059 3378 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3379 * the list. Normally the loop is executed just once. But if setting the
3380 * locale fails, inside the loop we add fallback trials to the array and so
3381 * will execute the loop multiple times */
c3fcd832
KW
3382 trial_locales[0] = setlocale_init;
3383 trial_locales_count = 1;
7d4bcc4a 3384
65ebb059
KW
3385 for (i= 0; i < trial_locales_count; i++) {
3386 const char * trial_locale = trial_locales[i];
3387
3388 if (i > 0) {
3389
3390 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3391 * when i==0, but I (khw) don't think that behavior makes much
3392 * sense */
3393 setlocale_failure = FALSE;
3394
7d4bcc4a 3395# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3396# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3397
65ebb059
KW
3398 /* On Windows machines, an entry of "" after the 0th means to use
3399 * the system default locale, which we now proceed to get. */
3400 if (strEQ(trial_locale, "")) {
3401 unsigned int j;
3402
3403 /* Note that this may change the locale, but we are going to do
3404 * that anyway just below */
837ce802 3405 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3406 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3407
7d4bcc4a 3408 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3409 * try */
3410 if (! system_default_locale) {
3411 goto next_iteration;
3412 }
3413 for (j = 0; j < trial_locales_count; j++) {
3414 if (strEQ(system_default_locale, trial_locales[j])) {
3415 goto next_iteration;
3416 }
3417 }
3418
3419 trial_locale = system_default_locale;
3420 }
ec0202b5
KW
3421# else
3422# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3423# endif
7d4bcc4a 3424# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3425
3426 } /* For i > 0 */
65ebb059 3427
7d4bcc4a
KW
3428# ifdef LC_ALL
3429
948523db
KW
3430 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3431 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3432 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3433 setlocale_failure = TRUE;
7cd8b568
KW
3434 }
3435 else {
3436 /* Since LC_ALL succeeded, it should have changed all the other
3437 * categories it can to its value; so we massage things so that the
3438 * setlocales below just return their category's current values.
3439 * This adequately handles the case in NetBSD where LC_COLLATE may
3440 * not be defined for a locale, and setting it individually will
7d4bcc4a 3441 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3442 * the POSIX locale. */
3443 trial_locale = NULL;
3444 }
7d4bcc4a
KW
3445
3446# endif /* LC_ALL */
98994639 3447
e5f10d49
KW
3448 if (! setlocale_failure) {
3449 unsigned int j;
3450 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3451 curlocales[j]
3452 = savepv(do_setlocale_r(categories[j], trial_locale));
3453 if (! curlocales[j]) {
3454 setlocale_failure = TRUE;
3455 }
3456 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3457 }
c835d6be 3458
e5f10d49
KW
3459 if (! setlocale_failure) { /* All succeeded */
3460 break; /* Exit trial_locales loop */
49c85077 3461 }
65ebb059 3462 }
98994639 3463
49c85077
KW
3464 /* Here, something failed; will need to try a fallback. */
3465 ok = 0;
65ebb059 3466
49c85077
KW
3467 if (i == 0) {
3468 unsigned int j;
98994639 3469
65ebb059 3470 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3471
3472# ifdef LC_ALL
98994639 3473
49c85077
KW
3474 PerlIO_printf(Perl_error_log,
3475 "perl: warning: Setting locale failed.\n");
98994639 3476
7d4bcc4a 3477# else /* !LC_ALL */
98994639 3478
49c85077
KW
3479 PerlIO_printf(Perl_error_log,
3480 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3481
e5f10d49
KW
3482 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3483 if (! curlocales[j]) {
3484 PerlIO_printf(Perl_error_log, category_names[j]);
3485 }
3486 else {
3487 Safefree(curlocales[j]);
3488 }
3489 }
7d4bcc4a 3490
7d4bcc4a 3491# endif /* LC_ALL */
98994639 3492
49c85077
KW
3493 PerlIO_printf(Perl_error_log,
3494 "perl: warning: Please check that your locale settings:\n");
98994639 3495
7d4bcc4a
KW
3496# ifdef __GLIBC__
3497
49c85077
KW
3498 PerlIO_printf(Perl_error_log,
3499 "\tLANGUAGE = %c%s%c,\n",
3500 language ? '"' : '(',
3501 language ? language : "unset",
3502 language ? '"' : ')');
7d4bcc4a 3503# endif
98994639 3504
49c85077
KW
3505 PerlIO_printf(Perl_error_log,
3506 "\tLC_ALL = %c%s%c,\n",
3507 lc_all ? '"' : '(',
3508 lc_all ? lc_all : "unset",
3509 lc_all ? '"' : ')');
98994639 3510
7d4bcc4a
KW
3511# if defined(USE_ENVIRON_ARRAY)
3512
49c85077 3513 {
cd999af9 3514 char **e;
d5e32b93
KW
3515
3516 /* Look through the environment for any variables of the
3517 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3518 * already handled above. These are assumed to be locale
3519 * settings. Output them and their values. */
cd999af9 3520 for (e = environ; *e; e++) {
d5e32b93
KW
3521 const STRLEN prefix_len = sizeof("LC_") - 1;
3522 STRLEN uppers_len;
3523
cd999af9 3524 if ( strBEGINs(*e, "LC_")
c8b388b0 3525 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3526 && (uppers_len = strspn(*e + prefix_len,
3527 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3528 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3529 {
3530 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3531 (int) (prefix_len + uppers_len), *e,
3532 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3533 }
3534 }
49c85077 3535 }
7d4bcc4a
KW
3536
3537# else
3538
49c85077
KW
3539 PerlIO_printf(Perl_error_log,
3540 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3541
3542# endif
98994639 3543
49c85077
KW
3544 PerlIO_printf(Perl_error_log,
3545 "\tLANG = %c%s%c\n",
3546 lang ? '"' : '(',
3547 lang ? lang : "unset",
3548 lang ? '"' : ')');
98994639 3549
49c85077
KW
3550 PerlIO_printf(Perl_error_log,
3551 " are supported and installed on your system.\n");
3552 }
98994639 3553
65ebb059 3554 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3555 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3556 * usually change the upper bound of the loop we are in.
3557 *
3558 * Since the system's default way of setting the locale has not
3559 * found one that works, We use Perl's defined ordering: LC_ALL,
3560 * LANG, and the C locale. We don't try the same locale twice, so
3561 * don't add to the list if already there. (On POSIX systems, the
3562 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3563 * but there's no harm done by doing it explicitly.
3564 *
3565 * Note that this tries the LC_ALL environment variable even on
3566 * systems which have no LC_ALL locale setting. This may or may
3567 * not have been originally intentional, but there's no real need
3568 * to change the behavior. */
65ebb059
KW
3569 if (lc_all) {
3570 for (j = 0; j < trial_locales_count; j++) {
3571 if (strEQ(lc_all, trial_locales[j])) {
3572 goto done_lc_all;
3573 }
3574 }
3575 trial_locales[trial_locales_count++] = lc_all;
3576 }
3577 done_lc_all:
98994639 3578
65ebb059
KW
3579 if (lang) {
3580 for (j = 0; j < trial_locales_count; j++) {
3581 if (strEQ(lang, trial_locales[j])) {
3582 goto done_lang;
3583 }
3584 }
3585 trial_locales[trial_locales_count++] = lang;
3586 }
3587 done_lang:
3588
7d4bcc4a
KW
3589# if defined(WIN32) && defined(LC_ALL)
3590
65ebb059
KW
3591 /* For Windows, we also try the system default locale before "C".
3592 * (If there exists a Windows without LC_ALL we skip this because
3593 * it gets too complicated. For those, the "C" is the next
3594 * fallback possibility). The "" is the same as the 0th element of
3595 * the array, but the code at the loop above knows to treat it
3596 * differently when not the 0th */
3597 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3598
3599# endif
65ebb059
KW
3600
3601 for (j = 0; j < trial_locales_count; j++) {
3602 if (strEQ("C", trial_locales[j])) {
3603 goto done_C;
3604 }
3605 }
3606 trial_locales[trial_locales_count++] = "C";
98994639 3607
65ebb059
KW
3608 done_C: ;
3609 } /* end of first time through the loop */
98994639 3610
7d4bcc4a
KW
3611# ifdef WIN32
3612
65ebb059 3613 next_iteration: ;
7d4bcc4a
KW
3614
3615# endif
65ebb059
KW
3616
3617 } /* end of looping through the trial locales */
3618
3619 if (ok < 1) { /* If we tried to fallback */
3620 const char* msg;
3621 if (! setlocale_failure) { /* fallback succeeded */
3622 msg = "Falling back to";
3623 }
3624 else { /* fallback failed */
e5f10d49 3625 unsigned int j;
98994639 3626
65ebb059
KW
3627 /* We dropped off the end of the loop, so have to decrement i to
3628 * get back to the value the last time through */
3629 i--;
98994639 3630
65ebb059
KW
3631 ok = -1;
3632 msg = "Failed to fall back to";
3633
3634 /* To continue, we should use whatever values we've got */
7d4bcc4a 3635
e5f10d49
KW
3636 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3637 Safefree(curlocales[j]);
3638 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3639 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3640 }
65ebb059
KW
3641 }
3642
3643 if (locwarn) {
3644 const char * description;
3645 const char * name = "";
3646 if (strEQ(trial_locales[i], "C")) {
3647 description = "the standard locale";
3648 name = "C";
3649 }
7d4bcc4a
KW
3650
3651# ifdef SYSTEM_DEFAULT_LOCALE
3652
65ebb059
KW
3653 else if (strEQ(trial_locales[i], "")) {
3654 description = "the system default locale";
3655 if (system_default_locale) {
3656 name = system_default_locale;
3657 }
3658 }
7d4bcc4a
KW
3659
3660# endif /* SYSTEM_DEFAULT_LOCALE */
3661
65ebb059
KW
3662 else {
3663 description = "a fallback locale";
3664 name = trial_locales[i];
3665 }
3666 if (name && strNE(name, "")) {
3667 PerlIO_printf(Perl_error_log,
3668 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3669 }
3670 else {
3671 PerlIO_printf(Perl_error_log,
3672 "perl: warning: %s %s.\n", msg, description);
3673 }
3674 }
3675 } /* End of tried to fallback */
98994639 3676
e5f10d49
KW
3677 /* Done with finding the locales; update our records */
3678
7d4bcc4a
KW
3679# ifdef USE_LOCALE_CTYPE
3680
948523db 3681 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3682
e5f10d49 3683# endif
7d4bcc4a
KW
3684# ifdef USE_LOCALE_COLLATE
3685
948523db 3686 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3687
e5f10d49 3688# endif
7d4bcc4a
KW
3689# ifdef USE_LOCALE_NUMERIC
3690
948523db 3691 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3692
3693# endif
3694
948523db 3695 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3696
e9bc6d6b 3697# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3698
3699 /* This caches whether each category's locale is UTF-8 or not. This
3700 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3701 * initialization time before any threads have started, but not later
3702 * unless thread-safe operations are used.
47280b20
KW
3703 * Caching means that if the program heeds our dictate not to change
3704 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3705 * it may get queried without having to change locales. If the
3706 * environment is such that all categories have the same locale, this
3707 * isn't needed, as the code will not change the locale; but this
3708 * handles the uncommon case where the environment has disparate
3709 * locales for the categories */
47280b20
KW
3710 (void) _is_cur_LC_category_utf8(categories[i]);
3711
3712# endif
3713
e5f10d49
KW
3714 Safefree(curlocales[i]);
3715 }
b310b053 3716
7d4bcc4a
KW
3717# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3718
49c85077 3719 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3720 * locale is UTF-8. The call to new_ctype() just above has already
3721 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3722 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3723 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3724 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3725 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3726
a05d7ebb 3727 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3728 This is an alternative to using the -C command line switch
3729 (the -C if present will override this). */
3730 {
dd374669 3731 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3732 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3733 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3734 PL_utf8cache = -1;
b310b053
JH
3735 }
3736
7d4bcc4a 3737# endif
7d4bcc4a
KW
3738# ifdef __GLIBC__
3739
175c4cf9 3740 Safefree(language);
7d4bcc4a
KW
3741
3742# endif
175c4cf9
KW
3743
3744 Safefree(lc_all);
3745 Safefree(lang);
3746
e3305790 3747#endif /* USE_LOCALE */
2fcc0ca9 3748#ifdef DEBUGGING
7d4bcc4a 3749
2fcc0ca9 3750 /* So won't continue to output stuff */
27cdc72e 3751 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3752
2fcc0ca9
KW
3753#endif
3754
98994639
HS
3755 return ok;
3756}
3757
98994639
HS
3758#ifdef USE_LOCALE_COLLATE
3759
a4a439fb 3760char *
a4a439fb
KW
3761Perl__mem_collxfrm(pTHX_ const char *input_string,
3762 STRLEN len, /* Length of 'input_string' */
3763 STRLEN *xlen, /* Set to length of returned string
3764 (not including the collation index
3765 prefix) */
3766 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3767 )
98994639 3768{
a4a439fb
KW
3769
3770 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3771 * differences. First, it handles embedded NULs. Second, it allocates a bit
3772 * more memory than needed for the transformed data itself. The real
55e5378d 3773 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3774 * the length of that, and doesn't include the collation index size.
3775 * Please see sv_collxfrm() to see how this is used. */
3776
55e5378d
KW
3777#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3778
6696cfa7
KW
3779 char * s = (char *) input_string;
3780 STRLEN s_strlen = strlen(input_string);
79f120c8 3781 char *xbuf = NULL;
55e5378d 3782 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3783 STRLEN length_in_chars;
c664130f 3784 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3785
a4a439fb
KW
3786 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3787
3788 /* Must be NUL-terminated */
3789 assert(*(input_string + len) == '\0');
7918f24d 3790
79f120c8
KW
3791 /* If this locale has defective collation, skip */
3792 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3793 DEBUG_L(PerlIO_printf(Perl_debug_log,
3794 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3795 goto bad;
3796 }
3797
6696cfa7
KW
3798 /* Replace any embedded NULs with the control that sorts before any others.
3799 * This will give as good as possible results on strings that don't
3800 * otherwise contain that character, but otherwise there may be
3801 * less-than-perfect results with that character and NUL. This is
fdc080f3 3802 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3803 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3804 NUL */
6696cfa7
KW
3805 char * e = s + len;
3806 char * sans_nuls;
fdc080f3 3807 STRLEN sans_nuls_len;
94762aa0 3808 int try_non_controls;
afc4976f
KW
3809 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3810 making sure 2nd byte is NUL.
3811 */
3812 STRLEN this_replacement_len;
3813
1e4c9676
KW
3814 /* If we don't know what non-NUL control character sorts lowest for
3815 * this locale, find it */
f28f4d2a 3816 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3817 int j;
afc4976f 3818 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3819 includes the collation index
3820 prefixed. */
3821
91c0e2e0 3822 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3823
3824 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3825 * NUL, in which case we instead look for any character. Controls
3826 * are preferred because collation order is, in general, context
3827 * sensitive, with adjoining characters affecting the order, and
3828 * controls are less likely to have such interactions, allowing the
3829 * NUL-replacement to stand on its own. (Another way to look at it
3830 * is to imagine what would happen if the NUL were replaced by a
3831 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3832 for (try_non_controls = 0;
3833 try_non_controls < 2;
3834 try_non_controls++)
3835 {
d4ff9586
KW
3836 /* Look through all legal code points (NUL isn't) */
3837 for (j = 1; j < 256; j++) {
3838 char * x; /* j's xfrm plus collation index */
3839 STRLEN x_len; /* length of 'x' */
3840 STRLEN trial_len = 1;
736a4fed 3841 char cur_source[] = { '\0', '\0' };
d4ff9586 3842
736a4fed
KW
3843 /* Skip non-controls the first time through the loop. The
3844 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3845 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3846 ? ! isCNTRL_L1(j)
3847 : ! isCNTRL_LC(j))
3848 {
d4ff9586 3849 continue;
6696cfa7 3850 }
6696cfa7 3851
736a4fed
KW
3852 /* Create a 1-char string of the current code point */
3853 cur_source[0] = (char) j;
3854
d4ff9586
KW
3855 /* Then transform it */
3856 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3857 0 /* The string is not in UTF-8 */);
6696cfa7 3858
1e4c9676 3859 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3860 * */