This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
s/safefree()/Safefree() in a few places
[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
5b64f24c
KW
56#ifdef I_WCTYPE
57# include <wctype.h>
58#endif
0dec74cd 59
2fcc0ca9
KW
60/* If the environment says to, we can output debugging information during
61 * initialization. This is done before option parsing, and before any thread
62 * creation, so can be a file-level static */
5a4b0634
KW
63#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
64# define debug_initialization 0
65# define DEBUG_INITIALIZATION_set(v)
66#else
2fcc0ca9 67static bool debug_initialization = FALSE;
5a4b0634 68# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
69#endif
70
48015184
KW
71
72/* Returns the Unix errno portion; ignoring any others. This is a macro here
73 * instead of putting it into perl.h, because unclear to khw what should be
74 * done generally. */
75#define GET_ERRNO saved_errno
76
291a84fb
KW
77/* strlen() of a literal string constant. We might want this more general,
78 * but using it in just this file for now. A problem with more generality is
79 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
80#define STRLENs(s) (sizeof("" s "") - 1)
81
63e5b0d7
KW
82/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
83 * return of setlocale(), then this is extremely likely to be the C or POSIX
84 * locale. However, the output of setlocale() is documented to be opaque, but
85 * the odds are extremely small that it would return these two strings for some
86 * other locale. Note that VMS in these two locales includes many non-ASCII
87 * characters as controls and punctuation (below are hex bytes):
88 * cntrl: 84-97 9B-9F
89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
90 * Oddly, none there are listed as alphas, though some represent alphabetics
91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
92#define isNAME_C_OR_POSIX(name) \
93 ( (name) != NULL \
94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
95 || strEQ((name), "POSIX")))
96
8ef6e574
KW
97#ifdef USE_LOCALE
98
47280b20
KW
99/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
100 * looked up. This is in the form of a C string: */
101
102#define UTF8NESS_SEP "\v"
103#define UTF8NESS_PREFIX "\f"
104
105/* So, the string looks like:
98994639 106 *
47280b20 107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 108 *
47280b20
KW
109 * where the digit 0 after the \a indicates that the locale starting just
110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
111
112STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
113STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
114
115#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
117
118/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
119 * kept there always. The remining portion of the cache is LRU, with the
120 * oldest looked-up locale at the tail end */
121
98994639
HS
122STATIC char *
123S_stdize_locale(pTHX_ char *locs)
124{
47280b20
KW
125 /* Standardize the locale name from a string returned by 'setlocale',
126 * possibly modifying that string.
127 *
128 * The typical return value of setlocale() is either
129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
131 * (the space-separated values represent the various sublocales,
132 * in some unspecified order). This is not handled by this function.
133 *
134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
135 * which is harmful for further use of the string in setlocale(). This
136 * function removes the trailing new line and everything up through the '='
137 * */
138
7452cf6a 139 const char * const s = strchr(locs, '=');
98994639
HS
140 bool okay = TRUE;
141
7918f24d
NC
142 PERL_ARGS_ASSERT_STDIZE_LOCALE;
143
8772537c
AL
144 if (s) {
145 const char * const t = strchr(s, '.');
98994639 146 okay = FALSE;
8772537c
AL
147 if (t) {
148 const char * const u = strchr(t, '\n');
149 if (u && (u[1] == 0)) {
150 const STRLEN len = u - s;
151 Move(s + 1, locs, len, char);
152 locs[len] = 0;
153 okay = TRUE;
98994639
HS
154 }
155 }
156 }
157
158 if (!okay)
159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
160
161 return locs;
162}
163
e5f10d49
KW
164/* Two parallel arrays; first the locale categories Perl uses on this system;
165 * the second array is their names. These arrays are in mostly arbitrary
166 * order. */
167
168const int categories[] = {
169
170# ifdef USE_LOCALE_NUMERIC
171 LC_NUMERIC,
172# endif
173# ifdef USE_LOCALE_CTYPE
174 LC_CTYPE,
175# endif
176# ifdef USE_LOCALE_COLLATE
177 LC_COLLATE,
178# endif
179# ifdef USE_LOCALE_TIME
180 LC_TIME,
181# endif
182# ifdef USE_LOCALE_MESSAGES
183 LC_MESSAGES,
184# endif
185# ifdef USE_LOCALE_MONETARY
186 LC_MONETARY,
187# endif
9821811f
KW
188# ifdef USE_LOCALE_ADDRESS
189 LC_ADDRESS,
190# endif
191# ifdef USE_LOCALE_IDENTIFICATION
192 LC_IDENTIFICATION,
193# endif
194# ifdef USE_LOCALE_MEASUREMENT
195 LC_MEASUREMENT,
196# endif
197# ifdef USE_LOCALE_PAPER
198 LC_PAPER,
199# endif
200# ifdef USE_LOCALE_TELEPHONE
201 LC_TELEPHONE,
202# endif
e5f10d49
KW
203# ifdef LC_ALL
204 LC_ALL,
205# endif
206 -1 /* Placeholder because C doesn't allow a
207 trailing comma, and it would get complicated
208 with all the #ifdef's */
209};
210
211/* The top-most real element is LC_ALL */
212
4ef8bdf9 213const char * const category_names[] = {
e5f10d49
KW
214
215# ifdef USE_LOCALE_NUMERIC
216 "LC_NUMERIC",
217# endif
218# ifdef USE_LOCALE_CTYPE
219 "LC_CTYPE",
220# endif
221# ifdef USE_LOCALE_COLLATE
222 "LC_COLLATE",
223# endif
224# ifdef USE_LOCALE_TIME
225 "LC_TIME",
226# endif
227# ifdef USE_LOCALE_MESSAGES
228 "LC_MESSAGES",
229# endif
230# ifdef USE_LOCALE_MONETARY
231 "LC_MONETARY",
232# endif
9821811f
KW
233# ifdef USE_LOCALE_ADDRESS
234 "LC_ADDRESS",
235# endif
236# ifdef USE_LOCALE_IDENTIFICATION
237 "LC_IDENTIFICATION",
238# endif
239# ifdef USE_LOCALE_MEASUREMENT
240 "LC_MEASUREMENT",
241# endif
242# ifdef USE_LOCALE_PAPER
243 "LC_PAPER",
244# endif
245# ifdef USE_LOCALE_TELEPHONE
246 "LC_TELEPHONE",
247# endif
e5f10d49
KW
248# ifdef LC_ALL
249 "LC_ALL",
250# endif
251 NULL /* Placeholder */
252 };
253
254# ifdef LC_ALL
255
256 /* On systems with LC_ALL, it is kept in the highest index position. (-2
257 * to account for the final unused placeholder element.) */
258# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
259
260# else
261
262 /* On systems without LC_ALL, we pretend it is there, one beyond the real
263 * top element, hence in the unused placeholder element. */
264# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
265
266# endif
267
268/* Pretending there is an LC_ALL element just above allows us to avoid most
269 * special cases. Most loops through these arrays in the code below are
270 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
271 * on either type of system. But the code must be written to not access the
948523db
KW
272 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
273 * checked for at compile time by using the #define LC_ALL_INDEX which is only
274 * defined if we do have LC_ALL. */
e5f10d49 275
b09aaf40
KW
276STATIC const char *
277S_category_name(const int category)
278{
279 unsigned int i;
280
281#ifdef LC_ALL
282
283 if (category == LC_ALL) {
284 return "LC_ALL";
285 }
286
287#endif
288
289 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
290 if (category == categories[i]) {
291 return category_names[i];
292 }
293 }
294
295 {
296 const char suffix[] = " (unknown)";
297 int temp = category;
298 Size_t length = sizeof(suffix) + 1;
299 char * unknown;
300 dTHX;
301
302 if (temp < 0) {
303 length++;
304 temp = - temp;
305 }
306
307 /* Calculate the number of digits */
308 while (temp >= 10) {
309 temp /= 10;
310 length++;
311 }
312
313 Newx(unknown, length, char);
314 my_snprintf(unknown, length, "%d%s", category, suffix);
315 SAVEFREEPV(unknown);
316 return unknown;
317 }
318}
319
948523db
KW
320/* Now create LC_foo_INDEX #defines for just those categories on this system */
321# ifdef USE_LOCALE_NUMERIC
322# define LC_NUMERIC_INDEX 0
323# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
324# else
325# define _DUMMY_NUMERIC -1
326# endif
327# ifdef USE_LOCALE_CTYPE
328# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
329# define _DUMMY_CTYPE LC_CTYPE_INDEX
330# else
331# define _DUMMY_CTYPE _DUMMY_NUMERIC
332# endif
333# ifdef USE_LOCALE_COLLATE
334# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
335# define _DUMMY_COLLATE LC_COLLATE_INDEX
336# else
8203b3c3 337# define _DUMMY_COLLATE _DUMMY_CTYPE
948523db
KW
338# endif
339# ifdef USE_LOCALE_TIME
340# define LC_TIME_INDEX _DUMMY_COLLATE + 1
341# define _DUMMY_TIME LC_TIME_INDEX
342# else
343# define _DUMMY_TIME _DUMMY_COLLATE
344# endif
345# ifdef USE_LOCALE_MESSAGES
346# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
347# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
348# else
349# define _DUMMY_MESSAGES _DUMMY_TIME
350# endif
351# ifdef USE_LOCALE_MONETARY
352# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
353# define _DUMMY_MONETARY LC_MONETARY_INDEX
354# else
355# define _DUMMY_MONETARY _DUMMY_MESSAGES
356# endif
9821811f
KW
357# ifdef USE_LOCALE_ADDRESS
358# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
359# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
360# else
361# define _DUMMY_ADDRESS _DUMMY_MONETARY
362# endif
363# ifdef USE_LOCALE_IDENTIFICATION
364# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
365# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
366# else
367# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
368# endif
369# ifdef USE_LOCALE_MEASUREMENT
370# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
371# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
372# else
373# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
374# endif
375# ifdef USE_LOCALE_PAPER
376# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
377# define _DUMMY_PAPER LC_PAPER_INDEX
378# else
379# define _DUMMY_PAPER _DUMMY_MEASUREMENT
380# endif
381# ifdef USE_LOCALE_TELEPHONE
382# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
383# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
384# else
385# define _DUMMY_TELEPHONE _DUMMY_PAPER
386# endif
948523db 387# ifdef LC_ALL
9821811f 388# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
389# endif
390#endif /* ifdef USE_LOCALE */
8ef6e574 391
d2b24094 392/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
393#ifdef WIN32
394# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
395#else
396# define my_setlocale(cat, locale) setlocale(cat, locale)
397#endif
398
399#ifndef USE_POSIX_2008_LOCALE
400
401/* "do_setlocale_c" is intended to be called when the category is a constant
402 * known at compile time; "do_setlocale_r", not known until run time */
403# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
404# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
405
406#else /* Below uses POSIX 2008 */
407
408/* We emulate setlocale with our own function. LC_foo is not valid for the
409 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
410 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
411 * proper offset into the array 'category_masks[]'. At runtime, we have to
412 * search through the array (as the actual numbers may not be small contiguous
413 * positive integers which would lend themselves to array lookup). */
414# define do_setlocale_c(cat, locale) \
415 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
416# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
417
418/* A third array, parallel to the ones above to map from category to its
419 * equivalent mask */
420const int category_masks[] = {
421# ifdef USE_LOCALE_NUMERIC
422 LC_NUMERIC_MASK,
423# endif
424# ifdef USE_LOCALE_CTYPE
425 LC_CTYPE_MASK,
426# endif
427# ifdef USE_LOCALE_COLLATE
428 LC_COLLATE_MASK,
429# endif
430# ifdef USE_LOCALE_TIME
431 LC_TIME_MASK,
432# endif
433# ifdef USE_LOCALE_MESSAGES
434 LC_MESSAGES_MASK,
435# endif
436# ifdef USE_LOCALE_MONETARY
437 LC_MONETARY_MASK,
438# endif
439# ifdef USE_LOCALE_ADDRESS
440 LC_ADDRESS_MASK,
441# endif
442# ifdef USE_LOCALE_IDENTIFICATION
443 LC_IDENTIFICATION_MASK,
444# endif
445# ifdef USE_LOCALE_MEASUREMENT
446 LC_MEASUREMENT_MASK,
447# endif
448# ifdef USE_LOCALE_PAPER
449 LC_PAPER_MASK,
450# endif
451# ifdef USE_LOCALE_TELEPHONE
452 LC_TELEPHONE_MASK,
453# endif
454 /* LC_ALL can't be turned off by a Configure
455 * option, and in Posix 2008, should always be
456 * here, so compile it in unconditionally.
457 * This could catch some glitches at compile
458 * time */
459 LC_ALL_MASK
460 };
461
462STATIC const char *
463S_emulate_setlocale(const int category,
464 const char * locale,
465 unsigned int index,
466 const bool is_index_valid
467 )
468{
469 /* This function effectively performs a setlocale() on just the current
470 * thread; thus it is thread-safe. It does this by using the POSIX 2008
471 * locale functions to emulate the behavior of setlocale(). Similar to
472 * regular setlocale(), the return from this function points to memory that
473 * can be overwritten by other system calls, so needs to be copied
474 * immediately if you need to retain it. The difference here is that
475 * system calls besides another setlocale() can overwrite it.
476 *
477 * By doing this, most locale-sensitive functions become thread-safe. The
478 * exceptions are mostly those that return a pointer to static memory.
479 *
480 * This function takes the same parameters, 'category' and 'locale', that
481 * the regular setlocale() function does, but it also takes two additional
482 * ones. This is because the 2008 functions don't use a category; instead
483 * they use a corresponding mask. Because this function operates in both
484 * worlds, it may need one or the other or both. This function can
485 * calculate the mask from the input category, but to avoid this
486 * calculation, if the caller knows at compile time what the mask is, it
487 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
488 * parameter is ignored.
489 *
490 * POSIX 2008, for some sick reason, chose not to provide a method to find
491 * the category name of a locale. Some vendors have created a
492 * querylocale() function to do just that. This function is a lot simpler
493 * to implement on systems that have this. Otherwise, we have to keep
494 * track of what the locale has been set to, so that we can return its
495 * name to emulate setlocale(). It's also possible for C code in some
496 * library to change the locale without us knowing it, though as of
497 * September 2017, there are no occurrences in CPAN of uselocale(). Some
498 * libraries do use setlocale(), but that changes the global locale, and
499 * threads using per-thread locales will just ignore those changes.
500 * Another problem is that without querylocale(), we have to guess at what
501 * was meant by setting a locale of "". We handle this by not actually
502 * ever setting to "" (unless querylocale exists), but to emulate what we
503 * think should happen for "".
504 */
505
506 int mask;
507 locale_t old_obj;
508 locale_t new_obj;
509 dTHX;
510
511# ifdef DEBUGGING
512
513 if (DEBUG_Lv_TEST || debug_initialization) {
514 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);
515 }
516
517# endif
518
519 /* If the input mask might be incorrect, calculate the correct one */
520 if (! is_index_valid) {
521 unsigned int i;
522
523# ifdef DEBUGGING
524
525 if (DEBUG_Lv_TEST || debug_initialization) {
526 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
527 }
528
529# endif
530
531 for (i = 0; i <= LC_ALL_INDEX; i++) {
532 if (category == categories[i]) {
533 index = i;
534 goto found_index;
535 }
536 }
537
538 /* Here, we don't know about this category, so can't handle it.
539 * Fallback to the early POSIX usages */
540 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
541 "Unknown locale category %d; can't set it to %s\n",
542 category, locale);
543 return NULL;
544
545 found_index: ;
546
547# ifdef DEBUGGING
548
549 if (DEBUG_Lv_TEST || debug_initialization) {
550 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
551 }
552
553# endif
554
555 }
556
557 mask = category_masks[index];
558
559# ifdef DEBUGGING
560
561 if (DEBUG_Lv_TEST || debug_initialization) {
562 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
563 }
564
565# endif
566
567 /* If just querying what the existing locale is ... */
568 if (locale == NULL) {
569 locale_t cur_obj = uselocale((locale_t) 0);
570
571# ifdef DEBUGGING
572
573 if (DEBUG_Lv_TEST || debug_initialization) {
574 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
575 }
576
577# endif
578
579 if (cur_obj == LC_GLOBAL_LOCALE) {
580 return my_setlocale(category, NULL);
581 }
582
583# ifdef HAS_QUERYLOCALE
584
585 return (char *) querylocale(mask, cur_obj);
586
d2b24094 587# else
ecdda939
KW
588
589 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
590 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
591
8e13243a
KW
592# if defined(_NL_LOCALE_NAME) \
593 && defined(DEBUGGING) \
594 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
595 /* On systems that accept any locale name, the real underlying locale
596 * is often returned by this internal function, so we can't use it */
e9bc6d6b
KW
597 {
598 /* Internal glibc for querylocale(), but doesn't handle
599 * empty-string ("") locale properly; who knows what other
3a732259 600 * glitches. Check for it now, under debug. */
e9bc6d6b
KW
601
602 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
603 uselocale((locale_t) 0));
604 /*
605 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
606 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
607 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
608 */
609 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
610 if ( strNE(PL_curlocales[index], temp_name)
611 && ! ( isNAME_C_OR_POSIX(temp_name)
612 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
613
614# ifdef USE_C_BACKTRACE
615
616 dump_c_backtrace(Perl_debug_log, 20, 1);
617
618# endif
619
620 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
621 " (%s) and what internal glibc thinks"
622 " (%s)\n", category_names[index],
623 PL_curlocales[index], temp_name);
624 }
625
626 return temp_name;
627 }
628 }
629
630# endif
631
632 /* Without querylocale(), we have to use our record-keeping we've
633 * done. */
634
635 if (category != LC_ALL) {
636
637# ifdef DEBUGGING
638
639 if (DEBUG_Lv_TEST || debug_initialization) {
640 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
641 }
642
643# endif
644
645 return PL_curlocales[index];
646 }
647 else { /* For LC_ALL */
648 unsigned int i;
649 Size_t names_len = 0;
650 char * all_string;
7c93a471 651 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
652
653 /* If we have a valid LC_ALL value, just return it */
654 if (PL_curlocales[LC_ALL_INDEX]) {
655
656# ifdef DEBUGGING
657
658 if (DEBUG_Lv_TEST || debug_initialization) {
659 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
660 }
661
662# endif
663
664 return PL_curlocales[LC_ALL_INDEX];
665 }
666
667 /* Otherwise, we need to construct a string of name=value pairs.
668 * We use the glibc syntax, like
669 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
670 * First calculate the needed size. Along the way, check if all
671 * the locale names are the same */
e9bc6d6b
KW
672 for (i = 0; i < LC_ALL_INDEX; i++) {
673
674# ifdef DEBUGGING
675
676 if (DEBUG_Lv_TEST || debug_initialization) {
677 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]);
678 }
679
680# endif
681
682 names_len += strlen(category_names[i])
683 + 1 /* '=' */
684 + strlen(PL_curlocales[i])
685 + 1; /* ';' */
7c93a471
KW
686
687 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
688 are_all_categories_the_same_locale = FALSE;
689 }
690 }
691
692 /* If they are the same, we don't actually have to construct the
693 * string; we just make the entry in LC_ALL_INDEX valid, and be
694 * that single name */
695 if (are_all_categories_the_same_locale) {
696 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
697 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 698 }
7c93a471 699
e9bc6d6b
KW
700 names_len++; /* Trailing '\0' */
701 SAVEFREEPV(Newx(all_string, names_len, char));
702 *all_string = '\0';
703
704 /* Then fill in the string */
705 for (i = 0; i < LC_ALL_INDEX; i++) {
706
707# ifdef DEBUGGING
708
709 if (DEBUG_Lv_TEST || debug_initialization) {
710 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]);
711 }
712
713# endif
714
715 my_strlcat(all_string, category_names[i], names_len);
716 my_strlcat(all_string, "=", names_len);
717 my_strlcat(all_string, PL_curlocales[i], names_len);
718 my_strlcat(all_string, ";", names_len);
719 }
720
721# ifdef DEBUGGING
722
723 if (DEBUG_L_TEST || debug_initialization) {
724 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
725 }
726
727 #endif
728
729 return all_string;
730 }
731
732# ifdef EINVAL
733
734 SETERRNO(EINVAL, LIB_INVARG);
735
736# endif
737
738 return NULL;
739
d2b24094
KW
740# endif
741
e9bc6d6b
KW
742 }
743
f8115d60 744 /* Here, we are switching locales. */
e9bc6d6b
KW
745
746# ifndef HAS_QUERYLOCALE
747
748 if (strEQ(locale, "")) {
749
750 /* For non-querylocale() systems, we do the setting of "" ourselves to
751 * be sure that we really know what's going on. We follow the Linux
752 * documented behavior (but if that differs from the actual behavior,
753 * this won't work exactly as the OS implements). We go out and
754 * examine the environment based on our understanding of how the system
755 * works, and use that to figure things out */
756
757 const char * const lc_all = PerlEnv_getenv("LC_ALL");
758
759 /* Use any "LC_ALL" environment variable, as it overrides everything
760 * else. */
761 if (lc_all && strNE(lc_all, "")) {
762 locale = lc_all;
763 }
764 else {
765
766 /* Otherwise, we need to dig deeper. Unless overridden, the
767 * default is the LANG environment variable; if it doesn't exist,
768 * then "C" */
769
770 const char * default_name;
771
207cc8df 772 default_name = PerlEnv_getenv("LANG");
e9bc6d6b
KW
773
774 if (! default_name || strEQ(default_name, "")) {
929a7f8c 775 default_name = "C";
e9bc6d6b
KW
776 }
777 else if (PL_scopestack_ix != 0) {
207cc8df
DM
778 /* To minimize other threads messing with the environment,
779 * we copy the variable, making it a temporary. But this
780 * doesn't work upon program initialization before any
781 * scopes are created, and at this time, there's nothing
782 * else going on that would interfere. So skip the copy
783 * in that case */
784 default_name = savepv(default_name);
e9bc6d6b
KW
785 SAVEFREEPV(default_name);
786 }
787
788 if (category != LC_ALL) {
789 const char * const name = PerlEnv_getenv(category_names[index]);
790
791 /* Here we are setting a single category. Assume will have the
792 * default name */
793 locale = default_name;
794
795 /* But then look for an overriding environment variable */
796 if (name && strNE(name, "")) {
797 locale = name;
798 }
799 }
800 else {
801 bool did_override = FALSE;
802 unsigned int i;
803
804 /* Here, we are getting LC_ALL. Any categories that don't have
805 * a corresponding environment variable set should be set to
806 * LANG, or to "C" if there is no LANG. If no individual
807 * categories differ from this, we can just set LC_ALL. This
808 * is buggy on systems that have extra categories that we don't
809 * know about. If there is an environment variable that sets
810 * that category, we won't know to look for it, and so our use
811 * of LANG or "C" improperly overrides it. On the other hand,
812 * if we don't do what is done here, and there is no
813 * environment variable, the category's locale should be set to
814 * LANG or "C". So there is no good solution. khw thinks the
815 * best is to look at systems to see what categories they have,
816 * and include them, and then to assume that we know the
817 * complete set */
818
819 for (i = 0; i < LC_ALL_INDEX; i++) {
820 const char * const env_override
821 = savepv(PerlEnv_getenv(category_names[i]));
822 const char * this_locale = ( env_override
823 && strNE(env_override, ""))
824 ? env_override
825 : default_name;
6435f98d
KW
826 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
827 {
828 Safefree(env_override);
829 return NULL;
830 }
e9bc6d6b
KW
831
832 if (strNE(this_locale, default_name)) {
833 did_override = TRUE;
834 }
835
836 Safefree(env_override);
837 }
838
839 /* If all the categories are the same, we can set LC_ALL to
840 * that */
841 if (! did_override) {
842 locale = default_name;
843 }
844 else {
845
846 /* Here, LC_ALL is no longer valid, as some individual
847 * categories don't match it. We call ourselves
848 * recursively, as that will execute the code that
849 * generates the proper locale string for this situation.
850 * We don't do the remainder of this function, as that is
851 * to update our records, and we've just done that for the
852 * individual categories in the loop above, and doing so
853 * would cause LC_ALL to be done as well */
854 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
855 }
856 }
857 }
858 }
859 else if (strchr(locale, ';')) {
860
861 /* LC_ALL may actually incude a conglomeration of various categories.
862 * Without querylocale, this code uses the glibc (as of this writing)
863 * syntax for representing that, but that is not a stable API, and
864 * other platforms do it differently, so we have to handle all cases
865 * ourselves */
866
f0023d47 867 unsigned int i;
e9bc6d6b
KW
868 const char * s = locale;
869 const char * e = locale + strlen(locale);
870 const char * p = s;
871 const char * category_end;
872 const char * name_start;
873 const char * name_end;
874
f0023d47
KW
875 /* If the string that gives what to set doesn't include all categories,
876 * the omitted ones get set to "C". To get this behavior, first set
877 * all the individual categories to "C", and override the furnished
878 * ones below */
879 for (i = 0; i < LC_ALL_INDEX; i++) {
880 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
881 return NULL;
882 }
883 }
884
e9bc6d6b 885 while (s < e) {
e9bc6d6b
KW
886
887 /* Parse through the category */
888 while (isWORDCHAR(*p)) {
889 p++;
890 }
891 category_end = p;
892
893 if (*p++ != '=') {
894 Perl_croak(aTHX_
895 "panic: %s: %d: Unexpected character in locale name '%02X",
896 __FILE__, __LINE__, *(p-1));
897 }
898
899 /* Parse through the locale name */
900 name_start = p;
bee74f4b
KW
901 while (p < e && *p != ';') {
902 if (! isGRAPH(*p)) {
903 Perl_croak(aTHX_
904 "panic: %s: %d: Unexpected character in locale name '%02X",
905 __FILE__, __LINE__, *(p-1));
906 }
e9bc6d6b
KW
907 p++;
908 }
909 name_end = p;
910
bee74f4b
KW
911 /* Space past the semi-colon */
912 if (p < e) {
913 p++;
e9bc6d6b
KW
914 }
915
916 /* Find the index of the category name in our lists */
917 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 918 char * individ_locale;
e9bc6d6b
KW
919
920 /* Keep going if this isn't the index. The strnNE() avoids a
921 * Perl_form(), but would fail if ever a category name could be
922 * a substring of another one, like if there were a
923 * "LC_TIME_DATE" */
924 if strnNE(s, category_names[i], category_end - s) {
925 continue;
926 }
927
928 /* If this index is for the single category we're changing, we
929 * have found the locale to set it to. */
930 if (category == categories[i]) {
931 locale = Perl_form(aTHX_ "%.*s",
932 (int) (name_end - name_start),
933 name_start);
934 goto ready_to_set;
935 }
936
3c76a412 937 assert(category == LC_ALL);
bee74f4b
KW
938 individ_locale = Perl_form(aTHX_ "%.*s",
939 (int) (name_end - name_start), name_start);
6435f98d
KW
940 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
941 {
942 return NULL;
943 }
e9bc6d6b
KW
944 }
945
946 s = p;
947 }
948
949 /* Here we have set all the individual categories by recursive calls.
950 * These collectively should have fixed up LC_ALL, so can just query
951 * what that now is */
952 assert(category == LC_ALL);
953
954 return do_setlocale_c(LC_ALL, NULL);
955 }
956
957 ready_to_set: ;
958
f8115d60
KW
959 /* Here at the end of having to deal with the absence of querylocale().
960 * Some cases have already been fully handled by recursive calls to this
961 * function. But at this point, we haven't dealt with those, but are now
962 * prepared to, knowing what the locale name to set this category to is.
963 * This would have come for free if this system had had querylocale() */
964
e9bc6d6b
KW
965# endif /* end of ! querylocale */
966
f8115d60
KW
967 assert(PL_C_locale_obj);
968
969 /* Switching locales generally entails freeing the current one's space (at
970 * the C library's discretion). We need to stop using that locale before
971 * the switch. So switch to a known locale object that we don't otherwise
972 * mess with. This returns the locale object in effect at the time of the
973 * switch. */
974 old_obj = uselocale(PL_C_locale_obj);
975
976# ifdef DEBUGGING
977
978 if (DEBUG_Lv_TEST || debug_initialization) {
979 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
980 }
981
982# endif
983
984 if (! old_obj) {
985
986# ifdef DEBUGGING
987
988 if (DEBUG_L_TEST || debug_initialization) {
989 dSAVE_ERRNO;
990 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
991 RESTORE_ERRNO;
992 }
993
994# endif
995
996 return NULL;
997 }
998
999# ifdef DEBUGGING
1000
1001 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1002 PerlIO_printf(Perl_debug_log,
1003 "%s:%d: emulate_setlocale now using %p\n",
1004 __FILE__, __LINE__, PL_C_locale_obj);
f8115d60
KW
1005 }
1006
1007# endif
1008
70bd6bc8
KW
1009 /* If we are switching to the LC_ALL C locale, it already exists. Use
1010 * it instead of trying to create a new locale */
1011 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1012
1013# ifdef DEBUGGING
1014
1015 if (DEBUG_Lv_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log,
1017 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1018 }
1019
1020# endif
1021
1022 new_obj = PL_C_locale_obj;
1023
1024 /* We already had switched to the C locale in preparation for freeing
b22e9937 1025 * 'old_obj' */
70bd6bc8
KW
1026 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1027 freelocale(old_obj);
1028 }
1029 }
1030 else {
b22e9937
KW
1031 /* If we weren't in a thread safe locale, set so that newlocale() below
1032 * which uses 'old_obj', uses an empty one. Same for our reserved C
1033 * object. The latter is defensive coding, so that, even if there is
1034 * some bug, we will never end up trying to modify either of these, as
1035 * if passed to newlocale(), they can be. */
1036 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1037 old_obj = (locale_t) 0;
1038 }
f8115d60 1039
b22e9937
KW
1040 /* Ready to create a new locale by modification of the exising one */
1041 new_obj = newlocale(mask, locale, old_obj);
e9bc6d6b 1042
b22e9937
KW
1043 if (! new_obj) {
1044 dSAVE_ERRNO;
e9bc6d6b
KW
1045
1046# ifdef DEBUGGING
1047
b22e9937
KW
1048 if (DEBUG_L_TEST || debug_initialization) {
1049 PerlIO_printf(Perl_debug_log,
1050 "%s:%d: emulate_setlocale creating new object"
1051 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1052 }
e9bc6d6b
KW
1053
1054# endif
1055
b22e9937 1056 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1057
1058# ifdef DEBUGGING
1059
b22e9937
KW
1060 if (DEBUG_L_TEST || debug_initialization) {
1061 PerlIO_printf(Perl_debug_log,
1062 "%s:%d: switching back failed: %d\n",
1063 __FILE__, __LINE__, GET_ERRNO);
1064 }
e9bc6d6b
KW
1065
1066# endif
1067
b22e9937
KW
1068 }
1069 RESTORE_ERRNO;
1070 return NULL;
e9bc6d6b 1071 }
e9bc6d6b
KW
1072
1073# ifdef DEBUGGING
1074
b22e9937
KW
1075 if (DEBUG_Lv_TEST || debug_initialization) {
1076 PerlIO_printf(Perl_debug_log,
1077 "%s:%d: emulate_setlocale created %p",
1078 __FILE__, __LINE__, new_obj);
1079 if (old_obj) {
1080 PerlIO_printf(Perl_debug_log,
1081 "; should have freed %p", old_obj);
1082 }
1083 PerlIO_printf(Perl_debug_log, "\n");
19ee3daf 1084 }
e9bc6d6b
KW
1085
1086# endif
1087
b22e9937
KW
1088 /* And switch into it */
1089 if (! uselocale(new_obj)) {
1090 dSAVE_ERRNO;
e9bc6d6b
KW
1091
1092# ifdef DEBUGGING
1093
b22e9937
KW
1094 if (DEBUG_L_TEST || debug_initialization) {
1095 PerlIO_printf(Perl_debug_log,
1096 "%s:%d: emulate_setlocale switching to new object"
1097 " failed\n", __FILE__, __LINE__);
1098 }
e9bc6d6b
KW
1099
1100# endif
1101
b22e9937 1102 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1103
1104# ifdef DEBUGGING
1105
b22e9937
KW
1106 if (DEBUG_L_TEST || debug_initialization) {
1107 PerlIO_printf(Perl_debug_log,
1108 "%s:%d: switching back failed: %d\n",
1109 __FILE__, __LINE__, GET_ERRNO);
1110 }
e9bc6d6b
KW
1111
1112# endif
1113
b22e9937
KW
1114 }
1115 freelocale(new_obj);
1116 RESTORE_ERRNO;
1117 return NULL;
e9bc6d6b 1118 }
70bd6bc8 1119 }
e9bc6d6b
KW
1120
1121# ifdef DEBUGGING
1122
1123 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1124 PerlIO_printf(Perl_debug_log,
1125 "%s:%d: emulate_setlocale now using %p\n",
1126 __FILE__, __LINE__, new_obj);
e9bc6d6b
KW
1127 }
1128
1129# endif
1130
1131 /* We are done, except for updating our records (if the system doesn't keep
1132 * them) and in the case of locale "", we don't actually know what the
1133 * locale that got switched to is, as it came from the environment. So
1134 * have to find it */
1135
1136# ifdef HAS_QUERYLOCALE
1137
1138 if (strEQ(locale, "")) {
1139 locale = querylocale(mask, new_obj);
1140 }
1141
1142# else
1143
1144 /* Here, 'locale' is the return value */
1145
1146 /* Without querylocale(), we have to update our records */
1147
1148 if (category == LC_ALL) {
1149 unsigned int i;
1150
1151 /* For LC_ALL, we change all individual categories to correspond */
1152 /* PL_curlocales is a parallel array, so has same
1153 * length as 'categories' */
1154 for (i = 0; i <= LC_ALL_INDEX; i++) {
1155 Safefree(PL_curlocales[i]);
1156 PL_curlocales[i] = savepv(locale);
1157 }
1158 }
1159 else {
1160
1161 /* For a single category, if it's not the same as the one in LC_ALL, we
1162 * nullify LC_ALL */
1163
1164 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1165 Safefree(PL_curlocales[LC_ALL_INDEX]);
1166 PL_curlocales[LC_ALL_INDEX] = NULL;
1167 }
1168
1169 /* Then update the category's record */
1170 Safefree(PL_curlocales[index]);
1171 PL_curlocales[index] = savepv(locale);
1172 }
1173
1174# endif
1175
1176 return locale;
1177}
1178
1179#endif /* USE_POSIX_2008_LOCALE */
1180
1181#if 0 /* Code that was to emulate thread-safe locales on platforms that
1182 didn't natively support them */
1183
1184/* The way this would work is that we would keep a per-thread list of the
1185 * correct locale for that thread. Any operation that was locale-sensitive
1186 * would have to be changed so that it would look like this:
1187 *
1188 * LOCALE_LOCK;
1189 * setlocale to the correct locale for this operation
1190 * do operation
1191 * LOCALE_UNLOCK
1192 *
1193 * This leaves the global locale in the most recently used operation's, but it
1194 * was locked long enough to get the result. If that result is static, it
1195 * needs to be copied before the unlock.
1196 *
1197 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1198 * the setup, but are no-ops when not needed, and similarly,
1199 * END_LOCALE_DEPENDENT_OP for the tear-down
1200 *
1201 * But every call to a locale-sensitive function would have to be changed, and
1202 * if a module didn't cooperate by using the mutex, things would break.
1203 *
1204 * This code was abandoned before being completed or tested, and is left as-is
1205*/
1206
1207# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1208# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1209
1210STATIC char *
1211S_locking_setlocale(pTHX_
1212 const int category,
1213 const char * locale,
1214 int index,
1215 const bool is_index_valid
1216 )
1217{
1218 /* This function kind of performs a setlocale() on just the current thread;
1219 * thus it is kind of thread-safe. It does this by keeping a thread-level
1220 * array of the current locales for each category. Every time a locale is
1221 * switched to, it does the switch globally, but updates the thread's
1222 * array. A query as to what the current locale is just returns the
1223 * appropriate element from the array, and doesn't actually call the system
1224 * setlocale(). The saving into the array is done in an uninterruptible
1225 * section of code, so is unaffected by whatever any other threads might be
1226 * doing.
1227 *
1228 * All locale-sensitive operations must work by first starting a critical
1229 * section, then switching to the thread's locale as kept by this function,
1230 * and then doing the operation, then ending the critical section. Thus,
1231 * each gets done in the appropriate locale. simulating thread-safety.
1232 *
1233 * This function takes the same parameters, 'category' and 'locale', that
1234 * the regular setlocale() function does, but it also takes two additional
1235 * ones. This is because as described earlier. If we know on input the
1236 * index corresponding to the category into the array where we store the
1237 * current locales, we don't have to calculate it. If the caller knows at
1238 * compile time what the index is, it it can pass it, setting
1239 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1240 *
1241 */
1242
1243 /* If the input index might be incorrect, calculate the correct one */
1244 if (! is_index_valid) {
1245 unsigned int i;
1246
1247 if (DEBUG_Lv_TEST || debug_initialization) {
1248 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1249 }
1250
1251 for (i = 0; i <= LC_ALL_INDEX; i++) {
1252 if (category == categories[i]) {
1253 index = i;
1254 goto found_index;
1255 }
1256 }
1257
1258 /* Here, we don't know about this category, so can't handle it.
1259 * XXX best we can do is to unsafely set this
1260 * XXX warning */
1261
1262 return my_setlocale(category, locale);
1263
1264 found_index: ;
1265
1266 if (DEBUG_Lv_TEST || debug_initialization) {
1267 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1268 }
1269 }
1270
1271 /* For a query, just return what's in our records */
1272 if (new_locale == NULL) {
1273 return curlocales[index];
1274 }
1275
1276
1277 /* Otherwise, we need to do the switch, and save the result, all in a
1278 * critical section */
1279
1280 Safefree(curlocales[[index]]);
1281
1282 /* It might be that this is called from an already-locked section of code.
1283 * We would have to detect and skip the LOCK/UNLOCK if so */
1284 LOCALE_LOCK;
1285
1286 curlocales[index] = savepv(my_setlocale(category, new_locale));
1287
1288 if (strEQ(new_locale, "")) {
1289
1290#ifdef LC_ALL
1291
1292 /* The locale values come from the environment, and may not all be the
1293 * same, so for LC_ALL, we have to update all the others, while the
1294 * mutex is still locked */
1295
1296 if (category == LC_ALL) {
1297 unsigned int i;
1298 for (i = 0; i < LC_ALL_INDEX) {
1299 curlocales[i] = my_setlocale(categories[i], NULL);
1300 }
1301 }
1302 }
1303
1304#endif
1305
1306 LOCALE_UNLOCK;
1307
1308 return curlocales[index];
1309}
1310
1311#endif
d36adde0 1312#ifdef USE_LOCALE
837ce802 1313
a4f00dcc 1314STATIC void
86799d2d 1315S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1316{
86799d2d
KW
1317 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1318 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1319
86799d2d
KW
1320#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1321 || defined(HAS_NL_LANGINFO))
98994639 1322
87f8e8e7 1323 const char * radix = (use_locale)
4e6826bf 1324 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1325 /* FALSE => already in dest locale */
1326 : ".";
2213a3be 1327
87f8e8e7 1328 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1329
87f8e8e7
KW
1330 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1331 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1332 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1333 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1334 && _is_cur_LC_category_utf8(LC_NUMERIC))
1335 {
1336 SvUTF8_on(PL_numeric_radix_sv);
1337 }
86799d2d
KW
1338
1339# ifdef DEBUGGING
7d4bcc4a 1340
2fcc0ca9
KW
1341 if (DEBUG_L_TEST || debug_initialization) {
1342 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1343 SvPVX(PL_numeric_radix_sv),
1344 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1345 }
69014004 1346
86799d2d 1347# endif
d36adde0
KW
1348#else
1349
1350 PERL_UNUSED_ARG(use_locale);
1351
86799d2d 1352#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1353
98994639
HS
1354}
1355
398eeea9
KW
1356STATIC void
1357S_new_numeric(pTHX_ const char *newnum)
98994639 1358{
7d4bcc4a
KW
1359
1360#ifndef USE_LOCALE_NUMERIC
1361
1362 PERL_UNUSED_ARG(newnum);
1363
1364#else
0d071d52 1365
291a84fb 1366 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1367 * core Perl this and that 'newnum' is the name of the new locale.
1368 * It installs this locale as the current underlying default.
1369 *
1370 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1371 * set_numeric_underlying() and set_numeric_standard() functions, which
1372 * should probably not be called directly, but only via macros like
0d071d52
KW
1373 * SET_NUMERIC_STANDARD() in perl.h.
1374 *
1375 * The toggling is necessary mainly so that a non-dot radix decimal point
1376 * character can be output, while allowing internal calculations to use a
1377 * dot.
1378 *
1379 * This sets several interpreter-level variables:
bb304765 1380 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1381 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1382 * that the current locale is the program's underlying
1383 * locale
1384 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1385 * that the current locale is the C locale or
1386 * indistinguishable from the C locale. If non-zero, it
1387 * is in C; if > 1, it means it may not be toggled away
7738054c 1388 * from C.
4c68b815
KW
1389 * PL_numeric_underlying_is_standard A bool kept by this function
1390 * indicating that the underlying locale and the standard
1391 * C locale are indistinguishable for the purposes of
1392 * LC_NUMERIC. This happens when both of the above two
1393 * variables are true at the same time. (Toggling is a
1394 * no-op under these circumstances.) This variable is
1395 * used to avoid having to recalculate.
398eeea9 1396 */
0d071d52 1397
b03f34cf 1398 char *save_newnum;
98994639
HS
1399
1400 if (! newnum) {
43c5f42d
NC
1401 Safefree(PL_numeric_name);
1402 PL_numeric_name = NULL;
98994639 1403 PL_numeric_standard = TRUE;
892e6465 1404 PL_numeric_underlying = TRUE;
4c68b815 1405 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1406 return;
1407 }
1408
b03f34cf 1409 save_newnum = stdize_locale(savepv(newnum));
892e6465 1410 PL_numeric_underlying = TRUE;
4c68b815
KW
1411 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1412
69c5e0db
KW
1413#ifndef TS_W32_BROKEN_LOCALECONV
1414
4c68b815 1415 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1416 * them. But on broken Windows systems calling my_nl_langinfo() for
1417 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1418 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1419 if (! PL_numeric_standard) {
4e6826bf 1420 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1421 FALSE /* Don't toggle locale */ ))
4e6826bf 1422 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1423 }
abe1abcf 1424
69c5e0db
KW
1425#endif
1426
4c68b815 1427 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1428 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1429 Safefree(PL_numeric_name);
b03f34cf 1430 PL_numeric_name = save_newnum;
b03f34cf 1431 }
abe1abcf
KW
1432 else {
1433 Safefree(save_newnum);
1434 }
4c28b29c 1435
4c68b815
KW
1436 PL_numeric_underlying_is_standard = PL_numeric_standard;
1437
7e5377f7 1438# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1439
1440 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1441 PL_numeric_name,
1442 PL_underlying_numeric_obj);
1443
1444#endif
1445
4c68b815
KW
1446 if (DEBUG_L_TEST || debug_initialization) {
1447 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1448 }
1449
4c28b29c
KW
1450 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1451 * have to worry about the radix being a non-dot. (Core operations that
1452 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1453 if (PL_numeric_standard) {
1454 set_numeric_radix(0);
1455 }
1456 else {
1457 set_numeric_standard();
1458 }
4c28b29c 1459
98994639 1460#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1461
98994639
HS
1462}
1463
1464void
1465Perl_set_numeric_standard(pTHX)
1466{
7d4bcc4a 1467
98994639 1468#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1469
28c1bf33
KW
1470 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1471 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1472 * macro avoids calling this routine if toggling isn't necessary according
1473 * to our records (which could be wrong if some XS code has changed the
1474 * locale behind our back) */
0d071d52 1475
7d4bcc4a
KW
1476# ifdef DEBUGGING
1477
2fcc0ca9
KW
1478 if (DEBUG_L_TEST || debug_initialization) {
1479 PerlIO_printf(Perl_debug_log,
7bb8f702 1480 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1481 }
98994639 1482
7d4bcc4a 1483# endif
7bb8f702
KW
1484
1485 do_setlocale_c(LC_NUMERIC, "C");
1486 PL_numeric_standard = TRUE;
1487 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1488 set_numeric_radix(0);
1489
98994639 1490#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1491
98994639
HS
1492}
1493
1494void
5792c642 1495Perl_set_numeric_underlying(pTHX)
98994639 1496{
7d4bcc4a 1497
98994639 1498#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1499
28c1bf33 1500 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1501 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1502 * instead of calling this directly. The macro avoids calling this routine
1503 * if toggling isn't necessary according to our records (which could be
1504 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1505
7d4bcc4a
KW
1506# ifdef DEBUGGING
1507
2fcc0ca9
KW
1508 if (DEBUG_L_TEST || debug_initialization) {
1509 PerlIO_printf(Perl_debug_log,
7bb8f702 1510 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1511 PL_numeric_name);
1512 }
98994639 1513
7d4bcc4a 1514# endif
7bb8f702
KW
1515
1516 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1517 PL_numeric_standard = PL_numeric_underlying_is_standard;
1518 PL_numeric_underlying = TRUE;
1519 set_numeric_radix(! PL_numeric_standard);
1520
98994639 1521#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1522
98994639
HS
1523}
1524
1525/*
1526 * Set up for a new ctype locale.
1527 */
a4f00dcc
KW
1528STATIC void
1529S_new_ctype(pTHX_ const char *newctype)
98994639 1530{
7d4bcc4a
KW
1531
1532#ifndef USE_LOCALE_CTYPE
1533
7d4bcc4a
KW
1534 PERL_UNUSED_ARG(newctype);
1535 PERL_UNUSED_CONTEXT;
1536
1537#else
0d071d52 1538
291a84fb 1539 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1540 * core Perl this and that 'newctype' is the name of the new locale.
1541 *
1542 * This function sets up the folding arrays for all 256 bytes, assuming
1543 * that tofold() is tolc() since fold case is not a concept in POSIX,
1544 *
1545 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1546 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1547 * this function should be called directly only from this file and from
0d071d52
KW
1548 * POSIX::setlocale() */
1549
27da23d5 1550 dVAR;
013f4e03 1551 unsigned int i;
98994639 1552
8b7358b9
KW
1553 /* Don't check for problems if we are suppressing the warnings */
1554 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
30d8090d 1555 bool maybe_utf8_turkic = FALSE;
8b7358b9 1556
7918f24d
NC
1557 PERL_ARGS_ASSERT_NEW_CTYPE;
1558
215c5139
KW
1559 /* We will replace any bad locale warning with 1) nothing if the new one is
1560 * ok; or 2) a new warning for the bad new locale */
1561 if (PL_warn_locale) {
1562 SvREFCNT_dec_NN(PL_warn_locale);
1563 PL_warn_locale = NULL;
1564 }
1565
c1284011 1566 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1567
1568 /* A UTF-8 locale gets standard rules. But note that code still has to
1569 * handle this specially because of the three problematic code points */
1570 if (PL_in_utf8_CTYPE_locale) {
1571 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
30d8090d
KW
1572
1573 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1574 * Turkic. Make sure these two are the only anomalies. (We don't use
1575 * towupper and towlower because they aren't in C89.) */
5b64f24c
KW
1576
1577#if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1578
1579 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1580
1581#else
1582
30d8090d 1583 if (toupper('i') == 'i' && tolower('I') == 'I') {
5b64f24c
KW
1584
1585#endif
30d8090d
KW
1586 check_for_problems = TRUE;
1587 maybe_utf8_turkic = TRUE;
1588 }
31f05a37 1589 }
8b7358b9
KW
1590
1591 /* We don't populate the other lists if a UTF-8 locale, but do check that
1592 * everything works as expected, unless checking turned off */
1593 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1594 /* Assume enough space for every character being bad. 4 spaces each
1595 * for the 94 printable characters that are output like "'x' "; and 5
1596 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1597 * NUL */
8b7358b9 1598 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1599 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1600 to start */
1601 unsigned int bad_count = 0; /* Count of bad characters */
1602
baa60164 1603 for (i = 0; i < 256; i++) {
8b7358b9 1604 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1605 if (isupper(i))
1606 PL_fold_locale[i] = (U8) tolower(i);
1607 else if (islower(i))
1608 PL_fold_locale[i] = (U8) toupper(i);
1609 else
1610 PL_fold_locale[i] = (U8) i;
8b7358b9 1611 }
8c6180a9
KW
1612
1613 /* If checking for locale problems, see if the native ASCII-range
1614 * printables plus \n and \t are in their expected categories in
1615 * the new locale. If not, this could mean big trouble, upending
1616 * Perl's and most programs' assumptions, like having a
1617 * metacharacter with special meaning become a \w. Fortunately,
1618 * it's very rare to find locales that aren't supersets of ASCII
1619 * nowadays. It isn't a problem for most controls to be changed
1620 * into something else; we check only \n and \t, though perhaps \r
1621 * could be an issue as well. */
7d4bcc4a 1622 if ( check_for_problems
8c6180a9
KW
1623 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1624 {
8b7358b9 1625 bool is_bad = FALSE;
6726b4f4 1626 char name[4] = { '\0' };
8b7358b9
KW
1627
1628 /* Convert the name into a string */
6726b4f4 1629 if (isGRAPH_A(i)) {
8b7358b9
KW
1630 name[0] = i;
1631 name[1] = '\0';
1632 }
1633 else if (i == '\n') {
6726b4f4
KW
1634 my_strlcpy(name, "\\n", sizeof(name));
1635 }
1636 else if (i == '\t') {
1637 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1638 }
1639 else {
6726b4f4
KW
1640 assert(i == ' ');
1641 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1642 }
1643
1644 /* Check each possibe class */
1645 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1646 is_bad = TRUE;
1647 DEBUG_L(PerlIO_printf(Perl_debug_log,
1648 "isalnum('%s') unexpectedly is %d\n",
1649 name, cBOOL(isalnum(i))));
1650 }
1651 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1652 is_bad = TRUE;
1653 DEBUG_L(PerlIO_printf(Perl_debug_log,
1654 "isalpha('%s') unexpectedly is %d\n",
1655 name, cBOOL(isalpha(i))));
1656 }
1657 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1658 is_bad = TRUE;
1659 DEBUG_L(PerlIO_printf(Perl_debug_log,
1660 "isdigit('%s') unexpectedly is %d\n",
1661 name, cBOOL(isdigit(i))));
1662 }
1663 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1664 is_bad = TRUE;
1665 DEBUG_L(PerlIO_printf(Perl_debug_log,
1666 "isgraph('%s') unexpectedly is %d\n",
1667 name, cBOOL(isgraph(i))));
1668 }
1669 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1670 is_bad = TRUE;
1671 DEBUG_L(PerlIO_printf(Perl_debug_log,
1672 "islower('%s') unexpectedly is %d\n",
1673 name, cBOOL(islower(i))));
1674 }
1675 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1676 is_bad = TRUE;
1677 DEBUG_L(PerlIO_printf(Perl_debug_log,
1678 "isprint('%s') unexpectedly is %d\n",
1679 name, cBOOL(isprint(i))));
1680 }
1681 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1682 is_bad = TRUE;
1683 DEBUG_L(PerlIO_printf(Perl_debug_log,
1684 "ispunct('%s') unexpectedly is %d\n",
1685 name, cBOOL(ispunct(i))));
1686 }
1687 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1688 is_bad = TRUE;
1689 DEBUG_L(PerlIO_printf(Perl_debug_log,
1690 "isspace('%s') unexpectedly is %d\n",
1691 name, cBOOL(isspace(i))));
1692 }
1693 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1694 is_bad = TRUE;
1695 DEBUG_L(PerlIO_printf(Perl_debug_log,
1696 "isupper('%s') unexpectedly is %d\n",
1697 name, cBOOL(isupper(i))));
1698 }
1699 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1700 is_bad = TRUE;
1701 DEBUG_L(PerlIO_printf(Perl_debug_log,
1702 "isxdigit('%s') unexpectedly is %d\n",
1703 name, cBOOL(isxdigit(i))));
1704 }
1705 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1706 is_bad = TRUE;
1707 DEBUG_L(PerlIO_printf(Perl_debug_log,
1708 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1709 name, tolower(i), (int) toLOWER_A(i)));
1710 }
1711 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1712 is_bad = TRUE;
1713 DEBUG_L(PerlIO_printf(Perl_debug_log,
1714 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1715 name, toupper(i), (int) toUPPER_A(i)));
1716 }
1717 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1718 is_bad = TRUE;
1719 DEBUG_L(PerlIO_printf(Perl_debug_log,
1720 "'\\n' (=%02X) is not a control\n", (int) i));
1721 }
1722
1723 /* Add to the list; Separate multiple entries with a blank */
1724 if (is_bad) {
1725 if (bad_count) {
1726 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1727 }
8b7358b9
KW
1728 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1729 bad_count++;
8c6180a9
KW
1730 }
1731 }
1732 }
1733
30d8090d
KW
1734 if (bad_count == 2 && maybe_utf8_turkic) {
1735 bad_count = 0;
1736 *bad_chars_list = '\0';
1737 PL_fold_locale['I'] = 'I';
1738 PL_fold_locale['i'] = 'i';
1739 PL_in_utf8_turkic_locale = TRUE;
1740 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1741 __FILE__, __LINE__, newctype));
1742 }
1743 else {
b8df1494 1744 PL_in_utf8_turkic_locale = FALSE;
30d8090d 1745 }
b8df1494 1746
7d4bcc4a
KW
1747# ifdef MB_CUR_MAX
1748
8c6180a9 1749 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1750 * this locale requires more than one byte, there are going to be
8c6180a9 1751 * problems. */
9c8a6dc2
KW
1752 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1753 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1754 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1755
8b7358b9
KW
1756 if ( check_for_problems && MB_CUR_MAX > 1
1757 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1758
1759 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1760 * locale. Just assume that the implementation for them (plus
1761 * for POSIX) is correct and the > 1 value is spurious. (Since
1762 * these are specially handled to never be considered UTF-8
1763 * locales, as long as this is the only problem, everything
1764 * should work fine */
1765 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1766 {
8c6180a9
KW
1767 multi_byte_locale = TRUE;
1768 }
7d4bcc4a
KW
1769
1770# endif
8c6180a9 1771
30d8090d
KW
1772 /* If we found problems and we want them output, do so */
1773 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1774 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1775 {
8b7358b9
KW
1776 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1777 PL_warn_locale = Perl_newSVpvf(aTHX_
1778 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1779 " which have\nunexpected meanings: %s\nThe Perl program"
1780 " will use the expected meanings",
8b7358b9 1781 newctype, bad_chars_list);
8b7358b9
KW
1782 }
1783 else {
1784 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1785 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1786 newctype,
8c6180a9
KW
1787 (multi_byte_locale)
1788 ? " Some characters in it are not recognized by"
1789 " Perl."
1790 : "",
1791 (bad_count)
1792 ? "\nThe following characters (and maybe others)"
1793 " may not have the same meaning as the Perl"
1794 " program expects:\n"
1795 : "",
1796 (bad_count)
1797 ? bad_chars_list
1798 : ""
1799 );
8b7358b9
KW
1800 }
1801
b119c2be
KW
1802# ifdef HAS_NL_LANGINFO
1803
1804 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1805 /* parameter FALSE is a don't care here */
4e6826bf 1806 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1807
1808# endif
1809
8b7358b9
KW
1810 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1811
cc9eaeb0 1812 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1813 * output the message now. If not in that scope, we save the
1814 * message to be output at the first operation using this locale,
1815 * if that actually happens. Most programs don't use locales, so
1816 * they are immune to bad ones. */
cc9eaeb0 1817 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1818
780fcc9f
KW
1819 /* The '0' below suppresses a bogus gcc compiler warning */
1820 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1821
bddebb56
KW
1822 if (IN_LC(LC_CTYPE)) {
1823 SvREFCNT_dec_NN(PL_warn_locale);
1824 PL_warn_locale = NULL;
1825 }
780fcc9f 1826 }
baa60164 1827 }
31f05a37 1828 }
98994639
HS
1829
1830#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1831
98994639
HS
1832}
1833
98994639 1834void
2726666d
KW
1835Perl__warn_problematic_locale()
1836{
2726666d
KW
1837
1838#ifdef USE_LOCALE_CTYPE
1839
5f04a188
KW
1840 dTHX;
1841
1842 /* Internal-to-core function that outputs the message in PL_warn_locale,
1843 * and then NULLS it. Should be called only through the macro
1844 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1845
2726666d 1846 if (PL_warn_locale) {
2726666d
KW
1847 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1848 SvPVX(PL_warn_locale),
1849 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1850 SvREFCNT_dec_NN(PL_warn_locale);
1851 PL_warn_locale = NULL;
1852 }
1853
1854#endif
1855
1856}
1857
a4f00dcc
KW
1858STATIC void
1859S_new_collate(pTHX_ const char *newcoll)
98994639 1860{
7d4bcc4a
KW
1861
1862#ifndef USE_LOCALE_COLLATE
1863
1864 PERL_UNUSED_ARG(newcoll);
1865 PERL_UNUSED_CONTEXT;
1866
1867#else
0d071d52 1868
291a84fb 1869 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1870 * core Perl this and that 'newcoll' is the name of the new locale.
1871 *
d35fca5f
KW
1872 * The design of locale collation is that every locale change is given an
1873 * index 'PL_collation_ix'. The first time a string particpates in an
1874 * operation that requires collation while locale collation is active, it
1875 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1876 * magic includes the collation index, and the transformation of the string
1877 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1878 * instead of the string itself. If a string changes, the magic is
1879 * cleared. The next time the locale changes, the index is incremented,
1880 * and so we know during a comparison that the transformation is not
1881 * necessarily still valid, and so is recomputed. Note that if the locale
1882 * changes enough times, the index could wrap (a U32), and it is possible
1883 * that a transformation would improperly be considered valid, leading to
1884 * an unlikely bug */
0d071d52 1885
98994639
HS
1886 if (! newcoll) {
1887 if (PL_collation_name) {
1888 ++PL_collation_ix;
1889 Safefree(PL_collation_name);
1890 PL_collation_name = NULL;
1891 }
1892 PL_collation_standard = TRUE;
00bf60ca 1893 is_standard_collation:
98994639
HS
1894 PL_collxfrm_base = 0;
1895 PL_collxfrm_mult = 2;
165a1c52 1896 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1897 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1898 PL_strxfrm_max_cp = 0;
98994639
HS
1899 return;
1900 }
1901
d35fca5f 1902 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1903 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1904 ++PL_collation_ix;
1905 Safefree(PL_collation_name);
1906 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1907 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1908 if (PL_collation_standard) {
1909 goto is_standard_collation;
1910 }
98994639 1911
165a1c52 1912 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1913 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1914 PL_strxfrm_max_cp = 0;
165a1c52 1915
59c018b9
KW
1916 /* A locale collation definition includes primary, secondary, tertiary,
1917 * etc. weights for each character. To sort, the primary weights are
1918 * used, and only if they compare equal, then the secondary weights are
1919 * used, and only if they compare equal, then the tertiary, etc.
1920 *
1921 * strxfrm() works by taking the input string, say ABC, and creating an
1922 * output transformed string consisting of first the primary weights,
1923 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1924 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1925 * may not have weights at every level. In our example, let's say B
1926 * doesn't have a tertiary weight, and A doesn't have a secondary
1927 * weight. The constructed string is then going to be
1928 * A¹B¹C¹ B²C² A³C³ ....
1929 * This has the desired effect that strcmp() will look at the secondary
1930 * or tertiary weights only if the strings compare equal at all higher
1931 * priority weights. The spaces shown here, like in
c342d20e 1932 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1933 * are not just for readability. In the general case, these must
1934 * actually be bytes, which we will call here 'separator weights'; and
1935 * they must be smaller than any other weight value, but since these
1936 * are C strings, only the terminating one can be a NUL (some
1937 * implementations may include a non-NUL separator weight just before
1938 * the NUL). Implementations tend to reserve 01 for the separator
1939 * weights. They are needed so that a shorter string's secondary
1940 * weights won't be misconstrued as primary weights of a longer string,
1941 * etc. By making them smaller than any other weight, the shorter
1942 * string will sort first. (Actually, if all secondary weights are
1943 * smaller than all primary ones, there is no need for a separator
1944 * weight between those two levels, etc.)
1945 *
1946 * The length of the transformed string is roughly a linear function of
1947 * the input string. It's not exactly linear because some characters
1948 * don't have weights at all levels. When we call strxfrm() we have to
1949 * allocate some memory to hold the transformed string. The
1950 * calculations below try to find coefficients 'm' and 'b' for this
1951 * locale so that m*x + b equals how much space we need, given the size
1952 * of the input string in 'x'. If we calculate too small, we increase
1953 * the size as needed, and call strxfrm() again, but it is better to
1954 * get it right the first time to avoid wasted expensive string
1955 * transformations. */
1956
98994639 1957 {
79f120c8
KW
1958 /* We use the string below to find how long the tranformation of it
1959 * is. Almost all locales are supersets of ASCII, or at least the
1960 * ASCII letters. We use all of them, half upper half lower,
1961 * because if we used fewer, we might hit just the ones that are
1962 * outliers in a particular locale. Most of the strings being
1963 * collated will contain a preponderance of letters, and even if
1964 * they are above-ASCII, they are likely to have the same number of
1965 * weight levels as the ASCII ones. It turns out that digits tend
1966 * to have fewer levels, and some punctuation has more, but those
1967 * are relatively sparse in text, and khw believes this gives a
1968 * reasonable result, but it could be changed if experience so
1969 * dictates. */
1970 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1971 char * x_longer; /* Transformed 'longer' */
1972 Size_t x_len_longer; /* Length of 'x_longer' */
1973
1974 char * x_shorter; /* We also transform a substring of 'longer' */
1975 Size_t x_len_shorter;
1976
a4a439fb 1977 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1978 * are interested only in its length). It is used because it has
1979 * the intelligence to handle all cases, but to work, it needs some
1980 * values of 'm' and 'b' to get it started. For the purposes of
1981 * this calculation we use a very conservative estimate of 'm' and
1982 * 'b'. This assumes a weight can be multiple bytes, enough to
1983 * hold any UV on the platform, and there are 5 levels, 4 weight
1984 * bytes, and a trailing NUL. */
1985 PL_collxfrm_base = 5;
1986 PL_collxfrm_mult = 5 * sizeof(UV);
1987
1988 /* Find out how long the transformation really is */
a4a439fb
KW
1989 x_longer = _mem_collxfrm(longer,
1990 sizeof(longer) - 1,
1991 &x_len_longer,
1992
1993 /* We avoid converting to UTF-8 in the
1994 * called function by telling it the
1995 * string is in UTF-8 if the locale is a
1996 * UTF-8 one. Since the string passed
1997 * here is invariant under UTF-8, we can
1998 * claim it's UTF-8 even though it isn't.
1999 * */
2000 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2001 Safefree(x_longer);
2002
2003 /* Find out how long the transformation of a substring of 'longer'
2004 * is. Together the lengths of these transformations are
2005 * sufficient to calculate 'm' and 'b'. The substring is all of
2006 * 'longer' except the first character. This minimizes the chances
2007 * of being swayed by outliers */
a4a439fb 2008 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 2009 sizeof(longer) - 2,
a4a439fb
KW
2010 &x_len_shorter,
2011 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2012 Safefree(x_shorter);
2013
2014 /* If the results are nonsensical for this simple test, the whole
2015 * locale definition is suspect. Mark it so that locale collation
2016 * is not active at all for it. XXX Should we warn? */
2017 if ( x_len_shorter == 0
2018 || x_len_longer == 0
2019 || x_len_shorter >= x_len_longer)
2020 {
2021 PL_collxfrm_mult = 0;
2022 PL_collxfrm_base = 0;
2023 }
2024 else {
2025 SSize_t base; /* Temporary */
2026
2027 /* We have both: m * strlen(longer) + b = x_len_longer
2028 * m * strlen(shorter) + b = x_len_shorter;
2029 * subtracting yields:
2030 * m * (strlen(longer) - strlen(shorter))
2031 * = x_len_longer - x_len_shorter
2032 * But we have set things up so that 'shorter' is 1 byte smaller
2033 * than 'longer'. Hence:
2034 * m = x_len_longer - x_len_shorter
2035 *
2036 * But if something went wrong, make sure the multiplier is at
2037 * least 1.
2038 */
2039 if (x_len_longer > x_len_shorter) {
2040 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2041 }
2042 else {
2043 PL_collxfrm_mult = 1;
2044 }
2045
2046 /* mx + b = len
2047 * so: b = len - mx
2048 * but in case something has gone wrong, make sure it is
2049 * non-negative */
2050 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2051 if (base < 0) {
2052 base = 0;
2053 }
2054
2055 /* Add 1 for the trailing NUL */
2056 PL_collxfrm_base = base + 1;
2057 }
58eebef2 2058
7d4bcc4a
KW
2059# ifdef DEBUGGING
2060
58eebef2
KW
2061 if (DEBUG_L_TEST || debug_initialization) {
2062 PerlIO_printf(Perl_debug_log,
b07929e4
KW
2063 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2064 "x_len_longer=%zu,"
2065 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
2066 __FILE__, __LINE__,
2067 PL_in_utf8_COLLATE_locale,
2068 x_len_shorter, x_len_longer,
2069 PL_collxfrm_mult, PL_collxfrm_base);
2070 }
7d4bcc4a
KW
2071# endif
2072
98994639
HS
2073 }
2074 }
2075
2076#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 2077
98994639
HS
2078}
2079
d36adde0
KW
2080#endif
2081
d2b24094 2082#ifdef WIN32
b8cc575c 2083
a4f00dcc 2084STATIC char *
b8cc575c 2085S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2086{
2087 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2088 * difference between the two unless the input locale is "", which normally
2089 * means on Windows to get the machine default, which is set via the
2090 * computer's "Regional and Language Options" (or its current equivalent).
2091 * In POSIX, it instead means to find the locale from the user's
2092 * environment. This routine changes the Windows behavior to first look in
2093 * the environment, and, if anything is found, use that instead of going to
2094 * the machine default. If there is no environment override, the machine
2095 * default is used, by calling the real setlocale() with "".
2096 *
2097 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2098 * use the particular category's variable if set; otherwise to use the LANG
2099 * variable. */
b385bb4d 2100
175c4cf9 2101 bool override_LC_ALL = FALSE;
89f7b9aa 2102 char * result;
e5f10d49 2103 unsigned int i;
89f7b9aa 2104
b385bb4d 2105 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2106
2107# ifdef LC_ALL
2108
b385bb4d
KW
2109 locale = PerlEnv_getenv("LC_ALL");
2110 if (! locale) {
e5f10d49
KW
2111 if (category == LC_ALL) {
2112 override_LC_ALL = TRUE;
2113 }
2114 else {
7d4bcc4a
KW
2115
2116# endif
7d4bcc4a 2117
e5f10d49
KW
2118 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2119 if (category == categories[i]) {
2120 locale = PerlEnv_getenv(category_names[i]);
2121 goto found_locale;
2122 }
2123 }
7d4bcc4a 2124
b385bb4d 2125 locale = PerlEnv_getenv("LANG");
481465ea 2126 if (! locale) {
b385bb4d
KW
2127 locale = "";
2128 }
e5f10d49
KW
2129
2130 found_locale: ;
7d4bcc4a
KW
2131
2132# ifdef LC_ALL
2133
e5f10d49 2134 }
b385bb4d 2135 }
7d4bcc4a
KW
2136
2137# endif
2138
b385bb4d
KW
2139 }
2140
89f7b9aa 2141 result = setlocale(category, locale);
48015184
KW
2142 DEBUG_L(STMT_START {
2143 dSAVE_ERRNO;
2144 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2145 setlocale_debug_string(category, locale, result));
2146 RESTORE_ERRNO;
2147 } STMT_END);
89f7b9aa 2148
481465ea 2149 if (! override_LC_ALL) {
89f7b9aa
KW
2150 return result;
2151 }
2152
dfd77d7a 2153 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2154 * LANG variable or the system default if there is no LANG. But these have
2155 * lower priority than the other LC_foo variables, so override it for each
2156 * one that is set. (If they are set to "", it means to use the same thing
2157 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2158
948523db 2159 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2160 result = PerlEnv_getenv(category_names[i]);
2161 if (result && strNE(result, "")) {
2162 setlocale(categories[i], result);
2163 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2164 __FILE__, __LINE__,
2165 setlocale_debug_string(categories[i], result, "not captured")));
2166 }
89f7b9aa 2167 }
7d4bcc4a 2168
bbc98134 2169 result = setlocale(LC_ALL, NULL);
48015184
KW
2170 DEBUG_L(STMT_START {
2171 dSAVE_ERRNO;
2172 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2173 __FILE__, __LINE__,
48015184
KW
2174 setlocale_debug_string(LC_ALL, NULL, result));
2175 RESTORE_ERRNO;
2176 } STMT_END);
89f7b9aa 2177
bbc98134 2178 return result;
b385bb4d
KW
2179}
2180
2181#endif
2182
9aac5db8
KW
2183/*
2184
2185=head1 Locale-related functions and macros
2186
2187=for apidoc Perl_setlocale
2188
2189This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2190taking the same parameters, and returning the same information, except that it
9487427b
KW
2191returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2192instead return C<C> if the underlying locale has a non-dot decimal point
2193character, or a non-empty thousands separator for displaying floating point
2194numbers. This is because perl keeps that locale category such that it has a
2195dot and empty separator, changing the locale briefly during the operations
2196where the underlying one is required. C<Perl_setlocale> knows about this, and
2197compensates; regular C<setlocale> doesn't.
9aac5db8 2198
e9bc6d6b 2199Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2200declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2201C<const> (presumably because its API was specified long ago, and can't be
2202updated; it is illegal to change the information C<setlocale> returns; doing
2203so leads to segfaults.)
9aac5db8 2204
e9bc6d6b
KW
2205Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2206C<setlocale> can be completely ineffective on some platforms under some
2207configurations.
2208
9aac5db8 2209C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2210where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2211the system C<setlocale()> is ineffective, returning the wrong information, and
2212failing to actually change the locale. C<Perl_setlocale>, however works
2213properly in all circumstances.
9aac5db8
KW
2214
2215The return points to a per-thread static buffer, which is overwritten the next
2216time C<Perl_setlocale> is called from the same thread.
2217
2218=cut
2219
2220*/
2221
2222const char *
2223Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2224{
2225 /* This wraps POSIX::setlocale() */
2226
d36adde0
KW
2227#ifdef NO_LOCALE
2228
2229 PERL_UNUSED_ARG(category);
2230 PERL_UNUSED_ARG(locale);
2231
2232 return "C";
2233
2234#else
2235
9aac5db8
KW
2236 const char * retval;
2237 const char * newlocale;
2238 dSAVEDERRNO;
a4f00dcc 2239 dTHX;
d36adde0 2240 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2241
a4f00dcc
KW
2242#ifdef USE_LOCALE_NUMERIC
2243
291a84fb
KW
2244 /* A NULL locale means only query what the current one is. We have the
2245 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2246 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2247 * the correct results. All other categories don't require special
2248 * handling */
a4f00dcc
KW
2249 if (locale == NULL) {
2250 if (category == LC_NUMERIC) {
9aac5db8
KW
2251
2252 /* We don't have to copy this return value, as it is a per-thread
2253 * variable, and won't change until a future setlocale */
2254 return PL_numeric_name;
a4f00dcc
KW
2255 }
2256
7d4bcc4a 2257# ifdef LC_ALL
a4f00dcc 2258
9aac5db8
KW
2259 else if (category == LC_ALL) {
2260 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2261 }
2262
7d4bcc4a 2263# endif
a4f00dcc
KW
2264
2265 }
2266
2267#endif
2268
33e5a354
KW
2269 retval = save_to_buffer(do_setlocale_r(category, locale),
2270 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2271 SAVE_ERRNO;
2272
2273#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2274
2275 if (locale == NULL && category == LC_ALL) {
2276 RESTORE_LC_NUMERIC();
2277 }
2278
2279#endif
a4f00dcc
KW
2280
2281 DEBUG_L(PerlIO_printf(Perl_debug_log,
2282 "%s:%d: %s\n", __FILE__, __LINE__,
2283 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2284
9aac5db8
KW
2285 RESTORE_ERRNO;
2286
2287 if (! retval) {
a4f00dcc
KW
2288 return NULL;
2289 }
2290
9aac5db8 2291 /* If locale == NULL, we are just querying the state */
a4f00dcc 2292 if (locale == NULL) {
a4f00dcc
KW
2293 return retval;
2294 }
a4f00dcc 2295
1159483a
KW
2296 /* Now that have switched locales, we have to update our records to
2297 * correspond. */
a4f00dcc 2298
1159483a 2299 switch (category) {
a4f00dcc 2300
1159483a 2301#ifdef USE_LOCALE_CTYPE
a4f00dcc 2302
1159483a
KW
2303 case LC_CTYPE:
2304 new_ctype(retval);
2305 break;
a4f00dcc 2306
1159483a 2307#endif
a4f00dcc
KW
2308#ifdef USE_LOCALE_COLLATE
2309
1159483a
KW
2310 case LC_COLLATE:
2311 new_collate(retval);
2312 break;
a4f00dcc 2313
1159483a 2314#endif
a4f00dcc
KW
2315#ifdef USE_LOCALE_NUMERIC
2316
1159483a
KW
2317 case LC_NUMERIC:
2318 new_numeric(retval);
2319 break;
a4f00dcc 2320
1159483a
KW
2321#endif
2322#ifdef LC_ALL
a4f00dcc 2323
1159483a 2324 case LC_ALL:
a4f00dcc 2325
1159483a
KW
2326 /* LC_ALL updates all the things we care about. The values may not
2327 * be the same as 'retval', as the locale "" may have set things
2328 * individually */
a4f00dcc 2329
1159483a 2330# ifdef USE_LOCALE_CTYPE
a4f00dcc 2331
b0bde642 2332 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2333 new_ctype(newlocale);
b0bde642 2334 Safefree(newlocale);
a4f00dcc 2335
1159483a
KW
2336# endif /* USE_LOCALE_CTYPE */
2337# ifdef USE_LOCALE_COLLATE
2338
b0bde642 2339 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2340 new_collate(newlocale);
b0bde642 2341 Safefree(newlocale);
a4f00dcc 2342
7d4bcc4a 2343# endif
1159483a 2344# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2345
b0bde642 2346 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2347 new_numeric(newlocale);
b0bde642 2348 Safefree(newlocale);
a4f00dcc 2349
1159483a
KW
2350# endif /* USE_LOCALE_NUMERIC */
2351#endif /* LC_ALL */
a4f00dcc 2352
1159483a
KW
2353 default:
2354 break;
a4f00dcc
KW
2355 }
2356
2357 return retval;
2358
d36adde0
KW
2359#endif
2360
f7416781
KW
2361}
2362
2363PERL_STATIC_INLINE const char *
2364S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2365{
2366 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2367 * growing it if necessary */
2368
0b6697c5 2369 Size_t string_size;
f7416781
KW
2370
2371 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2372
0b6697c5
KW
2373 if (! string) {
2374 return NULL;
2375 }
2376
2377 string_size = strlen(string) + offset + 1;
2378
f7416781
KW
2379 if (*buf_size == 0) {
2380 Newx(*buf, string_size, char);
2381 *buf_size = string_size;
2382 }
2383 else if (string_size > *buf_size) {
2384 Renew(*buf, string_size, char);
2385 *buf_size = string_size;
2386 }
2387
2388 Copy(string, *buf + offset, string_size - offset, char);
2389 return *buf;
2390}
2391
2392/*
2393
f7416781
KW
2394=for apidoc Perl_langinfo
2395
ce181bf6 2396This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2397taking the same C<item> parameter values, and returning the same information.
2398But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2399of Perl's locale handling from your code, and can be used on systems that lack
2400a native C<nl_langinfo>.
2401
2402Expanding on these:
2403
2404=over
2405
2406=item *
2407
ce181bf6
KW
2408The reason it isn't quite a drop-in replacement is actually an advantage. The
2409only difference is that it returns S<C<const char *>>, whereas plain
2410C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2411forbidden to write into the buffer. By declaring this C<const>, the compiler
2412enforces this restriction, so if it is violated, you know at compilation time,
2413rather than getting segfaults at runtime.
2414
2415=item *
2416
69e2275e 2417It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2418without you having to write extra code. The reason for the extra code would be
2419because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2420kept set by Perl so that the radix is a dot, and the separator is the empty
2421string, no matter what the underlying locale is supposed to be, and so to get
2422the expected results, you have to temporarily toggle into the underlying
2423locale, and later toggle back. (You could use plain C<nl_langinfo> and
2424C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2425the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2426(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2427(decimal point) character to be a dot.)
ce181bf6
KW
2428
2429=item *
2430
2431The system function it replaces can have its static return buffer trashed,
2432not only by a subesequent call to that function, but by a C<freelocale>,
2433C<setlocale>, or other locale change. The returned buffer of this function is
2434not changed until the next call to it, so the buffer is never in a trashed
2435state.
f7416781
KW
2436
2437=item *
2438
ce181bf6
KW
2439Its return buffer is per-thread, so it also is never overwritten by a call to
2440this function from another thread; unlike the function it replaces.
2441
2442=item *
2443
2444But most importantly, it works on systems that don't have C<nl_langinfo>, such
2445as Windows, hence makes your code more portable. Of the fifty-some possible
2446items specified by the POSIX 2008 standard,
f7416781 2447L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2448only one is completely unimplemented, though on non-Windows platforms, another
2449significant one is also not implemented). It uses various techniques to
2450recover the other items, including calling C<L<localeconv(3)>>, and
2451C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2452available. Later C<strftime()> versions have additional capabilities; C<""> is
2453returned for those not available on your system.
ce181bf6
KW
2454
2455It is important to note that when called with an item that is recovered by
2456using C<localeconv>, the buffer from any previous explicit call to
2457C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2458contents if you need to access them after a call to this function. (But note
2459that you might not want to be using C<localeconv()> directly anyway, because of
2460issues like the ones listed in the second item of this list (above) for
2461C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2462call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2463unpack).
9f0bc911 2464
6201c220
KW
2465The details for those items which may deviate from what this emulation returns
2466and what a native C<nl_langinfo()> would return are specified in
2467L<I18N::Langinfo>.
f7416781 2468
ce181bf6
KW
2469=back
2470
f7416781
KW
2471When using C<Perl_langinfo> on systems that don't have a native
2472C<nl_langinfo()>, you must
2473
2474 #include "perl_langinfo.h"
2475
2476before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2477C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2478C<langinfo.h> would try to import into the namespace for code that doesn't need
2479it.)
f7416781 2480
f7416781
KW
2481The original impetus for C<Perl_langinfo()> was so that code that needs to
2482find out the current currency symbol, floating point radix character, or digit
2483grouping separator can use, on all systems, the simpler and more
2484thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2485pain to make thread-friendly. For other fields returned by C<localeconv>, it
2486is better to use the methods given in L<perlcall> to call
2487L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2488
2489=cut
2490
2491*/
2492
2493const char *
2494#ifdef HAS_NL_LANGINFO
2495Perl_langinfo(const nl_item item)
2496#else
2497Perl_langinfo(const int item)
2498#endif
2499{
f61748ac
KW
2500 return my_nl_langinfo(item, TRUE);
2501}
2502
59a15af0 2503STATIC const char *
f61748ac
KW
2504#ifdef HAS_NL_LANGINFO
2505S_my_nl_langinfo(const nl_item item, bool toggle)
2506#else
2507S_my_nl_langinfo(const int item, bool toggle)
2508#endif
2509{
ae74815b 2510 dTHX;
0b6697c5 2511 const char * retval;
f7416781 2512
d36adde0
KW
2513#ifdef USE_LOCALE_NUMERIC
2514
5a854ab3
KW
2515 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2516 * two items, and only if not already there */
4e6826bf 2517 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2518 || PL_numeric_underlying))
d36adde0
KW
2519
2520#endif /* No toggling needed if not using LC_NUMERIC */
2521
5a854ab3 2522 toggle = FALSE;
5a854ab3 2523
ab340fff 2524#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2525# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2526 || ! defined(HAS_POSIX_2008_LOCALE) \
2527 || ! defined(DUPLOCALE)
f7416781 2528
ab340fff 2529 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2530 * for those items dependent on it. This must be copied to a buffer before
2531 * switching back, as some systems destroy the buffer when setlocale() is
2532 * called */
f7416781 2533
038d3702
KW
2534 {
2535 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2536
b2fee59c 2537 if (toggle) {
038d3702 2538 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2539 }
f7416781 2540
5acc4454
KW
2541 LOCALE_LOCK; /* Prevent interference from another thread executing
2542 this code section (the only call to nl_langinfo in
2543 the core) */
2544
ee90eb2d 2545
628ff75a
KW
2546 /* Copy to a per-thread buffer, which is also one that won't be
2547 * destroyed by a subsequent setlocale(), such as the
2548 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2549 retval = save_to_buffer(nl_langinfo(item),
2550 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2551
5acc4454
KW
2552 LOCALE_UNLOCK;
2553
b2fee59c 2554 if (toggle) {
038d3702 2555 RESTORE_LC_NUMERIC();
b2fee59c 2556 }
f7416781
KW
2557 }
2558
ab340fff
KW
2559# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2560
5a854ab3 2561 {
b2fee59c
KW
2562 bool do_free = FALSE;
2563 locale_t cur = uselocale((locale_t) 0);
ab340fff 2564
b2fee59c
KW
2565 if (cur == LC_GLOBAL_LOCALE) {
2566 cur = duplocale(LC_GLOBAL_LOCALE);
2567 do_free = TRUE;
2568 }
ab340fff 2569
d36adde0
KW
2570# ifdef USE_LOCALE_NUMERIC
2571
b2fee59c 2572 if (toggle) {
e1aa2579
KW
2573 if (PL_underlying_numeric_obj) {
2574 cur = PL_underlying_numeric_obj;
2575 }
2576 else {
2577 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2578 do_free = TRUE;
2579 }
b2fee59c 2580 }
ab340fff 2581
d36adde0
KW
2582# endif
2583
628ff75a
KW
2584 /* We have to save it to a buffer, because the freelocale() just below
2585 * can invalidate the internal one */
0b6697c5
KW
2586 retval = save_to_buffer(nl_langinfo_l(item, cur),
2587 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2588
b2fee59c
KW
2589 if (do_free) {
2590 freelocale(cur);
2591 }
5a854ab3 2592 }
ab340fff 2593
c1566110
KW
2594# endif
2595
0b6697c5 2596 if (strEQ(retval, "")) {
4e6826bf 2597 if (item == YESSTR) {
c1566110
KW
2598 return "yes";
2599 }
4e6826bf 2600 if (item == NOSTR) {
c1566110
KW
2601 return "no";
2602 }
2603 }
2604
0b6697c5 2605 return retval;
ab340fff 2606
f7416781 2607#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2608
2609 {
2610
f7416781
KW
2611# ifdef HAS_LOCALECONV
2612
43dd6b15 2613 const struct lconv* lc;
628ff75a 2614 const char * temp;
038d3702 2615 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2616
69c5e0db
KW
2617# ifdef TS_W32_BROKEN_LOCALECONV
2618
2619 const char * save_global;
2620 const char * save_thread;
2621 int needed_size;
2622 char * ptr;
2623 char * e;
2624 char * item_start;
2625
2626# endif
f7416781
KW
2627# endif
2628# ifdef HAS_STRFTIME
2629
43dd6b15
KW
2630 struct tm tm;
2631 bool return_format = FALSE; /* Return the %format, not the value */
2632 const char * format;
f7416781
KW
2633
2634# endif
2635
43dd6b15
KW
2636 /* We copy the results to a per-thread buffer, even if not
2637 * multi-threaded. This is in part to simplify this code, and partly
2638 * because we need a buffer anyway for strftime(), and partly because a
2639 * call of localeconv() could otherwise wipe out the buffer, and the
2640 * programmer would not be expecting this, as this is a nl_langinfo()
2641 * substitute after all, so s/he might be thinking their localeconv()
2642 * is safe until another localeconv() call. */
f7416781 2643
43dd6b15
KW
2644 switch (item) {
2645 Size_t len;
f7416781 2646
8d72e74e 2647 /* This is unimplemented */
4e6826bf 2648 case ERA: /* For use with strftime() %E modifier */
f7416781 2649
43dd6b15
KW
2650 default:
2651 return "";
f7416781 2652
43dd6b15 2653 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2654 case YESEXPR: return "^[+1yY]";
2655 case YESSTR: return "yes";
2656 case NOEXPR: return "^[-0nN]";
2657 case NOSTR: return "no";
2658
8d72e74e
KW
2659 case CODESET:
2660
2661# ifndef WIN32
2662
2663 /* On non-windows, this is unimplemented, in part because of
2664 * inconsistencies between vendors. The Darwin native
2665 * nl_langinfo() implementation simply looks at everything past
2666 * any dot in the name, but that doesn't work for other
2667 * vendors. Many Linux locales that don't have UTF-8 in their
2668 * names really are UTF-8, for example; z/OS locales that do
2669 * have UTF-8 in their names, aren't really UTF-8 */
2670 return "";
2671
2672# else
2673
2674 { /* But on Windows, the name does seem to be consistent, so
2675 use that. */
2676 const char * p;
2677 const char * first;
2678 Size_t offset = 0;
2679 const char * name = my_setlocale(LC_CTYPE, NULL);
2680
2681 if (isNAME_C_OR_POSIX(name)) {
2682 return "ANSI_X3.4-1968";
2683 }
2684
2685 /* Find the dot in the locale name */
2686 first = (const char *) strchr(name, '.');
2687 if (! first) {
2688 first = name;
2689 goto has_nondigit;
2690 }
2691
2692 /* Look at everything past the dot */
2693 first++;
2694 p = first;
2695
2696 while (*p) {
2697 if (! isDIGIT(*p)) {
2698 goto has_nondigit;
2699 }
2700
2701 p++;
2702 }
2703
2704 /* Here everything past the dot is a digit. Treat it as a
2705 * code page */
18503cbd 2706 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2707 &PL_langinfo_bufsize, 0);
8d72e74e 2708 offset = STRLENs("CP");
f7416781 2709
8d72e74e
KW
2710 has_nondigit:
2711
2712 retval = save_to_buffer(first, &PL_langinfo_buf,
2713 &PL_langinfo_bufsize, offset);
2714 }
2715
2716 break;
2717
2718# endif
f7416781
KW
2719# ifdef HAS_LOCALECONV
2720
4e6826bf 2721 case CRNCYSTR:
f7416781 2722
43dd6b15
KW
2723 /* We don't bother with localeconv_l() because any system that
2724 * has it is likely to also have nl_langinfo() */
291a84fb 2725
69c5e0db
KW
2726 LOCALE_LOCK_V; /* Prevent interference with other threads
2727 using localeconv() */
2728
2729# ifdef TS_W32_BROKEN_LOCALECONV
2730
2731 /* This is a workaround for a Windows bug prior to VS 15.
2732 * What we do here is, while locked, switch to the global
2733 * locale so localeconv() works; then switch back just before
2734 * the unlock. This can screw things up if some thread is
2735 * already using the global locale while assuming no other is.
2736 * A different workaround would be to call GetCurrencyFormat on
2737 * a known value, and parse it; patches welcome
2738 *
2739 * We have to use LC_ALL instead of LC_MONETARY because of
2740 * another bug in Windows */
2741
2742 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2743 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2744 save_global= savepv(my_setlocale(LC_ALL, NULL));
2745 my_setlocale(LC_ALL, save_thread);
2746
2747# endif
5acc4454 2748
43dd6b15
KW
2749 lc = localeconv();
2750 if ( ! lc
2751 || ! lc->currency_symbol
2752 || strEQ("", lc->currency_symbol))
2753 {
69c5e0db 2754 LOCALE_UNLOCK_V;
43dd6b15
KW
2755 return "";
2756 }
f7416781 2757
43dd6b15 2758 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2759 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2760 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2761 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2762 { /* khw couldn't figure out how the localedef specifications
2763 would show that the $ should replace the radix; this is
2764 just a guess as to how it might work.*/
a34edee3 2765 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2766 }
2767 else if (lc->p_cs_precedes) {
a34edee3 2768 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2769 }
2770 else {
a34edee3 2771 PL_langinfo_buf[0] = '+';
43dd6b15 2772 }
f7416781 2773
69c5e0db
KW
2774# ifdef TS_W32_BROKEN_LOCALECONV
2775
2776 my_setlocale(LC_ALL, save_global);
2777 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2778 my_setlocale(LC_ALL, save_thread);
2779 Safefree(save_global);
2780 Safefree(save_thread);
2781
2782# endif
2783
2784 LOCALE_UNLOCK_V;
43dd6b15 2785 break;
f7416781 2786
69c5e0db
KW
2787# ifdef TS_W32_BROKEN_LOCALECONV
2788
4e6826bf 2789 case RADIXCHAR:
69c5e0db
KW
2790
2791 /* For this, we output a known simple floating point number to
2792 * a buffer, and parse it, looking for the radix */
2793
2794 if (toggle) {
2795 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2796 }
2797
2798 if (PL_langinfo_bufsize < 10) {
2799 PL_langinfo_bufsize = 10;
2800 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2801 }
2802
2803 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2804 "%.1f", 1.5);
2805 if (needed_size >= (int) PL_langinfo_bufsize) {
2806 PL_langinfo_bufsize = needed_size + 1;
2807 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2808 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2809 "%.1f", 1.5);
2810 assert(needed_size < (int) PL_langinfo_bufsize);
2811 }
2812
2813 ptr = PL_langinfo_buf;
2814 e = PL_langinfo_buf + PL_langinfo_bufsize;
2815
2816 /* Find the '1' */
2817 while (ptr < e && *ptr != '1') {
2818 ptr++;
2819 }
2820 ptr++;
2821
2822 /* Find the '5' */
2823 item_start = ptr;
2824 while (ptr < e && *ptr != '5') {
2825 ptr++;
2826 }
2827
2828 /* Everything in between is the radix string */
2829 if (ptr >= e) {
2830 PL_langinfo_buf[0] = '?';
2831 PL_langinfo_buf[1] = '\0';
2832 }
2833 else {
2834 *ptr = '\0';
2835 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2836 }
2837
2838 if (toggle) {
2839 RESTORE_LC_NUMERIC();
2840 }
2841
2842 retval = PL_langinfo_buf;
2843 break;
2844
2845# else
2846
2847 case RADIXCHAR: /* No special handling needed */
2848
2849# endif
2850
4e6826bf 2851 case THOUSEP:
f7416781 2852
43dd6b15 2853 if (toggle) {
038d3702 2854 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2855 }
f7416781 2856
69c5e0db
KW
2857 LOCALE_LOCK_V; /* Prevent interference with other threads
2858 using localeconv() */
2859
2860# ifdef TS_W32_BROKEN_LOCALECONV
2861
2862 /* This should only be for the thousands separator. A
2863 * different work around would be to use GetNumberFormat on a
2864 * known value and parse the result to find the separator */
2865 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2866 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2867 save_global = savepv(my_setlocale(LC_ALL, NULL));
2868 my_setlocale(LC_ALL, save_thread);
2869# if 0
2870 /* This is the start of code that for broken Windows replaces
2871 * the above and below code, and instead calls
2872 * GetNumberFormat() and then would parse that to find the
2873 * thousands separator. It needs to handle UTF-16 vs -8
2874 * issues. */
2875
2876 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2877 DEBUG_L(PerlIO_printf(Perl_debug_log,
2878 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2879 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2880
2881# endif
2882# endif
5acc4454 2883
43dd6b15
KW
2884 lc = localeconv();
2885 if (! lc) {
628ff75a 2886 temp = "";
33394adc 2887 }
43dd6b15 2888 else {
4e6826bf 2889 temp = (item == RADIXCHAR)
43dd6b15
KW
2890 ? lc->decimal_point
2891 : lc->thousands_sep;
628ff75a
KW
2892 if (! temp) {
2893 temp = "";
43dd6b15
KW
2894 }
2895 }
f7416781 2896
0b6697c5
KW
2897 retval = save_to_buffer(temp, &PL_langinfo_buf,
2898 &PL_langinfo_bufsize, 0);
f7416781 2899
69c5e0db
KW
2900# ifdef TS_W32_BROKEN_LOCALECONV
2901
2902 my_setlocale(LC_ALL, save_global);
2903 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2904 my_setlocale(LC_ALL, save_thread);
2905 Safefree(save_global);
2906 Safefree(save_thread);
2907
2908# endif
2909
2910 LOCALE_UNLOCK_V;
5acc4454 2911
43dd6b15 2912 if (toggle) {
038d3702 2913 RESTORE_LC_NUMERIC();
43dd6b15 2914 }
f7416781 2915
43dd6b15 2916 break;
f7416781
KW
2917
2918# endif
2919# ifdef HAS_STRFTIME
2920
43dd6b15
KW
2921 /* These are defined by C89, so we assume that strftime supports
2922 * them, and so are returned unconditionally; they may not be what
2923 * the locale actually says, but should give good enough results
2924 * for someone using them as formats (as opposed to trying to parse
2925 * them to figure out what the locale says). The other format
2926 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2927 case D_FMT: return "%x";
2928 case T_FMT: return "%X";
2929 case D_T_FMT: return "%c";
43dd6b15
KW
2930
2931 /* These formats are only available in later strfmtime's */
4e6826bf 2932 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2933
2934 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2935 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2936 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2937 case ALT_DIGITS:
2938 case AM_STR: case PM_STR:
2939 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2940 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2941 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2942 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2943 case DAY_5: case DAY_6: case DAY_7:
2944 case MON_1: case MON_2: case MON_3: case MON_4:
2945 case MON_5: case MON_6: case MON_7: case MON_8:
2946 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2947
2948 LOCALE_LOCK;
2949
2950 init_tm(&tm); /* Precaution against core dumps */
2951 tm.tm_sec = 30;
2952 tm.tm_min = 30;
2953 tm.tm_hour = 6;
2954 tm.tm_year = 2017 - 1900;
2955 tm.tm_wday = 0;
2956 tm.tm_mon = 0;
2957 switch (item) {
2958 default:
2959 LOCALE_UNLOCK;
2960 Perl_croak(aTHX_
2961 "panic: %s: %d: switch case: %d problem",
2962 __FILE__, __LINE__, item);
2963 NOT_REACHED; /* NOTREACHED */
2964
4e6826bf
KW
2965 case PM_STR: tm.tm_hour = 18;
2966 case AM_STR:
43dd6b15
KW
2967 format = "%p";
2968 break;
2969
4e6826bf
KW
2970 case ABDAY_7: tm.tm_wday++;
2971 case ABDAY_6: tm.tm_wday++;
2972 case ABDAY_5: tm.tm_wday++;
2973 case ABDAY_4: tm.tm_wday++;
2974 case ABDAY_3: tm.tm_wday++;
2975 case ABDAY_2: tm.tm_wday++;
2976 case ABDAY_1:
43dd6b15
KW
2977 format = "%a";
2978 break;
2979
4e6826bf
KW
2980 case DAY_7: tm.tm_wday++;
2981 case DAY_6: tm.tm_wday++;
2982 case DAY_5: tm.tm_wday++;
2983 case DAY_4: tm.tm_wday++;
2984 case DAY_3: tm.tm_wday++;
2985 case DAY_2: tm.tm_wday++;
2986 case DAY_1:
43dd6b15
KW
2987 format = "%A";
2988 break;
2989
4e6826bf
KW
2990 case ABMON_12: tm.tm_mon++;
2991 case ABMON_11: tm.tm_mon++;
2992 case ABMON_10: tm.tm_mon++;
2993 case ABMON_9: tm.tm_mon++;
2994 case ABMON_8: tm.tm_mon++;
2995 case ABMON_7: tm.tm_mon++;
2996 case ABMON_6: tm.tm_mon++;
2997 case ABMON_5: tm.tm_mon++;
2998 case ABMON_4: tm.tm_mon++;
2999 case ABMON_3: tm.tm_mon++;
3000 case ABMON_2: tm.tm_mon++;
3001 case ABMON_1:
43dd6b15
KW
3002 format = "%b";
3003 break;
3004
4e6826bf
KW
3005 case MON_12: tm.tm_mon++;
3006 case MON_11: tm.tm_mon++;
3007 case MON_10: tm.tm_mon++;
3008 case MON_9: tm.tm_mon++;
3009 case MON_8: tm.tm_mon++;
3010 case MON_7: tm.tm_mon++;
3011 case MON_6: tm.tm_mon++;
3012 case MON_5: tm.tm_mon++;
3013 case MON_4: tm.tm_mon++;
3014 case MON_3: tm.tm_mon++;
3015 case MON_2: tm.tm_mon++;
3016 case MON_1:
43dd6b15
KW
3017 format = "%B";
3018 break;
3019
4e6826bf 3020 case T_FMT_AMPM:
43dd6b15
KW
3021 format = "%r";
3022 return_format = TRUE;
3023 break;
3024
4e6826bf 3025 case ERA_D_FMT:
43dd6b15
KW
3026 format = "%Ex";
3027 return_format = TRUE;
3028 break;
3029
4e6826bf 3030 case ERA_T_FMT:
43dd6b15
KW
3031 format = "%EX";
3032 return_format = TRUE;
3033 break;
3034
4e6826bf 3035 case ERA_D_T_FMT:
43dd6b15
KW
3036 format = "%Ec";
3037 return_format = TRUE;
3038 break;
3039
4e6826bf 3040 case ALT_DIGITS:
43dd6b15
KW
3041 tm.tm_wday = 0;
3042 format = "%Ow"; /* Find the alternate digit for 0 */
3043 break;
3044 }
f7416781 3045
43dd6b15
KW
3046 /* We can't use my_strftime() because it doesn't look at
3047 * tm_wday */
3048 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3049 format, &tm))
3050 {
3051 /* A zero return means one of:
3052 * a) there wasn't enough space in PL_langinfo_buf
3053 * b) the format, like a plain %p, returns empty
3054 * c) it was an illegal format, though some
3055 * implementations of strftime will just return the
3056 * illegal format as a plain character sequence.
3057 *
3058 * To quickly test for case 'b)', try again but precede
3059 * the format with a plain character. If that result is
3060 * still empty, the problem is either 'a)' or 'c)' */
3061
3062 Size_t format_size = strlen(format) + 1;
3063 Size_t mod_size = format_size + 1;
3064 char * mod_format;
3065 char * temp_result;
3066
3067 Newx(mod_format, mod_size, char);
3068 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 3069 *mod_format = ' ';
43dd6b15
KW
3070 my_strlcpy(mod_format + 1, format, mod_size);
3071 len = strftime(temp_result,
3072 PL_langinfo_bufsize,
3073 mod_format, &tm);
3074 Safefree(mod_format);
3075 Safefree(temp_result);
3076
3077 /* If 'len' is non-zero, it means that we had a case like
3078 * %p which means the current locale doesn't use a.m. or
3079 * p.m., and that is valid */
3080 if (len == 0) {
3081
3082 /* Here, still didn't work. If we get well beyond a
3083 * reasonable size, bail out to prevent an infinite
3084 * loop. */
3085
3086 if (PL_langinfo_bufsize > 100 * format_size) {
3087 *PL_langinfo_buf = '\0';
3088 }
3089 else {
3090 /* Double the buffer size to retry; Add 1 in case
3091 * original was 0, so we aren't stuck at 0. */
3092 PL_langinfo_bufsize *= 2;
3093 PL_langinfo_bufsize++;
3094 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3095 continue;
3096 }
3097 }
f7416781 3098
f7416781 3099 break;
43dd6b15 3100 }
f7416781 3101
43dd6b15
KW
3102 /* Here, we got a result.
3103 *
3104 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3105 * alternate format for wday 0. If the value is the same as
3106 * the normal 0, there isn't an alternate, so clear the buffer.
3107 * */
4e6826bf 3108 if ( item == ALT_DIGITS
43dd6b15
KW
3109 && strEQ(PL_langinfo_buf, "0"))
3110 {
3111 *PL_langinfo_buf = '\0';
3112 }
f7416781 3113
43dd6b15
KW
3114 /* ALT_DIGITS is problematic. Experiments on it showed that
3115 * strftime() did not always work properly when going from
3116 * alt-9 to alt-10. Only a few locales have this item defined,
3117 * and in all of them on Linux that khw was able to find,
3118 * nl_langinfo() merely returned the alt-0 character, possibly
3119 * doubled. Most Unicode digits are in blocks of 10
3120 * consecutive code points, so that is sufficient information
3121 * for those scripts, as we can infer alt-1, alt-2, .... But
3122 * for a Japanese locale, a CJK ideographic 0 is returned, and
3123 * the CJK digits are not in code point order, so you can't
3124 * really infer anything. The localedef for this locale did
3125 * specify the succeeding digits, so that strftime() works
3126 * properly on them, without needing to infer anything. But
3127 * the nl_langinfo() return did not give sufficient information
3128 * for the caller to understand what's going on. So until
3129 * there is evidence that it should work differently, this
3130 * returns the alt-0 string for ALT_DIGITS.
3131 *
3132 * wday was chosen because its range is all a single digit.
3133 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3134
43dd6b15 3135 LOCALE_UNLOCK;
f7416781 3136
d1b150d9
KW
3137 retval = PL_langinfo_buf;
3138
43dd6b15
KW
3139 /* If to return the format, not the value, overwrite the buffer
3140 * with it. But some strftime()s will keep the original format
3141 * if illegal, so change those to "" */
3142 if (return_format) {
3143 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3144 *PL_langinfo_buf = '\0';
3145 }
43dd6b15 3146 else {
0b6697c5
KW
3147 retval = save_to_buffer(format, &PL_langinfo_buf,
3148 &PL_langinfo_bufsize, 0);
f7416781
KW
3149 }
3150 }
3151
3152 break;
f7416781
KW
3153
3154# endif
3155
43dd6b15 3156 }
f7416781
KW
3157 }
3158
0b6697c5 3159 return retval;
f7416781
KW
3160
3161#endif
3162
a4f00dcc 3163}
b385bb4d 3164
98994639
HS
3165/*
3166 * Initialize locale awareness.
3167 */
3168int
3169Perl_init_i18nl10n(pTHX_ int printwarn)
3170{
0e92a118
KW
3171 /* printwarn is
3172 *
3173 * 0 if not to output warning when setup locale is bad
3174 * 1 if to output warning based on value of PERL_BADLANG
3175 * >1 if to output regardless of PERL_BADLANG
3176 *
3177 * returns
98994639 3178 * 1 = set ok or not applicable,
0e92a118
KW
3179 * 0 = fallback to a locale of lower priority
3180 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3181 *
3182 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3183 * set, debugging information is output.
3184 *
3185 * This looks more complicated than it is, mainly due to the #ifdefs.
3186 *
3187 * We try to set LC_ALL to the value determined by the environment. If
3188 * there is no LC_ALL on this platform, we try the individual categories we
3189 * know about. If this works, we are done.
3190 *
3191 * But if it doesn't work, we have to do something else. We search the
3192 * environment variables ourselves instead of relying on the system to do
3193 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3194 * think there is one), and the ultimate fallback "C". This is all done in
3195 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3196 * more complex. The 'trial_locales' array is initialized with just one
3197 * element; it causes the behavior described in the paragraph above this to
3198 * happen. If that fails, we add elements to 'trial_locales', and do extra
3199 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3200 *
3201 * On Ultrix, the locale MUST come from the environment, so there is
3202 * preliminary code to set it. I (khw) am not sure that it is necessary,
3203 * and that this couldn't be folded into the loop, but barring any real
3204 * platforms to test on, it's staying as-is
3205 *
3206 * A slight complication is that in embedded Perls, the locale may already
3207 * be set-up, and we don't want to get it from the normal environment
3208 * variables. This is handled by having a special environment variable
3209 * indicate we're in this situation. We simply set setlocale's 2nd
3210 * parameter to be a NULL instead of "". That indicates to setlocale that
3211 * it is not to change anything, but to return the current value,
3212 * effectively initializing perl's db to what the locale already is.
3213 *
3214 * We play the same trick with NULL if a LC_ALL succeeds. We call
3215 * setlocale() on the individual categores with NULL to get their existing
3216 * values for our db, instead of trying to change them.
3217 * */
98994639 3218
1565c085
DM
3219 dVAR;
3220
0e92a118
KW
3221 int ok = 1;
3222
7d4bcc4a
KW
3223#ifndef USE_LOCALE
3224
3225 PERL_UNUSED_ARG(printwarn);
3226
3227#else /* USE_LOCALE */
7d4bcc4a
KW
3228# ifdef __GLIBC__
3229
175c4cf9 3230 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3231
3232# endif
65ebb059 3233
ccd65d51
KW
3234 /* NULL uses the existing already set up locale */
3235 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3236 ? NULL
3237 : "";
c3fcd832
KW
3238 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3239 unsigned int trial_locales_count;
175c4cf9
KW
3240 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3241 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3242 bool setlocale_failure = FALSE;
65ebb059 3243 unsigned int i;
175c4cf9
KW
3244
3245 /* A later getenv() could zap this, so only use here */
3246 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3247
3248 const bool locwarn = (printwarn > 1
e5f10d49
KW
3249 || ( printwarn
3250 && ( ! bad_lang_use_once
22ff3130 3251 || (
e5f10d49
KW
3252 /* disallow with "" or "0" */
3253 *bad_lang_use_once
3254 && strNE("0", bad_lang_use_once)))));
ea92aad8 3255
291a84fb 3256 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3257 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3258
3259 /* current locale for given category; should have been copied so aren't
3260 * volatile */
8de4332b 3261 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3262
7d4bcc4a
KW
3263# ifdef WIN32
3264
6bce99ee
JH
3265 /* In some systems you can find out the system default locale
3266 * and use that as the fallback locale. */
7d4bcc4a
KW
3267# define SYSTEM_DEFAULT_LOCALE
3268# endif
3269# ifdef SYSTEM_DEFAULT_LOCALE
3270
65ebb059 3271 const char *system_default_locale = NULL;
98994639 3272
7d4bcc4a 3273# endif
948523db
KW
3274
3275# ifndef DEBUGGING
3276# define DEBUG_LOCALE_INIT(a,b,c)
3277# else
7d4bcc4a 3278
8298454c 3279 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3280
3281# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3282 STMT_START { \
3283 if (debug_initialization) { \
3284 PerlIO_printf(Perl_debug_log, \
3285 "%s:%d: %s\n", \
3286 __FILE__, __LINE__, \
a4f00dcc 3287 setlocale_debug_string(category, \
2fcc0ca9
KW
3288 locale, \
3289 result)); \
3290 } \
3291 } STMT_END
2fcc0ca9 3292
948523db
KW
3293/* Make sure the parallel arrays are properly set up */
3294# ifdef USE_LOCALE_NUMERIC
3295 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3296 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3297# ifdef USE_POSIX_2008_LOCALE
3298 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3299# endif
948523db
KW
3300# endif
3301# ifdef USE_LOCALE_CTYPE
3302 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3303 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3304# ifdef USE_POSIX_2008_LOCALE
3305 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3306# endif
948523db
KW
3307# endif
3308# ifdef USE_LOCALE_COLLATE
3309 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3310 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3311# ifdef USE_POSIX_2008_LOCALE
3312 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3313# endif
948523db
KW
3314# endif
3315# ifdef USE_LOCALE_TIME
3316 assert(categories[LC_TIME_INDEX] == LC_TIME);
3317 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3318# ifdef USE_POSIX_2008_LOCALE
3319 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3320# endif
948523db
KW
3321# endif
3322# ifdef USE_LOCALE_MESSAGES
3323 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3324 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3325# ifdef USE_POSIX_2008_LOCALE
3326 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3327# endif
948523db
KW
3328# endif
3329# ifdef USE_LOCALE_MONETARY
3330 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3331 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3332# ifdef USE_POSIX_2008_LOCALE
3333 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3334# endif
948523db 3335# endif
9821811f
KW
3336# ifdef USE_LOCALE_ADDRESS
3337 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3338 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3339# ifdef USE_POSIX_2008_LOCALE
3340 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3341# endif
9821811f
KW
3342# endif
3343# ifdef USE_LOCALE_IDENTIFICATION
3344 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3345 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3346# ifdef USE_POSIX_2008_LOCALE
3347 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3348# endif
9821811f
KW
3349# endif
3350# ifdef USE_LOCALE_MEASUREMENT
3351 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3352 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3353# ifdef USE_POSIX_2008_LOCALE
3354 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3355# endif
9821811f
KW
3356# endif
3357# ifdef USE_LOCALE_PAPER
3358 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3359 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3360# ifdef USE_POSIX_2008_LOCALE
3361 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3362# endif
9821811f
KW
3363# endif
3364# ifdef USE_LOCALE_TELEPHONE
3365 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3366 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3367# ifdef USE_POSIX_2008_LOCALE
3368 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3369# endif
9821811f 3370# endif
948523db
KW
3371# ifdef LC_ALL
3372 assert(categories[LC_ALL_INDEX] == LC_ALL);
3373 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3374 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3375# ifdef USE_POSIX_2008_LOCALE
3376 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3377# endif
948523db
KW
3378# endif
3379# endif /* DEBUGGING */
47280b20
KW
3380
3381 /* Initialize the cache of the program's UTF-8ness for the always known
3382 * locales C and POSIX */
3383 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3384 sizeof(PL_locale_utf8ness));
3385
e9bc6d6b
KW
3386# ifdef USE_THREAD_SAFE_LOCALE
3387# ifdef WIN32
3388
3389 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3390
3391# endif
3392# endif
39e69e77 3393# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3394
3395 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3396 if (! PL_C_locale_obj) {
3397 Perl_croak_nocontext(
3398 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3399 }
3400 if (DEBUG_Lv_TEST || debug_initialization) {
3401 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3402 }
3403
3404# endif
3405
78f7c09f
KW
3406# ifdef USE_LOCALE_NUMERIC
3407
3ca88433
KW
3408 PL_numeric_radix_sv = newSVpvs(".");
3409
78f7c09f
KW
3410# endif
3411
e9bc6d6b
KW
3412# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3413
3414 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3415 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3416
3417# endif
b56c4436 3418# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3419
3420 /*
3421 * Ultrix setlocale(..., "") fails if there are no environment
3422 * variables from which to get a locale name.
3423 */
3424
b56c4436
KW
3425# ifndef LC_ALL
3426# error Ultrix without LC_ALL not implemented
3427# else
7d4bcc4a 3428
b56c4436
KW
3429 {
3430 bool done = FALSE;
ea92aad8
KW
3431 if (lang) {
3432 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3433 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3434 if (sl_result[LC_ALL_INDEX])
3435 done = TRUE;
3436 else
e5f10d49 3437 setlocale_failure = TRUE;
ea92aad8
KW
3438 }
3439 if (! setlocale_failure) {
3440 const char * locale_param;
3441 for (i = 0; i < LC_ALL_INDEX; i++) {
3442 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3443 ? setlocale_init
3444 : NULL;
3445 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3446 if (! sl_result[i]) {
3447 setlocale_failure = TRUE;
3448 }
3449 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3450 }
c835d6be 3451 }
7d4bcc4a
KW
3452 }
3453
3454# endif /* LC_ALL */
e5f10d49 3455# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3456
65ebb059 3457 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3458 * the list. Normally the loop is executed just once. But if setting the
3459 * locale fails, inside the loop we add fallback trials to the array and so
3460 * will execute the loop multiple times */
c3fcd832
KW
3461 trial_locales[0] = setlocale_init;
3462 trial_locales_count = 1;
7d4bcc4a 3463
65ebb059
KW
3464 for (i= 0; i < trial_locales_count; i++) {
3465 const char * trial_locale = trial_locales[i];
3466
3467 if (i > 0) {
3468
3469 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3470 * when i==0, but I (khw) don't think that behavior makes much
3471 * sense */
3472 setlocale_failure = FALSE;
3473
7d4bcc4a 3474# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3475# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3476
65ebb059
KW
3477 /* On Windows machines, an entry of "" after the 0th means to use
3478 * the system default locale, which we now proceed to get. */
3479 if (strEQ(trial_locale, "")) {
3480 unsigned int j;
3481
3482 /* Note that this may change the locale, but we are going to do
3483 * that anyway just below */
837ce802 3484 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3485 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3486
7d4bcc4a 3487 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3488 * try */
3489 if (! system_default_locale) {
3490 goto next_iteration;
3491 }
3492 for (j = 0; j < trial_locales_count; j++) {
3493 if (strEQ(system_default_locale, trial_locales[j])) {
3494 goto next_iteration;
3495 }
3496 }
3497
3498 trial_locale = system_default_locale;
3499 }
ec0202b5
KW
3500# else
3501# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3502# endif
7d4bcc4a 3503# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3504
3505 } /* For i > 0 */
65ebb059 3506
7d4bcc4a
KW
3507# ifdef LC_ALL
3508
948523db
KW
3509 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3510 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3511 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3512 setlocale_failure = TRUE;
7cd8b568
KW
3513 }
3514 else {
3515 /* Since LC_ALL succeeded, it should have changed all the other
3516 * categories it can to its value; so we massage things so that the
3517 * setlocales below just return their category's current values.
3518 * This adequately handles the case in NetBSD where LC_COLLATE may
3519 * not be defined for a locale, and setting it individually will
7d4bcc4a 3520 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3521 * the POSIX locale. */
3522 trial_locale = NULL;
3523 }
7d4bcc4a
KW
3524
3525# endif /* LC_ALL */
98994639 3526
e5f10d49
KW
3527 if (! setlocale_failure) {
3528 unsigned int j;
3529 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3530 curlocales[j]
3531 = savepv(do_setlocale_r(categories[j], trial_locale));
3532 if (! curlocales[j]) {
3533 setlocale_failure = TRUE;
3534 }
3535 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3536 }
c835d6be 3537
e5f10d49
KW
3538 if (! setlocale_failure) { /* All succeeded */
3539 break; /* Exit trial_locales loop */
49c85077 3540 }
65ebb059 3541 }
98994639 3542
49c85077
KW
3543 /* Here, something failed; will need to try a fallback. */
3544 ok = 0;
65ebb059 3545
49c85077
KW
3546 if (i == 0) {
3547 unsigned int j;
98994639 3548
65ebb059 3549 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3550
3551# ifdef LC_ALL
98994639 3552
49c85077
KW
3553 PerlIO_printf(Perl_error_log,
3554 "perl: warning: Setting locale failed.\n");
98994639 3555
7d4bcc4a 3556# else /* !LC_ALL */
98994639 3557
49c85077
KW
3558 PerlIO_printf(Perl_error_log,
3559 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3560
e5f10d49
KW
3561 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3562 if (! curlocales[j]) {
3563 PerlIO_printf(Perl_error_log, category_names[j]);
3564 }
3565 else {
3566 Safefree(curlocales[j]);
3567 }
3568 }
7d4bcc4a 3569
7d4bcc4a 3570# endif /* LC_ALL */
98994639 3571
49c85077
KW
3572 PerlIO_printf(Perl_error_log,
3573 "perl: warning: Please check that your locale settings:\n");
98994639 3574
7d4bcc4a
KW
3575# ifdef __GLIBC__
3576
49c85077
KW
3577 PerlIO_printf(Perl_error_log,
3578 "\tLANGUAGE = %c%s%c,\n",
3579 language ? '"' : '(',
3580 language ? language : "unset",
3581 language ? '"' : ')');
7d4bcc4a 3582# endif
98994639 3583
49c85077
KW
3584 PerlIO_printf(Perl_error_log,
3585 "\tLC_ALL = %c%s%c,\n",
3586 lc_all ? '"' : '(',
3587 lc_all ? lc_all : "unset",
3588 lc_all ? '"' : ')');
98994639 3589
7d4bcc4a
KW
3590# if defined(USE_ENVIRON_ARRAY)
3591
49c85077 3592 {
cd999af9 3593 char **e;
d5e32b93
KW
3594
3595 /* Look through the environment for any variables of the
3596 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3597 * already handled above. These are assumed to be locale
3598 * settings. Output them and their values. */
cd999af9 3599 for (e = environ; *e; e++) {
d5e32b93
KW
3600 const STRLEN prefix_len = sizeof("LC_") - 1;
3601 STRLEN uppers_len;
3602
cd999af9 3603 if ( strBEGINs(*e, "LC_")
c8b388b0 3604 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3605 && (uppers_len = strspn(*e + prefix_len,
3606 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3607 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3608 {
3609 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3610 (int) (prefix_len + uppers_len), *e,
3611 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3612 }
3613 }
49c85077 3614 }
7d4bcc4a
KW
3615
3616# else
3617
49c85077
KW
3618 PerlIO_printf(Perl_error_log,
3619 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3620
3621# endif
98994639 3622
49c85077
KW
3623 PerlIO_printf(Perl_error_log,
3624 "\tLANG = %c%s%c\n",
3625 lang ? '"' : '(',
3626 lang ? lang : "unset",
3627 lang ? '"' : ')');
98994639 3628
49c85077
KW
3629 PerlIO_printf(Perl_error_log,
3630 " are supported and installed on your system.\n");
3631 }
98994639 3632
65ebb059 3633 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3634 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3635 * usually change the upper bound of the loop we are in.
3636 *
3637 * Since the system's default way of setting the locale has not
3638 * found one that works, We use Perl's defined ordering: LC_ALL,
3639 * LANG, and the C locale. We don't try the same locale twice, so
3640 * don't add to the list if already there. (On POSIX systems, the
3641 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3642 * but there's no harm done by doing it explicitly.
3643 *
3644 * Note that this tries the LC_ALL environment variable even on
3645 * systems which have no LC_ALL locale setting. This may or may
3646 * not have been originally intentional, but there's no real need
3647 * to change the behavior. */
65ebb059
KW
3648 if (lc_all) {
3649 for (j = 0; j < trial_locales_count; j++) {
3650 if (strEQ(lc_all, trial_locales[j])) {
3651 goto done_lc_all;
3652 }
3653 }
3654 trial_locales[trial_locales_count++] = lc_all;
3655 }
3656 done_lc_all:
98994639 3657
65ebb059
KW
3658 if (lang) {
3659 for (j = 0; j < trial_locales_count; j++) {
3660 if (strEQ(lang, trial_locales[j])) {
3661 goto done_lang;
3662 }
3663 }
3664 trial_locales[trial_locales_count++] = lang;
3665 }
3666 done_lang:
3667
7d4bcc4a
KW
3668# if defined(WIN32) && defined(LC_ALL)
3669
65ebb059
KW
3670 /* For Windows, we also try the system default locale before "C".
3671 * (If there exists a Windows without LC_ALL we skip this because
3672 * it gets too complicated. For those, the "C" is the next
3673 * fallback possibility). The "" is the same as the 0th element of
3674 * the array, but the code at the loop above knows to treat it
3675 * differently when not the 0th */
3676 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3677
3678# endif
65ebb059
KW
3679
3680 for (j = 0; j < trial_locales_count; j++) {
3681 if (strEQ("C", trial_locales[j])) {
3682 goto done_C;
3683 }
3684 }
3685 trial_locales[trial_locales_count++] = "C";
98994639 3686
65ebb059
KW
3687 done_C: ;
3688 } /* end of first time through the loop */
98994639 3689
7d4bcc4a
KW
3690# ifdef WIN32
3691
65ebb059 3692 next_iteration: ;
7d4bcc4a
KW
3693
3694# endif
65ebb059
KW
3695
3696 } /* end of looping through the trial locales */
3697
3698 if (ok < 1) { /* If we tried to fallback */
3699 const char* msg;
3700 if (! setlocale_failure) { /* fallback succeeded */
3701 msg = "Falling back to";
3702 }
3703 else { /* fallback failed */
e5f10d49 3704 unsigned int j;
98994639 3705
65ebb059
KW
3706 /* We dropped off the end of the loop, so have to decrement i to
3707 * get back to the value the last time through */
3708 i--;
98994639 3709
65ebb059
KW
3710 ok = -1;
3711 msg = "Failed to fall back to";
3712
3713 /* To continue, we should use whatever values we've got */
7d4bcc4a 3714
e5f10d49
KW
3715 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3716 Safefree(curlocales[j]);
3717 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3718 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3719 }
65ebb059
KW
3720 }
3721
3722 if (locwarn) {
3723 const char * description;
3724 const char * name = "";
3725 if (strEQ(trial_locales[i], "C")) {
3726 description = "the standard locale";
3727 name = "C";
3728 }
7d4bcc4a
KW
3729
3730# ifdef SYSTEM_DEFAULT_LOCALE
3731
65ebb059
KW
3732 else if (strEQ(trial_locales[i], "")) {
3733 description = "the system default locale";
3734 if (system_default_locale) {
3735 name = system_default_locale;
3736 }
3737 }
7d4bcc4a
KW
3738
3739# endif /* SYSTEM_DEFAULT_LOCALE */
3740
65ebb059
KW
3741 else {
3742 description = "a fallback locale";
3743 name = trial_locales[i];
3744 }
3745 if (name && strNE(name, "")) {
3746 PerlIO_printf(Perl_error_log,
3747 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3748 }
3749 else {
3750 PerlIO_printf(Perl_error_log,
3751 "perl: warning: %s %s.\n", msg, description);
3752 }
3753 }
3754 } /* End of tried to fallback */
98994639 3755
e5f10d49
KW
3756 /* Done with finding the locales; update our records */
3757
7d4bcc4a
KW
3758# ifdef USE_LOCALE_CTYPE
3759
948523db 3760 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3761
e5f10d49 3762# endif
7d4bcc4a
KW
3763# ifdef USE_LOCALE_COLLATE
3764
948523db 3765 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3766
e5f10d49 3767# endif
7d4bcc4a
KW
3768# ifdef USE_LOCALE_NUMERIC
3769
948523db 3770 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3771
3772# endif
3773
948523db 3774 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3775
e9bc6d6b 3776# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3777
3778 /* This caches whether each category's locale is UTF-8 or not. This
3779 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3780 * initialization time before any threads have started, but not later
3781 * unless thread-safe operations are used.
47280b20
KW
3782 * Caching means that if the program heeds our dictate not to change
3783 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3784 * it may get queried without having to change locales. If the
3785 * environment is such that all categories have the same locale, this
3786 * isn't needed, as the code will not change the locale; but this
3787 * handles the uncommon case where the environment has disparate
3788 * locales for the categories */
47280b20
KW
3789 (void) _is_cur_LC_category_utf8(categories[i]);
3790
3791# endif
3792
e5f10d49
KW
3793 Safefree(curlocales[i]);
3794 }
b310b053 3795
7d4bcc4a
KW
3796# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3797
49c85077 3798 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3799 * locale is UTF-8. The call to new_ctype() just above has already
3800 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3801 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3802 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3803 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3804 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3805
a05d7ebb 3806 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3807 This is an alternative to using the -C command line switch
3808 (the -C if present will override this). */
3809 {
dd374669 3810 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3811 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3812 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3813 PL_utf8cache = -1;
b310b053
JH
3814 }
3815
7d4bcc4a 3816# endif
7d4bcc4a
KW
3817# ifdef __GLIBC__
3818
175c4cf9 3819 Safefree(language);
7d4bcc4a
KW
3820
3821# endif
175c4cf9
KW
3822
3823 Safefree(lc_all);
3824 Safefree(lang);
3825
e3305790 3826#endif /* USE_LOCALE */
2fcc0ca9 3827#ifdef DEBUGGING
7d4bcc4a 3828
2fcc0ca9 3829 /* So won't continue to output stuff */
27cdc72e 3830 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3831
2fcc0ca9
KW
3832#endif
3833
98994639
HS
3834 return ok;
3835}
3836
98994639
HS
3837#ifdef USE_LOCALE_COLLATE
3838
a4a439fb 3839char *
a4a439fb
KW
3840Perl__mem_collxfrm(pTHX_ const char *input_string,
3841 STRLEN len, /* Length of 'input_string' */
3842 STRLEN *xlen, /* Set to length of returned string
3843 (not including the collation index
3844 prefix) */
3845 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3846 )
98994639 3847{
a4a439fb
KW
3848
3849 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3850 * differences. First, it handles embedded NULs. Second, it allocates a bit
3851 * more memory than needed for the transformed data itself. The real
55e5378d 3852 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3853