This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add thread safety to some environment accesses
[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)
45cef8fb 405# define FIX_GLIBC_LC_MESSAGES_BUG(i)
e9bc6d6b
KW
406
407#else /* Below uses POSIX 2008 */
408
409/* We emulate setlocale with our own function. LC_foo is not valid for the
410 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
411 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
412 * proper offset into the array 'category_masks[]'. At runtime, we have to
413 * search through the array (as the actual numbers may not be small contiguous
414 * positive integers which would lend themselves to array lookup). */
415# define do_setlocale_c(cat, locale) \
416 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
417# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
418
45cef8fb
KW
419# if ! defined(__GLIBC__) || ! defined(USE_LOCALE_MESSAGES)
420
421# define FIX_GLIBC_LC_MESSAGES_BUG(i)
422
423# else /* Invalidate glibc cache of loaded translations, see [perl #134264] */
424
425# include <libintl.h>
426# define FIX_GLIBC_LC_MESSAGES_BUG(i) \
427 STMT_START { \
428 if ((i) == LC_MESSAGES_INDEX) { \
429 textdomain(textdomain(NULL)); \
430 } \
431 } STMT_END
432
433# endif
434
e9bc6d6b
KW
435/* A third array, parallel to the ones above to map from category to its
436 * equivalent mask */
437const int category_masks[] = {
438# ifdef USE_LOCALE_NUMERIC
439 LC_NUMERIC_MASK,
440# endif
441# ifdef USE_LOCALE_CTYPE
442 LC_CTYPE_MASK,
443# endif
444# ifdef USE_LOCALE_COLLATE
445 LC_COLLATE_MASK,
446# endif
447# ifdef USE_LOCALE_TIME
448 LC_TIME_MASK,
449# endif
450# ifdef USE_LOCALE_MESSAGES
451 LC_MESSAGES_MASK,
452# endif
453# ifdef USE_LOCALE_MONETARY
454 LC_MONETARY_MASK,
455# endif
456# ifdef USE_LOCALE_ADDRESS
457 LC_ADDRESS_MASK,
458# endif
459# ifdef USE_LOCALE_IDENTIFICATION
460 LC_IDENTIFICATION_MASK,
461# endif
462# ifdef USE_LOCALE_MEASUREMENT
463 LC_MEASUREMENT_MASK,
464# endif
465# ifdef USE_LOCALE_PAPER
466 LC_PAPER_MASK,
467# endif
468# ifdef USE_LOCALE_TELEPHONE
469 LC_TELEPHONE_MASK,
470# endif
471 /* LC_ALL can't be turned off by a Configure
472 * option, and in Posix 2008, should always be
473 * here, so compile it in unconditionally.
474 * This could catch some glitches at compile
475 * time */
476 LC_ALL_MASK
477 };
478
479STATIC const char *
480S_emulate_setlocale(const int category,
481 const char * locale,
482 unsigned int index,
483 const bool is_index_valid
484 )
485{
486 /* This function effectively performs a setlocale() on just the current
487 * thread; thus it is thread-safe. It does this by using the POSIX 2008
488 * locale functions to emulate the behavior of setlocale(). Similar to
489 * regular setlocale(), the return from this function points to memory that
490 * can be overwritten by other system calls, so needs to be copied
491 * immediately if you need to retain it. The difference here is that
492 * system calls besides another setlocale() can overwrite it.
493 *
494 * By doing this, most locale-sensitive functions become thread-safe. The
495 * exceptions are mostly those that return a pointer to static memory.
496 *
497 * This function takes the same parameters, 'category' and 'locale', that
498 * the regular setlocale() function does, but it also takes two additional
499 * ones. This is because the 2008 functions don't use a category; instead
500 * they use a corresponding mask. Because this function operates in both
501 * worlds, it may need one or the other or both. This function can
502 * calculate the mask from the input category, but to avoid this
503 * calculation, if the caller knows at compile time what the mask is, it
504 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
505 * parameter is ignored.
506 *
507 * POSIX 2008, for some sick reason, chose not to provide a method to find
508 * the category name of a locale. Some vendors have created a
509 * querylocale() function to do just that. This function is a lot simpler
510 * to implement on systems that have this. Otherwise, we have to keep
511 * track of what the locale has been set to, so that we can return its
512 * name to emulate setlocale(). It's also possible for C code in some
513 * library to change the locale without us knowing it, though as of
514 * September 2017, there are no occurrences in CPAN of uselocale(). Some
515 * libraries do use setlocale(), but that changes the global locale, and
516 * threads using per-thread locales will just ignore those changes.
517 * Another problem is that without querylocale(), we have to guess at what
518 * was meant by setting a locale of "". We handle this by not actually
519 * ever setting to "" (unless querylocale exists), but to emulate what we
520 * think should happen for "".
521 */
522
523 int mask;
524 locale_t old_obj;
525 locale_t new_obj;
526 dTHX;
527
528# ifdef DEBUGGING
529
530 if (DEBUG_Lv_TEST || debug_initialization) {
531 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);
532 }
533
534# endif
535
536 /* If the input mask might be incorrect, calculate the correct one */
537 if (! is_index_valid) {
538 unsigned int i;
539
540# ifdef DEBUGGING
541
542 if (DEBUG_Lv_TEST || debug_initialization) {
543 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
544 }
545
546# endif
547
548 for (i = 0; i <= LC_ALL_INDEX; i++) {
549 if (category == categories[i]) {
550 index = i;
551 goto found_index;
552 }
553 }
554
555 /* Here, we don't know about this category, so can't handle it.
556 * Fallback to the early POSIX usages */
557 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
558 "Unknown locale category %d; can't set it to %s\n",
559 category, locale);
560 return NULL;
561
562 found_index: ;
563
564# ifdef DEBUGGING
565
566 if (DEBUG_Lv_TEST || debug_initialization) {
567 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
568 }
569
570# endif
571
572 }
573
574 mask = category_masks[index];
575
576# ifdef DEBUGGING
577
578 if (DEBUG_Lv_TEST || debug_initialization) {
579 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
580 }
581
582# endif
583
584 /* If just querying what the existing locale is ... */
585 if (locale == NULL) {
586 locale_t cur_obj = uselocale((locale_t) 0);
587
588# ifdef DEBUGGING
589
590 if (DEBUG_Lv_TEST || debug_initialization) {
591 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
592 }
593
594# endif
595
596 if (cur_obj == LC_GLOBAL_LOCALE) {
597 return my_setlocale(category, NULL);
598 }
599
600# ifdef HAS_QUERYLOCALE
601
602 return (char *) querylocale(mask, cur_obj);
603
d2b24094 604# else
ecdda939
KW
605
606 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
607 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
608
8e13243a
KW
609# if defined(_NL_LOCALE_NAME) \
610 && defined(DEBUGGING) \
611 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
612 /* On systems that accept any locale name, the real underlying locale
613 * is often returned by this internal function, so we can't use it */
e9bc6d6b
KW
614 {
615 /* Internal glibc for querylocale(), but doesn't handle
616 * empty-string ("") locale properly; who knows what other
3a732259 617 * glitches. Check for it now, under debug. */
e9bc6d6b
KW
618
619 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
620 uselocale((locale_t) 0));
621 /*
622 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
623 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
624 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
625 */
626 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
627 if ( strNE(PL_curlocales[index], temp_name)
628 && ! ( isNAME_C_OR_POSIX(temp_name)
629 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
630
631# ifdef USE_C_BACKTRACE
632
633 dump_c_backtrace(Perl_debug_log, 20, 1);
634
635# endif
636
637 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
638 " (%s) and what internal glibc thinks"
639 " (%s)\n", category_names[index],
640 PL_curlocales[index], temp_name);
641 }
642
643 return temp_name;
644 }
645 }
646
647# endif
648
649 /* Without querylocale(), we have to use our record-keeping we've
650 * done. */
651
652 if (category != LC_ALL) {
653
654# ifdef DEBUGGING
655
656 if (DEBUG_Lv_TEST || debug_initialization) {
657 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
658 }
659
660# endif
661
662 return PL_curlocales[index];
663 }
664 else { /* For LC_ALL */
665 unsigned int i;
666 Size_t names_len = 0;
667 char * all_string;
7c93a471 668 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
669
670 /* If we have a valid LC_ALL value, just return it */
671 if (PL_curlocales[LC_ALL_INDEX]) {
672
673# ifdef DEBUGGING
674
675 if (DEBUG_Lv_TEST || debug_initialization) {
676 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
677 }
678
679# endif
680
681 return PL_curlocales[LC_ALL_INDEX];
682 }
683
684 /* Otherwise, we need to construct a string of name=value pairs.
685 * We use the glibc syntax, like
686 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
687 * First calculate the needed size. Along the way, check if all
688 * the locale names are the same */
e9bc6d6b
KW
689 for (i = 0; i < LC_ALL_INDEX; i++) {
690
691# ifdef DEBUGGING
692
693 if (DEBUG_Lv_TEST || debug_initialization) {
694 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]);
695 }
696
697# endif
698
699 names_len += strlen(category_names[i])
700 + 1 /* '=' */
701 + strlen(PL_curlocales[i])
702 + 1; /* ';' */
7c93a471
KW
703
704 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
705 are_all_categories_the_same_locale = FALSE;
706 }
707 }
708
709 /* If they are the same, we don't actually have to construct the
710 * string; we just make the entry in LC_ALL_INDEX valid, and be
711 * that single name */
712 if (are_all_categories_the_same_locale) {
713 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
714 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 715 }
7c93a471 716
e9bc6d6b
KW
717 names_len++; /* Trailing '\0' */
718 SAVEFREEPV(Newx(all_string, names_len, char));
719 *all_string = '\0';
720
721 /* Then fill in the string */
722 for (i = 0; i < LC_ALL_INDEX; i++) {
723
724# ifdef DEBUGGING
725
726 if (DEBUG_Lv_TEST || debug_initialization) {
727 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]);
728 }
729
730# endif
731
732 my_strlcat(all_string, category_names[i], names_len);
733 my_strlcat(all_string, "=", names_len);
734 my_strlcat(all_string, PL_curlocales[i], names_len);
735 my_strlcat(all_string, ";", names_len);
736 }
737
738# ifdef DEBUGGING
739
740 if (DEBUG_L_TEST || debug_initialization) {
741 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
742 }
743
744 #endif
745
746 return all_string;
747 }
748
749# ifdef EINVAL
750
751 SETERRNO(EINVAL, LIB_INVARG);
752
753# endif
754
755 return NULL;
756
d2b24094
KW
757# endif
758
6aba5c5e 759 } /* End of this being setlocale(LC_foo, NULL) */
e9bc6d6b 760
f8115d60 761 /* Here, we are switching locales. */
e9bc6d6b
KW
762
763# ifndef HAS_QUERYLOCALE
764
765 if (strEQ(locale, "")) {
766
767 /* For non-querylocale() systems, we do the setting of "" ourselves to
768 * be sure that we really know what's going on. We follow the Linux
769 * documented behavior (but if that differs from the actual behavior,
770 * this won't work exactly as the OS implements). We go out and
771 * examine the environment based on our understanding of how the system
772 * works, and use that to figure things out */
773
774 const char * const lc_all = PerlEnv_getenv("LC_ALL");
775
776 /* Use any "LC_ALL" environment variable, as it overrides everything
777 * else. */
778 if (lc_all && strNE(lc_all, "")) {
779 locale = lc_all;
780 }
781 else {
782
783 /* Otherwise, we need to dig deeper. Unless overridden, the
784 * default is the LANG environment variable; if it doesn't exist,
785 * then "C" */
786
787 const char * default_name;
788
207cc8df 789 default_name = PerlEnv_getenv("LANG");
e9bc6d6b
KW
790
791 if (! default_name || strEQ(default_name, "")) {
929a7f8c 792 default_name = "C";
e9bc6d6b 793 }
e9bc6d6b
KW
794
795 if (category != LC_ALL) {
796 const char * const name = PerlEnv_getenv(category_names[index]);
797
798 /* Here we are setting a single category. Assume will have the
799 * default name */
800 locale = default_name;
801
802 /* But then look for an overriding environment variable */
803 if (name && strNE(name, "")) {
804 locale = name;
805 }
806 }
807 else {
808 bool did_override = FALSE;
809 unsigned int i;
810
811 /* Here, we are getting LC_ALL. Any categories that don't have
812 * a corresponding environment variable set should be set to
813 * LANG, or to "C" if there is no LANG. If no individual
814 * categories differ from this, we can just set LC_ALL. This
815 * is buggy on systems that have extra categories that we don't
816 * know about. If there is an environment variable that sets
817 * that category, we won't know to look for it, and so our use
818 * of LANG or "C" improperly overrides it. On the other hand,
819 * if we don't do what is done here, and there is no
820 * environment variable, the category's locale should be set to
821 * LANG or "C". So there is no good solution. khw thinks the
822 * best is to look at systems to see what categories they have,
823 * and include them, and then to assume that we know the
824 * complete set */
825
826 for (i = 0; i < LC_ALL_INDEX; i++) {
827 const char * const env_override
24f3e849 828 = PerlEnv_getenv(category_names[i]);
e9bc6d6b
KW
829 const char * this_locale = ( env_override
830 && strNE(env_override, ""))
831 ? env_override
832 : default_name;
6435f98d
KW
833 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
834 {
6435f98d
KW
835 return NULL;
836 }
e9bc6d6b
KW
837
838 if (strNE(this_locale, default_name)) {
839 did_override = TRUE;
840 }
e9bc6d6b
KW
841 }
842
843 /* If all the categories are the same, we can set LC_ALL to
844 * that */
845 if (! did_override) {
846 locale = default_name;
847 }
848 else {
849
850 /* Here, LC_ALL is no longer valid, as some individual
851 * categories don't match it. We call ourselves
852 * recursively, as that will execute the code that
853 * generates the proper locale string for this situation.
854 * We don't do the remainder of this function, as that is
855 * to update our records, and we've just done that for the
856 * individual categories in the loop above, and doing so
857 * would cause LC_ALL to be done as well */
858 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
859 }
860 }
861 }
6aba5c5e 862 } /* End of this being setlocale(LC_foo, "") */
e9bc6d6b
KW
863 else if (strchr(locale, ';')) {
864
865 /* LC_ALL may actually incude a conglomeration of various categories.
866 * Without querylocale, this code uses the glibc (as of this writing)
867 * syntax for representing that, but that is not a stable API, and
868 * other platforms do it differently, so we have to handle all cases
869 * ourselves */
870
f0023d47 871 unsigned int i;
e9bc6d6b
KW
872 const char * s = locale;
873 const char * e = locale + strlen(locale);
874 const char * p = s;
875 const char * category_end;
876 const char * name_start;
877 const char * name_end;
878
f0023d47
KW
879 /* If the string that gives what to set doesn't include all categories,
880 * the omitted ones get set to "C". To get this behavior, first set
881 * all the individual categories to "C", and override the furnished
882 * ones below */
883 for (i = 0; i < LC_ALL_INDEX; i++) {
884 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
885 return NULL;
886 }
887 }
888
e9bc6d6b 889 while (s < e) {
e9bc6d6b
KW
890
891 /* Parse through the category */
892 while (isWORDCHAR(*p)) {
893 p++;
894 }
895 category_end = p;
896
897 if (*p++ != '=') {
898 Perl_croak(aTHX_
899 "panic: %s: %d: Unexpected character in locale name '%02X",
900 __FILE__, __LINE__, *(p-1));
901 }
902
903 /* Parse through the locale name */
904 name_start = p;
bee74f4b
KW
905 while (p < e && *p != ';') {
906 if (! isGRAPH(*p)) {
907 Perl_croak(aTHX_
908 "panic: %s: %d: Unexpected character in locale name '%02X",
909 __FILE__, __LINE__, *(p-1));
910 }
e9bc6d6b
KW
911 p++;
912 }
913 name_end = p;
914
bee74f4b
KW
915 /* Space past the semi-colon */
916 if (p < e) {
917 p++;
e9bc6d6b
KW
918 }
919
920 /* Find the index of the category name in our lists */
921 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 922 char * individ_locale;
e9bc6d6b
KW
923
924 /* Keep going if this isn't the index. The strnNE() avoids a
925 * Perl_form(), but would fail if ever a category name could be
926 * a substring of another one, like if there were a
927 * "LC_TIME_DATE" */
928 if strnNE(s, category_names[i], category_end - s) {
929 continue;
930 }
931
932 /* If this index is for the single category we're changing, we
933 * have found the locale to set it to. */
934 if (category == categories[i]) {
935 locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start),
937 name_start);
938 goto ready_to_set;
939 }
940
3c76a412 941 assert(category == LC_ALL);
bee74f4b
KW
942 individ_locale = Perl_form(aTHX_ "%.*s",
943 (int) (name_end - name_start), name_start);
6435f98d
KW
944 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
945 {
946 return NULL;
947 }
e9bc6d6b
KW
948 }
949
950 s = p;
951 }
952
953 /* Here we have set all the individual categories by recursive calls.
954 * These collectively should have fixed up LC_ALL, so can just query
955 * what that now is */
956 assert(category == LC_ALL);
957
958 return do_setlocale_c(LC_ALL, NULL);
6aba5c5e
KW
959 } /* End of this being setlocale(LC_ALL,
960 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
e9bc6d6b
KW
961
962 ready_to_set: ;
963
f8115d60
KW
964 /* Here at the end of having to deal with the absence of querylocale().
965 * Some cases have already been fully handled by recursive calls to this
966 * function. But at this point, we haven't dealt with those, but are now
967 * prepared to, knowing what the locale name to set this category to is.
968 * This would have come for free if this system had had querylocale() */
969
e9bc6d6b
KW
970# endif /* end of ! querylocale */
971
f8115d60
KW
972 assert(PL_C_locale_obj);
973
974 /* Switching locales generally entails freeing the current one's space (at
975 * the C library's discretion). We need to stop using that locale before
976 * the switch. So switch to a known locale object that we don't otherwise
977 * mess with. This returns the locale object in effect at the time of the
978 * switch. */
979 old_obj = uselocale(PL_C_locale_obj);
980
981# ifdef DEBUGGING
982
983 if (DEBUG_Lv_TEST || debug_initialization) {
984 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
985 }
986
987# endif
988
989 if (! old_obj) {
990
991# ifdef DEBUGGING
992
993 if (DEBUG_L_TEST || debug_initialization) {
994 dSAVE_ERRNO;
995 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
996 RESTORE_ERRNO;
997 }
998
999# endif
1000
1001 return NULL;
1002 }
1003
1004# ifdef DEBUGGING
1005
1006 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1007 PerlIO_printf(Perl_debug_log,
1008 "%s:%d: emulate_setlocale now using %p\n",
1009 __FILE__, __LINE__, PL_C_locale_obj);
f8115d60
KW
1010 }
1011
1012# endif
1013
6aba5c5e
KW
1014 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1015 * in fact, we already have switched to it (in preparation for what
1016 * normally is to come). But since we're already there, continue to use
70bd6bc8
KW
1017 * it instead of trying to create a new locale */
1018 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1019
1020# ifdef DEBUGGING
1021
1022 if (DEBUG_Lv_TEST || debug_initialization) {
1023 PerlIO_printf(Perl_debug_log,
1024 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1025 }
1026
1027# endif
1028
1029 new_obj = PL_C_locale_obj;
1030
1031 /* We already had switched to the C locale in preparation for freeing
b22e9937 1032 * 'old_obj' */
70bd6bc8
KW
1033 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1034 freelocale(old_obj);
1035 }
1036 }
1037 else {
b22e9937
KW
1038 /* If we weren't in a thread safe locale, set so that newlocale() below
1039 * which uses 'old_obj', uses an empty one. Same for our reserved C
1040 * object. The latter is defensive coding, so that, even if there is
1041 * some bug, we will never end up trying to modify either of these, as
1042 * if passed to newlocale(), they can be. */
1043 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1044 old_obj = (locale_t) 0;
1045 }
f8115d60 1046
b22e9937
KW
1047 /* Ready to create a new locale by modification of the exising one */
1048 new_obj = newlocale(mask, locale, old_obj);
e9bc6d6b 1049
b22e9937
KW
1050 if (! new_obj) {
1051 dSAVE_ERRNO;
e9bc6d6b
KW
1052
1053# ifdef DEBUGGING
1054
b22e9937
KW
1055 if (DEBUG_L_TEST || debug_initialization) {
1056 PerlIO_printf(Perl_debug_log,
1057 "%s:%d: emulate_setlocale creating new object"
1058 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1059 }
e9bc6d6b
KW
1060
1061# endif
1062
b22e9937 1063 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1064
1065# ifdef DEBUGGING
1066
b22e9937
KW
1067 if (DEBUG_L_TEST || debug_initialization) {
1068 PerlIO_printf(Perl_debug_log,
1069 "%s:%d: switching back failed: %d\n",
1070 __FILE__, __LINE__, GET_ERRNO);
1071 }
e9bc6d6b
KW
1072
1073# endif
1074
b22e9937
KW
1075 }
1076 RESTORE_ERRNO;
1077 return NULL;
e9bc6d6b 1078 }
e9bc6d6b
KW
1079
1080# ifdef DEBUGGING
1081
b22e9937
KW
1082 if (DEBUG_Lv_TEST || debug_initialization) {
1083 PerlIO_printf(Perl_debug_log,
1084 "%s:%d: emulate_setlocale created %p",
1085 __FILE__, __LINE__, new_obj);
1086 if (old_obj) {
1087 PerlIO_printf(Perl_debug_log,
1088 "; should have freed %p", old_obj);
1089 }
1090 PerlIO_printf(Perl_debug_log, "\n");
19ee3daf 1091 }
e9bc6d6b
KW
1092
1093# endif
1094
b22e9937
KW
1095 /* And switch into it */
1096 if (! uselocale(new_obj)) {
1097 dSAVE_ERRNO;
e9bc6d6b
KW
1098
1099# ifdef DEBUGGING
1100
b22e9937
KW
1101 if (DEBUG_L_TEST || debug_initialization) {
1102 PerlIO_printf(Perl_debug_log,
1103 "%s:%d: emulate_setlocale switching to new object"
1104 " failed\n", __FILE__, __LINE__);
1105 }
e9bc6d6b
KW
1106
1107# endif
1108
b22e9937 1109 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1110
1111# ifdef DEBUGGING
1112
b22e9937
KW
1113 if (DEBUG_L_TEST || debug_initialization) {
1114 PerlIO_printf(Perl_debug_log,
1115 "%s:%d: switching back failed: %d\n",
1116 __FILE__, __LINE__, GET_ERRNO);
1117 }
e9bc6d6b
KW
1118
1119# endif
1120
b22e9937
KW
1121 }
1122 freelocale(new_obj);
1123 RESTORE_ERRNO;
1124 return NULL;
e9bc6d6b 1125 }
70bd6bc8 1126 }
e9bc6d6b
KW
1127
1128# ifdef DEBUGGING
1129
1130 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1131 PerlIO_printf(Perl_debug_log,
1132 "%s:%d: emulate_setlocale now using %p\n",
1133 __FILE__, __LINE__, new_obj);
e9bc6d6b
KW
1134 }
1135
1136# endif
1137
1138 /* We are done, except for updating our records (if the system doesn't keep
1139 * them) and in the case of locale "", we don't actually know what the
1140 * locale that got switched to is, as it came from the environment. So
1141 * have to find it */
1142
1143# ifdef HAS_QUERYLOCALE
1144
1145 if (strEQ(locale, "")) {
1146 locale = querylocale(mask, new_obj);
1147 }
1148
1149# else
1150
1151 /* Here, 'locale' is the return value */
1152
1153 /* Without querylocale(), we have to update our records */
1154
1155 if (category == LC_ALL) {
1156 unsigned int i;
1157
1158 /* For LC_ALL, we change all individual categories to correspond */
1159 /* PL_curlocales is a parallel array, so has same
1160 * length as 'categories' */
1161 for (i = 0; i <= LC_ALL_INDEX; i++) {
1162 Safefree(PL_curlocales[i]);
1163 PL_curlocales[i] = savepv(locale);
1164 }
45cef8fb
KW
1165
1166 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
e9bc6d6b
KW
1167 }
1168 else {
1169
1170 /* For a single category, if it's not the same as the one in LC_ALL, we
1171 * nullify LC_ALL */
1172
1173 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1174 Safefree(PL_curlocales[LC_ALL_INDEX]);
1175 PL_curlocales[LC_ALL_INDEX] = NULL;
1176 }
1177
1178 /* Then update the category's record */
1179 Safefree(PL_curlocales[index]);
1180 PL_curlocales[index] = savepv(locale);
45cef8fb
KW
1181
1182 FIX_GLIBC_LC_MESSAGES_BUG(index);
e9bc6d6b
KW
1183 }
1184
1185# endif
1186
1187 return locale;
1188}
1189
1190#endif /* USE_POSIX_2008_LOCALE */
1191
1192#if 0 /* Code that was to emulate thread-safe locales on platforms that
1193 didn't natively support them */
1194
1195/* The way this would work is that we would keep a per-thread list of the
1196 * correct locale for that thread. Any operation that was locale-sensitive
1197 * would have to be changed so that it would look like this:
1198 *
1199 * LOCALE_LOCK;
1200 * setlocale to the correct locale for this operation
1201 * do operation
1202 * LOCALE_UNLOCK
1203 *
1204 * This leaves the global locale in the most recently used operation's, but it
1205 * was locked long enough to get the result. If that result is static, it
1206 * needs to be copied before the unlock.
1207 *
1208 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1209 * the setup, but are no-ops when not needed, and similarly,
1210 * END_LOCALE_DEPENDENT_OP for the tear-down
1211 *
1212 * But every call to a locale-sensitive function would have to be changed, and
1213 * if a module didn't cooperate by using the mutex, things would break.
1214 *
1215 * This code was abandoned before being completed or tested, and is left as-is
1216*/
1217
1218# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1219# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1220
1221STATIC char *
1222S_locking_setlocale(pTHX_
1223 const int category,
1224 const char * locale,
1225 int index,
1226 const bool is_index_valid
1227 )
1228{
1229 /* This function kind of performs a setlocale() on just the current thread;
1230 * thus it is kind of thread-safe. It does this by keeping a thread-level
1231 * array of the current locales for each category. Every time a locale is
1232 * switched to, it does the switch globally, but updates the thread's
1233 * array. A query as to what the current locale is just returns the
1234 * appropriate element from the array, and doesn't actually call the system
1235 * setlocale(). The saving into the array is done in an uninterruptible
1236 * section of code, so is unaffected by whatever any other threads might be
1237 * doing.
1238 *
1239 * All locale-sensitive operations must work by first starting a critical
1240 * section, then switching to the thread's locale as kept by this function,
1241 * and then doing the operation, then ending the critical section. Thus,
1242 * each gets done in the appropriate locale. simulating thread-safety.
1243 *
1244 * This function takes the same parameters, 'category' and 'locale', that
1245 * the regular setlocale() function does, but it also takes two additional
1246 * ones. This is because as described earlier. If we know on input the
1247 * index corresponding to the category into the array where we store the
1248 * current locales, we don't have to calculate it. If the caller knows at
06cc5386 1249 * compile time what the index is, it can pass it, setting
e9bc6d6b
KW
1250 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1251 *
1252 */
1253
1254 /* If the input index might be incorrect, calculate the correct one */
1255 if (! is_index_valid) {
1256 unsigned int i;
1257
1258 if (DEBUG_Lv_TEST || debug_initialization) {
1259 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1260 }
1261
1262 for (i = 0; i <= LC_ALL_INDEX; i++) {
1263 if (category == categories[i]) {
1264 index = i;
1265 goto found_index;
1266 }
1267 }
1268
1269 /* Here, we don't know about this category, so can't handle it.
1270 * XXX best we can do is to unsafely set this
1271 * XXX warning */
1272
1273 return my_setlocale(category, locale);
1274
1275 found_index: ;
1276
1277 if (DEBUG_Lv_TEST || debug_initialization) {
1278 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1279 }
1280 }
1281
1282 /* For a query, just return what's in our records */
1283 if (new_locale == NULL) {
1284 return curlocales[index];
1285 }
1286
1287
1288 /* Otherwise, we need to do the switch, and save the result, all in a
1289 * critical section */
1290
1291 Safefree(curlocales[[index]]);
1292
1293 /* It might be that this is called from an already-locked section of code.
1294 * We would have to detect and skip the LOCK/UNLOCK if so */
1295 LOCALE_LOCK;
1296
1297 curlocales[index] = savepv(my_setlocale(category, new_locale));
1298
1299 if (strEQ(new_locale, "")) {
1300
1301#ifdef LC_ALL
1302
1303 /* The locale values come from the environment, and may not all be the
1304 * same, so for LC_ALL, we have to update all the others, while the
1305 * mutex is still locked */
1306
1307 if (category == LC_ALL) {
1308 unsigned int i;
1309 for (i = 0; i < LC_ALL_INDEX) {
1310 curlocales[i] = my_setlocale(categories[i], NULL);
1311 }
1312 }
1313 }
1314
1315#endif
1316
1317 LOCALE_UNLOCK;
1318
1319 return curlocales[index];
1320}
1321
1322#endif
d36adde0 1323#ifdef USE_LOCALE
837ce802 1324
a4f00dcc 1325STATIC void
86799d2d 1326S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1327{
86799d2d
KW
1328 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1329 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1330
86799d2d
KW
1331#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1332 || defined(HAS_NL_LANGINFO))
98994639 1333
87f8e8e7 1334 const char * radix = (use_locale)
4e6826bf 1335 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1336 /* FALSE => already in dest locale */
1337 : ".";
2213a3be 1338
87f8e8e7 1339 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1340
87f8e8e7
KW
1341 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1342 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1343 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1344 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1345 && _is_cur_LC_category_utf8(LC_NUMERIC))
1346 {
1347 SvUTF8_on(PL_numeric_radix_sv);
1348 }
86799d2d
KW
1349
1350# ifdef DEBUGGING
7d4bcc4a 1351
2fcc0ca9
KW
1352 if (DEBUG_L_TEST || debug_initialization) {
1353 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1354 SvPVX(PL_numeric_radix_sv),
1355 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1356 }
69014004 1357
86799d2d 1358# endif
d36adde0
KW
1359#else
1360
1361 PERL_UNUSED_ARG(use_locale);
1362
86799d2d 1363#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1364
98994639
HS
1365}
1366
398eeea9
KW
1367STATIC void
1368S_new_numeric(pTHX_ const char *newnum)
98994639 1369{
7d4bcc4a
KW
1370
1371#ifndef USE_LOCALE_NUMERIC
1372
1373 PERL_UNUSED_ARG(newnum);
1374
1375#else
0d071d52 1376
291a84fb 1377 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1378 * core Perl this and that 'newnum' is the name of the new locale.
1379 * It installs this locale as the current underlying default.
1380 *
1381 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1382 * set_numeric_underlying() and set_numeric_standard() functions, which
1383 * should probably not be called directly, but only via macros like
0d071d52
KW
1384 * SET_NUMERIC_STANDARD() in perl.h.
1385 *
1386 * The toggling is necessary mainly so that a non-dot radix decimal point
1387 * character can be output, while allowing internal calculations to use a
1388 * dot.
1389 *
1390 * This sets several interpreter-level variables:
bb304765 1391 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1392 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1393 * that the current locale is the program's underlying
1394 * locale
1395 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1396 * that the current locale is the C locale or
1397 * indistinguishable from the C locale. If non-zero, it
1398 * is in C; if > 1, it means it may not be toggled away
7738054c 1399 * from C.
4c68b815
KW
1400 * PL_numeric_underlying_is_standard A bool kept by this function
1401 * indicating that the underlying locale and the standard
1402 * C locale are indistinguishable for the purposes of
1403 * LC_NUMERIC. This happens when both of the above two
1404 * variables are true at the same time. (Toggling is a
1405 * no-op under these circumstances.) This variable is
1406 * used to avoid having to recalculate.
398eeea9 1407 */
0d071d52 1408
b03f34cf 1409 char *save_newnum;
98994639
HS
1410
1411 if (! newnum) {
43c5f42d
NC
1412 Safefree(PL_numeric_name);
1413 PL_numeric_name = NULL;
98994639 1414 PL_numeric_standard = TRUE;
892e6465 1415 PL_numeric_underlying = TRUE;
4c68b815 1416 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1417 return;
1418 }
1419
b03f34cf 1420 save_newnum = stdize_locale(savepv(newnum));
892e6465 1421 PL_numeric_underlying = TRUE;
4c68b815
KW
1422 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1423
69c5e0db
KW
1424#ifndef TS_W32_BROKEN_LOCALECONV
1425
4c68b815 1426 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1427 * them. But on broken Windows systems calling my_nl_langinfo() for
1428 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1429 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1430 if (! PL_numeric_standard) {
4e6826bf 1431 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1432 FALSE /* Don't toggle locale */ ))
4e6826bf 1433 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1434 }
abe1abcf 1435
69c5e0db
KW
1436#endif
1437
4c68b815 1438 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1439 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1440 Safefree(PL_numeric_name);
b03f34cf 1441 PL_numeric_name = save_newnum;
b03f34cf 1442 }
abe1abcf
KW
1443 else {
1444 Safefree(save_newnum);
1445 }
4c28b29c 1446
4c68b815
KW
1447 PL_numeric_underlying_is_standard = PL_numeric_standard;
1448
7e5377f7 1449# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1450
1451 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1452 PL_numeric_name,
1453 PL_underlying_numeric_obj);
1454
1455#endif
1456
4c68b815
KW
1457 if (DEBUG_L_TEST || debug_initialization) {
1458 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1459 }
1460
4c28b29c
KW
1461 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1462 * have to worry about the radix being a non-dot. (Core operations that
1463 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1464 if (PL_numeric_standard) {
1465 set_numeric_radix(0);
1466 }
1467 else {
1468 set_numeric_standard();
1469 }
4c28b29c 1470
98994639 1471#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1472
98994639
HS
1473}
1474
1475void
1476Perl_set_numeric_standard(pTHX)
1477{
7d4bcc4a 1478
98994639 1479#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1480
28c1bf33
KW
1481 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1482 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1483 * macro avoids calling this routine if toggling isn't necessary according
1484 * to our records (which could be wrong if some XS code has changed the
1485 * locale behind our back) */
0d071d52 1486
7d4bcc4a
KW
1487# ifdef DEBUGGING
1488
2fcc0ca9
KW
1489 if (DEBUG_L_TEST || debug_initialization) {
1490 PerlIO_printf(Perl_debug_log,
7bb8f702 1491 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1492 }
98994639 1493
7d4bcc4a 1494# endif
7bb8f702
KW
1495
1496 do_setlocale_c(LC_NUMERIC, "C");
1497 PL_numeric_standard = TRUE;
1498 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1499 set_numeric_radix(0);
1500
98994639 1501#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1502
98994639
HS
1503}
1504
1505void
5792c642 1506Perl_set_numeric_underlying(pTHX)
98994639 1507{
7d4bcc4a 1508
98994639 1509#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1510
28c1bf33 1511 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1512 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1513 * instead of calling this directly. The macro avoids calling this routine
1514 * if toggling isn't necessary according to our records (which could be
1515 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1516
7d4bcc4a
KW
1517# ifdef DEBUGGING
1518
2fcc0ca9
KW
1519 if (DEBUG_L_TEST || debug_initialization) {
1520 PerlIO_printf(Perl_debug_log,
7bb8f702 1521 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1522 PL_numeric_name);
1523 }
98994639 1524
7d4bcc4a 1525# endif
7bb8f702
KW
1526
1527 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1528 PL_numeric_standard = PL_numeric_underlying_is_standard;
1529 PL_numeric_underlying = TRUE;
1530 set_numeric_radix(! PL_numeric_standard);
1531
98994639 1532#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1533
98994639
HS
1534}
1535
1536/*
1537 * Set up for a new ctype locale.
1538 */
a4f00dcc
KW
1539STATIC void
1540S_new_ctype(pTHX_ const char *newctype)
98994639 1541{
7d4bcc4a
KW
1542
1543#ifndef USE_LOCALE_CTYPE
1544
7d4bcc4a
KW
1545 PERL_UNUSED_ARG(newctype);
1546 PERL_UNUSED_CONTEXT;
1547
1548#else
0d071d52 1549
291a84fb 1550 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1551 * core Perl this and that 'newctype' is the name of the new locale.
1552 *
1553 * This function sets up the folding arrays for all 256 bytes, assuming
1554 * that tofold() is tolc() since fold case is not a concept in POSIX,
1555 *
1556 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1557 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1558 * this function should be called directly only from this file and from
0d071d52
KW
1559 * POSIX::setlocale() */
1560
27da23d5 1561 dVAR;
013f4e03 1562 unsigned int i;
98994639 1563
8b7358b9
KW
1564 /* Don't check for problems if we are suppressing the warnings */
1565 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
30d8090d 1566 bool maybe_utf8_turkic = FALSE;
8b7358b9 1567
7918f24d
NC
1568 PERL_ARGS_ASSERT_NEW_CTYPE;
1569
215c5139
KW
1570 /* We will replace any bad locale warning with 1) nothing if the new one is
1571 * ok; or 2) a new warning for the bad new locale */
1572 if (PL_warn_locale) {
1573 SvREFCNT_dec_NN(PL_warn_locale);
1574 PL_warn_locale = NULL;
1575 }
1576
c1284011 1577 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1578
1579 /* A UTF-8 locale gets standard rules. But note that code still has to
1580 * handle this specially because of the three problematic code points */
1581 if (PL_in_utf8_CTYPE_locale) {
1582 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
30d8090d
KW
1583
1584 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1585 * Turkic. Make sure these two are the only anomalies. (We don't use
1586 * towupper and towlower because they aren't in C89.) */
5b64f24c
KW
1587
1588#if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1589
1590 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1591
1592#else
1593
30d8090d 1594 if (toupper('i') == 'i' && tolower('I') == 'I') {
5b64f24c
KW
1595
1596#endif
30d8090d
KW
1597 check_for_problems = TRUE;
1598 maybe_utf8_turkic = TRUE;
1599 }
31f05a37 1600 }
8b7358b9
KW
1601
1602 /* We don't populate the other lists if a UTF-8 locale, but do check that
1603 * everything works as expected, unless checking turned off */
1604 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1605 /* Assume enough space for every character being bad. 4 spaces each
1606 * for the 94 printable characters that are output like "'x' "; and 5
1607 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1608 * NUL */
8b7358b9 1609 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1610 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1611 to start */
1612 unsigned int bad_count = 0; /* Count of bad characters */
1613
baa60164 1614 for (i = 0; i < 256; i++) {
8b7358b9 1615 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1616 if (isupper(i))
1617 PL_fold_locale[i] = (U8) tolower(i);
1618 else if (islower(i))
1619 PL_fold_locale[i] = (U8) toupper(i);
1620 else
1621 PL_fold_locale[i] = (U8) i;
8b7358b9 1622 }
8c6180a9
KW
1623
1624 /* If checking for locale problems, see if the native ASCII-range
1625 * printables plus \n and \t are in their expected categories in
1626 * the new locale. If not, this could mean big trouble, upending
1627 * Perl's and most programs' assumptions, like having a
1628 * metacharacter with special meaning become a \w. Fortunately,
1629 * it's very rare to find locales that aren't supersets of ASCII
1630 * nowadays. It isn't a problem for most controls to be changed
1631 * into something else; we check only \n and \t, though perhaps \r
1632 * could be an issue as well. */
7d4bcc4a 1633 if ( check_for_problems
8c6180a9
KW
1634 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1635 {
8b7358b9 1636 bool is_bad = FALSE;
6726b4f4 1637 char name[4] = { '\0' };
8b7358b9
KW
1638
1639 /* Convert the name into a string */
6726b4f4 1640 if (isGRAPH_A(i)) {
8b7358b9
KW
1641 name[0] = i;
1642 name[1] = '\0';
1643 }
1644 else if (i == '\n') {
6726b4f4
KW
1645 my_strlcpy(name, "\\n", sizeof(name));
1646 }
1647 else if (i == '\t') {
1648 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1649 }
1650 else {
6726b4f4
KW
1651 assert(i == ' ');
1652 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1653 }
1654
1655 /* Check each possibe class */
1656 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1657 is_bad = TRUE;
1658 DEBUG_L(PerlIO_printf(Perl_debug_log,
1659 "isalnum('%s') unexpectedly is %d\n",
1660 name, cBOOL(isalnum(i))));
1661 }
1662 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1663 is_bad = TRUE;
1664 DEBUG_L(PerlIO_printf(Perl_debug_log,
1665 "isalpha('%s') unexpectedly is %d\n",
1666 name, cBOOL(isalpha(i))));
1667 }
1668 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1669 is_bad = TRUE;
1670 DEBUG_L(PerlIO_printf(Perl_debug_log,
1671 "isdigit('%s') unexpectedly is %d\n",
1672 name, cBOOL(isdigit(i))));
1673 }
1674 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1675 is_bad = TRUE;
1676 DEBUG_L(PerlIO_printf(Perl_debug_log,
1677 "isgraph('%s') unexpectedly is %d\n",
1678 name, cBOOL(isgraph(i))));
1679 }
1680 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1681 is_bad = TRUE;
1682 DEBUG_L(PerlIO_printf(Perl_debug_log,
1683 "islower('%s') unexpectedly is %d\n",
1684 name, cBOOL(islower(i))));
1685 }
1686 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1687 is_bad = TRUE;
1688 DEBUG_L(PerlIO_printf(Perl_debug_log,
1689 "isprint('%s') unexpectedly is %d\n",
1690 name, cBOOL(isprint(i))));
1691 }
1692 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1693 is_bad = TRUE;
1694 DEBUG_L(PerlIO_printf(Perl_debug_log,
1695 "ispunct('%s') unexpectedly is %d\n",
1696 name, cBOOL(ispunct(i))));
1697 }
1698 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1699 is_bad = TRUE;
1700 DEBUG_L(PerlIO_printf(Perl_debug_log,
1701 "isspace('%s') unexpectedly is %d\n",
1702 name, cBOOL(isspace(i))));
1703 }
1704 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1705 is_bad = TRUE;
1706 DEBUG_L(PerlIO_printf(Perl_debug_log,
1707 "isupper('%s') unexpectedly is %d\n",
1708 name, cBOOL(isupper(i))));
1709 }
1710 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1711 is_bad = TRUE;
1712 DEBUG_L(PerlIO_printf(Perl_debug_log,
1713 "isxdigit('%s') unexpectedly is %d\n",
1714 name, cBOOL(isxdigit(i))));
1715 }
65c15174 1716 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
8b7358b9
KW
1717 is_bad = TRUE;
1718 DEBUG_L(PerlIO_printf(Perl_debug_log,
1719 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1720 name, tolower(i), (int) toLOWER_A(i)));
1721 }
65c15174 1722 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
8b7358b9
KW
1723 is_bad = TRUE;
1724 DEBUG_L(PerlIO_printf(Perl_debug_log,
1725 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1726 name, toupper(i), (int) toUPPER_A(i)));
1727 }
1728 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1729 is_bad = TRUE;
1730 DEBUG_L(PerlIO_printf(Perl_debug_log,
1731 "'\\n' (=%02X) is not a control\n", (int) i));
1732 }
1733
1734 /* Add to the list; Separate multiple entries with a blank */
1735 if (is_bad) {
1736 if (bad_count) {
1737 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1738 }
8b7358b9
KW
1739 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1740 bad_count++;
8c6180a9
KW
1741 }
1742 }
1743 }
1744
30d8090d
KW
1745 if (bad_count == 2 && maybe_utf8_turkic) {
1746 bad_count = 0;
1747 *bad_chars_list = '\0';
1748 PL_fold_locale['I'] = 'I';
1749 PL_fold_locale['i'] = 'i';
1750 PL_in_utf8_turkic_locale = TRUE;
1751 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1752 __FILE__, __LINE__, newctype));
1753 }
1754 else {
b8df1494 1755 PL_in_utf8_turkic_locale = FALSE;
30d8090d 1756 }
b8df1494 1757
7d4bcc4a
KW
1758# ifdef MB_CUR_MAX
1759
8c6180a9 1760 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1761 * this locale requires more than one byte, there are going to be
8c6180a9 1762 * problems. */
9c8a6dc2
KW
1763 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1764 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1765 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1766
8b7358b9
KW
1767 if ( check_for_problems && MB_CUR_MAX > 1
1768 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1769
1770 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1771 * locale. Just assume that the implementation for them (plus
1772 * for POSIX) is correct and the > 1 value is spurious. (Since
1773 * these are specially handled to never be considered UTF-8
1774 * locales, as long as this is the only problem, everything
1775 * should work fine */
1776 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1777 {
8c6180a9
KW
1778 multi_byte_locale = TRUE;
1779 }
7d4bcc4a
KW
1780
1781# endif
8c6180a9 1782
30d8090d
KW
1783 /* If we found problems and we want them output, do so */
1784 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1785 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1786 {
8b7358b9
KW
1787 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1788 PL_warn_locale = Perl_newSVpvf(aTHX_
1789 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1790 " which have\nunexpected meanings: %s\nThe Perl program"
1791 " will use the expected meanings",
8b7358b9 1792 newctype, bad_chars_list);
8b7358b9
KW
1793 }
1794 else {
1795 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1796 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1797 newctype,
8c6180a9
KW
1798 (multi_byte_locale)
1799 ? " Some characters in it are not recognized by"
1800 " Perl."
1801 : "",
1802 (bad_count)
1803 ? "\nThe following characters (and maybe others)"
1804 " may not have the same meaning as the Perl"
1805 " program expects:\n"
1806 : "",
1807 (bad_count)
1808 ? bad_chars_list
1809 : ""
1810 );
8b7358b9
KW
1811 }
1812
b119c2be
KW
1813# ifdef HAS_NL_LANGINFO
1814
1815 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1816 /* parameter FALSE is a don't care here */
4e6826bf 1817 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1818
1819# endif
1820
8b7358b9
KW
1821 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1822
cc9eaeb0 1823 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1824 * output the message now. If not in that scope, we save the
1825 * message to be output at the first operation using this locale,
1826 * if that actually happens. Most programs don't use locales, so
1827 * they are immune to bad ones. */
cc9eaeb0 1828 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1829
780fcc9f
KW
1830 /* The '0' below suppresses a bogus gcc compiler warning */
1831 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1832
bddebb56
KW
1833 if (IN_LC(LC_CTYPE)) {
1834 SvREFCNT_dec_NN(PL_warn_locale);
1835 PL_warn_locale = NULL;
1836 }
780fcc9f 1837 }
baa60164 1838 }
31f05a37 1839 }
98994639
HS
1840
1841#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1842
98994639
HS
1843}
1844
98994639 1845void
2726666d
KW
1846Perl__warn_problematic_locale()
1847{
2726666d
KW
1848
1849#ifdef USE_LOCALE_CTYPE
1850
5f04a188
KW
1851 dTHX;
1852
1853 /* Internal-to-core function that outputs the message in PL_warn_locale,
1854 * and then NULLS it. Should be called only through the macro
1855 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1856
2726666d 1857 if (PL_warn_locale) {
2726666d
KW
1858 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1859 SvPVX(PL_warn_locale),
1860 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1861 SvREFCNT_dec_NN(PL_warn_locale);
1862 PL_warn_locale = NULL;
1863 }
1864
1865#endif
1866
1867}
1868
a4f00dcc
KW
1869STATIC void
1870S_new_collate(pTHX_ const char *newcoll)
98994639 1871{
7d4bcc4a
KW
1872
1873#ifndef USE_LOCALE_COLLATE
1874
1875 PERL_UNUSED_ARG(newcoll);
1876 PERL_UNUSED_CONTEXT;
1877
1878#else
0d071d52 1879
291a84fb 1880 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1881 * core Perl this and that 'newcoll' is the name of the new locale.
1882 *
d35fca5f
KW
1883 * The design of locale collation is that every locale change is given an
1884 * index 'PL_collation_ix'. The first time a string particpates in an
1885 * operation that requires collation while locale collation is active, it
1886 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1887 * magic includes the collation index, and the transformation of the string
1888 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1889 * instead of the string itself. If a string changes, the magic is
1890 * cleared. The next time the locale changes, the index is incremented,
1891 * and so we know during a comparison that the transformation is not
1892 * necessarily still valid, and so is recomputed. Note that if the locale
1893 * changes enough times, the index could wrap (a U32), and it is possible
1894 * that a transformation would improperly be considered valid, leading to
1895 * an unlikely bug */
0d071d52 1896
98994639
HS
1897 if (! newcoll) {
1898 if (PL_collation_name) {
1899 ++PL_collation_ix;
1900 Safefree(PL_collation_name);
1901 PL_collation_name = NULL;
1902 }
1903 PL_collation_standard = TRUE;
00bf60ca 1904 is_standard_collation:
98994639
HS
1905 PL_collxfrm_base = 0;
1906 PL_collxfrm_mult = 2;
165a1c52 1907 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1908 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1909 PL_strxfrm_max_cp = 0;
98994639
HS
1910 return;
1911 }
1912
d35fca5f 1913 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1914 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1915 ++PL_collation_ix;
1916 Safefree(PL_collation_name);
1917 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1918 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1919 if (PL_collation_standard) {
1920 goto is_standard_collation;
1921 }
98994639 1922
165a1c52 1923 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1924 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1925 PL_strxfrm_max_cp = 0;
165a1c52 1926
59c018b9
KW
1927 /* A locale collation definition includes primary, secondary, tertiary,
1928 * etc. weights for each character. To sort, the primary weights are
1929 * used, and only if they compare equal, then the secondary weights are
1930 * used, and only if they compare equal, then the tertiary, etc.
1931 *
1932 * strxfrm() works by taking the input string, say ABC, and creating an
1933 * output transformed string consisting of first the primary weights,
1934 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1935 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1936 * may not have weights at every level. In our example, let's say B
1937 * doesn't have a tertiary weight, and A doesn't have a secondary
1938 * weight. The constructed string is then going to be
1939 * A¹B¹C¹ B²C² A³C³ ....
1940 * This has the desired effect that strcmp() will look at the secondary
1941 * or tertiary weights only if the strings compare equal at all higher
1942 * priority weights. The spaces shown here, like in
c342d20e 1943 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1944 * are not just for readability. In the general case, these must
1945 * actually be bytes, which we will call here 'separator weights'; and
1946 * they must be smaller than any other weight value, but since these
1947 * are C strings, only the terminating one can be a NUL (some
1948 * implementations may include a non-NUL separator weight just before
1949 * the NUL). Implementations tend to reserve 01 for the separator
1950 * weights. They are needed so that a shorter string's secondary
1951 * weights won't be misconstrued as primary weights of a longer string,
1952 * etc. By making them smaller than any other weight, the shorter
1953 * string will sort first. (Actually, if all secondary weights are
1954 * smaller than all primary ones, there is no need for a separator
1955 * weight between those two levels, etc.)
1956 *
1957 * The length of the transformed string is roughly a linear function of
1958 * the input string. It's not exactly linear because some characters
1959 * don't have weights at all levels. When we call strxfrm() we have to
1960 * allocate some memory to hold the transformed string. The
1961 * calculations below try to find coefficients 'm' and 'b' for this
1962 * locale so that m*x + b equals how much space we need, given the size
1963 * of the input string in 'x'. If we calculate too small, we increase
1964 * the size as needed, and call strxfrm() again, but it is better to
1965 * get it right the first time to avoid wasted expensive string
1966 * transformations. */
1967
98994639 1968 {
79f120c8
KW
1969 /* We use the string below to find how long the tranformation of it
1970 * is. Almost all locales are supersets of ASCII, or at least the
1971 * ASCII letters. We use all of them, half upper half lower,
1972 * because if we used fewer, we might hit just the ones that are
1973 * outliers in a particular locale. Most of the strings being
1974 * collated will contain a preponderance of letters, and even if
1975 * they are above-ASCII, they are likely to have the same number of
1976 * weight levels as the ASCII ones. It turns out that digits tend
1977 * to have fewer levels, and some punctuation has more, but those
1978 * are relatively sparse in text, and khw believes this gives a
1979 * reasonable result, but it could be changed if experience so
1980 * dictates. */
1981 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1982 char * x_longer; /* Transformed 'longer' */
1983 Size_t x_len_longer; /* Length of 'x_longer' */
1984
1985 char * x_shorter; /* We also transform a substring of 'longer' */
1986 Size_t x_len_shorter;
1987
a4a439fb 1988 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1989 * are interested only in its length). It is used because it has
1990 * the intelligence to handle all cases, but to work, it needs some
1991 * values of 'm' and 'b' to get it started. For the purposes of
1992 * this calculation we use a very conservative estimate of 'm' and
1993 * 'b'. This assumes a weight can be multiple bytes, enough to
1994 * hold any UV on the platform, and there are 5 levels, 4 weight
1995 * bytes, and a trailing NUL. */
1996 PL_collxfrm_base = 5;
1997 PL_collxfrm_mult = 5 * sizeof(UV);
1998
1999 /* Find out how long the transformation really is */
a4a439fb
KW
2000 x_longer = _mem_collxfrm(longer,
2001 sizeof(longer) - 1,
2002 &x_len_longer,
2003
2004 /* We avoid converting to UTF-8 in the
2005 * called function by telling it the
2006 * string is in UTF-8 if the locale is a
2007 * UTF-8 one. Since the string passed
2008 * here is invariant under UTF-8, we can
2009 * claim it's UTF-8 even though it isn't.
2010 * */
2011 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2012 Safefree(x_longer);
2013
2014 /* Find out how long the transformation of a substring of 'longer'
2015 * is. Together the lengths of these transformations are
2016 * sufficient to calculate 'm' and 'b'. The substring is all of
2017 * 'longer' except the first character. This minimizes the chances
2018 * of being swayed by outliers */
a4a439fb 2019 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 2020 sizeof(longer) - 2,
a4a439fb
KW
2021 &x_len_shorter,
2022 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2023 Safefree(x_shorter);
2024
2025 /* If the results are nonsensical for this simple test, the whole
2026 * locale definition is suspect. Mark it so that locale collation
2027 * is not active at all for it. XXX Should we warn? */
2028 if ( x_len_shorter == 0
2029 || x_len_longer == 0
2030 || x_len_shorter >= x_len_longer)
2031 {
2032 PL_collxfrm_mult = 0;
2033 PL_collxfrm_base = 0;
2034 }
2035 else {
2036 SSize_t base; /* Temporary */
2037
2038 /* We have both: m * strlen(longer) + b = x_len_longer
2039 * m * strlen(shorter) + b = x_len_shorter;
2040 * subtracting yields:
2041 * m * (strlen(longer) - strlen(shorter))
2042 * = x_len_longer - x_len_shorter
2043 * But we have set things up so that 'shorter' is 1 byte smaller
2044 * than 'longer'. Hence:
2045 * m = x_len_longer - x_len_shorter
2046 *
2047 * But if something went wrong, make sure the multiplier is at
2048 * least 1.
2049 */
2050 if (x_len_longer > x_len_shorter) {
2051 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2052 }
2053 else {
2054 PL_collxfrm_mult = 1;
2055 }
2056
2057 /* mx + b = len
2058 * so: b = len - mx
2059 * but in case something has gone wrong, make sure it is
2060 * non-negative */
2061 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2062 if (base < 0) {
2063 base = 0;
2064 }
2065
2066 /* Add 1 for the trailing NUL */
2067 PL_collxfrm_base = base + 1;
2068 }
58eebef2 2069
7d4bcc4a
KW
2070# ifdef DEBUGGING
2071
58eebef2
KW
2072 if (DEBUG_L_TEST || debug_initialization) {
2073 PerlIO_printf(Perl_debug_log,
b07929e4
KW
2074 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2075 "x_len_longer=%zu,"
2076 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
2077 __FILE__, __LINE__,
2078 PL_in_utf8_COLLATE_locale,
2079 x_len_shorter, x_len_longer,
2080 PL_collxfrm_mult, PL_collxfrm_base);
2081 }
7d4bcc4a
KW
2082# endif
2083
98994639
HS
2084 }
2085 }
2086
2087#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 2088
98994639
HS
2089}
2090
d36adde0
KW
2091#endif
2092
d2b24094 2093#ifdef WIN32
b8cc575c 2094
6c332036
TC
2095#define USE_WSETLOCALE
2096
2097#ifdef USE_WSETLOCALE
2098
2099STATIC char *
2100S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2101 wchar_t *wlocale;
2102 wchar_t *wresult;
2103 char *result;
2104
2105 if (locale) {
2106 int req_size =
2107 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2108
2109 if (!req_size) {
2110 errno = EINVAL;
2111 return NULL;
2112 }
2113
2114 Newx(wlocale, req_size, wchar_t);
2115 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2116 Safefree(wlocale);
2117 errno = EINVAL;
2118 return NULL;
2119 }
2120 }
2121 else {
2122 wlocale = NULL;
2123 }
2124 wresult = _wsetlocale(category, wlocale);
2125 Safefree(wlocale);
2126 if (wresult) {
2127 int req_size =
2128 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2129 Newx(result, req_size, char);
2130 SAVEFREEPV(result); /* is there something better we can do here? */
2131 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2132 result, req_size, NULL, NULL)) {
2133 errno = EINVAL;
2134 return NULL;
2135 }
2136 }
2137 else {
2138 result = NULL;
2139 }
2140
2141 return result;
2142}
2143
2144#endif
2145
a4f00dcc 2146STATIC char *
b8cc575c 2147S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2148{
2149 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2150 * difference between the two unless the input locale is "", which normally
2151 * means on Windows to get the machine default, which is set via the
2152 * computer's "Regional and Language Options" (or its current equivalent).
2153 * In POSIX, it instead means to find the locale from the user's
2154 * environment. This routine changes the Windows behavior to first look in
2155 * the environment, and, if anything is found, use that instead of going to
2156 * the machine default. If there is no environment override, the machine
2157 * default is used, by calling the real setlocale() with "".
2158 *
2159 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2160 * use the particular category's variable if set; otherwise to use the LANG
2161 * variable. */
b385bb4d 2162
175c4cf9 2163 bool override_LC_ALL = FALSE;
89f7b9aa 2164 char * result;
e5f10d49 2165 unsigned int i;
89f7b9aa 2166
b385bb4d 2167 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2168
2169# ifdef LC_ALL
2170
b385bb4d
KW
2171 locale = PerlEnv_getenv("LC_ALL");
2172 if (! locale) {
e5f10d49
KW
2173 if (category == LC_ALL) {
2174 override_LC_ALL = TRUE;
2175 }
2176 else {
7d4bcc4a
KW
2177
2178# endif
7d4bcc4a 2179
e5f10d49
KW
2180 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2181 if (category == categories[i]) {
2182 locale = PerlEnv_getenv(category_names[i]);
2183 goto found_locale;
2184 }
2185 }
7d4bcc4a 2186
b385bb4d 2187 locale = PerlEnv_getenv("LANG");
481465ea 2188 if (! locale) {
b385bb4d
KW
2189 locale = "";
2190 }
e5f10d49
KW
2191
2192 found_locale: ;
7d4bcc4a
KW
2193
2194# ifdef LC_ALL
2195
e5f10d49 2196 }
b385bb4d 2197 }
7d4bcc4a
KW
2198
2199# endif
2200
b385bb4d
KW
2201 }
2202
6c332036
TC
2203#ifdef USE_WSETLOCALE
2204 result = S_wrap_wsetlocale(aTHX_ category, locale);
2205#else
89f7b9aa 2206 result = setlocale(category, locale);
6c332036 2207#endif
48015184
KW
2208 DEBUG_L(STMT_START {
2209 dSAVE_ERRNO;
2210 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2211 setlocale_debug_string(category, locale, result));
2212 RESTORE_ERRNO;
2213 } STMT_END);
89f7b9aa 2214
481465ea 2215 if (! override_LC_ALL) {
89f7b9aa
KW
2216 return result;
2217 }
2218
dfd77d7a 2219 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2220 * LANG variable or the system default if there is no LANG. But these have
2221 * lower priority than the other LC_foo variables, so override it for each
2222 * one that is set. (If they are set to "", it means to use the same thing
2223 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2224
948523db 2225 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2226 result = PerlEnv_getenv(category_names[i]);
2227 if (result && strNE(result, "")) {
6c332036 2228#ifdef USE_WSETLOCALE
17a1e9ac 2229 S_wrap_wsetlocale(aTHX_ categories[i], result);
6c332036 2230#else
17a1e9ac 2231 setlocale(categories[i], result);
6c332036 2232#endif
e5f10d49
KW
2233 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2234 __FILE__, __LINE__,
2235 setlocale_debug_string(categories[i], result, "not captured")));
2236 }
89f7b9aa 2237 }
7d4bcc4a 2238
bbc98134 2239 result = setlocale(LC_ALL, NULL);
48015184
KW
2240 DEBUG_L(STMT_START {
2241 dSAVE_ERRNO;
2242 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2243 __FILE__, __LINE__,
48015184
KW
2244 setlocale_debug_string(LC_ALL, NULL, result));
2245 RESTORE_ERRNO;
2246 } STMT_END);
89f7b9aa 2247
bbc98134 2248 return result;
b385bb4d
KW
2249}
2250
2251#endif
2252
9aac5db8
KW
2253/*
2254
2255=head1 Locale-related functions and macros
2256
2257=for apidoc Perl_setlocale
2258
2259This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2260taking the same parameters, and returning the same information, except that it
9487427b
KW
2261returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2262instead return C<C> if the underlying locale has a non-dot decimal point
2263character, or a non-empty thousands separator for displaying floating point
2264numbers. This is because perl keeps that locale category such that it has a
2265dot and empty separator, changing the locale briefly during the operations
2266where the underlying one is required. C<Perl_setlocale> knows about this, and
2267compensates; regular C<setlocale> doesn't.
9aac5db8 2268
e9bc6d6b 2269Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2270declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2271C<const> (presumably because its API was specified long ago, and can't be
2272updated; it is illegal to change the information C<setlocale> returns; doing
2273so leads to segfaults.)
9aac5db8 2274
e9bc6d6b
KW
2275Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2276C<setlocale> can be completely ineffective on some platforms under some
2277configurations.
2278
9aac5db8 2279C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2280where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2281the system C<setlocale()> is ineffective, returning the wrong information, and
2282failing to actually change the locale. C<Perl_setlocale>, however works
2283properly in all circumstances.
9aac5db8
KW
2284
2285The return points to a per-thread static buffer, which is overwritten the next
2286time C<Perl_setlocale> is called from the same thread.
2287
2288=cut
2289
2290*/
2291
2292const char *
2293Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2294{
2295 /* This wraps POSIX::setlocale() */
2296
52129632 2297#ifndef USE_LOCALE
d36adde0
KW
2298
2299 PERL_UNUSED_ARG(category);
2300 PERL_UNUSED_ARG(locale);
2301
2302 return "C";
2303
2304#else
2305
9aac5db8
KW
2306 const char * retval;
2307 const char * newlocale;
2308 dSAVEDERRNO;
a4f00dcc 2309 dTHX;
d36adde0 2310 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2311
a4f00dcc
KW
2312#ifdef USE_LOCALE_NUMERIC
2313
291a84fb
KW
2314 /* A NULL locale means only query what the current one is. We have the
2315 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2316 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2317 * the correct results. All other categories don't require special
2318 * handling */
a4f00dcc
KW
2319 if (locale == NULL) {
2320 if (category == LC_NUMERIC) {
9aac5db8
KW
2321
2322 /* We don't have to copy this return value, as it is a per-thread
2323 * variable, and won't change until a future setlocale */
2324 return PL_numeric_name;
a4f00dcc
KW
2325 }
2326
7d4bcc4a 2327# ifdef LC_ALL
a4f00dcc 2328
9aac5db8
KW
2329 else if (category == LC_ALL) {
2330 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2331 }
2332
7d4bcc4a 2333# endif
a4f00dcc
KW
2334
2335 }
2336
2337#endif
2338
33e5a354
KW
2339 retval = save_to_buffer(do_setlocale_r(category, locale),
2340 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2341 SAVE_ERRNO;
2342
2343#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2344
2345 if (locale == NULL && category == LC_ALL) {
2346 RESTORE_LC_NUMERIC();
2347 }
2348
2349#endif
a4f00dcc
KW
2350
2351 DEBUG_L(PerlIO_printf(Perl_debug_log,
2352 "%s:%d: %s\n", __FILE__, __LINE__,
2353 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2354
9aac5db8
KW
2355 RESTORE_ERRNO;
2356
2357 if (! retval) {
a4f00dcc
KW
2358 return NULL;
2359 }
2360
9aac5db8 2361 /* If locale == NULL, we are just querying the state */
a4f00dcc 2362 if (locale == NULL) {
a4f00dcc
KW
2363 return retval;
2364 }
a4f00dcc 2365
1159483a
KW
2366 /* Now that have switched locales, we have to update our records to
2367 * correspond. */
a4f00dcc 2368
1159483a 2369 switch (category) {
a4f00dcc 2370
1159483a 2371#ifdef USE_LOCALE_CTYPE
a4f00dcc 2372
1159483a
KW
2373 case LC_CTYPE:
2374 new_ctype(retval);
2375 break;
a4f00dcc 2376
1159483a 2377#endif
a4f00dcc
KW
2378#ifdef USE_LOCALE_COLLATE
2379
1159483a
KW
2380 case LC_COLLATE:
2381 new_collate(retval);
2382 break;
a4f00dcc 2383
1159483a 2384#endif
a4f00dcc
KW
2385#ifdef USE_LOCALE_NUMERIC
2386
1159483a
KW
2387 case LC_NUMERIC:
2388 new_numeric(retval);
2389 break;
a4f00dcc 2390
1159483a
KW
2391#endif
2392#ifdef LC_ALL
a4f00dcc 2393
1159483a 2394 case LC_ALL:
a4f00dcc 2395
1159483a
KW
2396 /* LC_ALL updates all the things we care about. The values may not
2397 * be the same as 'retval', as the locale "" may have set things
2398 * individually */
a4f00dcc 2399
1159483a 2400# ifdef USE_LOCALE_CTYPE
a4f00dcc 2401
b0bde642 2402 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2403 new_ctype(newlocale);
b0bde642 2404 Safefree(newlocale);
a4f00dcc 2405
1159483a
KW
2406# endif /* USE_LOCALE_CTYPE */
2407# ifdef USE_LOCALE_COLLATE
2408
b0bde642 2409 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2410 new_collate(newlocale);
b0bde642 2411 Safefree(newlocale);
a4f00dcc 2412
7d4bcc4a 2413# endif
1159483a 2414# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2415
b0bde642 2416 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2417 new_numeric(newlocale);
b0bde642 2418 Safefree(newlocale);
a4f00dcc 2419
1159483a
KW
2420# endif /* USE_LOCALE_NUMERIC */
2421#endif /* LC_ALL */
a4f00dcc 2422
1159483a
KW
2423 default:
2424 break;
a4f00dcc
KW
2425 }
2426
2427 return retval;
2428
d36adde0
KW
2429#endif
2430
f7416781
KW
2431}
2432
2433PERL_STATIC_INLINE const char *
2434S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2435{
2436 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2437 * growing it if necessary */
2438
0b6697c5 2439 Size_t string_size;
f7416781
KW
2440
2441 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2442
0b6697c5
KW
2443 if (! string) {
2444 return NULL;
2445 }
2446
2447 string_size = strlen(string) + offset + 1;
2448
f7416781
KW
2449 if (*buf_size == 0) {
2450 Newx(*buf, string_size, char);
2451 *buf_size = string_size;
2452 }
2453 else if (string_size > *buf_size) {
2454 Renew(*buf, string_size, char);
2455 *buf_size = string_size;
2456 }
2457
2458 Copy(string, *buf + offset, string_size - offset, char);
2459 return *buf;
2460}
2461
2462/*
2463
f7416781
KW
2464=for apidoc Perl_langinfo
2465
ce181bf6 2466This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2467taking the same C<item> parameter values, and returning the same information.
2468But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2469of Perl's locale handling from your code, and can be used on systems that lack
2470a native C<nl_langinfo>.
2471
2472Expanding on these:
2473
2474=over
2475
2476=item *
2477
ce181bf6
KW
2478The reason it isn't quite a drop-in replacement is actually an advantage. The
2479only difference is that it returns S<C<const char *>>, whereas plain
2480C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2481forbidden to write into the buffer. By declaring this C<const>, the compiler
2482enforces this restriction, so if it is violated, you know at compilation time,
2483rather than getting segfaults at runtime.
2484
2485=item *
2486
69e2275e 2487It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2488without you having to write extra code. The reason for the extra code would be
2489because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2490kept set by Perl so that the radix is a dot, and the separator is the empty
2491string, no matter what the underlying locale is supposed to be, and so to get
2492the expected results, you have to temporarily toggle into the underlying
2493locale, and later toggle back. (You could use plain C<nl_langinfo> and
2494C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2495the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2496(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2497(decimal point) character to be a dot.)
ce181bf6
KW
2498
2499=item *
2500
2501The system function it replaces can have its static return buffer trashed,
2502not only by a subesequent call to that function, but by a C<freelocale>,
2503C<setlocale>, or other locale change. The returned buffer of this function is
2504not changed until the next call to it, so the buffer is never in a trashed
2505state.
f7416781
KW
2506
2507=item *
2508
ce181bf6
KW
2509Its return buffer is per-thread, so it also is never overwritten by a call to
2510this function from another thread; unlike the function it replaces.
2511
2512=item *
2513
2514But most importantly, it works on systems that don't have C<nl_langinfo>, such
2515as Windows, hence makes your code more portable. Of the fifty-some possible
2516items specified by the POSIX 2008 standard,
f7416781 2517L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2518only one is completely unimplemented, though on non-Windows platforms, another
2519significant one is also not implemented). It uses various techniques to
2520recover the other items, including calling C<L<localeconv(3)>>, and
2521C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2522available. Later C<strftime()> versions have additional capabilities; C<""> is
2523returned for those not available on your system.
ce181bf6
KW
2524
2525It is important to note that when called with an item that is recovered by
2526using C<localeconv>, the buffer from any previous explicit call to
2527C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2528contents if you need to access them after a call to this function. (But note
2529that you might not want to be using C<localeconv()> directly anyway, because of
2530issues like the ones listed in the second item of this list (above) for
2531C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2532call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2533unpack).
9f0bc911 2534
6201c220
KW
2535The details for those items which may deviate from what this emulation returns
2536and what a native C<nl_langinfo()> would return are specified in
2537L<I18N::Langinfo>.
f7416781 2538
ce181bf6
KW
2539=back
2540
f7416781
KW
2541When using C<Perl_langinfo> on systems that don't have a native
2542C<nl_langinfo()>, you must
2543
2544 #include "perl_langinfo.h"
2545
2546before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2547C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2548C<langinfo.h> would try to import into the namespace for code that doesn't need
2549it.)
f7416781 2550
f7416781
KW
2551The original impetus for C<Perl_langinfo()> was so that code that needs to
2552find out the current currency symbol, floating point radix character, or digit
2553grouping separator can use, on all systems, the simpler and more
2554thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2555pain to make thread-friendly. For other fields returned by C<localeconv>, it
2556is better to use the methods given in L<perlcall> to call
2557L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2558
2559=cut
2560
2561*/
2562
2563const char *
2564#ifdef HAS_NL_LANGINFO
2565Perl_langinfo(const nl_item item)
2566#else
2567Perl_langinfo(const int item)
2568#endif
2569{
f61748ac
KW
2570 return my_nl_langinfo(item, TRUE);
2571}
2572
59a15af0 2573STATIC const char *
f61748ac
KW
2574#ifdef HAS_NL_LANGINFO
2575S_my_nl_langinfo(const nl_item item, bool toggle)
2576#else
2577S_my_nl_langinfo(const int item, bool toggle)
2578#endif
2579{
ae74815b 2580 dTHX;
0b6697c5 2581 const char * retval;
f7416781 2582
d36adde0
KW
2583#ifdef USE_LOCALE_NUMERIC
2584
5a854ab3
KW
2585 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2586 * two items, and only if not already there */
4e6826bf 2587 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2588 || PL_numeric_underlying))
d36adde0
KW
2589
2590#endif /* No toggling needed if not using LC_NUMERIC */
2591
5a854ab3 2592 toggle = FALSE;
5a854ab3 2593
ab340fff 2594#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2595# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2596 || ! defined(HAS_POSIX_2008_LOCALE) \
2597 || ! defined(DUPLOCALE)
f7416781 2598
ab340fff 2599 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2600 * for those items dependent on it. This must be copied to a buffer before
2601 * switching back, as some systems destroy the buffer when setlocale() is
2602 * called */
f7416781 2603
038d3702
KW
2604 {
2605 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2606
b2fee59c 2607 if (toggle) {
038d3702 2608 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2609 }
f7416781 2610
5acc4454
KW
2611 LOCALE_LOCK; /* Prevent interference from another thread executing
2612 this code section (the only call to nl_langinfo in
2613 the core) */
2614
ee90eb2d 2615
628ff75a
KW
2616 /* Copy to a per-thread buffer, which is also one that won't be
2617 * destroyed by a subsequent setlocale(), such as the
2618 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2619 retval = save_to_buffer(nl_langinfo(item),
2620 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2621
5acc4454
KW
2622 LOCALE_UNLOCK;
2623
b2fee59c 2624 if (toggle) {
038d3702 2625 RESTORE_LC_NUMERIC();
b2fee59c 2626 }
f7416781
KW
2627 }
2628
ab340fff
KW
2629# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2630
5a854ab3 2631 {
b2fee59c
KW
2632 bool do_free = FALSE;
2633 locale_t cur = uselocale((locale_t) 0);
ab340fff 2634
b2fee59c
KW
2635 if (cur == LC_GLOBAL_LOCALE) {
2636 cur = duplocale(LC_GLOBAL_LOCALE);
2637 do_free = TRUE;
2638 }
ab340fff 2639
d36adde0
KW
2640# ifdef USE_LOCALE_NUMERIC
2641
b2fee59c 2642 if (toggle) {
e1aa2579
KW
2643 if (PL_underlying_numeric_obj) {
2644 cur = PL_underlying_numeric_obj;
2645 }
2646 else {
2647 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2648 do_free = TRUE;
2649 }
b2fee59c 2650 }
ab340fff 2651
d36adde0
KW
2652# endif
2653
628ff75a
KW
2654 /* We have to save it to a buffer, because the freelocale() just below
2655 * can invalidate the internal one */
0b6697c5
KW
2656 retval = save_to_buffer(nl_langinfo_l(item, cur),
2657 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2658
b2fee59c
KW
2659 if (do_free) {
2660 freelocale(cur);
2661 }
5a854ab3 2662 }
ab340fff 2663
c1566110
KW
2664# endif
2665
0b6697c5 2666 if (strEQ(retval, "")) {
4e6826bf 2667 if (item == YESSTR) {
c1566110
KW
2668 return "yes";
2669 }
4e6826bf 2670 if (item == NOSTR) {
c1566110
KW
2671 return "no";
2672 }
2673 }
2674
0b6697c5 2675 return retval;
ab340fff 2676
f7416781 2677#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2678
2679 {
2680
f7416781
KW
2681# ifdef HAS_LOCALECONV
2682
43dd6b15 2683 const struct lconv* lc;
628ff75a 2684 const char * temp;
038d3702 2685 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2686
69c5e0db
KW
2687# ifdef TS_W32_BROKEN_LOCALECONV
2688
2689 const char * save_global;
2690 const char * save_thread;
2691 int needed_size;
2692 char * ptr;
2693 char * e;
2694 char * item_start;
2695
2696# endif
f7416781
KW
2697# endif
2698# ifdef HAS_STRFTIME
2699
43dd6b15
KW
2700 struct tm tm;
2701 bool return_format = FALSE; /* Return the %format, not the value */
2702 const char * format;
f7416781
KW
2703
2704# endif
2705
43dd6b15
KW
2706 /* We copy the results to a per-thread buffer, even if not
2707 * multi-threaded. This is in part to simplify this code, and partly
2708 * because we need a buffer anyway for strftime(), and partly because a
2709 * call of localeconv() could otherwise wipe out the buffer, and the
2710 * programmer would not be expecting this, as this is a nl_langinfo()
2711 * substitute after all, so s/he might be thinking their localeconv()
2712 * is safe until another localeconv() call. */
f7416781 2713
43dd6b15
KW
2714 switch (item) {
2715 Size_t len;
f7416781 2716
8d72e74e 2717 /* This is unimplemented */
4e6826bf 2718 case ERA: /* For use with strftime() %E modifier */
f7416781 2719
43dd6b15
KW
2720 default:
2721 return "";
f7416781 2722
43dd6b15 2723 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2724 case YESEXPR: return "^[+1yY]";
2725 case YESSTR: return "yes";
2726 case NOEXPR: return "^[-0nN]";
2727 case NOSTR: return "no";
2728
8d72e74e
KW
2729 case CODESET:
2730
2731# ifndef WIN32
2732
2733 /* On non-windows, this is unimplemented, in part because of
2734 * inconsistencies between vendors. The Darwin native
2735 * nl_langinfo() implementation simply looks at everything past
2736 * any dot in the name, but that doesn't work for other
2737 * vendors. Many Linux locales that don't have UTF-8 in their
2738 * names really are UTF-8, for example; z/OS locales that do
2739 * have UTF-8 in their names, aren't really UTF-8 */
2740 return "";
2741
2742# else
2743
2744 { /* But on Windows, the name does seem to be consistent, so
2745 use that. */
2746 const char * p;
2747 const char * first;
2748 Size_t offset = 0;
2749 const char * name = my_setlocale(LC_CTYPE, NULL);
2750
2751 if (isNAME_C_OR_POSIX(name)) {
2752 return "ANSI_X3.4-1968";
2753 }
2754
2755 /* Find the dot in the locale name */
2756 first = (const char *) strchr(name, '.');
2757 if (! first) {
2758 first = name;
2759 goto has_nondigit;
2760 }
2761
2762 /* Look at everything past the dot */
2763 first++;
2764 p = first;
2765
2766 while (*p) {
2767 if (! isDIGIT(*p)) {
2768 goto has_nondigit;
2769 }
2770
2771 p++;
2772 }
2773
2774 /* Here everything past the dot is a digit. Treat it as a
2775 * code page */
18503cbd 2776 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2777 &PL_langinfo_bufsize, 0);
8d72e74e 2778 offset = STRLENs("CP");
f7416781 2779
8d72e74e
KW
2780 has_nondigit:
2781
2782 retval = save_to_buffer(first, &PL_langinfo_buf,
2783 &PL_langinfo_bufsize, offset);
2784 }
2785
2786 break;
2787
2788# endif
f7416781
KW
2789# ifdef HAS_LOCALECONV
2790
4e6826bf 2791 case CRNCYSTR:
f7416781 2792
43dd6b15
KW
2793 /* We don't bother with localeconv_l() because any system that
2794 * has it is likely to also have nl_langinfo() */
291a84fb 2795
69c5e0db
KW
2796 LOCALE_LOCK_V; /* Prevent interference with other threads
2797 using localeconv() */
2798
2799# ifdef TS_W32_BROKEN_LOCALECONV
2800
2801 /* This is a workaround for a Windows bug prior to VS 15.
2802 * What we do here is, while locked, switch to the global
2803 * locale so localeconv() works; then switch back just before
2804 * the unlock. This can screw things up if some thread is
2805 * already using the global locale while assuming no other is.
2806 * A different workaround would be to call GetCurrencyFormat on
2807 * a known value, and parse it; patches welcome
2808 *
2809 * We have to use LC_ALL instead of LC_MONETARY because of
2810 * another bug in Windows */
2811
2812 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2813 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2814 save_global= savepv(my_setlocale(LC_ALL, NULL));
2815 my_setlocale(LC_ALL, save_thread);
2816
2817# endif
5acc4454 2818
43dd6b15
KW
2819 lc = localeconv();
2820 if ( ! lc
2821 || ! lc->currency_symbol
2822 || strEQ("", lc->currency_symbol))
2823 {
69c5e0db 2824 LOCALE_UNLOCK_V;
43dd6b15
KW
2825 return "";
2826 }
f7416781 2827
43dd6b15 2828 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2829 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2830 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2831 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2832 { /* khw couldn't figure out how the localedef specifications
2833 would show that the $ should replace the radix; this is
2834 just a guess as to how it might work.*/
a34edee3 2835 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2836 }
2837 else if (lc->p_cs_precedes) {
a34edee3 2838 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2839 }
2840 else {
a34edee3 2841 PL_langinfo_buf[0] = '+';
43dd6b15 2842 }
f7416781 2843
69c5e0db
KW
2844# ifdef TS_W32_BROKEN_LOCALECONV
2845
2846 my_setlocale(LC_ALL, save_global);
2847 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2848 my_setlocale(LC_ALL, save_thread);
2849 Safefree(save_global);
2850 Safefree(save_thread);
2851
2852# endif
2853
2854 LOCALE_UNLOCK_V;
43dd6b15 2855 break;
f7416781 2856
69c5e0db
KW
2857# ifdef TS_W32_BROKEN_LOCALECONV
2858
4e6826bf 2859 case RADIXCHAR:
69c5e0db
KW
2860
2861 /* For this, we output a known simple floating point number to
2862 * a buffer, and parse it, looking for the radix */
2863
2864 if (toggle) {
2865 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2866 }
2867
2868 if (PL_langinfo_bufsize < 10) {
2869 PL_langinfo_bufsize = 10;
2870 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2871 }
2872
2873 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2874 "%.1f", 1.5);
2875 if (needed_size >= (int) PL_langinfo_bufsize) {
2876 PL_langinfo_bufsize = needed_size + 1;
2877 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2878 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2879 "%.1f", 1.5);
2880 assert(needed_size < (int) PL_langinfo_bufsize);
2881 }
2882
2883 ptr = PL_langinfo_buf;
2884 e = PL_langinfo_buf + PL_langinfo_bufsize;
2885
2886 /* Find the '1' */
2887 while (ptr < e && *ptr != '1') {
2888 ptr++;
2889 }
2890 ptr++;
2891
2892 /* Find the '5' */
2893 item_start = ptr;
2894 while (ptr < e && *ptr != '5') {
2895 ptr++;
2896 }
2897
2898 /* Everything in between is the radix string */
2899 if (ptr >= e) {
2900 PL_langinfo_buf[0] = '?';
2901 PL_langinfo_buf[1] = '\0';
2902 }
2903 else {
2904 *ptr = '\0';
2905 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2906 }
2907
2908 if (toggle) {
2909 RESTORE_LC_NUMERIC();
2910 }
2911
2912 retval = PL_langinfo_buf;
2913 break;
2914
2915# else
2916
2917 case RADIXCHAR: /* No special handling needed */
2918
2919# endif
2920
4e6826bf 2921 case THOUSEP:
f7416781 2922
43dd6b15 2923 if (toggle) {
038d3702 2924 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2925 }
f7416781 2926
69c5e0db
KW
2927 LOCALE_LOCK_V; /* Prevent interference with other threads
2928 using localeconv() */
2929
2930# ifdef TS_W32_BROKEN_LOCALECONV
2931
2932 /* This should only be for the thousands separator. A
2933 * different work around would be to use GetNumberFormat on a
2934 * known value and parse the result to find the separator */
2935 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2936 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2937 save_global = savepv(my_setlocale(LC_ALL, NULL));
2938 my_setlocale(LC_ALL, save_thread);
2939# if 0
2940 /* This is the start of code that for broken Windows replaces
2941 * the above and below code, and instead calls
2942 * GetNumberFormat() and then would parse that to find the
2943 * thousands separator. It needs to handle UTF-16 vs -8
2944 * issues. */
2945
2946 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2947 DEBUG_L(PerlIO_printf(Perl_debug_log,
2948 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2949 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2950
2951# endif
2952# endif
5acc4454 2953
43dd6b15
KW
2954 lc = localeconv();
2955 if (! lc) {
628ff75a 2956 temp = "";
33394adc 2957 }
43dd6b15 2958 else {
4e6826bf 2959 temp = (item == RADIXCHAR)
43dd6b15
KW
2960 ? lc->decimal_point
2961 : lc->thousands_sep;
628ff75a
KW
2962 if (! temp) {
2963 temp = "";
43dd6b15
KW
2964 }
2965 }
f7416781 2966
0b6697c5
KW
2967 retval = save_to_buffer(temp, &PL_langinfo_buf,
2968 &PL_langinfo_bufsize, 0);
f7416781 2969
69c5e0db
KW
2970# ifdef TS_W32_BROKEN_LOCALECONV
2971
2972 my_setlocale(LC_ALL, save_global);
2973 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2974 my_setlocale(LC_ALL, save_thread);
2975 Safefree(save_global);
2976 Safefree(save_thread);
2977
2978# endif
2979
2980 LOCALE_UNLOCK_V;
5acc4454 2981
43dd6b15 2982 if (toggle) {
038d3702 2983 RESTORE_LC_NUMERIC();
43dd6b15 2984 }
f7416781 2985
43dd6b15 2986 break;
f7416781
KW
2987
2988# endif
2989# ifdef HAS_STRFTIME
2990
43dd6b15
KW
2991 /* These are defined by C89, so we assume that strftime supports
2992 * them, and so are returned unconditionally; they may not be what
2993 * the locale actually says, but should give good enough results
2994 * for someone using them as formats (as opposed to trying to parse
2995 * them to figure out what the locale says). The other format
2996 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2997 case D_FMT: return "%x";
2998 case T_FMT: return "%X";
2999 case D_T_FMT: return "%c";
43dd6b15
KW
3000
3001 /* These formats are only available in later strfmtime's */
4e6826bf 3002 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
3003
3004 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
3005 case ABDAY_1: case ABDAY_2: case ABDAY_3:
3006 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
3007 case ALT_DIGITS:
3008 case AM_STR: case PM_STR:
3009 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
3010 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
3011 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
3012 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
3013 case DAY_5: case DAY_6: case DAY_7:
3014 case MON_1: case MON_2: case MON_3: case MON_4:
3015 case MON_5: case MON_6: case MON_7: case MON_8:
3016 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
3017
3018 LOCALE_LOCK;
3019
3020 init_tm(&tm); /* Precaution against core dumps */
3021 tm.tm_sec = 30;
3022 tm.tm_min = 30;
3023 tm.tm_hour = 6;
3024 tm.tm_year = 2017 - 1900;
3025 tm.tm_wday = 0;
3026 tm.tm_mon = 0;
3027 switch (item) {
3028 default:
3029 LOCALE_UNLOCK;
3030 Perl_croak(aTHX_
3031 "panic: %s: %d: switch case: %d problem",
3032 __FILE__, __LINE__, item);
3033 NOT_REACHED; /* NOTREACHED */
3034
4e6826bf
KW
3035 case PM_STR: tm.tm_hour = 18;
3036 case AM_STR:
43dd6b15
KW
3037 format = "%p";
3038 break;
3039
4e6826bf
KW
3040 case ABDAY_7: tm.tm_wday++;
3041 case ABDAY_6: tm.tm_wday++;
3042 case ABDAY_5: tm.tm_wday++;
3043 case ABDAY_4: tm.tm_wday++;
3044 case ABDAY_3: tm.tm_wday++;
3045 case ABDAY_2: tm.tm_wday++;
3046 case ABDAY_1:
43dd6b15
KW
3047 format = "%a";
3048 break;
3049
4e6826bf
KW
3050 case DAY_7: tm.tm_wday++;
3051 case DAY_6: tm.tm_wday++;
3052 case DAY_5: tm.tm_wday++;
3053 case DAY_4: tm.tm_wday++;
3054 case DAY_3: tm.tm_wday++;
3055 case DAY_2: tm.tm_wday++;
3056 case DAY_1:
43dd6b15
KW
3057 format = "%A";
3058 break;
3059
4e6826bf
KW
3060 case ABMON_12: tm.tm_mon++;
3061 case ABMON_11: tm.tm_mon++;
3062 case ABMON_10: tm.tm_mon++;
3063 case ABMON_9: tm.tm_mon++;
3064 case ABMON_8: tm.tm_mon++;
3065 case ABMON_7: tm.tm_mon++;
3066 case ABMON_6: tm.tm_mon++;
3067 case ABMON_5: tm.tm_mon++;
3068 case ABMON_4: tm.tm_mon++;
3069 case ABMON_3: tm.tm_mon++;
3070 case ABMON_2: tm.tm_mon++;
3071 case ABMON_1:
43dd6b15
KW
3072 format = "%b";
3073 break;
3074
4e6826bf
KW
3075 case MON_12: tm.tm_mon++;
3076 case MON_11: tm.tm_mon++;
3077 case MON_10: tm.tm_mon++;
3078 case MON_9: tm.tm_mon++;
3079 case MON_8: tm.tm_mon++;
3080 case MON_7: tm.tm_mon++;
3081 case MON_6: tm.tm_mon++;
3082 case MON_5: tm.tm_mon++;
3083 case MON_4: tm.tm_mon++;
3084 case MON_3: tm.tm_mon++;
3085 case MON_2: tm.tm_mon++;
3086 case MON_1:
43dd6b15
KW
3087 format = "%B";
3088 break;
3089
4e6826bf 3090 case T_FMT_AMPM:
43dd6b15
KW
3091 format = "%r";
3092 return_format = TRUE;
3093 break;
3094
4e6826bf 3095 case ERA_D_FMT:
43dd6b15
KW
3096 format = "%Ex";
3097 return_format = TRUE;
3098 break;
3099
4e6826bf 3100 case ERA_T_FMT:
43dd6b15
KW
3101 format = "%EX";
3102 return_format = TRUE;
3103 break;
3104
4e6826bf 3105 case ERA_D_T_FMT:
43dd6b15
KW
3106 format = "%Ec";
3107 return_format = TRUE;
3108 break;
3109
4e6826bf 3110 case ALT_DIGITS:
43dd6b15
KW
3111 tm.tm_wday = 0;
3112 format = "%Ow"; /* Find the alternate digit for 0 */
3113 break;
3114 }
f7416781 3115
43dd6b15
KW
3116 /* We can't use my_strftime() because it doesn't look at
3117 * tm_wday */
3118 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3119 format, &tm))
3120 {
3121 /* A zero return means one of:
3122 * a) there wasn't enough space in PL_langinfo_buf
3123 * b) the format, like a plain %p, returns empty
3124 * c) it was an illegal format, though some
3125 * implementations of strftime will just return the
3126 * illegal format as a plain character sequence.
3127 *
3128 * To quickly test for case 'b)', try again but precede
3129 * the format with a plain character. If that result is
3130 * still empty, the problem is either 'a)' or 'c)' */
3131
3132 Size_t format_size = strlen(format) + 1;
3133 Size_t mod_size = format_size + 1;
3134 char * mod_format;
3135 char * temp_result;
3136
3137 Newx(mod_format, mod_size, char);
3138 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 3139 *mod_format = ' ';
43dd6b15
KW
3140 my_strlcpy(mod_format + 1, format, mod_size);
3141 len = strftime(temp_result,
3142 PL_langinfo_bufsize,
3143 mod_format, &tm);
3144 Safefree(mod_format);
3145 Safefree(temp_result);
3146
3147 /* If 'len' is non-zero, it means that we had a case like
3148 * %p which means the current locale doesn't use a.m. or
3149 * p.m., and that is valid */
3150 if (len == 0) {
3151
3152 /* Here, still didn't work. If we get well beyond a
3153 * reasonable size, bail out to prevent an infinite
3154 * loop. */
3155
3156 if (PL_langinfo_bufsize > 100 * format_size) {
3157 *PL_langinfo_buf = '\0';
3158 }
3159 else {
3160 /* Double the buffer size to retry; Add 1 in case
3161 * original was 0, so we aren't stuck at 0. */
3162 PL_langinfo_bufsize *= 2;
3163 PL_langinfo_bufsize++;
3164 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3165 continue;
3166 }
3167 }
f7416781 3168
f7416781 3169 break;
43dd6b15 3170 }
f7416781 3171
43dd6b15
KW
3172 /* Here, we got a result.
3173 *
3174 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3175 * alternate format for wday 0. If the value is the same as
3176 * the normal 0, there isn't an alternate, so clear the buffer.
3177 * */
4e6826bf 3178 if ( item == ALT_DIGITS
43dd6b15
KW
3179 && strEQ(PL_langinfo_buf, "0"))
3180 {
3181 *PL_langinfo_buf = '\0';
3182 }
f7416781 3183
43dd6b15
KW
3184 /* ALT_DIGITS is problematic. Experiments on it showed that
3185 * strftime() did not always work properly when going from
3186 * alt-9 to alt-10. Only a few locales have this item defined,
3187 * and in all of them on Linux that khw was able to find,
3188 * nl_langinfo() merely returned the alt-0 character, possibly
3189 * doubled. Most Unicode digits are in blocks of 10
3190 * consecutive code points, so that is sufficient information
3191 * for those scripts, as we can infer alt-1, alt-2, .... But
3192 * for a Japanese locale, a CJK ideographic 0 is returned, and
3193 * the CJK digits are not in code point order, so you can't
3194 * really infer anything. The localedef for this locale did
3195 * specify the succeeding digits, so that strftime() works
3196 * properly on them, without needing to infer anything. But
3197 * the nl_langinfo() return did not give sufficient information
3198 * for the caller to understand what's going on. So until
3199 * there is evidence that it should work differently, this
3200 * returns the alt-0 string for ALT_DIGITS.
3201 *
3202 * wday was chosen because its range is all a single digit.
3203 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3204
43dd6b15 3205 LOCALE_UNLOCK;
f7416781 3206
d1b150d9
KW
3207 retval = PL_langinfo_buf;
3208
43dd6b15
KW
3209 /* If to return the format, not the value, overwrite the buffer
3210 * with it. But some strftime()s will keep the original format
3211 * if illegal, so change those to "" */
3212 if (return_format) {
3213 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3214 *PL_langinfo_buf = '\0';
3215 }
43dd6b15 3216 else {
0b6697c5
KW
3217 retval = save_to_buffer(format, &PL_langinfo_buf,
3218 &PL_langinfo_bufsize, 0);
f7416781
KW
3219 }
3220 }
3221
3222 break;
f7416781
KW
3223
3224# endif
3225
43dd6b15 3226 }
f7416781
KW
3227 }
3228
0b6697c5 3229 return retval;
f7416781
KW
3230
3231#endif
3232
a4f00dcc 3233}
b385bb4d 3234
98994639
HS
3235/*
3236 * Initialize locale awareness.
3237 */
3238int
3239Perl_init_i18nl10n(pTHX_ int printwarn)
3240{
0e92a118
KW
3241 /* printwarn is
3242 *
3243 * 0 if not to output warning when setup locale is bad
3244 * 1 if to output warning based on value of PERL_BADLANG
3245 * >1 if to output regardless of PERL_BADLANG
3246 *
3247 * returns
98994639 3248 * 1 = set ok or not applicable,
0e92a118
KW
3249 * 0 = fallback to a locale of lower priority
3250 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3251 *
3252 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3253 * set, debugging information is output.
3254 *
3255 * This looks more complicated than it is, mainly due to the #ifdefs.
3256 *
3257 * We try to set LC_ALL to the value determined by the environment. If
3258 * there is no LC_ALL on this platform, we try the individual categories we
3259 * know about. If this works, we are done.
3260 *
3261 * But if it doesn't work, we have to do something else. We search the
3262 * environment variables ourselves instead of relying on the system to do
3263 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3264 * think there is one), and the ultimate fallback "C". This is all done in
3265 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3266 * more complex. The 'trial_locales' array is initialized with just one
3267 * element; it causes the behavior described in the paragraph above this to
3268 * happen. If that fails, we add elements to 'trial_locales', and do extra
3269 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3270 *
3271 * On Ultrix, the locale MUST come from the environment, so there is
3272 * preliminary code to set it. I (khw) am not sure that it is necessary,
3273 * and that this couldn't be folded into the loop, but barring any real
3274 * platforms to test on, it's staying as-is
3275 *
3276 * A slight complication is that in embedded Perls, the locale may already
3277 * be set-up, and we don't want to get it from the normal environment
3278 * variables. This is handled by having a special environment variable
3279 * indicate we're in this situation. We simply set setlocale's 2nd
3280 * parameter to be a NULL instead of "". That indicates to setlocale that
3281 * it is not to change anything, but to return the current value,
3282 * effectively initializing perl's db to what the locale already is.
3283 *
3284 * We play the same trick with NULL if a LC_ALL succeeds. We call
3285 * setlocale() on the individual categores with NULL to get their existing
3286 * values for our db, instead of trying to change them.
3287 * */
98994639 3288
1565c085
DM
3289 dVAR;
3290
0e92a118
KW
3291 int ok = 1;
3292
7d4bcc4a
KW
3293#ifndef USE_LOCALE
3294
3295 PERL_UNUSED_ARG(printwarn);
3296
3297#else /* USE_LOCALE */
7d4bcc4a
KW
3298# ifdef __GLIBC__
3299
24f3e849 3300 const char * const language = PerlEnv_getenv("LANGUAGE");
7d4bcc4a
KW
3301
3302# endif
65ebb059 3303
ccd65d51
KW
3304 /* NULL uses the existing already set up locale */
3305 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3306 ? NULL
3307 : "";
c3fcd832
KW
3308 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3309 unsigned int trial_locales_count;
24f3e849
KW
3310 const char * const lc_all = PerlEnv_getenv("LC_ALL");
3311 const char * const lang = PerlEnv_getenv("LANG");
98994639 3312 bool setlocale_failure = FALSE;
65ebb059 3313 unsigned int i;
175c4cf9
KW
3314
3315 /* A later getenv() could zap this, so only use here */
3316 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3317
3318 const bool locwarn = (printwarn > 1
e5f10d49
KW
3319 || ( printwarn
3320 && ( ! bad_lang_use_once
22ff3130 3321 || (
e5f10d49
KW
3322 /* disallow with "" or "0" */
3323 *bad_lang_use_once
3324 && strNE("0", bad_lang_use_once)))));
ea92aad8 3325
291a84fb 3326 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3327 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3328
3329 /* current locale for given category; should have been copied so aren't
3330 * volatile */
8de4332b 3331 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3332
7d4bcc4a
KW
3333# ifdef WIN32
3334
6bce99ee
JH
3335 /* In some systems you can find out the system default locale
3336 * and use that as the fallback locale. */
7d4bcc4a
KW
3337# define SYSTEM_DEFAULT_LOCALE
3338# endif
3339# ifdef SYSTEM_DEFAULT_LOCALE
3340
65ebb059 3341 const char *system_default_locale = NULL;
98994639 3342
7d4bcc4a 3343# endif
948523db
KW
3344
3345# ifndef DEBUGGING
3346# define DEBUG_LOCALE_INIT(a,b,c)
3347# else
7d4bcc4a 3348
8298454c 3349 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3350
3351# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3352 STMT_START { \
3353 if (debug_initialization) { \
3354 PerlIO_printf(Perl_debug_log, \
3355 "%s:%d: %s\n", \
3356 __FILE__, __LINE__, \
a4f00dcc 3357 setlocale_debug_string(category, \
2fcc0ca9
KW
3358 locale, \
3359 result)); \
3360 } \
3361 } STMT_END
2fcc0ca9 3362
948523db
KW
3363/* Make sure the parallel arrays are properly set up */
3364# ifdef USE_LOCALE_NUMERIC
3365 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3366 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3367# ifdef USE_POSIX_2008_LOCALE
3368 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3369# endif
948523db
KW
3370# endif
3371# ifdef USE_LOCALE_CTYPE
3372 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3373 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3374# ifdef USE_POSIX_2008_LOCALE
3375 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3376# endif
948523db
KW
3377# endif
3378# ifdef USE_LOCALE_COLLATE
3379 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3380 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3381# ifdef USE_POSIX_2008_LOCALE
3382 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3383# endif
948523db
KW
3384# endif
3385# ifdef USE_LOCALE_TIME
3386 assert(categories[LC_TIME_INDEX] == LC_TIME);
3387 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3388# ifdef USE_POSIX_2008_LOCALE
3389 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3390# endif
948523db
KW
3391# endif
3392# ifdef USE_LOCALE_MESSAGES
3393 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3394 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3395# ifdef USE_POSIX_2008_LOCALE
3396 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3397# endif
948523db
KW
3398# endif
3399# ifdef USE_LOCALE_MONETARY
3400 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3401 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3402# ifdef USE_POSIX_2008_LOCALE
3403 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3404# endif
948523db 3405# endif
9821811f
KW
3406# ifdef USE_LOCALE_ADDRESS
3407 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3408 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3409# ifdef USE_POSIX_2008_LOCALE
3410 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3411# endif
9821811f
KW
3412# endif
3413# ifdef USE_LOCALE_IDENTIFICATION
3414 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3415 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3416# ifdef USE_POSIX_2008_LOCALE
3417 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3418# endif
9821811f
KW
3419# endif
3420# ifdef USE_LOCALE_MEASUREMENT
3421 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3422 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3423# ifdef USE_POSIX_2008_LOCALE
3424 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3425# endif
9821811f
KW
3426# endif
3427# ifdef USE_LOCALE_PAPER
3428 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3429 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3430# ifdef USE_POSIX_2008_LOCALE
3431 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3432# endif
9821811f
KW
3433# endif
3434# ifdef USE_LOCALE_TELEPHONE
3435 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3436 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3437# ifdef USE_POSIX_2008_LOCALE
3438 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3439# endif
9821811f 3440# endif
948523db
KW
3441# ifdef LC_ALL
3442 assert(categories[LC_ALL_INDEX] == LC_ALL);
3443 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3444 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3445# ifdef USE_POSIX_2008_LOCALE
3446 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3447# endif
948523db
KW
3448# endif
3449# endif /* DEBUGGING */
47280b20 3450
5a6637f0
KW
3451 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
3452 * why these particular incantations are used. */
d2c9cb53
KW
3453#ifdef HAS_MBRLEN
3454 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
3455#endif
5a6637f0
KW
3456#ifdef HAS_MBRTOWC
3457 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
3458#endif
3459#ifdef HAS_WCTOMBR
3460 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
3461#endif
d2c9cb53 3462
47280b20
KW
3463 /* Initialize the cache of the program's UTF-8ness for the always known
3464 * locales C and POSIX */
3465 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3466 sizeof(PL_locale_utf8ness));
3467
e9bc6d6b
KW
3468# ifdef USE_THREAD_SAFE_LOCALE
3469# ifdef WIN32
3470
3471 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3472
3473# endif
3474# endif
39e69e77 3475# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3476
3477 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3478 if (! PL_C_locale_obj) {
3479 Perl_croak_nocontext(
3480 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3481 }
3482 if (DEBUG_Lv_TEST || debug_initialization) {
3483 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3484 }
3485
3486# endif
3487
78f7c09f
KW
3488# ifdef USE_LOCALE_NUMERIC
3489
3ca88433
KW
3490 PL_numeric_radix_sv = newSVpvs(".");
3491
78f7c09f
KW
3492# endif
3493
e9bc6d6b
KW
3494# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3495
3496 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3497 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3498
3499# endif
b56c4436 3500# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3501
3502 /*
3503 * Ultrix setlocale(..., "") fails if there are no environment
3504 * variables from which to get a locale name.
3505 */
3506
b56c4436
KW
3507# ifndef LC_ALL
3508# error Ultrix without LC_ALL not implemented
3509# else
7d4bcc4a 3510
b56c4436
KW
3511 {
3512 bool done = FALSE;
ea92aad8
KW
3513 if (lang) {
3514 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3515 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3516 if (sl_result[LC_ALL_INDEX])
3517 done = TRUE;
3518 else
e5f10d49 3519 setlocale_failure = TRUE;
ea92aad8
KW
3520 }
3521 if (! setlocale_failure) {
3522 const char * locale_param;
3523 for (i = 0; i < LC_ALL_INDEX; i++) {
3524 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3525 ? setlocale_init
3526 : NULL;
3527 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3528 if (! sl_result[i]) {
3529 setlocale_failure = TRUE;
3530 }
3531 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3532 }
c835d6be 3533 }
7d4bcc4a
KW
3534 }
3535
3536# endif /* LC_ALL */
e5f10d49 3537# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3538
65ebb059 3539 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3540 * the list. Normally the loop is executed just once. But if setting the
3541 * locale fails, inside the loop we add fallback trials to the array and so
3542 * will execute the loop multiple times */
c3fcd832
KW
3543 trial_locales[0] = setlocale_init;
3544 trial_locales_count = 1;
7d4bcc4a 3545
65ebb059
KW
3546 for (i= 0; i < trial_locales_count; i++) {
3547 const char * trial_locale = trial_locales[i];
3548
3549 if (i > 0) {
3550
3551 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3552 * when i==0, but I (khw) don't think that behavior makes much
3553 * sense */
3554 setlocale_failure = FALSE;
3555
7d4bcc4a 3556# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3557# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3558
65ebb059
KW
3559 /* On Windows machines, an entry of "" after the 0th means to use
3560 * the system default locale, which we now proceed to get. */
3561 if (strEQ(trial_locale, "")) {
3562 unsigned int j;
3563
3564 /* Note that this may change the locale, but we are going to do
3565 * that anyway just below */
837ce802 3566 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3567 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3568
7d4bcc4a 3569 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3570 * try */
3571 if (! system_default_locale) {
3572 goto next_iteration;
3573 }
3574 for (j = 0; j < trial_locales_count; j++) {
3575 if (strEQ(system_default_locale, trial_locales[j])) {
3576 goto next_iteration;
3577 }
3578 }
3579
3580 trial_locale = system_default_locale;
3581 }
ec0202b5
KW
3582# else
3583# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3584# endif
7d4bcc4a 3585# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3586
3587 } /* For i > 0 */
65ebb059 3588
7d4bcc4a
KW
3589# ifdef LC_ALL
3590
948523db
KW
3591 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3592 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3593 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3594 setlocale_failure = TRUE;
7cd8b568
KW
3595 }
3596 else {
3597 /* Since LC_ALL succeeded, it should have changed all the other
3598 * categories it can to its value; so we massage things so that the
3599 * setlocales below just return their category's current values.
3600 * This adequately handles the case in NetBSD where LC_COLLATE may
3601 * not be defined for a locale, and setting it individually will
7d4bcc4a 3602 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3603 * the POSIX locale. */
3604 trial_locale = NULL;
3605 }
7d4bcc4a
KW
3606
3607# endif /* LC_ALL */
98994639 3608
e5f10d49
KW
3609 if (! setlocale_failure) {
3610 unsigned int j;
3611 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3612 curlocales[j]
3613 = savepv(do_setlocale_r(categories[j], trial_locale));
3614 if (! curlocales[j]) {
3615 setlocale_failure = TRUE;
3616 }
3617 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3618 }
c835d6be 3619
e5f10d49
KW
3620 if (! setlocale_failure) { /* All succeeded */
3621 break; /* Exit trial_locales loop */
49c85077 3622 }
65ebb059 3623 }
98994639 3624
49c85077
KW
3625 /* Here, something failed; will need to try a fallback. */
3626 ok = 0;
65ebb059 3627
49c85077
KW
3628 if (i == 0) {
3629 unsigned int j;
98994639 3630
65ebb059 3631 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3632
3633# ifdef LC_ALL
98994639 3634
49c85077
KW
3635 PerlIO_printf(Perl_error_log,
3636 "perl: warning: Setting locale failed.\n");
98994639 3637
7d4bcc4a 3638# else /* !LC_ALL */
98994639 3639
49c85077
KW
3640 PerlIO_printf(Perl_error_log,
3641 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3642
e5f10d49
KW
3643 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3644 if (! curlocales[j]) {
3645 PerlIO_printf(Perl_error_log, category_names[j]);
3646 }
3647 else {
3648 Safefree(curlocales[j]);
3649 }
3650 }
7d4bcc4a 3651
7d4bcc4a 3652# endif /* LC_ALL */
98994639 3653
49c85077
KW
3654 PerlIO_printf(Perl_error_log,
3655 "perl: warning: Please check that your locale settings:\n");
98994639 3656
7d4bcc4a
KW
3657# ifdef __GLIBC__
3658
49c85077
KW
3659 PerlIO_printf(Perl_error_log,
3660 "\tLANGUAGE = %c%s%c,\n",
3661 language ? '"' : '(',
3662 language ? language : "unset",
3663 language ? '"' : ')');
7d4bcc4a 3664# endif
98994639 3665
49c85077
KW
3666 PerlIO_printf(Perl_error_log,
3667 "\tLC_ALL = %c%s%c,\n",
3668 lc_all ? '"' : '(',
3669 lc_all ? lc_all : "unset",
3670 lc_all ? '"' : ')');
98994639 3671
7d4bcc4a
KW
3672# if defined(USE_ENVIRON_ARRAY)
3673
49c85077 3674 {
cd999af9 3675 char **e;
d5e32b93
KW
3676
3677 /* Look through the environment for any variables of the
3678 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3679 * already handled above. These are assumed to be locale
3680 * settings. Output them and their values. */
cd999af9 3681 for (e = environ; *e; e++) {
d5e32b93
KW
3682 const STRLEN prefix_len = sizeof("LC_") - 1;
3683 STRLEN uppers_len;
3684
cd999af9 3685 if ( strBEGINs(*e, "LC_")
c8b388b0 3686 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3687 && (uppers_len = strspn(*e + prefix_len,
3688 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3689 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3690 {
3691 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3692 (int) (prefix_len + uppers_len), *e,
3693 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3694 }
3695 }
49c85077 3696 }
7d4bcc4a
KW
3697
3698# else
3699
49c85077
KW
3700 PerlIO_printf(Perl_error_log,
3701 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3702
3703# endif
98994639 3704
49c85077
KW
3705 PerlIO_printf(Perl_error_log,
3706 "\tLANG = %c%s%c\n",
3707 lang ? '"' : '(',
3708 lang ? lang : "unset",
3709 lang ? '"' : ')');
98994639 3710
49c85077
KW
3711 PerlIO_printf(Perl_error_log,
3712 " are supported and installed on your system.\n");
3713 }
98994639 3714
65ebb059 3715 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3716 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3717 * usually change the upper bound of the loop we are in.
3718 *
3719 * Since the system's default way of setting the locale has not
3720 * found one that works, We use Perl's defined ordering: LC_ALL,
3721 * LANG, and the C locale. We don't try the same locale twice, so
3722 * don't add to the list if already there. (On POSIX systems, the
3723 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3724 * but there's no harm done by doing it explicitly.
3725 *
3726 * Note that this tries the LC_ALL environment variable even on
3727 * systems which have no LC_ALL locale setting. This may or may
3728 * not have been originally intentional, but there's no real need
3729 * to change the behavior. */
65ebb059
KW
3730 if (lc_all) {
3731 for (j = 0; j < trial_locales_count; j++) {
3732 if (strEQ(lc_all, trial_locales[j])) {
3733 goto done_lc_all;
3734 }
3735 }
3736 trial_locales[trial_locales_count++] = lc_all;
3737 }
3738 done_lc_all:
98994639 3739
65ebb059
KW
3740 if (lang) {
3741 for (j = 0; j < trial_locales_count; j++) {
3742 if (strEQ(lang, trial_locales[j])) {
3743 goto done_lang;
3744 }
3745 }
3746 trial_locales[trial_locales_count++] = lang;
3747 }
3748 done_lang:
3749
7d4bcc4a
KW
3750# if defined(WIN32) && defined(LC_ALL)
3751
65ebb059
KW
3752 /* For Windows, we also try the system default locale before "C".
3753 * (If there exists a Windows without LC_ALL we skip this because
3754 * it gets too complicated. For those, the "C" is the next
3755 * fallback possibility). The "" is the same as the 0th element of
3756 * the array, but the code at the loop above knows to treat it
3757 * differently when not the 0th */
3758 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3759
3760# endif
65ebb059
KW
3761
3762 for (j = 0; j < trial_locales_count; j++) {
3763 if (strEQ("C", trial_locales[j])) {
3764 goto done_C;
3765 }
3766 }
3767 trial_locales[trial_locales_count++] = "C";
98994639 3768
65ebb059
KW
3769 done_C: ;
3770 } /* end of first time through the loop */
98994639 3771
7d4bcc4a
KW
3772# ifdef WIN32
3773
65ebb059 3774 next_iteration: ;
7d4bcc4a
KW
3775
3776# endif
65ebb059
KW
3777
3778 } /* end of looping through the trial locales */
3779
3780 if (ok < 1) { /* If we tried to fallback */
3781 const char* msg;
3782 if (! setlocale_failure) { /* fallback succeeded */
3783 msg = "Falling back to";
3784 }
3785 else { /* fallback failed */
e5f10d49 3786 unsigned int j;
98994639 3787
65ebb059
KW
3788 /* We dropped off the end of the loop, so have to decrement i to
3789 * get back to the value the last time through */
3790 i--;
98994639 3791
65ebb059
KW
3792 ok = -1;
3793 msg = "Failed to fall back to";
3794
3795 /* To continue, we should use whatever values we've got */
7d4bcc4a 3796
e5f10d49
KW
3797 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3798 Safefree(curlocales[j]);
3799 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3800 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3801 }
65ebb059
KW
3802 }
3803
3804 if (locwarn) {
3805 const char * description;
3806 const char * name = "";
3807 if (strEQ(trial_locales[i], "C")) {
3808 description = "the standard locale";
3809 name = "C";
3810 }
7d4bcc4a
KW
3811
3812# ifdef SYSTEM_DEFAULT_LOCALE
3813
65ebb059
KW
3814 else if (strEQ(trial_locales[i], "")) {
3815 description = "the system default locale";
3816 if (system_default_locale) {
3817 name = system_default_locale;
3818 }
3819 }
7d4bcc4a
KW
3820
3821# endif /* SYSTEM_DEFAULT_LOCALE */
3822
65ebb059
KW
3823 else {
3824 description = "a fallback locale";
3825 name = trial_locales[i];
3826 }
3827 if (name && strNE(name, "")) {
3828 PerlIO_printf(Perl_error_log,
3829 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3830 }
3831 else {
3832 PerlIO_printf(Perl_error_log,
3833 "perl: warning: %s %s.\n", msg, description);
3834 }
3835 }
3836 } /* End of tried to fallback */
98994639 3837
e5f10d49
KW
3838 /* Done with finding the locales; update our records */
3839
7d4bcc4a
KW
3840# ifdef USE_LOCALE_CTYPE
3841
948523db 3842 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3843
e5f10d49 3844# endif
7d4bcc4a
KW
3845# ifdef USE_LOCALE_COLLATE
3846
948523db 3847 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3848
e5f10d49 3849# endif
7d4bcc4a
KW
3850# ifdef USE_LOCALE_NUMERIC
3851
948523db 3852 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3853
3854# endif
3855
948523db 3856 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3857
e9bc6d6b