This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Bump version to 0.25.
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
e9bc6d6b
KW
35 *
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
166f8a29
DM
44 */
45
98994639
HS
46#include "EXTERN.h"
47#define PERL_IN_LOCALE_C
f7416781 48#include "perl_langinfo.h"
98994639
HS
49#include "perl.h"
50
a4af207c
JH
51#include "reentr.h"
52
0dec74cd
KW
53#ifdef I_WCHAR
54# include <wchar.h>
55#endif
56
2fcc0ca9
KW
57/* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
5a4b0634
KW
60#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61# define debug_initialization 0
62# define DEBUG_INITIALIZATION_set(v)
63#else
2fcc0ca9 64static bool debug_initialization = FALSE;
5a4b0634 65# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
66#endif
67
48015184
KW
68
69/* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
71 * done generally. */
72#define GET_ERRNO saved_errno
73
291a84fb
KW
74/* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
77#define STRLENs(s) (sizeof("" s "") - 1)
78
63e5b0d7
KW
79/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
85 * cntrl: 84-97 9B-9F
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89#define isNAME_C_OR_POSIX(name) \
90 ( (name) != NULL \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
93
8ef6e574
KW
94#ifdef USE_LOCALE
95
47280b20
KW
96/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
98
99#define UTF8NESS_SEP "\v"
100#define UTF8NESS_PREFIX "\f"
101
102/* So, the string looks like:
98994639 103 *
47280b20 104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 105 *
47280b20
KW
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
108
109STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
111
112#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
114
115/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
118
98994639
HS
119STATIC char *
120S_stdize_locale(pTHX_ char *locs)
121{
47280b20
KW
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
124 *
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
130 *
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
134 * */
135
7452cf6a 136 const char * const s = strchr(locs, '=');
98994639
HS
137 bool okay = TRUE;
138
7918f24d
NC
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
140
8772537c
AL
141 if (s) {
142 const char * const t = strchr(s, '.');
98994639 143 okay = FALSE;
8772537c
AL
144 if (t) {
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
149 locs[len] = 0;
150 okay = TRUE;
98994639
HS
151 }
152 }
153 }
154
155 if (!okay)
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
157
158 return locs;
159}
160
e5f10d49
KW
161/* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
163 * order. */
164
165const int categories[] = {
166
167# ifdef USE_LOCALE_NUMERIC
168 LC_NUMERIC,
169# endif
170# ifdef USE_LOCALE_CTYPE
171 LC_CTYPE,
172# endif
173# ifdef USE_LOCALE_COLLATE
174 LC_COLLATE,
175# endif
176# ifdef USE_LOCALE_TIME
177 LC_TIME,
178# endif
179# ifdef USE_LOCALE_MESSAGES
180 LC_MESSAGES,
181# endif
182# ifdef USE_LOCALE_MONETARY
183 LC_MONETARY,
184# endif
9821811f
KW
185# ifdef USE_LOCALE_ADDRESS
186 LC_ADDRESS,
187# endif
188# ifdef USE_LOCALE_IDENTIFICATION
189 LC_IDENTIFICATION,
190# endif
191# ifdef USE_LOCALE_MEASUREMENT
192 LC_MEASUREMENT,
193# endif
194# ifdef USE_LOCALE_PAPER
195 LC_PAPER,
196# endif
197# ifdef USE_LOCALE_TELEPHONE
198 LC_TELEPHONE,
199# endif
e5f10d49
KW
200# ifdef LC_ALL
201 LC_ALL,
202# endif
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
206};
207
208/* The top-most real element is LC_ALL */
209
210const char * category_names[] = {
211
212# ifdef USE_LOCALE_NUMERIC
213 "LC_NUMERIC",
214# endif
215# ifdef USE_LOCALE_CTYPE
216 "LC_CTYPE",
217# endif
218# ifdef USE_LOCALE_COLLATE
219 "LC_COLLATE",
220# endif
221# ifdef USE_LOCALE_TIME
222 "LC_TIME",
223# endif
224# ifdef USE_LOCALE_MESSAGES
225 "LC_MESSAGES",
226# endif
227# ifdef USE_LOCALE_MONETARY
228 "LC_MONETARY",
229# endif
9821811f
KW
230# ifdef USE_LOCALE_ADDRESS
231 "LC_ADDRESS",
232# endif
233# ifdef USE_LOCALE_IDENTIFICATION
234 "LC_IDENTIFICATION",
235# endif
236# ifdef USE_LOCALE_MEASUREMENT
237 "LC_MEASUREMENT",
238# endif
239# ifdef USE_LOCALE_PAPER
240 "LC_PAPER",
241# endif
242# ifdef USE_LOCALE_TELEPHONE
243 "LC_TELEPHONE",
244# endif
e5f10d49
KW
245# ifdef LC_ALL
246 "LC_ALL",
247# endif
248 NULL /* Placeholder */
249 };
250
251# ifdef LC_ALL
252
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
256
257# else
258
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
262
263# endif
264
265/* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
948523db
KW
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
e5f10d49 272
b09aaf40
KW
273STATIC const char *
274S_category_name(const int category)
275{
276 unsigned int i;
277
278#ifdef LC_ALL
279
280 if (category == LC_ALL) {
281 return "LC_ALL";
282 }
283
284#endif
285
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
289 }
290 }
291
292 {
293 const char suffix[] = " (unknown)";
294 int temp = category;
295 Size_t length = sizeof(suffix) + 1;
296 char * unknown;
297 dTHX;
298
299 if (temp < 0) {
300 length++;
301 temp = - temp;
302 }
303
304 /* Calculate the number of digits */
305 while (temp >= 10) {
306 temp /= 10;
307 length++;
308 }
309
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
312 SAVEFREEPV(unknown);
313 return unknown;
314 }
315}
316
948523db
KW
317/* Now create LC_foo_INDEX #defines for just those categories on this system */
318# ifdef USE_LOCALE_NUMERIC
319# define LC_NUMERIC_INDEX 0
320# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
321# else
322# define _DUMMY_NUMERIC -1
323# endif
324# ifdef USE_LOCALE_CTYPE
325# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326# define _DUMMY_CTYPE LC_CTYPE_INDEX
327# else
328# define _DUMMY_CTYPE _DUMMY_NUMERIC
329# endif
330# ifdef USE_LOCALE_COLLATE
331# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332# define _DUMMY_COLLATE LC_COLLATE_INDEX
333# else
334# define _DUMMY_COLLATE _DUMMY_COLLATE
335# endif
336# ifdef USE_LOCALE_TIME
337# define LC_TIME_INDEX _DUMMY_COLLATE + 1
338# define _DUMMY_TIME LC_TIME_INDEX
339# else
340# define _DUMMY_TIME _DUMMY_COLLATE
341# endif
342# ifdef USE_LOCALE_MESSAGES
343# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
345# else
346# define _DUMMY_MESSAGES _DUMMY_TIME
347# endif
348# ifdef USE_LOCALE_MONETARY
349# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350# define _DUMMY_MONETARY LC_MONETARY_INDEX
351# else
352# define _DUMMY_MONETARY _DUMMY_MESSAGES
353# endif
9821811f
KW
354# ifdef USE_LOCALE_ADDRESS
355# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
357# else
358# define _DUMMY_ADDRESS _DUMMY_MONETARY
359# endif
360# ifdef USE_LOCALE_IDENTIFICATION
361# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
363# else
364# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
365# endif
366# ifdef USE_LOCALE_MEASUREMENT
367# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
369# else
370# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
371# endif
372# ifdef USE_LOCALE_PAPER
373# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374# define _DUMMY_PAPER LC_PAPER_INDEX
375# else
376# define _DUMMY_PAPER _DUMMY_MEASUREMENT
377# endif
378# ifdef USE_LOCALE_TELEPHONE
379# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
381# else
382# define _DUMMY_TELEPHONE _DUMMY_PAPER
383# endif
948523db 384# ifdef LC_ALL
9821811f 385# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
386# endif
387#endif /* ifdef USE_LOCALE */
8ef6e574 388
d2b24094 389/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
390#ifdef WIN32
391# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
392#else
393# define my_setlocale(cat, locale) setlocale(cat, locale)
394#endif
395
396#ifndef USE_POSIX_2008_LOCALE
397
398/* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
402
403#else /* Below uses POSIX 2008 */
404
405/* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411# define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
414
415/* A third array, parallel to the ones above to map from category to its
416 * equivalent mask */
417const int category_masks[] = {
418# ifdef USE_LOCALE_NUMERIC
419 LC_NUMERIC_MASK,
420# endif
421# ifdef USE_LOCALE_CTYPE
422 LC_CTYPE_MASK,
423# endif
424# ifdef USE_LOCALE_COLLATE
425 LC_COLLATE_MASK,
426# endif
427# ifdef USE_LOCALE_TIME
428 LC_TIME_MASK,
429# endif
430# ifdef USE_LOCALE_MESSAGES
431 LC_MESSAGES_MASK,
432# endif
433# ifdef USE_LOCALE_MONETARY
434 LC_MONETARY_MASK,
435# endif
436# ifdef USE_LOCALE_ADDRESS
437 LC_ADDRESS_MASK,
438# endif
439# ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
441# endif
442# ifdef USE_LOCALE_MEASUREMENT
443 LC_MEASUREMENT_MASK,
444# endif
445# ifdef USE_LOCALE_PAPER
446 LC_PAPER_MASK,
447# endif
448# ifdef USE_LOCALE_TELEPHONE
449 LC_TELEPHONE_MASK,
450# endif
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
455 * time */
456 LC_ALL_MASK
457 };
458
459STATIC const char *
460S_emulate_setlocale(const int category,
461 const char * locale,
462 unsigned int index,
463 const bool is_index_valid
464 )
465{
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
473 *
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
476 *
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
486 *
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
501 */
502
503 int mask;
504 locale_t old_obj;
505 locale_t new_obj;
506 dTHX;
507
508# ifdef DEBUGGING
509
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
512 }
513
514# endif
515
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
518 unsigned int i;
519
520# ifdef DEBUGGING
521
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
524 }
525
526# endif
527
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
530 index = i;
531 goto found_index;
532 }
533 }
534
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
539 category, locale);
540 return NULL;
541
542 found_index: ;
543
544# ifdef DEBUGGING
545
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
548 }
549
550# endif
551
552 }
553
554 mask = category_masks[index];
555
556# ifdef DEBUGGING
557
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
560 }
561
562# endif
563
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
567
568# ifdef DEBUGGING
569
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
572 }
573
574# endif
575
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
578 }
579
580# ifdef HAS_QUERYLOCALE
581
582 return (char *) querylocale(mask, cur_obj);
583
d2b24094 584# else
ecdda939
KW
585
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
588
e9bc6d6b
KW
589# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
590
591 {
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
595
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
598 /*
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
602 */
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
607
608# ifdef USE_C_BACKTRACE
609
610 dump_c_backtrace(Perl_debug_log, 20, 1);
611
612# endif
613
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
618 }
619
620 return temp_name;
621 }
622 }
623
624# endif
625
626 /* Without querylocale(), we have to use our record-keeping we've
627 * done. */
628
629 if (category != LC_ALL) {
630
631# ifdef DEBUGGING
632
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 }
636
637# endif
638
639 return PL_curlocales[index];
640 }
641 else { /* For LC_ALL */
642 unsigned int i;
643 Size_t names_len = 0;
644 char * all_string;
7c93a471 645 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
646
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
649
650# ifdef DEBUGGING
651
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
654 }
655
656# endif
657
658 return PL_curlocales[LC_ALL_INDEX];
659 }
660
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
e9bc6d6b
KW
666 for (i = 0; i < LC_ALL_INDEX; i++) {
667
668# ifdef DEBUGGING
669
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
672 }
673
674# endif
675
676 names_len += strlen(category_names[i])
677 + 1 /* '=' */
678 + strlen(PL_curlocales[i])
679 + 1; /* ';' */
7c93a471
KW
680
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
683 }
684 }
685
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 692 }
7c93a471 693
e9bc6d6b
KW
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
696 *all_string = '\0';
697
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
700
701# ifdef DEBUGGING
702
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
705 }
706
707# endif
708
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
713 }
714
715# ifdef DEBUGGING
716
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
719 }
720
721 #endif
722
723 return all_string;
724 }
725
726# ifdef EINVAL
727
728 SETERRNO(EINVAL, LIB_INVARG);
729
730# endif
731
732 return NULL;
733
d2b24094
KW
734# endif
735
e9bc6d6b
KW
736 }
737
f8115d60 738 /* Here, we are switching locales. */
e9bc6d6b
KW
739
740# ifndef HAS_QUERYLOCALE
741
742 if (strEQ(locale, "")) {
743
744 /* For non-querylocale() systems, we do the setting of "" ourselves to
745 * be sure that we really know what's going on. We follow the Linux
746 * documented behavior (but if that differs from the actual behavior,
747 * this won't work exactly as the OS implements). We go out and
748 * examine the environment based on our understanding of how the system
749 * works, and use that to figure things out */
750
751 const char * const lc_all = PerlEnv_getenv("LC_ALL");
752
753 /* Use any "LC_ALL" environment variable, as it overrides everything
754 * else. */
755 if (lc_all && strNE(lc_all, "")) {
756 locale = lc_all;
757 }
758 else {
759
760 /* Otherwise, we need to dig deeper. Unless overridden, the
761 * default is the LANG environment variable; if it doesn't exist,
762 * then "C" */
763
764 const char * default_name;
765
766 /* To minimize other threads messing with the environment, we copy
767 * the variable, making it a temporary. But this doesn't work upon
768 * program initialization before any scopes are created, and at
769 * this time, there's nothing else going on that would interfere.
770 * So skip the copy in that case */
771 if (PL_scopestack_ix == 0) {
772 default_name = PerlEnv_getenv("LANG");
773 }
774 else {
775 default_name = savepv(PerlEnv_getenv("LANG"));
776 }
777
778 if (! default_name || strEQ(default_name, "")) {
929a7f8c 779 default_name = "C";
e9bc6d6b
KW
780 }
781 else if (PL_scopestack_ix != 0) {
782 SAVEFREEPV(default_name);
783 }
784
785 if (category != LC_ALL) {
786 const char * const name = PerlEnv_getenv(category_names[index]);
787
788 /* Here we are setting a single category. Assume will have the
789 * default name */
790 locale = default_name;
791
792 /* But then look for an overriding environment variable */
793 if (name && strNE(name, "")) {
794 locale = name;
795 }
796 }
797 else {
798 bool did_override = FALSE;
799 unsigned int i;
800
801 /* Here, we are getting LC_ALL. Any categories that don't have
802 * a corresponding environment variable set should be set to
803 * LANG, or to "C" if there is no LANG. If no individual
804 * categories differ from this, we can just set LC_ALL. This
805 * is buggy on systems that have extra categories that we don't
806 * know about. If there is an environment variable that sets
807 * that category, we won't know to look for it, and so our use
808 * of LANG or "C" improperly overrides it. On the other hand,
809 * if we don't do what is done here, and there is no
810 * environment variable, the category's locale should be set to
811 * LANG or "C". So there is no good solution. khw thinks the
812 * best is to look at systems to see what categories they have,
813 * and include them, and then to assume that we know the
814 * complete set */
815
816 for (i = 0; i < LC_ALL_INDEX; i++) {
817 const char * const env_override
818 = savepv(PerlEnv_getenv(category_names[i]));
819 const char * this_locale = ( env_override
820 && strNE(env_override, ""))
821 ? env_override
822 : default_name;
6435f98d
KW
823 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
824 {
825 Safefree(env_override);
826 return NULL;
827 }
e9bc6d6b
KW
828
829 if (strNE(this_locale, default_name)) {
830 did_override = TRUE;
831 }
832
833 Safefree(env_override);
834 }
835
836 /* If all the categories are the same, we can set LC_ALL to
837 * that */
838 if (! did_override) {
839 locale = default_name;
840 }
841 else {
842
843 /* Here, LC_ALL is no longer valid, as some individual
844 * categories don't match it. We call ourselves
845 * recursively, as that will execute the code that
846 * generates the proper locale string for this situation.
847 * We don't do the remainder of this function, as that is
848 * to update our records, and we've just done that for the
849 * individual categories in the loop above, and doing so
850 * would cause LC_ALL to be done as well */
851 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
852 }
853 }
854 }
855 }
856 else if (strchr(locale, ';')) {
857
858 /* LC_ALL may actually incude a conglomeration of various categories.
859 * Without querylocale, this code uses the glibc (as of this writing)
860 * syntax for representing that, but that is not a stable API, and
861 * other platforms do it differently, so we have to handle all cases
862 * ourselves */
863
f0023d47 864 unsigned int i;
e9bc6d6b
KW
865 const char * s = locale;
866 const char * e = locale + strlen(locale);
867 const char * p = s;
868 const char * category_end;
869 const char * name_start;
870 const char * name_end;
871
f0023d47
KW
872 /* If the string that gives what to set doesn't include all categories,
873 * the omitted ones get set to "C". To get this behavior, first set
874 * all the individual categories to "C", and override the furnished
875 * ones below */
876 for (i = 0; i < LC_ALL_INDEX; i++) {
877 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
878 return NULL;
879 }
880 }
881
e9bc6d6b 882 while (s < e) {
e9bc6d6b
KW
883
884 /* Parse through the category */
885 while (isWORDCHAR(*p)) {
886 p++;
887 }
888 category_end = p;
889
890 if (*p++ != '=') {
891 Perl_croak(aTHX_
892 "panic: %s: %d: Unexpected character in locale name '%02X",
893 __FILE__, __LINE__, *(p-1));
894 }
895
896 /* Parse through the locale name */
897 name_start = p;
bee74f4b
KW
898 while (p < e && *p != ';') {
899 if (! isGRAPH(*p)) {
900 Perl_croak(aTHX_
901 "panic: %s: %d: Unexpected character in locale name '%02X",
902 __FILE__, __LINE__, *(p-1));
903 }
e9bc6d6b
KW
904 p++;
905 }
906 name_end = p;
907
bee74f4b
KW
908 /* Space past the semi-colon */
909 if (p < e) {
910 p++;
e9bc6d6b
KW
911 }
912
913 /* Find the index of the category name in our lists */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 915 char * individ_locale;
e9bc6d6b
KW
916
917 /* Keep going if this isn't the index. The strnNE() avoids a
918 * Perl_form(), but would fail if ever a category name could be
919 * a substring of another one, like if there were a
920 * "LC_TIME_DATE" */
921 if strnNE(s, category_names[i], category_end - s) {
922 continue;
923 }
924
925 /* If this index is for the single category we're changing, we
926 * have found the locale to set it to. */
927 if (category == categories[i]) {
928 locale = Perl_form(aTHX_ "%.*s",
929 (int) (name_end - name_start),
930 name_start);
931 goto ready_to_set;
932 }
933
3c76a412 934 assert(category == LC_ALL);
bee74f4b
KW
935 individ_locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start), name_start);
6435f98d
KW
937 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
938 {
939 return NULL;
940 }
e9bc6d6b
KW
941 }
942
943 s = p;
944 }
945
946 /* Here we have set all the individual categories by recursive calls.
947 * These collectively should have fixed up LC_ALL, so can just query
948 * what that now is */
949 assert(category == LC_ALL);
950
951 return do_setlocale_c(LC_ALL, NULL);
952 }
953
954 ready_to_set: ;
955
f8115d60
KW
956 /* Here at the end of having to deal with the absence of querylocale().
957 * Some cases have already been fully handled by recursive calls to this
958 * function. But at this point, we haven't dealt with those, but are now
959 * prepared to, knowing what the locale name to set this category to is.
960 * This would have come for free if this system had had querylocale() */
961
e9bc6d6b
KW
962# endif /* end of ! querylocale */
963
f8115d60
KW
964 assert(PL_C_locale_obj);
965
966 /* Switching locales generally entails freeing the current one's space (at
967 * the C library's discretion). We need to stop using that locale before
968 * the switch. So switch to a known locale object that we don't otherwise
969 * mess with. This returns the locale object in effect at the time of the
970 * switch. */
971 old_obj = uselocale(PL_C_locale_obj);
972
973# ifdef DEBUGGING
974
975 if (DEBUG_Lv_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
977 }
978
979# endif
980
981 if (! old_obj) {
982
983# ifdef DEBUGGING
984
985 if (DEBUG_L_TEST || debug_initialization) {
986 dSAVE_ERRNO;
987 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
988 RESTORE_ERRNO;
989 }
990
991# endif
992
993 return NULL;
994 }
995
996# ifdef DEBUGGING
997
998 if (DEBUG_Lv_TEST || debug_initialization) {
999 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1000 }
1001
1002# endif
1003
1004 /* If we weren't in a thread safe locale, set so that newlocale() below
1005 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1006 The latter is defensive coding, so that, even if there is some bug, we
1007 will never end up trying to modify either of these, as if passed to
1008 newlocale(), they can be. */
1009 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1010 old_obj = (locale_t) 0;
1011 }
1012
e9bc6d6b
KW
1013 /* Ready to create a new locale by modification of the exising one */
1014 new_obj = newlocale(mask, locale, old_obj);
1015
1016 if (! new_obj) {
1017 dSAVE_ERRNO;
1018
1019# ifdef DEBUGGING
1020
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1023 }
1024
1025# endif
1026
1027 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1028
1029# ifdef DEBUGGING
1030
1031 if (DEBUG_L_TEST || debug_initialization) {
1032 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1033 }
1034
1035# endif
1036
1037 }
1038 RESTORE_ERRNO;
1039 return NULL;
1040 }
1041
1042# ifdef DEBUGGING
1043
1044 if (DEBUG_Lv_TEST || debug_initialization) {
d15b6e8b 1045 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p; should have freed %p\n", __FILE__, __LINE__, new_obj, old_obj);
e9bc6d6b
KW
1046 }
1047
1048# endif
1049
1050 /* And switch into it */
1051 if (! uselocale(new_obj)) {
1052 dSAVE_ERRNO;
1053
1054# ifdef DEBUGGING
1055
1056 if (DEBUG_L_TEST || debug_initialization) {
1057 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1058 }
1059
1060# endif
1061
1062 if (! uselocale(old_obj)) {
1063
1064# ifdef DEBUGGING
1065
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1068 }
1069
1070# endif
1071
1072 }
1073 freelocale(new_obj);
1074 RESTORE_ERRNO;
1075 return NULL;
1076 }
1077
1078# ifdef DEBUGGING
1079
1080 if (DEBUG_Lv_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1082 }
1083
1084# endif
1085
1086 /* We are done, except for updating our records (if the system doesn't keep
1087 * them) and in the case of locale "", we don't actually know what the
1088 * locale that got switched to is, as it came from the environment. So
1089 * have to find it */
1090
1091# ifdef HAS_QUERYLOCALE
1092
1093 if (strEQ(locale, "")) {
1094 locale = querylocale(mask, new_obj);
1095 }
1096
1097# else
1098
1099 /* Here, 'locale' is the return value */
1100
1101 /* Without querylocale(), we have to update our records */
1102
1103 if (category == LC_ALL) {
1104 unsigned int i;
1105
1106 /* For LC_ALL, we change all individual categories to correspond */
1107 /* PL_curlocales is a parallel array, so has same
1108 * length as 'categories' */
1109 for (i = 0; i <= LC_ALL_INDEX; i++) {
1110 Safefree(PL_curlocales[i]);
1111 PL_curlocales[i] = savepv(locale);
1112 }
1113 }
1114 else {
1115
1116 /* For a single category, if it's not the same as the one in LC_ALL, we
1117 * nullify LC_ALL */
1118
1119 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1120 Safefree(PL_curlocales[LC_ALL_INDEX]);
1121 PL_curlocales[LC_ALL_INDEX] = NULL;
1122 }
1123
1124 /* Then update the category's record */
1125 Safefree(PL_curlocales[index]);
1126 PL_curlocales[index] = savepv(locale);
1127 }
1128
1129# endif
1130
1131 return locale;
1132}
1133
1134#endif /* USE_POSIX_2008_LOCALE */
1135
1136#if 0 /* Code that was to emulate thread-safe locales on platforms that
1137 didn't natively support them */
1138
1139/* The way this would work is that we would keep a per-thread list of the
1140 * correct locale for that thread. Any operation that was locale-sensitive
1141 * would have to be changed so that it would look like this:
1142 *
1143 * LOCALE_LOCK;
1144 * setlocale to the correct locale for this operation
1145 * do operation
1146 * LOCALE_UNLOCK
1147 *
1148 * This leaves the global locale in the most recently used operation's, but it
1149 * was locked long enough to get the result. If that result is static, it
1150 * needs to be copied before the unlock.
1151 *
1152 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1153 * the setup, but are no-ops when not needed, and similarly,
1154 * END_LOCALE_DEPENDENT_OP for the tear-down
1155 *
1156 * But every call to a locale-sensitive function would have to be changed, and
1157 * if a module didn't cooperate by using the mutex, things would break.
1158 *
1159 * This code was abandoned before being completed or tested, and is left as-is
1160*/
1161
1162# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1163# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1164
1165STATIC char *
1166S_locking_setlocale(pTHX_
1167 const int category,
1168 const char * locale,
1169 int index,
1170 const bool is_index_valid
1171 )
1172{
1173 /* This function kind of performs a setlocale() on just the current thread;
1174 * thus it is kind of thread-safe. It does this by keeping a thread-level
1175 * array of the current locales for each category. Every time a locale is
1176 * switched to, it does the switch globally, but updates the thread's
1177 * array. A query as to what the current locale is just returns the
1178 * appropriate element from the array, and doesn't actually call the system
1179 * setlocale(). The saving into the array is done in an uninterruptible
1180 * section of code, so is unaffected by whatever any other threads might be
1181 * doing.
1182 *
1183 * All locale-sensitive operations must work by first starting a critical
1184 * section, then switching to the thread's locale as kept by this function,
1185 * and then doing the operation, then ending the critical section. Thus,
1186 * each gets done in the appropriate locale. simulating thread-safety.
1187 *
1188 * This function takes the same parameters, 'category' and 'locale', that
1189 * the regular setlocale() function does, but it also takes two additional
1190 * ones. This is because as described earlier. If we know on input the
1191 * index corresponding to the category into the array where we store the
1192 * current locales, we don't have to calculate it. If the caller knows at
1193 * compile time what the index is, it it can pass it, setting
1194 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1195 *
1196 */
1197
1198 /* If the input index might be incorrect, calculate the correct one */
1199 if (! is_index_valid) {
1200 unsigned int i;
1201
1202 if (DEBUG_Lv_TEST || debug_initialization) {
1203 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1204 }
1205
1206 for (i = 0; i <= LC_ALL_INDEX; i++) {
1207 if (category == categories[i]) {
1208 index = i;
1209 goto found_index;
1210 }
1211 }
1212
1213 /* Here, we don't know about this category, so can't handle it.
1214 * XXX best we can do is to unsafely set this
1215 * XXX warning */
1216
1217 return my_setlocale(category, locale);
1218
1219 found_index: ;
1220
1221 if (DEBUG_Lv_TEST || debug_initialization) {
1222 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1223 }
1224 }
1225
1226 /* For a query, just return what's in our records */
1227 if (new_locale == NULL) {
1228 return curlocales[index];
1229 }
1230
1231
1232 /* Otherwise, we need to do the switch, and save the result, all in a
1233 * critical section */
1234
1235 Safefree(curlocales[[index]]);
1236
1237 /* It might be that this is called from an already-locked section of code.
1238 * We would have to detect and skip the LOCK/UNLOCK if so */
1239 LOCALE_LOCK;
1240
1241 curlocales[index] = savepv(my_setlocale(category, new_locale));
1242
1243 if (strEQ(new_locale, "")) {
1244
1245#ifdef LC_ALL
1246
1247 /* The locale values come from the environment, and may not all be the
1248 * same, so for LC_ALL, we have to update all the others, while the
1249 * mutex is still locked */
1250
1251 if (category == LC_ALL) {
1252 unsigned int i;
1253 for (i = 0; i < LC_ALL_INDEX) {
1254 curlocales[i] = my_setlocale(categories[i], NULL);
1255 }
1256 }
1257 }
1258
1259#endif
1260
1261 LOCALE_UNLOCK;
1262
1263 return curlocales[index];
1264}
1265
1266#endif
837ce802 1267
a4f00dcc 1268STATIC void
86799d2d 1269S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1270{
86799d2d
KW
1271 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1272 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1273
86799d2d
KW
1274#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1275 || defined(HAS_NL_LANGINFO))
98994639 1276
87f8e8e7 1277 const char * radix = (use_locale)
4e6826bf 1278 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1279 /* FALSE => already in dest locale */
1280 : ".";
2213a3be 1281
87f8e8e7 1282 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1283
87f8e8e7
KW
1284 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1285 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1286 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1287 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1288 && _is_cur_LC_category_utf8(LC_NUMERIC))
1289 {
1290 SvUTF8_on(PL_numeric_radix_sv);
1291 }
86799d2d
KW
1292
1293# ifdef DEBUGGING
7d4bcc4a 1294
2fcc0ca9
KW
1295 if (DEBUG_L_TEST || debug_initialization) {
1296 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1297 SvPVX(PL_numeric_radix_sv),
1298 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1299 }
69014004 1300
86799d2d
KW
1301# endif
1302#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1303
98994639
HS
1304}
1305
398eeea9
KW
1306STATIC void
1307S_new_numeric(pTHX_ const char *newnum)
98994639 1308{
7d4bcc4a
KW
1309
1310#ifndef USE_LOCALE_NUMERIC
1311
1312 PERL_UNUSED_ARG(newnum);
1313
1314#else
0d071d52 1315
291a84fb 1316 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1317 * core Perl this and that 'newnum' is the name of the new locale.
1318 * It installs this locale as the current underlying default.
1319 *
1320 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1321 * set_numeric_underlying() and set_numeric_standard() functions, which
1322 * should probably not be called directly, but only via macros like
0d071d52
KW
1323 * SET_NUMERIC_STANDARD() in perl.h.
1324 *
1325 * The toggling is necessary mainly so that a non-dot radix decimal point
1326 * character can be output, while allowing internal calculations to use a
1327 * dot.
1328 *
1329 * This sets several interpreter-level variables:
bb304765 1330 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1331 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1332 * that the current locale is the program's underlying
1333 * locale
1334 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1335 * that the current locale is the C locale or
1336 * indistinguishable from the C locale. If non-zero, it
1337 * is in C; if > 1, it means it may not be toggled away
7738054c 1338 * from C.
4c68b815
KW
1339 * PL_numeric_underlying_is_standard A bool kept by this function
1340 * indicating that the underlying locale and the standard
1341 * C locale are indistinguishable for the purposes of
1342 * LC_NUMERIC. This happens when both of the above two
1343 * variables are true at the same time. (Toggling is a
1344 * no-op under these circumstances.) This variable is
1345 * used to avoid having to recalculate.
398eeea9 1346 */
0d071d52 1347
b03f34cf 1348 char *save_newnum;
98994639
HS
1349
1350 if (! newnum) {
43c5f42d
NC
1351 Safefree(PL_numeric_name);
1352 PL_numeric_name = NULL;
98994639 1353 PL_numeric_standard = TRUE;
892e6465 1354 PL_numeric_underlying = TRUE;
4c68b815 1355 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1356 return;
1357 }
1358
b03f34cf 1359 save_newnum = stdize_locale(savepv(newnum));
892e6465 1360 PL_numeric_underlying = TRUE;
4c68b815
KW
1361 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1362
69c5e0db
KW
1363#ifndef TS_W32_BROKEN_LOCALECONV
1364
4c68b815 1365 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1366 * them. But on broken Windows systems calling my_nl_langinfo() for
1367 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1368 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1369 if (! PL_numeric_standard) {
4e6826bf 1370 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1371 FALSE /* Don't toggle locale */ ))
4e6826bf 1372 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1373 }
abe1abcf 1374
69c5e0db
KW
1375#endif
1376
4c68b815 1377 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1378 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1379 Safefree(PL_numeric_name);
b03f34cf 1380 PL_numeric_name = save_newnum;
b03f34cf 1381 }
abe1abcf
KW
1382 else {
1383 Safefree(save_newnum);
1384 }
4c28b29c 1385
4c68b815
KW
1386 PL_numeric_underlying_is_standard = PL_numeric_standard;
1387
7e5377f7 1388# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1389
1390 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1391 PL_numeric_name,
1392 PL_underlying_numeric_obj);
1393
1394#endif
1395
4c68b815
KW
1396 if (DEBUG_L_TEST || debug_initialization) {
1397 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1398 }
1399
4c28b29c
KW
1400 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1401 * have to worry about the radix being a non-dot. (Core operations that
1402 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1403 if (PL_numeric_standard) {
1404 set_numeric_radix(0);
1405 }
1406 else {
1407 set_numeric_standard();
1408 }
4c28b29c 1409
98994639 1410#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1411
98994639
HS
1412}
1413
1414void
1415Perl_set_numeric_standard(pTHX)
1416{
7d4bcc4a 1417
98994639 1418#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1419
28c1bf33
KW
1420 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1421 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1422 * macro avoids calling this routine if toggling isn't necessary according
1423 * to our records (which could be wrong if some XS code has changed the
1424 * locale behind our back) */
0d071d52 1425
837ce802 1426 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1427 PL_numeric_standard = TRUE;
4c68b815 1428 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1429 set_numeric_radix(0);
7d4bcc4a
KW
1430
1431# ifdef DEBUGGING
1432
2fcc0ca9
KW
1433 if (DEBUG_L_TEST || debug_initialization) {
1434 PerlIO_printf(Perl_debug_log,
58e4a467 1435 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1436 }
98994639 1437
7d4bcc4a 1438# endif
98994639 1439#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1440
98994639
HS
1441}
1442
1443void
5792c642 1444Perl_set_numeric_underlying(pTHX)
98994639 1445{
7d4bcc4a 1446
98994639 1447#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1448
28c1bf33 1449 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1450 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1451 * instead of calling this directly. The macro avoids calling this routine
1452 * if toggling isn't necessary according to our records (which could be
1453 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1454
837ce802 1455 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1456 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1457 PL_numeric_underlying = TRUE;
3c62fd46 1458 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1459
1460# ifdef DEBUGGING
1461
2fcc0ca9
KW
1462 if (DEBUG_L_TEST || debug_initialization) {
1463 PerlIO_printf(Perl_debug_log,
58e4a467 1464 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1465 PL_numeric_name);
1466 }
98994639 1467
7d4bcc4a 1468# endif
98994639 1469#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1470
98994639
HS
1471}
1472
1473/*
1474 * Set up for a new ctype locale.
1475 */
a4f00dcc
KW
1476STATIC void
1477S_new_ctype(pTHX_ const char *newctype)
98994639 1478{
7d4bcc4a
KW
1479
1480#ifndef USE_LOCALE_CTYPE
1481
1482 PERL_ARGS_ASSERT_NEW_CTYPE;
1483 PERL_UNUSED_ARG(newctype);
1484 PERL_UNUSED_CONTEXT;
1485
1486#else
0d071d52 1487
291a84fb 1488 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1489 * core Perl this and that 'newctype' is the name of the new locale.
1490 *
1491 * This function sets up the folding arrays for all 256 bytes, assuming
1492 * that tofold() is tolc() since fold case is not a concept in POSIX,
1493 *
1494 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1495 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1496 * this function should be called directly only from this file and from
0d071d52
KW
1497 * POSIX::setlocale() */
1498
27da23d5 1499 dVAR;
013f4e03 1500 unsigned int i;
98994639 1501
8b7358b9
KW
1502 /* Don't check for problems if we are suppressing the warnings */
1503 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1504
7918f24d
NC
1505 PERL_ARGS_ASSERT_NEW_CTYPE;
1506
215c5139
KW
1507 /* We will replace any bad locale warning with 1) nothing if the new one is
1508 * ok; or 2) a new warning for the bad new locale */
1509 if (PL_warn_locale) {
1510 SvREFCNT_dec_NN(PL_warn_locale);
1511 PL_warn_locale = NULL;
1512 }
1513
c1284011 1514 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1515
1516 /* A UTF-8 locale gets standard rules. But note that code still has to
1517 * handle this specially because of the three problematic code points */
1518 if (PL_in_utf8_CTYPE_locale) {
1519 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1520 }
8b7358b9
KW
1521
1522 /* We don't populate the other lists if a UTF-8 locale, but do check that
1523 * everything works as expected, unless checking turned off */
1524 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1525 /* Assume enough space for every character being bad. 4 spaces each
1526 * for the 94 printable characters that are output like "'x' "; and 5
1527 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1528 * NUL */
8b7358b9 1529 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1530 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1531 to start */
1532 unsigned int bad_count = 0; /* Count of bad characters */
1533
baa60164 1534 for (i = 0; i < 256; i++) {
8b7358b9 1535 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1536 if (isupper(i))
1537 PL_fold_locale[i] = (U8) tolower(i);
1538 else if (islower(i))
1539 PL_fold_locale[i] = (U8) toupper(i);
1540 else
1541 PL_fold_locale[i] = (U8) i;
8b7358b9 1542 }
8c6180a9
KW
1543
1544 /* If checking for locale problems, see if the native ASCII-range
1545 * printables plus \n and \t are in their expected categories in
1546 * the new locale. If not, this could mean big trouble, upending
1547 * Perl's and most programs' assumptions, like having a
1548 * metacharacter with special meaning become a \w. Fortunately,
1549 * it's very rare to find locales that aren't supersets of ASCII
1550 * nowadays. It isn't a problem for most controls to be changed
1551 * into something else; we check only \n and \t, though perhaps \r
1552 * could be an issue as well. */
7d4bcc4a 1553 if ( check_for_problems
8c6180a9
KW
1554 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1555 {
8b7358b9 1556 bool is_bad = FALSE;
6726b4f4 1557 char name[4] = { '\0' };
8b7358b9
KW
1558
1559 /* Convert the name into a string */
6726b4f4 1560 if (isGRAPH_A(i)) {
8b7358b9
KW
1561 name[0] = i;
1562 name[1] = '\0';
1563 }
1564 else if (i == '\n') {
6726b4f4
KW
1565 my_strlcpy(name, "\\n", sizeof(name));
1566 }
1567 else if (i == '\t') {
1568 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1569 }
1570 else {
6726b4f4
KW
1571 assert(i == ' ');
1572 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1573 }
1574
1575 /* Check each possibe class */
1576 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1577 is_bad = TRUE;
1578 DEBUG_L(PerlIO_printf(Perl_debug_log,
1579 "isalnum('%s') unexpectedly is %d\n",
1580 name, cBOOL(isalnum(i))));
1581 }
1582 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1583 is_bad = TRUE;
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isalpha('%s') unexpectedly is %d\n",
1586 name, cBOOL(isalpha(i))));
1587 }
1588 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1589 is_bad = TRUE;
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isdigit('%s') unexpectedly is %d\n",
1592 name, cBOOL(isdigit(i))));
1593 }
1594 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1595 is_bad = TRUE;
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isgraph('%s') unexpectedly is %d\n",
1598 name, cBOOL(isgraph(i))));
1599 }
1600 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1601 is_bad = TRUE;
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "islower('%s') unexpectedly is %d\n",
1604 name, cBOOL(islower(i))));
1605 }
1606 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1607 is_bad = TRUE;
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "isprint('%s') unexpectedly is %d\n",
1610 name, cBOOL(isprint(i))));
1611 }
1612 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1613 is_bad = TRUE;
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "ispunct('%s') unexpectedly is %d\n",
1616 name, cBOOL(ispunct(i))));
1617 }
1618 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1619 is_bad = TRUE;
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "isspace('%s') unexpectedly is %d\n",
1622 name, cBOOL(isspace(i))));
1623 }
1624 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1625 is_bad = TRUE;
1626 DEBUG_L(PerlIO_printf(Perl_debug_log,
1627 "isupper('%s') unexpectedly is %d\n",
1628 name, cBOOL(isupper(i))));
1629 }
1630 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1631 is_bad = TRUE;
1632 DEBUG_L(PerlIO_printf(Perl_debug_log,
1633 "isxdigit('%s') unexpectedly is %d\n",
1634 name, cBOOL(isxdigit(i))));
1635 }
1636 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1637 is_bad = TRUE;
1638 DEBUG_L(PerlIO_printf(Perl_debug_log,
1639 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1640 name, tolower(i), (int) toLOWER_A(i)));
1641 }
1642 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1643 is_bad = TRUE;
1644 DEBUG_L(PerlIO_printf(Perl_debug_log,
1645 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1646 name, toupper(i), (int) toUPPER_A(i)));
1647 }
1648 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1649 is_bad = TRUE;
1650 DEBUG_L(PerlIO_printf(Perl_debug_log,
1651 "'\\n' (=%02X) is not a control\n", (int) i));
1652 }
1653
1654 /* Add to the list; Separate multiple entries with a blank */
1655 if (is_bad) {
1656 if (bad_count) {
1657 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1658 }
8b7358b9
KW
1659 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1660 bad_count++;
8c6180a9
KW
1661 }
1662 }
1663 }
1664
7d4bcc4a
KW
1665# ifdef MB_CUR_MAX
1666
8c6180a9 1667 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1668 * this locale requires more than one byte, there are going to be
8c6180a9 1669 * problems. */
9c8a6dc2
KW
1670 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1671 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1672 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1673
8b7358b9
KW
1674 if ( check_for_problems && MB_CUR_MAX > 1
1675 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1676
1677 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1678 * locale. Just assume that the implementation for them (plus
1679 * for POSIX) is correct and the > 1 value is spurious. (Since
1680 * these are specially handled to never be considered UTF-8
1681 * locales, as long as this is the only problem, everything
1682 * should work fine */
1683 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1684 {
8c6180a9
KW
1685 multi_byte_locale = TRUE;
1686 }
7d4bcc4a
KW
1687
1688# endif
8c6180a9 1689
8b7358b9
KW
1690 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1691 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1692 PL_warn_locale = Perl_newSVpvf(aTHX_
1693 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1694 " which have\nunexpected meanings: %s\nThe Perl program"
1695 " will use the expected meanings",
8b7358b9 1696 newctype, bad_chars_list);
8b7358b9
KW
1697 }
1698 else {
1699 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1700 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1701 newctype,
8c6180a9
KW
1702 (multi_byte_locale)
1703 ? " Some characters in it are not recognized by"
1704 " Perl."
1705 : "",
1706 (bad_count)
1707 ? "\nThe following characters (and maybe others)"
1708 " may not have the same meaning as the Perl"
1709 " program expects:\n"
1710 : "",
1711 (bad_count)
1712 ? bad_chars_list
1713 : ""
1714 );
8b7358b9
KW
1715 }
1716
b119c2be
KW
1717# ifdef HAS_NL_LANGINFO
1718
1719 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1720 /* parameter FALSE is a don't care here */
4e6826bf 1721 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1722
1723# endif
1724
8b7358b9
KW
1725 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1726
cc9eaeb0 1727 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1728 * output the message now. If not in that scope, we save the
1729 * message to be output at the first operation using this locale,
1730 * if that actually happens. Most programs don't use locales, so
1731 * they are immune to bad ones. */
cc9eaeb0 1732 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1733
780fcc9f
KW
1734 /* The '0' below suppresses a bogus gcc compiler warning */
1735 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1736
bddebb56
KW
1737 if (IN_LC(LC_CTYPE)) {
1738 SvREFCNT_dec_NN(PL_warn_locale);
1739 PL_warn_locale = NULL;
1740 }
780fcc9f 1741 }
baa60164 1742 }
31f05a37 1743 }
98994639
HS
1744
1745#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1746
98994639
HS
1747}
1748
98994639 1749void
2726666d
KW
1750Perl__warn_problematic_locale()
1751{
2726666d
KW
1752
1753#ifdef USE_LOCALE_CTYPE
1754
5f04a188
KW
1755 dTHX;
1756
1757 /* Internal-to-core function that outputs the message in PL_warn_locale,
1758 * and then NULLS it. Should be called only through the macro
1759 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1760
2726666d 1761 if (PL_warn_locale) {
2726666d
KW
1762 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1763 SvPVX(PL_warn_locale),
1764 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1765 SvREFCNT_dec_NN(PL_warn_locale);
1766 PL_warn_locale = NULL;
1767 }
1768
1769#endif
1770
1771}
1772
a4f00dcc
KW
1773STATIC void
1774S_new_collate(pTHX_ const char *newcoll)
98994639 1775{
7d4bcc4a
KW
1776
1777#ifndef USE_LOCALE_COLLATE
1778
1779 PERL_UNUSED_ARG(newcoll);
1780 PERL_UNUSED_CONTEXT;
1781
1782#else
0d071d52 1783
291a84fb 1784 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1785 * core Perl this and that 'newcoll' is the name of the new locale.
1786 *
d35fca5f
KW
1787 * The design of locale collation is that every locale change is given an
1788 * index 'PL_collation_ix'. The first time a string particpates in an
1789 * operation that requires collation while locale collation is active, it
1790 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1791 * magic includes the collation index, and the transformation of the string
1792 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1793 * instead of the string itself. If a string changes, the magic is
1794 * cleared. The next time the locale changes, the index is incremented,
1795 * and so we know during a comparison that the transformation is not
1796 * necessarily still valid, and so is recomputed. Note that if the locale
1797 * changes enough times, the index could wrap (a U32), and it is possible
1798 * that a transformation would improperly be considered valid, leading to
1799 * an unlikely bug */
0d071d52 1800
98994639
HS
1801 if (! newcoll) {
1802 if (PL_collation_name) {
1803 ++PL_collation_ix;
1804 Safefree(PL_collation_name);
1805 PL_collation_name = NULL;
1806 }
1807 PL_collation_standard = TRUE;
00bf60ca 1808 is_standard_collation:
98994639
HS
1809 PL_collxfrm_base = 0;
1810 PL_collxfrm_mult = 2;
165a1c52 1811 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1812 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1813 PL_strxfrm_max_cp = 0;
98994639
HS
1814 return;
1815 }
1816
d35fca5f 1817 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1818 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1819 ++PL_collation_ix;
1820 Safefree(PL_collation_name);
1821 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1822 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1823 if (PL_collation_standard) {
1824 goto is_standard_collation;
1825 }
98994639 1826
165a1c52 1827 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1828 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1829 PL_strxfrm_max_cp = 0;
165a1c52 1830
59c018b9
KW
1831 /* A locale collation definition includes primary, secondary, tertiary,
1832 * etc. weights for each character. To sort, the primary weights are
1833 * used, and only if they compare equal, then the secondary weights are
1834 * used, and only if they compare equal, then the tertiary, etc.
1835 *
1836 * strxfrm() works by taking the input string, say ABC, and creating an
1837 * output transformed string consisting of first the primary weights,
1838 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1839 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1840 * may not have weights at every level. In our example, let's say B
1841 * doesn't have a tertiary weight, and A doesn't have a secondary
1842 * weight. The constructed string is then going to be
1843 * A¹B¹C¹ B²C² A³C³ ....
1844 * This has the desired effect that strcmp() will look at the secondary
1845 * or tertiary weights only if the strings compare equal at all higher
1846 * priority weights. The spaces shown here, like in
c342d20e 1847 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1848 * are not just for readability. In the general case, these must
1849 * actually be bytes, which we will call here 'separator weights'; and
1850 * they must be smaller than any other weight value, but since these
1851 * are C strings, only the terminating one can be a NUL (some
1852 * implementations may include a non-NUL separator weight just before
1853 * the NUL). Implementations tend to reserve 01 for the separator
1854 * weights. They are needed so that a shorter string's secondary
1855 * weights won't be misconstrued as primary weights of a longer string,
1856 * etc. By making them smaller than any other weight, the shorter
1857 * string will sort first. (Actually, if all secondary weights are
1858 * smaller than all primary ones, there is no need for a separator
1859 * weight between those two levels, etc.)
1860 *
1861 * The length of the transformed string is roughly a linear function of
1862 * the input string. It's not exactly linear because some characters
1863 * don't have weights at all levels. When we call strxfrm() we have to
1864 * allocate some memory to hold the transformed string. The
1865 * calculations below try to find coefficients 'm' and 'b' for this
1866 * locale so that m*x + b equals how much space we need, given the size
1867 * of the input string in 'x'. If we calculate too small, we increase
1868 * the size as needed, and call strxfrm() again, but it is better to
1869 * get it right the first time to avoid wasted expensive string
1870 * transformations. */
1871
98994639 1872 {
79f120c8
KW
1873 /* We use the string below to find how long the tranformation of it
1874 * is. Almost all locales are supersets of ASCII, or at least the
1875 * ASCII letters. We use all of them, half upper half lower,
1876 * because if we used fewer, we might hit just the ones that are
1877 * outliers in a particular locale. Most of the strings being
1878 * collated will contain a preponderance of letters, and even if
1879 * they are above-ASCII, they are likely to have the same number of
1880 * weight levels as the ASCII ones. It turns out that digits tend
1881 * to have fewer levels, and some punctuation has more, but those
1882 * are relatively sparse in text, and khw believes this gives a
1883 * reasonable result, but it could be changed if experience so
1884 * dictates. */
1885 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1886 char * x_longer; /* Transformed 'longer' */
1887 Size_t x_len_longer; /* Length of 'x_longer' */
1888
1889 char * x_shorter; /* We also transform a substring of 'longer' */
1890 Size_t x_len_shorter;
1891
a4a439fb 1892 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1893 * are interested only in its length). It is used because it has
1894 * the intelligence to handle all cases, but to work, it needs some
1895 * values of 'm' and 'b' to get it started. For the purposes of
1896 * this calculation we use a very conservative estimate of 'm' and
1897 * 'b'. This assumes a weight can be multiple bytes, enough to
1898 * hold any UV on the platform, and there are 5 levels, 4 weight
1899 * bytes, and a trailing NUL. */
1900 PL_collxfrm_base = 5;
1901 PL_collxfrm_mult = 5 * sizeof(UV);
1902
1903 /* Find out how long the transformation really is */
a4a439fb
KW
1904 x_longer = _mem_collxfrm(longer,
1905 sizeof(longer) - 1,
1906 &x_len_longer,
1907
1908 /* We avoid converting to UTF-8 in the
1909 * called function by telling it the
1910 * string is in UTF-8 if the locale is a
1911 * UTF-8 one. Since the string passed
1912 * here is invariant under UTF-8, we can
1913 * claim it's UTF-8 even though it isn't.
1914 * */
1915 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1916 Safefree(x_longer);
1917
1918 /* Find out how long the transformation of a substring of 'longer'
1919 * is. Together the lengths of these transformations are
1920 * sufficient to calculate 'm' and 'b'. The substring is all of
1921 * 'longer' except the first character. This minimizes the chances
1922 * of being swayed by outliers */
a4a439fb 1923 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1924 sizeof(longer) - 2,
a4a439fb
KW
1925 &x_len_shorter,
1926 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1927 Safefree(x_shorter);
1928
1929 /* If the results are nonsensical for this simple test, the whole
1930 * locale definition is suspect. Mark it so that locale collation
1931 * is not active at all for it. XXX Should we warn? */
1932 if ( x_len_shorter == 0
1933 || x_len_longer == 0
1934 || x_len_shorter >= x_len_longer)
1935 {
1936 PL_collxfrm_mult = 0;
1937 PL_collxfrm_base = 0;
1938 }
1939 else {
1940 SSize_t base; /* Temporary */
1941
1942 /* We have both: m * strlen(longer) + b = x_len_longer
1943 * m * strlen(shorter) + b = x_len_shorter;
1944 * subtracting yields:
1945 * m * (strlen(longer) - strlen(shorter))
1946 * = x_len_longer - x_len_shorter
1947 * But we have set things up so that 'shorter' is 1 byte smaller
1948 * than 'longer'. Hence:
1949 * m = x_len_longer - x_len_shorter
1950 *
1951 * But if something went wrong, make sure the multiplier is at
1952 * least 1.
1953 */
1954 if (x_len_longer > x_len_shorter) {
1955 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1956 }
1957 else {
1958 PL_collxfrm_mult = 1;
1959 }
1960
1961 /* mx + b = len
1962 * so: b = len - mx
1963 * but in case something has gone wrong, make sure it is
1964 * non-negative */
1965 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1966 if (base < 0) {
1967 base = 0;
1968 }
1969
1970 /* Add 1 for the trailing NUL */
1971 PL_collxfrm_base = base + 1;
1972 }
58eebef2 1973
7d4bcc4a
KW
1974# ifdef DEBUGGING
1975
58eebef2
KW
1976 if (DEBUG_L_TEST || debug_initialization) {
1977 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1978 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1979 "x_len_longer=%zu,"
1980 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1981 __FILE__, __LINE__,
1982 PL_in_utf8_COLLATE_locale,
1983 x_len_shorter, x_len_longer,
1984 PL_collxfrm_mult, PL_collxfrm_base);
1985 }
7d4bcc4a
KW
1986# endif
1987
98994639
HS
1988 }
1989 }
1990
1991#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1992
98994639
HS
1993}
1994
d2b24094 1995#ifdef WIN32
b8cc575c 1996
a4f00dcc 1997STATIC char *
b8cc575c 1998S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1999{
2000 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2001 * difference between the two unless the input locale is "", which normally
2002 * means on Windows to get the machine default, which is set via the
2003 * computer's "Regional and Language Options" (or its current equivalent).
2004 * In POSIX, it instead means to find the locale from the user's
2005 * environment. This routine changes the Windows behavior to first look in
2006 * the environment, and, if anything is found, use that instead of going to
2007 * the machine default. If there is no environment override, the machine
2008 * default is used, by calling the real setlocale() with "".
2009 *
2010 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2011 * use the particular category's variable if set; otherwise to use the LANG
2012 * variable. */
b385bb4d 2013
175c4cf9 2014 bool override_LC_ALL = FALSE;
89f7b9aa 2015 char * result;
e5f10d49 2016 unsigned int i;
89f7b9aa 2017
b385bb4d 2018 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2019
2020# ifdef LC_ALL
2021
b385bb4d
KW
2022 locale = PerlEnv_getenv("LC_ALL");
2023 if (! locale) {
e5f10d49
KW
2024 if (category == LC_ALL) {
2025 override_LC_ALL = TRUE;
2026 }
2027 else {
7d4bcc4a
KW
2028
2029# endif
7d4bcc4a 2030
e5f10d49
KW
2031 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2032 if (category == categories[i]) {
2033 locale = PerlEnv_getenv(category_names[i]);
2034 goto found_locale;
2035 }
2036 }
7d4bcc4a 2037
b385bb4d 2038 locale = PerlEnv_getenv("LANG");
481465ea 2039 if (! locale) {
b385bb4d
KW
2040 locale = "";
2041 }
e5f10d49
KW
2042
2043 found_locale: ;
7d4bcc4a
KW
2044
2045# ifdef LC_ALL
2046
e5f10d49 2047 }
b385bb4d 2048 }
7d4bcc4a
KW
2049
2050# endif
2051
b385bb4d
KW
2052 }
2053
89f7b9aa 2054 result = setlocale(category, locale);
48015184
KW
2055 DEBUG_L(STMT_START {
2056 dSAVE_ERRNO;
2057 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2058 setlocale_debug_string(category, locale, result));
2059 RESTORE_ERRNO;
2060 } STMT_END);
89f7b9aa 2061
481465ea 2062 if (! override_LC_ALL) {
89f7b9aa
KW
2063 return result;
2064 }
2065
dfd77d7a 2066 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2067 * LANG variable or the system default if there is no LANG. But these have
2068 * lower priority than the other LC_foo variables, so override it for each
2069 * one that is set. (If they are set to "", it means to use the same thing
2070 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2071
948523db 2072 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2073 result = PerlEnv_getenv(category_names[i]);
2074 if (result && strNE(result, "")) {
2075 setlocale(categories[i], result);
2076 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2077 __FILE__, __LINE__,
2078 setlocale_debug_string(categories[i], result, "not captured")));
2079 }
89f7b9aa 2080 }
7d4bcc4a 2081
bbc98134 2082 result = setlocale(LC_ALL, NULL);
48015184
KW
2083 DEBUG_L(STMT_START {
2084 dSAVE_ERRNO;
2085 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2086 __FILE__, __LINE__,
48015184
KW
2087 setlocale_debug_string(LC_ALL, NULL, result));
2088 RESTORE_ERRNO;
2089 } STMT_END);
89f7b9aa 2090
bbc98134 2091 return result;
b385bb4d
KW
2092}
2093
2094#endif
2095
9aac5db8
KW
2096/*
2097
2098=head1 Locale-related functions and macros
2099
2100=for apidoc Perl_setlocale
2101
2102This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2103taking the same parameters, and returning the same information, except that it
9487427b
KW
2104returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2105instead return C<C> if the underlying locale has a non-dot decimal point
2106character, or a non-empty thousands separator for displaying floating point
2107numbers. This is because perl keeps that locale category such that it has a
2108dot and empty separator, changing the locale briefly during the operations
2109where the underlying one is required. C<Perl_setlocale> knows about this, and
2110compensates; regular C<setlocale> doesn't.
9aac5db8 2111
e9bc6d6b 2112Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2113declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2114C<const> (presumably because its API was specified long ago, and can't be
2115updated; it is illegal to change the information C<setlocale> returns; doing
2116so leads to segfaults.)
9aac5db8 2117
e9bc6d6b
KW
2118Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2119C<setlocale> can be completely ineffective on some platforms under some
2120configurations.
2121
9aac5db8 2122C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2123where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2124the system C<setlocale()> is ineffective, returning the wrong information, and
2125failing to actually change the locale. C<Perl_setlocale>, however works
2126properly in all circumstances.
9aac5db8
KW
2127
2128The return points to a per-thread static buffer, which is overwritten the next
2129time C<Perl_setlocale> is called from the same thread.
2130
2131=cut
2132
2133*/
2134
2135const char *
2136Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2137{
2138 /* This wraps POSIX::setlocale() */
2139
9aac5db8
KW
2140 const char * retval;
2141 const char * newlocale;
2142 dSAVEDERRNO;
2143 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2144 dTHX;
2145
a4f00dcc
KW
2146#ifdef USE_LOCALE_NUMERIC
2147
291a84fb
KW
2148 /* A NULL locale means only query what the current one is. We have the
2149 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2150 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2151 * the correct results. All other categories don't require special
2152 * handling */
a4f00dcc
KW
2153 if (locale == NULL) {
2154 if (category == LC_NUMERIC) {
9aac5db8
KW
2155
2156 /* We don't have to copy this return value, as it is a per-thread
2157 * variable, and won't change until a future setlocale */
2158 return PL_numeric_name;
a4f00dcc
KW
2159 }
2160
7d4bcc4a 2161# ifdef LC_ALL
a4f00dcc 2162
9aac5db8
KW
2163 else if (category == LC_ALL) {
2164 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2165 }
2166
7d4bcc4a 2167# endif
a4f00dcc
KW
2168
2169 }
2170
2171#endif
2172
33e5a354
KW
2173 retval = save_to_buffer(do_setlocale_r(category, locale),
2174 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2175 SAVE_ERRNO;
2176
2177#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2178
2179 if (locale == NULL && category == LC_ALL) {
2180 RESTORE_LC_NUMERIC();
2181 }
2182
2183#endif
a4f00dcc
KW
2184
2185 DEBUG_L(PerlIO_printf(Perl_debug_log,
2186 "%s:%d: %s\n", __FILE__, __LINE__,
2187 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2188
9aac5db8
KW
2189 RESTORE_ERRNO;
2190
2191 if (! retval) {
a4f00dcc
KW
2192 return NULL;
2193 }
2194
9aac5db8 2195 /* If locale == NULL, we are just querying the state */
a4f00dcc 2196 if (locale == NULL) {
a4f00dcc
KW
2197 return retval;
2198 }
a4f00dcc 2199
1159483a
KW
2200 /* Now that have switched locales, we have to update our records to
2201 * correspond. */
a4f00dcc 2202
1159483a 2203 switch (category) {
a4f00dcc 2204
1159483a 2205#ifdef USE_LOCALE_CTYPE
a4f00dcc 2206
1159483a
KW
2207 case LC_CTYPE:
2208 new_ctype(retval);
2209 break;
a4f00dcc 2210
1159483a 2211#endif
a4f00dcc
KW
2212#ifdef USE_LOCALE_COLLATE
2213
1159483a
KW
2214 case LC_COLLATE:
2215 new_collate(retval);
2216 break;
a4f00dcc 2217
1159483a 2218#endif
a4f00dcc
KW
2219#ifdef USE_LOCALE_NUMERIC
2220
1159483a
KW
2221 case LC_NUMERIC:
2222 new_numeric(retval);
2223 break;
a4f00dcc 2224
1159483a
KW
2225#endif
2226#ifdef LC_ALL
a4f00dcc 2227
1159483a 2228 case LC_ALL:
a4f00dcc 2229
1159483a
KW
2230 /* LC_ALL updates all the things we care about. The values may not
2231 * be the same as 'retval', as the locale "" may have set things
2232 * individually */
a4f00dcc 2233
1159483a 2234# ifdef USE_LOCALE_CTYPE
a4f00dcc 2235
b0bde642 2236 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2237 new_ctype(newlocale);
b0bde642 2238 Safefree(newlocale);
a4f00dcc 2239
1159483a
KW
2240# endif /* USE_LOCALE_CTYPE */
2241# ifdef USE_LOCALE_COLLATE
2242
b0bde642 2243 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2244 new_collate(newlocale);
b0bde642 2245 Safefree(newlocale);
a4f00dcc 2246
7d4bcc4a 2247# endif
1159483a 2248# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2249
b0bde642 2250 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2251 new_numeric(newlocale);
b0bde642 2252 Safefree(newlocale);
a4f00dcc 2253
1159483a
KW
2254# endif /* USE_LOCALE_NUMERIC */
2255#endif /* LC_ALL */
a4f00dcc 2256
1159483a
KW
2257 default:
2258 break;
a4f00dcc
KW
2259 }
2260
2261 return retval;
2262
f7416781
KW
2263}
2264
2265PERL_STATIC_INLINE const char *
2266S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2267{
2268 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2269 * growing it if necessary */
2270
0b6697c5 2271 Size_t string_size;
f7416781
KW
2272
2273 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2274
0b6697c5
KW
2275 if (! string) {
2276 return NULL;
2277 }
2278
2279 string_size = strlen(string) + offset + 1;
2280
f7416781
KW
2281 if (*buf_size == 0) {
2282 Newx(*buf, string_size, char);
2283 *buf_size = string_size;
2284 }
2285 else if (string_size > *buf_size) {
2286 Renew(*buf, string_size, char);
2287 *buf_size = string_size;
2288 }
2289
2290 Copy(string, *buf + offset, string_size - offset, char);
2291 return *buf;
2292}
2293
2294/*
2295
f7416781
KW
2296=for apidoc Perl_langinfo
2297
ce181bf6 2298This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2299taking the same C<item> parameter values, and returning the same information.
2300But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2301of Perl's locale handling from your code, and can be used on systems that lack
2302a native C<nl_langinfo>.
2303
2304Expanding on these:
2305
2306=over
2307
2308=item *
2309
ce181bf6
KW
2310The reason it isn't quite a drop-in replacement is actually an advantage. The
2311only difference is that it returns S<C<const char *>>, whereas plain
2312C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2313forbidden to write into the buffer. By declaring this C<const>, the compiler
2314enforces this restriction, so if it is violated, you know at compilation time,
2315rather than getting segfaults at runtime.
2316
2317=item *
2318
69e2275e 2319It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2320without you having to write extra code. The reason for the extra code would be
2321because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2322kept set by Perl so that the radix is a dot, and the separator is the empty
2323string, no matter what the underlying locale is supposed to be, and so to get
2324the expected results, you have to temporarily toggle into the underlying
2325locale, and later toggle back. (You could use plain C<nl_langinfo> and
2326C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2327the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2328(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2329(decimal point) character to be a dot.)
ce181bf6
KW
2330
2331=item *
2332
2333The system function it replaces can have its static return buffer trashed,
2334not only by a subesequent call to that function, but by a C<freelocale>,
2335C<setlocale>, or other locale change. The returned buffer of this function is
2336not changed until the next call to it, so the buffer is never in a trashed
2337state.
f7416781
KW
2338
2339=item *
2340
ce181bf6
KW
2341Its return buffer is per-thread, so it also is never overwritten by a call to
2342this function from another thread; unlike the function it replaces.
2343
2344=item *
2345
2346But most importantly, it works on systems that don't have C<nl_langinfo>, such
2347as Windows, hence makes your code more portable. Of the fifty-some possible
2348items specified by the POSIX 2008 standard,
f7416781 2349L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2350only one is completely unimplemented, though on non-Windows platforms, another
2351significant one is also not implemented). It uses various techniques to
2352recover the other items, including calling C<L<localeconv(3)>>, and
2353C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2354available. Later C<strftime()> versions have additional capabilities; C<""> is
2355returned for those not available on your system.
ce181bf6
KW
2356
2357It is important to note that when called with an item that is recovered by
2358using C<localeconv>, the buffer from any previous explicit call to
2359C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2360contents if you need to access them after a call to this function. (But note
2361that you might not want to be using C<localeconv()> directly anyway, because of
2362issues like the ones listed in the second item of this list (above) for
2363C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2364call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2365unpack).
9f0bc911 2366
6201c220
KW
2367The details for those items which may deviate from what this emulation returns
2368and what a native C<nl_langinfo()> would return are specified in
2369L<I18N::Langinfo>.
f7416781 2370
ce181bf6
KW
2371=back
2372
f7416781
KW
2373When using C<Perl_langinfo> on systems that don't have a native
2374C<nl_langinfo()>, you must
2375
2376 #include "perl_langinfo.h"
2377
2378before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2379C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2380C<langinfo.h> would try to import into the namespace for code that doesn't need
2381it.)
f7416781 2382
f7416781
KW
2383The original impetus for C<Perl_langinfo()> was so that code that needs to
2384find out the current currency symbol, floating point radix character, or digit
2385grouping separator can use, on all systems, the simpler and more
2386thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2387pain to make thread-friendly. For other fields returned by C<localeconv>, it
2388is better to use the methods given in L<perlcall> to call
2389L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2390
2391=cut
2392
2393*/
2394
2395const char *
2396#ifdef HAS_NL_LANGINFO
2397Perl_langinfo(const nl_item item)
2398#else
2399Perl_langinfo(const int item)
2400#endif
2401{
f61748ac
KW
2402 return my_nl_langinfo(item, TRUE);
2403}
2404
59a15af0 2405STATIC const char *
f61748ac
KW
2406#ifdef HAS_NL_LANGINFO
2407S_my_nl_langinfo(const nl_item item, bool toggle)
2408#else
2409S_my_nl_langinfo(const int item, bool toggle)
2410#endif
2411{
ae74815b 2412 dTHX;
0b6697c5 2413 const char * retval;
f7416781 2414
5a854ab3
KW
2415 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2416 * two items, and only if not already there */
4e6826bf 2417 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3
KW
2418 || PL_numeric_underlying))
2419 {
2420 toggle = FALSE;
2421 }
2422
ab340fff 2423#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2424# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2425 || ! defined(HAS_POSIX_2008_LOCALE) \
2426 || ! defined(DUPLOCALE)
f7416781 2427
ab340fff 2428 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2429 * for those items dependent on it. This must be copied to a buffer before
2430 * switching back, as some systems destroy the buffer when setlocale() is
2431 * called */
f7416781 2432
038d3702
KW
2433 {
2434 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2435
b2fee59c 2436 if (toggle) {
038d3702 2437 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2438 }
f7416781 2439
5acc4454
KW
2440 LOCALE_LOCK; /* Prevent interference from another thread executing
2441 this code section (the only call to nl_langinfo in
2442 the core) */
2443
ee90eb2d 2444
628ff75a
KW
2445 /* Copy to a per-thread buffer, which is also one that won't be
2446 * destroyed by a subsequent setlocale(), such as the
2447 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2448 retval = save_to_buffer(nl_langinfo(item),
2449 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2450
5acc4454
KW
2451 LOCALE_UNLOCK;
2452
b2fee59c 2453 if (toggle) {
038d3702 2454 RESTORE_LC_NUMERIC();
b2fee59c 2455 }
f7416781
KW
2456 }
2457
ab340fff
KW
2458# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2459
5a854ab3 2460 {
b2fee59c
KW
2461 bool do_free = FALSE;
2462 locale_t cur = uselocale((locale_t) 0);
ab340fff 2463
b2fee59c
KW
2464 if (cur == LC_GLOBAL_LOCALE) {
2465 cur = duplocale(LC_GLOBAL_LOCALE);
2466 do_free = TRUE;
2467 }
ab340fff 2468
b2fee59c 2469 if (toggle) {
e1aa2579
KW
2470 if (PL_underlying_numeric_obj) {
2471 cur = PL_underlying_numeric_obj;
2472 }
2473 else {
2474 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2475 do_free = TRUE;
2476 }
b2fee59c 2477 }
ab340fff 2478
628ff75a
KW
2479 /* We have to save it to a buffer, because the freelocale() just below
2480 * can invalidate the internal one */
0b6697c5
KW
2481 retval = save_to_buffer(nl_langinfo_l(item, cur),
2482 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2483
b2fee59c
KW
2484 if (do_free) {
2485 freelocale(cur);
2486 }
5a854ab3 2487 }
ab340fff 2488
c1566110
KW
2489# endif
2490
0b6697c5 2491 if (strEQ(retval, "")) {
4e6826bf 2492 if (item == YESSTR) {
c1566110
KW
2493 return "yes";
2494 }
4e6826bf 2495 if (item == NOSTR) {
c1566110
KW
2496 return "no";
2497 }
2498 }
2499
0b6697c5 2500 return retval;
ab340fff 2501
f7416781 2502#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2503
2504 {
2505
f7416781
KW
2506# ifdef HAS_LOCALECONV
2507
43dd6b15 2508 const struct lconv* lc;
628ff75a 2509 const char * temp;
038d3702 2510 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2511
69c5e0db
KW
2512# ifdef TS_W32_BROKEN_LOCALECONV
2513
2514 const char * save_global;
2515 const char * save_thread;
2516 int needed_size;
2517 char * ptr;
2518 char * e;
2519 char * item_start;
2520
2521# endif
f7416781
KW
2522# endif
2523# ifdef HAS_STRFTIME
2524
43dd6b15
KW
2525 struct tm tm;
2526 bool return_format = FALSE; /* Return the %format, not the value */
2527 const char * format;
f7416781
KW
2528
2529# endif
2530
43dd6b15
KW
2531 /* We copy the results to a per-thread buffer, even if not
2532 * multi-threaded. This is in part to simplify this code, and partly
2533 * because we need a buffer anyway for strftime(), and partly because a
2534 * call of localeconv() could otherwise wipe out the buffer, and the
2535 * programmer would not be expecting this, as this is a nl_langinfo()
2536 * substitute after all, so s/he might be thinking their localeconv()
2537 * is safe until another localeconv() call. */
f7416781 2538
43dd6b15
KW
2539 switch (item) {
2540 Size_t len;
f7416781 2541
8d72e74e 2542 /* This is unimplemented */
4e6826bf 2543 case ERA: /* For use with strftime() %E modifier */
f7416781 2544
43dd6b15
KW
2545 default:
2546 return "";
f7416781 2547
43dd6b15 2548 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2549 case YESEXPR: return "^[+1yY]";
2550 case YESSTR: return "yes";
2551 case NOEXPR: return "^[-0nN]";
2552 case NOSTR: return "no";
2553
8d72e74e
KW
2554 case CODESET:
2555
2556# ifndef WIN32
2557
2558 /* On non-windows, this is unimplemented, in part because of
2559 * inconsistencies between vendors. The Darwin native
2560 * nl_langinfo() implementation simply looks at everything past
2561 * any dot in the name, but that doesn't work for other
2562 * vendors. Many Linux locales that don't have UTF-8 in their
2563 * names really are UTF-8, for example; z/OS locales that do
2564 * have UTF-8 in their names, aren't really UTF-8 */
2565 return "";
2566
2567# else
2568
2569 { /* But on Windows, the name does seem to be consistent, so
2570 use that. */
2571 const char * p;
2572 const char * first;
2573 Size_t offset = 0;
2574 const char * name = my_setlocale(LC_CTYPE, NULL);
2575
2576 if (isNAME_C_OR_POSIX(name)) {
2577 return "ANSI_X3.4-1968";
2578 }
2579
2580 /* Find the dot in the locale name */
2581 first = (const char *) strchr(name, '.');
2582 if (! first) {
2583 first = name;
2584 goto has_nondigit;
2585 }
2586
2587 /* Look at everything past the dot */
2588 first++;
2589 p = first;
2590
2591 while (*p) {
2592 if (! isDIGIT(*p)) {
2593 goto has_nondigit;
2594 }
2595
2596 p++;
2597 }
2598
2599 /* Here everything past the dot is a digit. Treat it as a
2600 * code page */
18503cbd 2601 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2602 &PL_langinfo_bufsize, 0);
8d72e74e 2603 offset = STRLENs("CP");
f7416781 2604
8d72e74e
KW
2605 has_nondigit:
2606
2607 retval = save_to_buffer(first, &PL_langinfo_buf,
2608 &PL_langinfo_bufsize, offset);
2609 }
2610
2611 break;
2612
2613# endif
f7416781
KW
2614# ifdef HAS_LOCALECONV
2615
4e6826bf 2616 case CRNCYSTR:
f7416781 2617
43dd6b15
KW
2618 /* We don't bother with localeconv_l() because any system that
2619 * has it is likely to also have nl_langinfo() */
291a84fb 2620
69c5e0db
KW
2621 LOCALE_LOCK_V; /* Prevent interference with other threads
2622 using localeconv() */
2623
2624# ifdef TS_W32_BROKEN_LOCALECONV
2625
2626 /* This is a workaround for a Windows bug prior to VS 15.
2627 * What we do here is, while locked, switch to the global
2628 * locale so localeconv() works; then switch back just before
2629 * the unlock. This can screw things up if some thread is
2630 * already using the global locale while assuming no other is.
2631 * A different workaround would be to call GetCurrencyFormat on
2632 * a known value, and parse it; patches welcome
2633 *
2634 * We have to use LC_ALL instead of LC_MONETARY because of
2635 * another bug in Windows */
2636
2637 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2638 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2639 save_global= savepv(my_setlocale(LC_ALL, NULL));
2640 my_setlocale(LC_ALL, save_thread);
2641
2642# endif
5acc4454 2643
43dd6b15
KW
2644 lc = localeconv();
2645 if ( ! lc
2646 || ! lc->currency_symbol
2647 || strEQ("", lc->currency_symbol))
2648 {
69c5e0db 2649 LOCALE_UNLOCK_V;
43dd6b15
KW
2650 return "";
2651 }
f7416781 2652
43dd6b15 2653 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2654 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2655 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2656 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2657 { /* khw couldn't figure out how the localedef specifications
2658 would show that the $ should replace the radix; this is
2659 just a guess as to how it might work.*/
a34edee3 2660 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2661 }
2662 else if (lc->p_cs_precedes) {
a34edee3 2663 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2664 }
2665 else {
a34edee3 2666 PL_langinfo_buf[0] = '+';
43dd6b15 2667 }
f7416781 2668
69c5e0db
KW
2669# ifdef TS_W32_BROKEN_LOCALECONV
2670
2671 my_setlocale(LC_ALL, save_global);
2672 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2673 my_setlocale(LC_ALL, save_thread);
2674 Safefree(save_global);
2675 Safefree(save_thread);
2676
2677# endif
2678
2679 LOCALE_UNLOCK_V;
43dd6b15 2680 break;
f7416781 2681
69c5e0db
KW
2682# ifdef TS_W32_BROKEN_LOCALECONV
2683
4e6826bf 2684 case RADIXCHAR:
69c5e0db
KW
2685
2686 /* For this, we output a known simple floating point number to
2687 * a buffer, and parse it, looking for the radix */
2688
2689 if (toggle) {
2690 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2691 }
2692
2693 if (PL_langinfo_bufsize < 10) {
2694 PL_langinfo_bufsize = 10;
2695 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2696 }
2697
2698 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2699 "%.1f", 1.5);
2700 if (needed_size >= (int) PL_langinfo_bufsize) {
2701 PL_langinfo_bufsize = needed_size + 1;
2702 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2703 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2704 "%.1f", 1.5);
2705 assert(needed_size < (int) PL_langinfo_bufsize);
2706 }
2707
2708 ptr = PL_langinfo_buf;
2709 e = PL_langinfo_buf + PL_langinfo_bufsize;
2710
2711 /* Find the '1' */
2712 while (ptr < e && *ptr != '1') {
2713 ptr++;
2714 }
2715 ptr++;
2716
2717 /* Find the '5' */
2718 item_start = ptr;
2719 while (ptr < e && *ptr != '5') {
2720 ptr++;
2721 }
2722
2723 /* Everything in between is the radix string */
2724 if (ptr >= e) {
2725 PL_langinfo_buf[0] = '?';
2726 PL_langinfo_buf[1] = '\0';
2727 }
2728 else {
2729 *ptr = '\0';
2730 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2731 }
2732
2733 if (toggle) {
2734 RESTORE_LC_NUMERIC();
2735 }
2736
2737 retval = PL_langinfo_buf;
2738 break;
2739
2740# else
2741
2742 case RADIXCHAR: /* No special handling needed */
2743
2744# endif
2745
4e6826bf 2746 case THOUSEP:
f7416781 2747
43dd6b15 2748 if (toggle) {
038d3702 2749 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2750 }
f7416781 2751
69c5e0db
KW
2752 LOCALE_LOCK_V; /* Prevent interference with other threads
2753 using localeconv() */
2754
2755# ifdef TS_W32_BROKEN_LOCALECONV
2756
2757 /* This should only be for the thousands separator. A
2758 * different work around would be to use GetNumberFormat on a
2759 * known value and parse the result to find the separator */
2760 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2761 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2762 save_global = savepv(my_setlocale(LC_ALL, NULL));
2763 my_setlocale(LC_ALL, save_thread);
2764# if 0
2765 /* This is the start of code that for broken Windows replaces
2766 * the above and below code, and instead calls
2767 * GetNumberFormat() and then would parse that to find the
2768 * thousands separator. It needs to handle UTF-16 vs -8
2769 * issues. */
2770
2771 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2772 DEBUG_L(PerlIO_printf(Perl_debug_log,
2773 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2774 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2775
2776# endif
2777# endif
5acc4454 2778
43dd6b15
KW
2779 lc = localeconv();
2780 if (! lc) {
628ff75a 2781 temp = "";
33394adc 2782 }
43dd6b15 2783 else {
4e6826bf 2784 temp = (item == RADIXCHAR)
43dd6b15
KW
2785 ? lc->decimal_point
2786 : lc->thousands_sep;
628ff75a
KW
2787 if (! temp) {
2788 temp = "";
43dd6b15
KW
2789 }
2790 }
f7416781 2791
0b6697c5
KW
2792 retval = save_to_buffer(temp, &PL_langinfo_buf,
2793 &PL_langinfo_bufsize, 0);
f7416781 2794
69c5e0db
KW
2795# ifdef TS_W32_BROKEN_LOCALECONV
2796
2797 my_setlocale(LC_ALL, save_global);
2798 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2799 my_setlocale(LC_ALL, save_thread);
2800 Safefree(save_global);
2801 Safefree(save_thread);
2802
2803# endif
2804
2805 LOCALE_UNLOCK_V;
5acc4454 2806
43dd6b15 2807 if (toggle) {
038d3702 2808 RESTORE_LC_NUMERIC();
43dd6b15 2809 }
f7416781 2810
43dd6b15 2811 break;
f7416781
KW
2812
2813# endif
2814# ifdef HAS_STRFTIME
2815
43dd6b15
KW
2816 /* These are defined by C89, so we assume that strftime supports
2817 * them, and so are returned unconditionally; they may not be what
2818 * the locale actually says, but should give good enough results
2819 * for someone using them as formats (as opposed to trying to parse
2820 * them to figure out what the locale says). The other format
2821 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2822 case D_FMT: return "%x";
2823 case T_FMT: return "%X";
2824 case D_T_FMT: return "%c";
43dd6b15
KW
2825
2826 /* These formats are only available in later strfmtime's */
4e6826bf 2827 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2828
2829 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2830 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2831 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2832 case ALT_DIGITS:
2833 case AM_STR: case PM_STR:
2834 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2835 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2836 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2837 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2838 case DAY_5: case DAY_6: case DAY_7:
2839 case MON_1: case MON_2: case MON_3: case MON_4:
2840 case MON_5: case MON_6: case MON_7: case MON_8:
2841 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2842
2843 LOCALE_LOCK;
2844
2845 init_tm(&tm); /* Precaution against core dumps */
2846 tm.tm_sec = 30;
2847 tm.tm_min = 30;
2848 tm.tm_hour = 6;
2849 tm.tm_year = 2017 - 1900;
2850 tm.tm_wday = 0;
2851 tm.tm_mon = 0;
2852 switch (item) {
2853 default:
2854 LOCALE_UNLOCK;
2855 Perl_croak(aTHX_
2856 "panic: %s: %d: switch case: %d problem",
2857 __FILE__, __LINE__, item);
2858 NOT_REACHED; /* NOTREACHED */
2859
4e6826bf
KW
2860 case PM_STR: tm.tm_hour = 18;
2861 case AM_STR:
43dd6b15
KW
2862 format = "%p";
2863 break;
2864
4e6826bf
KW
2865 case ABDAY_7: tm.tm_wday++;
2866 case ABDAY_6: tm.tm_wday++;
2867 case ABDAY_5: tm.tm_wday++;
2868 case ABDAY_4: tm.tm_wday++;
2869 case ABDAY_3: tm.tm_wday++;
2870 case ABDAY_2: tm.tm_wday++;
2871 case ABDAY_1:
43dd6b15
KW
2872 format = "%a";
2873 break;
2874
4e6826bf
KW
2875 case DAY_7: tm.tm_wday++;
2876 case DAY_6: tm.tm_wday++;
2877 case DAY_5: tm.tm_wday++;
2878 case DAY_4: tm.tm_wday++;
2879 case DAY_3: tm.tm_wday++;
2880 case DAY_2: tm.tm_wday++;
2881 case DAY_1:
43dd6b15
KW
2882 format = "%A";
2883 break;
2884
4e6826bf
KW
2885 case ABMON_12: tm.tm_mon++;
2886 case ABMON_11: tm.tm_mon++;
2887 case ABMON_10: tm.tm_mon++;
2888 case ABMON_9: tm.tm_mon++;
2889 case ABMON_8: tm.tm_mon++;
2890 case ABMON_7: tm.tm_mon++;
2891 case ABMON_6: tm.tm_mon++;
2892 case ABMON_5: tm.tm_mon++;
2893 case ABMON_4: tm.tm_mon++;
2894 case ABMON_3: tm.tm_mon++;
2895 case ABMON_2: tm.tm_mon++;
2896 case ABMON_1:
43dd6b15
KW
2897 format = "%b";
2898 break;
2899
4e6826bf
KW
2900 case MON_12: tm.tm_mon++;
2901 case MON_11: tm.tm_mon++;
2902 case MON_10: tm.tm_mon++;
2903 case MON_9: tm.tm_mon++;
2904 case MON_8: tm.tm_mon++;
2905 case MON_7: tm.tm_mon++;
2906 case MON_6: tm.tm_mon++;
2907 case MON_5: tm.tm_mon++;
2908 case MON_4: tm.tm_mon++;
2909 case MON_3: tm.tm_mon++;
2910 case MON_2: tm.tm_mon++;
2911 case MON_1:
43dd6b15
KW
2912 format = "%B";
2913 break;
2914
4e6826bf 2915 case T_FMT_AMPM:
43dd6b15
KW
2916 format = "%r";
2917 return_format = TRUE;
2918 break;
2919
4e6826bf 2920 case ERA_D_FMT:
43dd6b15
KW
2921 format = "%Ex";
2922 return_format = TRUE;
2923 break;
2924
4e6826bf 2925 case ERA_T_FMT:
43dd6b15
KW
2926 format = "%EX";
2927 return_format = TRUE;
2928 break;
2929
4e6826bf 2930 case ERA_D_T_FMT:
43dd6b15
KW
2931 format = "%Ec";
2932 return_format = TRUE;
2933 break;
2934
4e6826bf 2935 case ALT_DIGITS:
43dd6b15
KW
2936 tm.tm_wday = 0;
2937 format = "%Ow"; /* Find the alternate digit for 0 */
2938 break;
2939 }
f7416781 2940
43dd6b15
KW
2941 /* We can't use my_strftime() because it doesn't look at
2942 * tm_wday */
2943 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2944 format, &tm))
2945 {
2946 /* A zero return means one of:
2947 * a) there wasn't enough space in PL_langinfo_buf
2948 * b) the format, like a plain %p, returns empty
2949 * c) it was an illegal format, though some
2950 * implementations of strftime will just return the
2951 * illegal format as a plain character sequence.
2952 *
2953 * To quickly test for case 'b)', try again but precede
2954 * the format with a plain character. If that result is
2955 * still empty, the problem is either 'a)' or 'c)' */
2956
2957 Size_t format_size = strlen(format) + 1;
2958 Size_t mod_size = format_size + 1;
2959 char * mod_format;
2960 char * temp_result;
2961
2962 Newx(mod_format, mod_size, char);
2963 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2964 *mod_format = ' ';
43dd6b15
KW
2965 my_strlcpy(mod_format + 1, format, mod_size);
2966 len = strftime(temp_result,
2967 PL_langinfo_bufsize,
2968 mod_format, &tm);
2969 Safefree(mod_format);
2970 Safefree(temp_result);
2971
2972 /* If 'len' is non-zero, it means that we had a case like
2973 * %p which means the current locale doesn't use a.m. or
2974 * p.m., and that is valid */
2975 if (len == 0) {
2976
2977 /* Here, still didn't work. If we get well beyond a
2978 * reasonable size, bail out to prevent an infinite
2979 * loop. */
2980
2981 if (PL_langinfo_bufsize > 100 * format_size) {
2982 *PL_langinfo_buf = '\0';
2983 }
2984 else {
2985 /* Double the buffer size to retry; Add 1 in case
2986 * original was 0, so we aren't stuck at 0. */
2987 PL_langinfo_bufsize *= 2;
2988 PL_langinfo_bufsize++;
2989 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2990 continue;
2991 }
2992 }
f7416781 2993
f7416781 2994 break;
43dd6b15 2995 }
f7416781 2996
43dd6b15
KW
2997 /* Here, we got a result.
2998 *
2999 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3000 * alternate format for wday 0. If the value is the same as
3001 * the normal 0, there isn't an alternate, so clear the buffer.
3002 * */
4e6826bf 3003 if ( item == ALT_DIGITS
43dd6b15
KW
3004 && strEQ(PL_langinfo_buf, "0"))
3005 {
3006 *PL_langinfo_buf = '\0';
3007 }
f7416781 3008
43dd6b15
KW
3009 /* ALT_DIGITS is problematic. Experiments on it showed that
3010 * strftime() did not always work properly when going from
3011 * alt-9 to alt-10. Only a few locales have this item defined,
3012 * and in all of them on Linux that khw was able to find,
3013 * nl_langinfo() merely returned the alt-0 character, possibly
3014 * doubled. Most Unicode digits are in blocks of 10
3015 * consecutive code points, so that is sufficient information
3016 * for those scripts, as we can infer alt-1, alt-2, .... But
3017 * for a Japanese locale, a CJK ideographic 0 is returned, and
3018 * the CJK digits are not in code point order, so you can't
3019 * really infer anything. The localedef for this locale did
3020 * specify the succeeding digits, so that strftime() works
3021 * properly on them, without needing to infer anything. But
3022 * the nl_langinfo() return did not give sufficient information
3023 * for the caller to understand what's going on. So until
3024 * there is evidence that it should work differently, this
3025 * returns the alt-0 string for ALT_DIGITS.
3026 *
3027 * wday was chosen because its range is all a single digit.
3028 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3029
43dd6b15 3030 LOCALE_UNLOCK;
f7416781 3031
d1b150d9
KW
3032 retval = PL_langinfo_buf;
3033
43dd6b15
KW
3034 /* If to return the format, not the value, overwrite the buffer
3035 * with it. But some strftime()s will keep the original format
3036 * if illegal, so change those to "" */
3037 if (return_format) {
3038 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3039 *PL_langinfo_buf = '\0';
3040 }
43dd6b15 3041 else {
0b6697c5
KW
3042 retval = save_to_buffer(format, &PL_langinfo_buf,
3043 &PL_langinfo_bufsize, 0);
f7416781
KW
3044 }
3045 }
3046
3047 break;
f7416781
KW
3048
3049# endif
3050
43dd6b15 3051 }
f7416781
KW
3052 }
3053
0b6697c5 3054 return retval;
f7416781
KW
3055
3056#endif
3057
a4f00dcc 3058}
b385bb4d 3059
98994639
HS
3060/*
3061 * Initialize locale awareness.
3062 */
3063int
3064Perl_init_i18nl10n(pTHX_ int printwarn)
3065{
0e92a118
KW
3066 /* printwarn is
3067 *
3068 * 0 if not to output warning when setup locale is bad
3069 * 1 if to output warning based on value of PERL_BADLANG
3070 * >1 if to output regardless of PERL_BADLANG
3071 *
3072 * returns
98994639 3073 * 1 = set ok or not applicable,
0e92a118
KW
3074 * 0 = fallback to a locale of lower priority
3075 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3076 *
3077 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3078 * set, debugging information is output.
3079 *
3080 * This looks more complicated than it is, mainly due to the #ifdefs.
3081 *
3082 * We try to set LC_ALL to the value determined by the environment. If
3083 * there is no LC_ALL on this platform, we try the individual categories we
3084 * know about. If this works, we are done.
3085 *
3086 * But if it doesn't work, we have to do something else. We search the
3087 * environment variables ourselves instead of relying on the system to do
3088 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3089 * think there is one), and the ultimate fallback "C". This is all done in
3090 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3091 * more complex. The 'trial_locales' array is initialized with just one
3092 * element; it causes the behavior described in the paragraph above this to
3093 * happen. If that fails, we add elements to 'trial_locales', and do extra
3094 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3095 *
3096 * On Ultrix, the locale MUST come from the environment, so there is
3097 * preliminary code to set it. I (khw) am not sure that it is necessary,
3098 * and that this couldn't be folded into the loop, but barring any real
3099 * platforms to test on, it's staying as-is
3100 *
3101 * A slight complication is that in embedded Perls, the locale may already
3102 * be set-up, and we don't want to get it from the normal environment
3103 * variables. This is handled by having a special environment variable
3104 * indicate we're in this situation. We simply set setlocale's 2nd
3105 * parameter to be a NULL instead of "". That indicates to setlocale that
3106 * it is not to change anything, but to return the current value,
3107 * effectively initializing perl's db to what the locale already is.
3108 *
3109 * We play the same trick with NULL if a LC_ALL succeeds. We call
3110 * setlocale() on the individual categores with NULL to get their existing
3111 * values for our db, instead of trying to change them.
3112 * */
98994639 3113
0e92a118
KW
3114 int ok = 1;
3115
7d4bcc4a
KW
3116#ifndef USE_LOCALE
3117
3118 PERL_UNUSED_ARG(printwarn);
3119
3120#else /* USE_LOCALE */
7d4bcc4a
KW
3121# ifdef __GLIBC__
3122
175c4cf9 3123 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3124
3125# endif
65ebb059 3126
ccd65d51
KW
3127 /* NULL uses the existing already set up locale */
3128 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3129 ? NULL
3130 : "";
c3fcd832
KW
3131 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3132 unsigned int trial_locales_count;
175c4cf9
KW
3133 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3134 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3135 bool setlocale_failure = FALSE;
65ebb059 3136 unsigned int i;
175c4cf9
KW
3137
3138 /* A later getenv() could zap this, so only use here */
3139 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3140
3141 const bool locwarn = (printwarn > 1
e5f10d49
KW
3142 || ( printwarn
3143 && ( ! bad_lang_use_once
22ff3130 3144 || (
e5f10d49
KW
3145 /* disallow with "" or "0" */
3146 *bad_lang_use_once
3147 && strNE("0", bad_lang_use_once)))));
ea92aad8 3148
291a84fb 3149 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3150 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3151
3152 /* current locale for given category; should have been copied so aren't
3153 * volatile */
8de4332b 3154 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3155
7d4bcc4a
KW
3156# ifdef WIN32
3157
6bce99ee
JH
3158 /* In some systems you can find out the system default locale
3159 * and use that as the fallback locale. */
7d4bcc4a
KW
3160# define SYSTEM_DEFAULT_LOCALE
3161# endif
3162# ifdef SYSTEM_DEFAULT_LOCALE
3163
65ebb059 3164 const char *system_default_locale = NULL;
98994639 3165
7d4bcc4a 3166# endif
948523db
KW
3167
3168# ifndef DEBUGGING
3169# define DEBUG_LOCALE_INIT(a,b,c)
3170# else
7d4bcc4a 3171
8298454c 3172 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3173
3174# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3175 STMT_START { \
3176 if (debug_initialization) { \
3177 PerlIO_printf(Perl_debug_log, \
3178 "%s:%d: %s\n", \
3179 __FILE__, __LINE__, \
a4f00dcc 3180 setlocale_debug_string(category, \
2fcc0ca9
KW
3181 locale, \
3182 result)); \
3183 } \
3184 } STMT_END
2fcc0ca9 3185
948523db
KW
3186/* Make sure the parallel arrays are properly set up */
3187# ifdef USE_LOCALE_NUMERIC
3188 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3189 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3190# ifdef USE_POSIX_2008_LOCALE
3191 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3192# endif
948523db
KW
3193# endif
3194# ifdef USE_LOCALE_CTYPE
3195 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3196 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3197# ifdef USE_POSIX_2008_LOCALE
3198 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3199# endif
948523db
KW
3200# endif
3201# ifdef USE_LOCALE_COLLATE
3202 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3203 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3204# ifdef USE_POSIX_2008_LOCALE
3205 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3206# endif
948523db
KW
3207# endif
3208# ifdef USE_LOCALE_TIME
3209 assert(categories[LC_TIME_INDEX] == LC_TIME);
3210 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3211# ifdef USE_POSIX_2008_LOCALE
3212 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3213# endif
948523db
KW
3214# endif
3215# ifdef USE_LOCALE_MESSAGES
3216 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3217 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3218# ifdef USE_POSIX_2008_LOCALE
3219 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3220# endif
948523db
KW
3221# endif
3222# ifdef USE_LOCALE_MONETARY
3223 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3224 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3225# ifdef USE_POSIX_2008_LOCALE
3226 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3227# endif
948523db 3228# endif
9821811f
KW
3229# ifdef USE_LOCALE_ADDRESS
3230 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3231 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3232# ifdef USE_POSIX_2008_LOCALE
3233 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3234# endif
9821811f
KW
3235# endif
3236# ifdef USE_LOCALE_IDENTIFICATION
3237 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3238 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3239# ifdef USE_POSIX_2008_LOCALE
3240 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3241# endif
9821811f
KW
3242# endif
3243# ifdef USE_LOCALE_MEASUREMENT
3244 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3245 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3246# ifdef USE_POSIX_2008_LOCALE
3247 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3248# endif
9821811f
KW
3249# endif
3250# ifdef USE_LOCALE_PAPER
3251 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3252 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3253# ifdef USE_POSIX_2008_LOCALE
3254 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3255# endif
9821811f
KW
3256# endif
3257# ifdef USE_LOCALE_TELEPHONE
3258 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3259 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3260# ifdef USE_POSIX_2008_LOCALE
3261 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3262# endif
9821811f 3263# endif
948523db
KW
3264# ifdef LC_ALL
3265 assert(categories[LC_ALL_INDEX] == LC_ALL);
3266 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3267 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3268# ifdef USE_POSIX_2008_LOCALE
3269 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3270# endif
948523db
KW
3271# endif
3272# endif /* DEBUGGING */
47280b20
KW
3273
3274 /* Initialize the cache of the program's UTF-8ness for the always known
3275 * locales C and POSIX */
3276 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3277 sizeof(PL_locale_utf8ness));
3278
e9bc6d6b
KW
3279# ifdef USE_THREAD_SAFE_LOCALE
3280# ifdef WIN32
3281
3282 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3283
3284# endif
3285# endif
39e69e77 3286# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3287
3288 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3289 if (! PL_C_locale_obj) {
3290 Perl_croak_nocontext(
3291 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3292 }
3293 if (DEBUG_Lv_TEST || debug_initialization) {
3294 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3295 }
3296
3297# endif
3298
3ca88433
KW
3299 PL_numeric_radix_sv = newSVpvs(".");
3300
e9bc6d6b
KW
3301# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3302
3303 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3304 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3305
3306# endif
b56c4436 3307# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3308
3309 /*
3310 * Ultrix setlocale(..., "") fails if there are no environment
3311 * variables from which to get a locale name.
3312 */
3313
b56c4436
KW
3314# ifndef LC_ALL
3315# error Ultrix without LC_ALL not implemented
3316# else
7d4bcc4a 3317
b56c4436
KW
3318 {
3319 bool done = FALSE;
ea92aad8
KW
3320 if (lang) {
3321 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3322 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3323 if (sl_result[LC_ALL_INDEX])
3324 done = TRUE;
3325 else
e5f10d49 3326 setlocale_failure = TRUE;
ea92aad8
KW
3327 }
3328 if (! setlocale_failure) {
3329 const char * locale_param;
3330 for (i = 0; i < LC_ALL_INDEX; i++) {
3331 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3332 ? setlocale_init
3333 : NULL;
3334 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3335 if (! sl_result[i]) {
3336 setlocale_failure = TRUE;
3337 }
3338 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3339 }
c835d6be 3340 }
7d4bcc4a
KW
3341 }
3342
3343# endif /* LC_ALL */
e5f10d49 3344# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3345
65ebb059 3346 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3347 * the list. Normally the loop is executed just once. But if setting the
3348 * locale fails, inside the loop we add fallback trials to the array and so
3349 * will execute the loop multiple times */
c3fcd832
KW
3350 trial_locales[0] = setlocale_init;
3351 trial_locales_count = 1;
7d4bcc4a 3352
65ebb059
KW
3353 for (i= 0; i < trial_locales_count; i++) {
3354 const char * trial_locale = trial_locales[i];
3355
3356 if (i > 0) {
3357
3358 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3359 * when i==0, but I (khw) don't think that behavior makes much
3360 * sense */
3361 setlocale_failure = FALSE;
3362
7d4bcc4a 3363# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3364# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3365
65ebb059
KW
3366 /* On Windows machines, an entry of "" after the 0th means to use
3367 * the system default locale, which we now proceed to get. */
3368 if (strEQ(trial_locale, "")) {
3369 unsigned int j;
3370
3371 /* Note that this may change the locale, but we are going to do
3372 * that anyway just below */
837ce802 3373 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3374 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3375
7d4bcc4a 3376 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3377 * try */
3378 if (! system_default_locale) {
3379 goto next_iteration;
3380 }
3381 for (j = 0; j < trial_locales_count; j++) {
3382 if (strEQ(system_default_locale, trial_locales[j])) {
3383 goto next_iteration;
3384 }
3385 }
3386
3387 trial_locale = system_default_locale;
3388 }
ec0202b5
KW
3389# else
3390# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3391# endif
7d4bcc4a 3392# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3393
3394 } /* For i > 0 */
65ebb059 3395
7d4bcc4a
KW
3396# ifdef LC_ALL
3397
948523db
KW
3398 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3399 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3400 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3401 setlocale_failure = TRUE;
7cd8b568
KW
3402 }
3403 else {
3404 /* Since LC_ALL succeeded, it should have changed all the other
3405 * categories it can to its value; so we massage things so that the
3406 * setlocales below just return their category's current values.
3407 * This adequately handles the case in NetBSD where LC_COLLATE may
3408 * not be defined for a locale, and setting it individually will
7d4bcc4a 3409 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3410 * the POSIX locale. */
3411 trial_locale = NULL;
3412 }
7d4bcc4a
KW
3413
3414# endif /* LC_ALL */
98994639 3415
e5f10d49
KW
3416 if (! setlocale_failure) {
3417 unsigned int j;
3418 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3419 curlocales[j]
3420 = savepv(do_setlocale_r(categories[j], trial_locale));
3421 if (! curlocales[j]) {
3422 setlocale_failure = TRUE;
3423 }
3424 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3425 }
c835d6be 3426
e5f10d49
KW
3427 if (! setlocale_failure) { /* All succeeded */
3428 break; /* Exit trial_locales loop */
49c85077 3429 }
65ebb059 3430 }
98994639 3431
49c85077
KW
3432 /* Here, something failed; will need to try a fallback. */
3433 ok = 0;
65ebb059 3434
49c85077
KW
3435 if (i == 0) {
3436 unsigned int j;
98994639 3437
65ebb059 3438 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3439
3440# ifdef LC_ALL
98994639 3441
49c85077
KW
3442 PerlIO_printf(Perl_error_log,
3443 "perl: warning: Setting locale failed.\n");
98994639 3444
7d4bcc4a 3445# else /* !LC_ALL */
98994639 3446
49c85077
KW
3447 PerlIO_printf(Perl_error_log,
3448 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3449
e5f10d49
KW
3450 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3451 if (! curlocales[j]) {
3452 PerlIO_printf(Perl_error_log, category_names[j]);
3453 }
3454 else {
3455 Safefree(curlocales[j]);
3456 }
3457 }
7d4bcc4a 3458
7d4bcc4a 3459# endif /* LC_ALL */
98994639 3460
49c85077
KW
3461 PerlIO_printf(Perl_error_log,
3462 "perl: warning: Please check that your locale settings:\n");
98994639 3463
7d4bcc4a
KW
3464# ifdef __GLIBC__
3465
49c85077
KW
3466 PerlIO_printf(Perl_error_log,
3467 "\tLANGUAGE = %c%s%c,\n",
3468 language ? '"' : '(',
3469 language ? language : "unset",
3470 language ? '"' : ')');
7d4bcc4a 3471# endif
98994639 3472
49c85077
KW
3473 PerlIO_printf(Perl_error_log,
3474 "\tLC_ALL = %c%s%c,\n",
3475 lc_all ? '"' : '(',
3476 lc_all ? lc_all : "unset",
3477 lc_all ? '"' : ')');
98994639 3478
7d4bcc4a
KW
3479# if defined(USE_ENVIRON_ARRAY)
3480
49c85077 3481 {
cd999af9 3482 char **e;
d5e32b93
KW
3483
3484 /* Look through the environment for any variables of the
3485 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3486 * already handled above. These are assumed to be locale
3487 * settings. Output them and their values. */
cd999af9 3488 for (e = environ; *e; e++) {
d5e32b93
KW
3489 const STRLEN prefix_len = sizeof("LC_") - 1;
3490 STRLEN uppers_len;
3491
cd999af9 3492 if ( strBEGINs(*e, "LC_")
c8b388b0 3493 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3494 && (uppers_len = strspn(*e + prefix_len,
3495 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3496 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3497 {
3498 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3499 (int) (prefix_len + uppers_len), *e,
3500 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3501 }
3502 }
49c85077 3503 }
7d4bcc4a
KW
3504
3505# else
3506
49c85077
KW
3507 PerlIO_printf(Perl_error_log,
3508 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3509
3510# endif
98994639 3511
49c85077
KW
3512 PerlIO_printf(Perl_error_log,
3513 "\tLANG = %c%s%c\n",
3514 lang ? '"' : '(',
3515 lang ? lang : "unset",
3516 lang ? '"' : ')');
98994639 3517
49c85077
KW
3518 PerlIO_printf(Perl_error_log,
3519 " are supported and installed on your system.\n");
3520 }
98994639 3521
65ebb059 3522 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3523 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3524 * usually change the upper bound of the loop we are in.
3525 *
3526 * Since the system's default way of setting the locale has not
3527 * found one that works, We use Perl's defined ordering: LC_ALL,
3528 * LANG, and the C locale. We don't try the same locale twice, so
3529 * don't add to the list if already there. (On POSIX systems, the
3530 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3531 * but there's no harm done by doing it explicitly.
3532 *
3533 * Note that this tries the LC_ALL environment variable even on
3534 * systems which have no LC_ALL locale setting. This may or may
3535 * not have been originally intentional, but there's no real need
3536 * to change the behavior. */
65ebb059
KW
3537 if (lc_all) {
3538 for (j = 0; j < trial_locales_count; j++) {
3539 if (strEQ(lc_all, trial_locales[j])) {
3540 goto done_lc_all;
3541 }
3542 }
3543 trial_locales[trial_locales_count++] = lc_all;
3544 }
3545 done_lc_all:
98994639 3546
65ebb059
KW
3547 if (lang) {
3548 for (j = 0; j < trial_locales_count; j++) {
3549 if (strEQ(lang, trial_locales[j])) {
3550 goto done_lang;
3551 }
3552 }
3553 trial_locales[trial_locales_count++] = lang;
3554 }
3555 done_lang:
3556
7d4bcc4a
KW
3557# if defined(WIN32) && defined(LC_ALL)
3558
65ebb059
KW
3559 /* For Windows, we also try the system default locale before "C".
3560 * (If there exists a Windows without LC_ALL we skip this because
3561 * it gets too complicated. For those, the "C" is the next
3562 * fallback possibility). The "" is the same as the 0th element of
3563 * the array, but the code at the loop above knows to treat it
3564 * differently when not the 0th */
3565 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3566
3567# endif
65ebb059
KW
3568
3569 for (j = 0; j < trial_locales_count; j++) {
3570 if (strEQ("C", trial_locales[j])) {
3571 goto done_C;
3572 }
3573 }
3574 trial_locales[trial_locales_count++] = "C";
98994639 3575
65ebb059
KW
3576 done_C: ;
3577 } /* end of first time through the loop */
98994639 3578
7d4bcc4a
KW
3579# ifdef WIN32
3580
65ebb059 3581 next_iteration: ;
7d4bcc4a
KW
3582
3583# endif
65ebb059
KW
3584
3585 } /* end of looping through the trial locales */
3586
3587 if (ok < 1) { /* If we tried to fallback */
3588 const char* msg;
3589 if (! setlocale_failure) { /* fallback succeeded */
3590 msg = "Falling back to";
3591 }
3592 else { /* fallback failed */
e5f10d49 3593 unsigned int j;
98994639 3594
65ebb059
KW
3595 /* We dropped off the end of the loop, so have to decrement i to
3596 * get back to the value the last time through */
3597 i--;
98994639 3598
65ebb059
KW
3599 ok = -1;
3600 msg = "Failed to fall back to";
3601
3602 /* To continue, we should use whatever values we've got */
7d4bcc4a 3603
e5f10d49
KW
3604 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3605 Safefree(curlocales[j]);
3606 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3607 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3608 }
65ebb059
KW
3609 }
3610
3611 if (locwarn) {
3612 const char * description;
3613 const char * name = "";
3614 if (strEQ(trial_locales[i], "C")) {
3615 description = "the standard locale";
3616 name = "C";
3617 }
7d4bcc4a
KW
3618
3619# ifdef SYSTEM_DEFAULT_LOCALE
3620
65ebb059
KW
3621 else if (strEQ(trial_locales[i], "")) {
3622 description = "the system default locale";
3623 if (system_default_locale) {
3624 name = system_default_locale;
3625 }
3626 }
7d4bcc4a
KW
3627
3628# endif /* SYSTEM_DEFAULT_LOCALE */
3629
65ebb059
KW
3630 else {
3631 description = "a fallback locale";
3632 name = trial_locales[i];
3633 }
3634 if (name && strNE(name, "")) {
3635 PerlIO_printf(Perl_error_log,
3636 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3637 }
3638 else {
3639 PerlIO_printf(Perl_error_log,
3640 "perl: warning: %s %s.\n", msg, description);
3641 }
3642 }
3643 } /* End of tried to fallback */
98994639 3644
e5f10d49
KW
3645 /* Done with finding the locales; update our records */
3646
7d4bcc4a
KW
3647# ifdef USE_LOCALE_CTYPE
3648
948523db 3649 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3650
e5f10d49 3651# endif
7d4bcc4a
KW
3652# ifdef USE_LOCALE_COLLATE
3653
948523db 3654 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3655
e5f10d49 3656# endif
7d4bcc4a
KW
3657# ifdef USE_LOCALE_NUMERIC
3658
948523db 3659 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3660
3661# endif
3662
948523db 3663 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3664
e9bc6d6b 3665# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3666
3667 /* This caches whether each category's locale is UTF-8 or not. This
3668 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3669 * initialization time before any threads have started, but not later
3670 * unless thread-safe operations are used.
47280b20
KW
3671 * Caching means that if the program heeds our dictate not to change
3672 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3673 * it may get queried without having to change locales. If the
3674 * environment is such that all categories have the same locale, this
3675 * isn't needed, as the code will not change the locale; but this
3676 * handles the uncommon case where the environment has disparate
3677 * locales for the categories */
47280b20
KW
3678 (void) _is_cur_LC_category_utf8(categories[i]);
3679
3680# endif
3681
e5f10d49
KW
3682 Safefree(curlocales[i]);
3683 }
b310b053 3684
7d4bcc4a
KW
3685# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3686
49c85077 3687 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3688 * locale is UTF-8. The call to new_ctype() just above has already
3689 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3690 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3691 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3692 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3693 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3694
a05d7ebb 3695 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3696 This is an alternative to using the -C command line switch
3697 (the -C if present will override this). */
3698 {
dd374669 3699 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3700 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3701 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3702 PL_utf8cache = -1;
b310b053
JH
3703 }
3704
7d4bcc4a 3705# endif
7d4bcc4a
KW
3706# ifdef __GLIBC__
3707
175c4cf9 3708 Safefree(language);
7d4bcc4a
KW
3709
3710# endif
175c4cf9
KW
3711
3712 Safefree(lc_all);
3713 Safefree(lang);
3714
e3305790 3715#endif /* USE_LOCALE */
2fcc0ca9 3716#ifdef DEBUGGING
7d4bcc4a 3717
2fcc0ca9 3718 /* So won't continue to output stuff */
27cdc72e 3719 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3720
2fcc0ca9
KW
3721#endif
3722
98994639
HS
3723 return ok;
3724}
3725
98994639
HS
3726#ifdef USE_LOCALE_COLLATE
3727
a4a439fb 3728char *
a4a439fb
KW
3729Perl__mem_collxfrm(pTHX_ const char *input_string,
3730 STRLEN len, /* Length of 'input_string' */
3731 STRLEN *xlen, /* Set to length of returned string
3732 (not including the collation index
3733 prefix) */
3734 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3735 )
98994639 3736{
a4a439fb
KW
3737
3738 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3739 * differences. First, it handles embedded NULs. Second, it allocates a bit
3740 * more memory than needed for the transformed data itself. The real
55e5378d 3741 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3742 * the length of that, and doesn't include the collation index size.
3743 * Please see sv_collxfrm() to see how this is used. */
3744
55e5378d
KW
3745#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3746
6696cfa7
KW
3747 char * s = (char *) input_string;
3748 STRLEN s_strlen = strlen(input_string);
79f120c8 3749 char *xbuf = NULL;
55e5378d 3750 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3751 STRLEN length_in_chars;
c664130f 3752 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3753
a4a439fb
KW
3754 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3755
3756 /* Must be NUL-terminated */
3757 assert(*(input_string + len) == '\0');
7918f24d 3758
79f120c8
KW
3759 /* If this locale has defective collation, skip */
3760 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3761 DEBUG_L(PerlIO_printf(Perl_debug_log,
3762 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3763 goto bad;
3764 }
3765
6696cfa7
KW
3766 /* Replace any embedded NULs with the control that sorts before any others.
3767 * This will give as good as possible results on strings that don't
3768 * otherwise contain that character, but otherwise there may be
3769 * less-than-perfect results with that character and NUL. This is
fdc080f3 3770 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3771 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3772 NUL */
6696cfa7
KW
3773 char * e = s + len;
3774 char * sans_nuls;
fdc080f3 3775 STRLEN sans_nuls_len;
94762aa0 3776 int try_non_controls;
afc4976f
KW
3777 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3778 making sure 2nd byte is NUL.
3779 */
3780 STRLEN this_replacement_len;
3781
1e4c9676
KW
3782 /* If we don't know what non-NUL control character sorts lowest for
3783 * this locale, find it */
f28f4d2a 3784 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3785 int j;
afc4976f 3786 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3787 includes the collation index
3788 prefixed. */
3789
91c0e2e0 3790 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3791
3792 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3793 * NUL, in which case we instead look for any character. Controls
3794 * are preferred because collation order is, in general, context
3795 * sensitive, with adjoining characters affecting the order, and
3796 * controls are less likely to have such interactions, allowing the
3797 * NUL-replacement to stand on its own. (Another way to look at it
3798 * is to imagine what would happen if the NUL were replaced by a
3799 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3800 for (try_non_controls = 0;
3801 try_non_controls < 2;
3802 try_non_controls++)
3803 {
d4ff9586
KW
3804 /* Look through all legal code points (NUL isn't) */
3805 for (j = 1; j < 256; j++) {
3806 char * x; /* j's xfrm plus collation index */
3807 STRLEN x_len; /* length of 'x' */
3808 STRLEN trial_len = 1;
736a4fed 3809 char cur_source[] = { '\0', '\0' };
d4ff9586 3810
736a4fed
KW
3811 /* Skip non-controls the first time through the loop. The
3812 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3813 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3814 ? ! isCNTRL_L1(j)
3815 : ! isCNTRL_LC(j))
3816 {
d4ff9586 3817 continue;
6696cfa7 3818 }
6696cfa7 3819
736a4fed
KW
3820 /* Create a 1-char string of the current code point */
3821 cur_source[0] = (char) j;
3822
d4ff9586
KW
3823 /* Then transform it */
3824 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3825 0 /* The string is not in UTF-8 */);
6696cfa7 3826
1e4c9676 3827 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3828 * */
3829 if (! x) {
3830 continue;
3831 }
6696cfa7 3832
d4ff9586
KW
3833 /* If this character's transformation is lower than
3834 * the current lowest, this one becomes the lowest */
3835 if ( cur_min_x == NULL
3836 || strLT(x + COLLXFRM_HDR_LEN,
3837 cur_min_x + COLLXFRM_HDR_LEN))
3838 {
f28f4d2a 3839 PL_strxfrm_NUL_replacement = j;
d4ff9586 3840 cur_min_x = x;
d4ff9586
KW
3841 }
3842 else {
3843 Safefree(x);
3844 }
1e4c9676 3845 } /* end of loop through all 255 characters */
6696cfa7 3846
1e4c9676 3847 /* Stop looking if found */
94762aa0
KW
3848 if (cur_min_x) {
3849 break;
3850 }
3851
3852 /* Unlikely, but possible, if there aren't any controls that
3853 * work in the locale, repeat the loop, looking for any
3854 * character that works */
3855 DEBUG_L(PerlIO_printf(Perl_debug_log,
3856 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3857 } /* End of loop to try first the controls, then any char */
6696cfa7 3858