This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add mutex for accessing ENV
[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
KW
793 }
794 else if (PL_scopestack_ix != 0) {
207cc8df
DM
795 /* To minimize other threads messing with the environment,
796 * we copy the variable, making it a temporary. But this
797 * doesn't work upon program initialization before any
798 * scopes are created, and at this time, there's nothing
799 * else going on that would interfere. So skip the copy
800 * in that case */
801 default_name = savepv(default_name);
e9bc6d6b
KW
802 SAVEFREEPV(default_name);
803 }
804
805 if (category != LC_ALL) {
806 const char * const name = PerlEnv_getenv(category_names[index]);
807
808 /* Here we are setting a single category. Assume will have the
809 * default name */
810 locale = default_name;
811
812 /* But then look for an overriding environment variable */
813 if (name && strNE(name, "")) {
814 locale = name;
815 }
816 }
817 else {
818 bool did_override = FALSE;
819 unsigned int i;
820
821 /* Here, we are getting LC_ALL. Any categories that don't have
822 * a corresponding environment variable set should be set to
823 * LANG, or to "C" if there is no LANG. If no individual
824 * categories differ from this, we can just set LC_ALL. This
825 * is buggy on systems that have extra categories that we don't
826 * know about. If there is an environment variable that sets
827 * that category, we won't know to look for it, and so our use
828 * of LANG or "C" improperly overrides it. On the other hand,
829 * if we don't do what is done here, and there is no
830 * environment variable, the category's locale should be set to
831 * LANG or "C". So there is no good solution. khw thinks the
832 * best is to look at systems to see what categories they have,
833 * and include them, and then to assume that we know the
834 * complete set */
835
836 for (i = 0; i < LC_ALL_INDEX; i++) {
837 const char * const env_override
838 = savepv(PerlEnv_getenv(category_names[i]));
839 const char * this_locale = ( env_override
840 && strNE(env_override, ""))
841 ? env_override
842 : default_name;
6435f98d
KW
843 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
844 {
845 Safefree(env_override);
846 return NULL;
847 }
e9bc6d6b
KW
848
849 if (strNE(this_locale, default_name)) {
850 did_override = TRUE;
851 }
852
853 Safefree(env_override);
854 }
855
856 /* If all the categories are the same, we can set LC_ALL to
857 * that */
858 if (! did_override) {
859 locale = default_name;
860 }
861 else {
862
863 /* Here, LC_ALL is no longer valid, as some individual
864 * categories don't match it. We call ourselves
865 * recursively, as that will execute the code that
866 * generates the proper locale string for this situation.
867 * We don't do the remainder of this function, as that is
868 * to update our records, and we've just done that for the
869 * individual categories in the loop above, and doing so
870 * would cause LC_ALL to be done as well */
871 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
872 }
873 }
874 }
6aba5c5e 875 } /* End of this being setlocale(LC_foo, "") */
e9bc6d6b
KW
876 else if (strchr(locale, ';')) {
877
878 /* LC_ALL may actually incude a conglomeration of various categories.
879 * Without querylocale, this code uses the glibc (as of this writing)
880 * syntax for representing that, but that is not a stable API, and
881 * other platforms do it differently, so we have to handle all cases
882 * ourselves */
883
f0023d47 884 unsigned int i;
e9bc6d6b
KW
885 const char * s = locale;
886 const char * e = locale + strlen(locale);
887 const char * p = s;
888 const char * category_end;
889 const char * name_start;
890 const char * name_end;
891
f0023d47
KW
892 /* If the string that gives what to set doesn't include all categories,
893 * the omitted ones get set to "C". To get this behavior, first set
894 * all the individual categories to "C", and override the furnished
895 * ones below */
896 for (i = 0; i < LC_ALL_INDEX; i++) {
897 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
898 return NULL;
899 }
900 }
901
e9bc6d6b 902 while (s < e) {
e9bc6d6b
KW
903
904 /* Parse through the category */
905 while (isWORDCHAR(*p)) {
906 p++;
907 }
908 category_end = p;
909
910 if (*p++ != '=') {
911 Perl_croak(aTHX_
912 "panic: %s: %d: Unexpected character in locale name '%02X",
913 __FILE__, __LINE__, *(p-1));
914 }
915
916 /* Parse through the locale name */
917 name_start = p;
bee74f4b
KW
918 while (p < e && *p != ';') {
919 if (! isGRAPH(*p)) {
920 Perl_croak(aTHX_
921 "panic: %s: %d: Unexpected character in locale name '%02X",
922 __FILE__, __LINE__, *(p-1));
923 }
e9bc6d6b
KW
924 p++;
925 }
926 name_end = p;
927
bee74f4b
KW
928 /* Space past the semi-colon */
929 if (p < e) {
930 p++;
e9bc6d6b
KW
931 }
932
933 /* Find the index of the category name in our lists */
934 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 935 char * individ_locale;
e9bc6d6b
KW
936
937 /* Keep going if this isn't the index. The strnNE() avoids a
938 * Perl_form(), but would fail if ever a category name could be
939 * a substring of another one, like if there were a
940 * "LC_TIME_DATE" */
941 if strnNE(s, category_names[i], category_end - s) {
942 continue;
943 }
944
945 /* If this index is for the single category we're changing, we
946 * have found the locale to set it to. */
947 if (category == categories[i]) {
948 locale = Perl_form(aTHX_ "%.*s",
949 (int) (name_end - name_start),
950 name_start);
951 goto ready_to_set;
952 }
953
3c76a412 954 assert(category == LC_ALL);
bee74f4b
KW
955 individ_locale = Perl_form(aTHX_ "%.*s",
956 (int) (name_end - name_start), name_start);
6435f98d
KW
957 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
958 {
959 return NULL;
960 }
e9bc6d6b
KW
961 }
962
963 s = p;
964 }
965
966 /* Here we have set all the individual categories by recursive calls.
967 * These collectively should have fixed up LC_ALL, so can just query
968 * what that now is */
969 assert(category == LC_ALL);
970
971 return do_setlocale_c(LC_ALL, NULL);
6aba5c5e
KW
972 } /* End of this being setlocale(LC_ALL,
973 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
e9bc6d6b
KW
974
975 ready_to_set: ;
976
f8115d60
KW
977 /* Here at the end of having to deal with the absence of querylocale().
978 * Some cases have already been fully handled by recursive calls to this
979 * function. But at this point, we haven't dealt with those, but are now
980 * prepared to, knowing what the locale name to set this category to is.
981 * This would have come for free if this system had had querylocale() */
982
e9bc6d6b
KW
983# endif /* end of ! querylocale */
984
f8115d60
KW
985 assert(PL_C_locale_obj);
986
987 /* Switching locales generally entails freeing the current one's space (at
988 * the C library's discretion). We need to stop using that locale before
989 * the switch. So switch to a known locale object that we don't otherwise
990 * mess with. This returns the locale object in effect at the time of the
991 * switch. */
992 old_obj = uselocale(PL_C_locale_obj);
993
994# ifdef DEBUGGING
995
996 if (DEBUG_Lv_TEST || debug_initialization) {
997 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
998 }
999
1000# endif
1001
1002 if (! old_obj) {
1003
1004# ifdef DEBUGGING
1005
1006 if (DEBUG_L_TEST || debug_initialization) {
1007 dSAVE_ERRNO;
1008 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1009 RESTORE_ERRNO;
1010 }
1011
1012# endif
1013
1014 return NULL;
1015 }
1016
1017# ifdef DEBUGGING
1018
1019 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1020 PerlIO_printf(Perl_debug_log,
1021 "%s:%d: emulate_setlocale now using %p\n",
1022 __FILE__, __LINE__, PL_C_locale_obj);
f8115d60
KW
1023 }
1024
1025# endif
1026
6aba5c5e
KW
1027 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1028 * in fact, we already have switched to it (in preparation for what
1029 * normally is to come). But since we're already there, continue to use
70bd6bc8
KW
1030 * it instead of trying to create a new locale */
1031 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1032
1033# ifdef DEBUGGING
1034
1035 if (DEBUG_Lv_TEST || debug_initialization) {
1036 PerlIO_printf(Perl_debug_log,
1037 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1038 }
1039
1040# endif
1041
1042 new_obj = PL_C_locale_obj;
1043
1044 /* We already had switched to the C locale in preparation for freeing
b22e9937 1045 * 'old_obj' */
70bd6bc8
KW
1046 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1047 freelocale(old_obj);
1048 }
1049 }
1050 else {
b22e9937
KW
1051 /* If we weren't in a thread safe locale, set so that newlocale() below
1052 * which uses 'old_obj', uses an empty one. Same for our reserved C
1053 * object. The latter is defensive coding, so that, even if there is
1054 * some bug, we will never end up trying to modify either of these, as
1055 * if passed to newlocale(), they can be. */
1056 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1057 old_obj = (locale_t) 0;
1058 }
f8115d60 1059
b22e9937
KW
1060 /* Ready to create a new locale by modification of the exising one */
1061 new_obj = newlocale(mask, locale, old_obj);
e9bc6d6b 1062
b22e9937
KW
1063 if (! new_obj) {
1064 dSAVE_ERRNO;
e9bc6d6b
KW
1065
1066# ifdef DEBUGGING
1067
b22e9937
KW
1068 if (DEBUG_L_TEST || debug_initialization) {
1069 PerlIO_printf(Perl_debug_log,
1070 "%s:%d: emulate_setlocale creating new object"
1071 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1072 }
e9bc6d6b
KW
1073
1074# endif
1075
b22e9937 1076 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1077
1078# ifdef DEBUGGING
1079
b22e9937
KW
1080 if (DEBUG_L_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log,
1082 "%s:%d: switching back failed: %d\n",
1083 __FILE__, __LINE__, GET_ERRNO);
1084 }
e9bc6d6b
KW
1085
1086# endif
1087
b22e9937
KW
1088 }
1089 RESTORE_ERRNO;
1090 return NULL;
e9bc6d6b 1091 }
e9bc6d6b
KW
1092
1093# ifdef DEBUGGING
1094
b22e9937
KW
1095 if (DEBUG_Lv_TEST || debug_initialization) {
1096 PerlIO_printf(Perl_debug_log,
1097 "%s:%d: emulate_setlocale created %p",
1098 __FILE__, __LINE__, new_obj);
1099 if (old_obj) {
1100 PerlIO_printf(Perl_debug_log,
1101 "; should have freed %p", old_obj);
1102 }
1103 PerlIO_printf(Perl_debug_log, "\n");
19ee3daf 1104 }
e9bc6d6b
KW
1105
1106# endif
1107
b22e9937
KW
1108 /* And switch into it */
1109 if (! uselocale(new_obj)) {
1110 dSAVE_ERRNO;
e9bc6d6b
KW
1111
1112# ifdef DEBUGGING
1113
b22e9937
KW
1114 if (DEBUG_L_TEST || debug_initialization) {
1115 PerlIO_printf(Perl_debug_log,
1116 "%s:%d: emulate_setlocale switching to new object"
1117 " failed\n", __FILE__, __LINE__);
1118 }
e9bc6d6b
KW
1119
1120# endif
1121
b22e9937 1122 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1123
1124# ifdef DEBUGGING
1125
b22e9937
KW
1126 if (DEBUG_L_TEST || debug_initialization) {
1127 PerlIO_printf(Perl_debug_log,
1128 "%s:%d: switching back failed: %d\n",
1129 __FILE__, __LINE__, GET_ERRNO);
1130 }
e9bc6d6b
KW
1131
1132# endif
1133
b22e9937
KW
1134 }
1135 freelocale(new_obj);
1136 RESTORE_ERRNO;
1137 return NULL;
e9bc6d6b 1138 }
70bd6bc8 1139 }
e9bc6d6b
KW
1140
1141# ifdef DEBUGGING
1142
1143 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1144 PerlIO_printf(Perl_debug_log,
1145 "%s:%d: emulate_setlocale now using %p\n",
1146 __FILE__, __LINE__, new_obj);
e9bc6d6b
KW
1147 }
1148
1149# endif
1150
1151 /* We are done, except for updating our records (if the system doesn't keep
1152 * them) and in the case of locale "", we don't actually know what the
1153 * locale that got switched to is, as it came from the environment. So
1154 * have to find it */
1155
1156# ifdef HAS_QUERYLOCALE
1157
1158 if (strEQ(locale, "")) {
1159 locale = querylocale(mask, new_obj);
1160 }
1161
1162# else
1163
1164 /* Here, 'locale' is the return value */
1165
1166 /* Without querylocale(), we have to update our records */
1167
1168 if (category == LC_ALL) {
1169 unsigned int i;
1170
1171 /* For LC_ALL, we change all individual categories to correspond */
1172 /* PL_curlocales is a parallel array, so has same
1173 * length as 'categories' */
1174 for (i = 0; i <= LC_ALL_INDEX; i++) {
1175 Safefree(PL_curlocales[i]);
1176 PL_curlocales[i] = savepv(locale);
1177 }
45cef8fb
KW
1178
1179 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
e9bc6d6b
KW
1180 }
1181 else {
1182
1183 /* For a single category, if it's not the same as the one in LC_ALL, we
1184 * nullify LC_ALL */
1185
1186 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1187 Safefree(PL_curlocales[LC_ALL_INDEX]);
1188 PL_curlocales[LC_ALL_INDEX] = NULL;
1189 }
1190
1191 /* Then update the category's record */
1192 Safefree(PL_curlocales[index]);
1193 PL_curlocales[index] = savepv(locale);
45cef8fb
KW
1194
1195 FIX_GLIBC_LC_MESSAGES_BUG(index);
e9bc6d6b
KW
1196 }
1197
1198# endif
1199
1200 return locale;
1201}
1202
1203#endif /* USE_POSIX_2008_LOCALE */
1204
1205#if 0 /* Code that was to emulate thread-safe locales on platforms that
1206 didn't natively support them */
1207
1208/* The way this would work is that we would keep a per-thread list of the
1209 * correct locale for that thread. Any operation that was locale-sensitive
1210 * would have to be changed so that it would look like this:
1211 *
1212 * LOCALE_LOCK;
1213 * setlocale to the correct locale for this operation
1214 * do operation
1215 * LOCALE_UNLOCK
1216 *
1217 * This leaves the global locale in the most recently used operation's, but it
1218 * was locked long enough to get the result. If that result is static, it
1219 * needs to be copied before the unlock.
1220 *
1221 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1222 * the setup, but are no-ops when not needed, and similarly,
1223 * END_LOCALE_DEPENDENT_OP for the tear-down
1224 *
1225 * But every call to a locale-sensitive function would have to be changed, and
1226 * if a module didn't cooperate by using the mutex, things would break.
1227 *
1228 * This code was abandoned before being completed or tested, and is left as-is
1229*/
1230
1231# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1232# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1233
1234STATIC char *
1235S_locking_setlocale(pTHX_
1236 const int category,
1237 const char * locale,
1238 int index,
1239 const bool is_index_valid
1240 )
1241{
1242 /* This function kind of performs a setlocale() on just the current thread;
1243 * thus it is kind of thread-safe. It does this by keeping a thread-level
1244 * array of the current locales for each category. Every time a locale is
1245 * switched to, it does the switch globally, but updates the thread's
1246 * array. A query as to what the current locale is just returns the
1247 * appropriate element from the array, and doesn't actually call the system
1248 * setlocale(). The saving into the array is done in an uninterruptible
1249 * section of code, so is unaffected by whatever any other threads might be
1250 * doing.
1251 *
1252 * All locale-sensitive operations must work by first starting a critical
1253 * section, then switching to the thread's locale as kept by this function,
1254 * and then doing the operation, then ending the critical section. Thus,
1255 * each gets done in the appropriate locale. simulating thread-safety.
1256 *
1257 * This function takes the same parameters, 'category' and 'locale', that
1258 * the regular setlocale() function does, but it also takes two additional
1259 * ones. This is because as described earlier. If we know on input the
1260 * index corresponding to the category into the array where we store the
1261 * current locales, we don't have to calculate it. If the caller knows at
06cc5386 1262 * compile time what the index is, it can pass it, setting
e9bc6d6b
KW
1263 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1264 *
1265 */
1266
1267 /* If the input index might be incorrect, calculate the correct one */
1268 if (! is_index_valid) {
1269 unsigned int i;
1270
1271 if (DEBUG_Lv_TEST || debug_initialization) {
1272 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1273 }
1274
1275 for (i = 0; i <= LC_ALL_INDEX; i++) {
1276 if (category == categories[i]) {
1277 index = i;
1278 goto found_index;
1279 }
1280 }
1281
1282 /* Here, we don't know about this category, so can't handle it.
1283 * XXX best we can do is to unsafely set this
1284 * XXX warning */
1285
1286 return my_setlocale(category, locale);
1287
1288 found_index: ;
1289
1290 if (DEBUG_Lv_TEST || debug_initialization) {
1291 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1292 }
1293 }
1294
1295 /* For a query, just return what's in our records */
1296 if (new_locale == NULL) {
1297 return curlocales[index];
1298 }
1299
1300
1301 /* Otherwise, we need to do the switch, and save the result, all in a
1302 * critical section */
1303
1304 Safefree(curlocales[[index]]);
1305
1306 /* It might be that this is called from an already-locked section of code.
1307 * We would have to detect and skip the LOCK/UNLOCK if so */
1308 LOCALE_LOCK;
1309
1310 curlocales[index] = savepv(my_setlocale(category, new_locale));
1311
1312 if (strEQ(new_locale, "")) {
1313
1314#ifdef LC_ALL
1315
1316 /* The locale values come from the environment, and may not all be the
1317 * same, so for LC_ALL, we have to update all the others, while the
1318 * mutex is still locked */
1319
1320 if (category == LC_ALL) {
1321 unsigned int i;
1322 for (i = 0; i < LC_ALL_INDEX) {
1323 curlocales[i] = my_setlocale(categories[i], NULL);
1324 }
1325 }
1326 }
1327
1328#endif
1329
1330 LOCALE_UNLOCK;
1331
1332 return curlocales[index];
1333}
1334
1335#endif
d36adde0 1336#ifdef USE_LOCALE
837ce802 1337
a4f00dcc 1338STATIC void
86799d2d 1339S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1340{
86799d2d
KW
1341 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1342 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1343
86799d2d
KW
1344#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1345 || defined(HAS_NL_LANGINFO))
98994639 1346
87f8e8e7 1347 const char * radix = (use_locale)
4e6826bf 1348 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1349 /* FALSE => already in dest locale */
1350 : ".";
2213a3be 1351
87f8e8e7 1352 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1353
87f8e8e7
KW
1354 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1355 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1356 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1357 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1358 && _is_cur_LC_category_utf8(LC_NUMERIC))
1359 {
1360 SvUTF8_on(PL_numeric_radix_sv);
1361 }
86799d2d
KW
1362
1363# ifdef DEBUGGING
7d4bcc4a 1364
2fcc0ca9
KW
1365 if (DEBUG_L_TEST || debug_initialization) {
1366 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1367 SvPVX(PL_numeric_radix_sv),
1368 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1369 }
69014004 1370
86799d2d 1371# endif
d36adde0
KW
1372#else
1373
1374 PERL_UNUSED_ARG(use_locale);
1375
86799d2d 1376#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1377
98994639
HS
1378}
1379
398eeea9
KW
1380STATIC void
1381S_new_numeric(pTHX_ const char *newnum)
98994639 1382{
7d4bcc4a
KW
1383
1384#ifndef USE_LOCALE_NUMERIC
1385
1386 PERL_UNUSED_ARG(newnum);
1387
1388#else
0d071d52 1389
291a84fb 1390 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1391 * core Perl this and that 'newnum' is the name of the new locale.
1392 * It installs this locale as the current underlying default.
1393 *
1394 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1395 * set_numeric_underlying() and set_numeric_standard() functions, which
1396 * should probably not be called directly, but only via macros like
0d071d52
KW
1397 * SET_NUMERIC_STANDARD() in perl.h.
1398 *
1399 * The toggling is necessary mainly so that a non-dot radix decimal point
1400 * character can be output, while allowing internal calculations to use a
1401 * dot.
1402 *
1403 * This sets several interpreter-level variables:
bb304765 1404 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1405 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1406 * that the current locale is the program's underlying
1407 * locale
1408 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1409 * that the current locale is the C locale or
1410 * indistinguishable from the C locale. If non-zero, it
1411 * is in C; if > 1, it means it may not be toggled away
7738054c 1412 * from C.
4c68b815
KW
1413 * PL_numeric_underlying_is_standard A bool kept by this function
1414 * indicating that the underlying locale and the standard
1415 * C locale are indistinguishable for the purposes of
1416 * LC_NUMERIC. This happens when both of the above two
1417 * variables are true at the same time. (Toggling is a
1418 * no-op under these circumstances.) This variable is
1419 * used to avoid having to recalculate.
398eeea9 1420 */
0d071d52 1421
b03f34cf 1422 char *save_newnum;
98994639
HS
1423
1424 if (! newnum) {
43c5f42d
NC
1425 Safefree(PL_numeric_name);
1426 PL_numeric_name = NULL;
98994639 1427 PL_numeric_standard = TRUE;
892e6465 1428 PL_numeric_underlying = TRUE;
4c68b815 1429 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1430 return;
1431 }
1432
b03f34cf 1433 save_newnum = stdize_locale(savepv(newnum));
892e6465 1434 PL_numeric_underlying = TRUE;
4c68b815
KW
1435 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1436
69c5e0db
KW
1437#ifndef TS_W32_BROKEN_LOCALECONV
1438
4c68b815 1439 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1440 * them. But on broken Windows systems calling my_nl_langinfo() for
1441 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1442 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1443 if (! PL_numeric_standard) {
4e6826bf 1444 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1445 FALSE /* Don't toggle locale */ ))
4e6826bf 1446 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1447 }
abe1abcf 1448
69c5e0db
KW
1449#endif
1450
4c68b815 1451 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1452 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1453 Safefree(PL_numeric_name);
b03f34cf 1454 PL_numeric_name = save_newnum;
b03f34cf 1455 }
abe1abcf
KW
1456 else {
1457 Safefree(save_newnum);
1458 }
4c28b29c 1459
4c68b815
KW
1460 PL_numeric_underlying_is_standard = PL_numeric_standard;
1461
7e5377f7 1462# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1463
1464 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1465 PL_numeric_name,
1466 PL_underlying_numeric_obj);
1467
1468#endif
1469
4c68b815
KW
1470 if (DEBUG_L_TEST || debug_initialization) {
1471 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1472 }
1473
4c28b29c
KW
1474 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1475 * have to worry about the radix being a non-dot. (Core operations that
1476 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1477 if (PL_numeric_standard) {
1478 set_numeric_radix(0);
1479 }
1480 else {
1481 set_numeric_standard();
1482 }
4c28b29c 1483
98994639 1484#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1485
98994639
HS
1486}
1487
1488void
1489Perl_set_numeric_standard(pTHX)
1490{
7d4bcc4a 1491
98994639 1492#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1493
28c1bf33
KW
1494 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1495 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1496 * macro avoids calling this routine if toggling isn't necessary according
1497 * to our records (which could be wrong if some XS code has changed the
1498 * locale behind our back) */
0d071d52 1499
7d4bcc4a
KW
1500# ifdef DEBUGGING
1501
2fcc0ca9
KW
1502 if (DEBUG_L_TEST || debug_initialization) {
1503 PerlIO_printf(Perl_debug_log,
7bb8f702 1504 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1505 }
98994639 1506
7d4bcc4a 1507# endif
7bb8f702
KW
1508
1509 do_setlocale_c(LC_NUMERIC, "C");
1510 PL_numeric_standard = TRUE;
1511 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1512 set_numeric_radix(0);
1513
98994639 1514#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1515
98994639
HS
1516}
1517
1518void
5792c642 1519Perl_set_numeric_underlying(pTHX)
98994639 1520{
7d4bcc4a 1521
98994639 1522#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1523
28c1bf33 1524 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1525 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1526 * instead of calling this directly. The macro avoids calling this routine
1527 * if toggling isn't necessary according to our records (which could be
1528 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1529
7d4bcc4a
KW
1530# ifdef DEBUGGING
1531
2fcc0ca9
KW
1532 if (DEBUG_L_TEST || debug_initialization) {
1533 PerlIO_printf(Perl_debug_log,
7bb8f702 1534 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1535 PL_numeric_name);
1536 }
98994639 1537
7d4bcc4a 1538# endif
7bb8f702
KW
1539
1540 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1541 PL_numeric_standard = PL_numeric_underlying_is_standard;
1542 PL_numeric_underlying = TRUE;
1543 set_numeric_radix(! PL_numeric_standard);
1544
98994639 1545#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1546
98994639
HS
1547}
1548
1549/*
1550 * Set up for a new ctype locale.
1551 */
a4f00dcc
KW
1552STATIC void
1553S_new_ctype(pTHX_ const char *newctype)
98994639 1554{
7d4bcc4a
KW
1555
1556#ifndef USE_LOCALE_CTYPE
1557
7d4bcc4a
KW
1558 PERL_UNUSED_ARG(newctype);
1559 PERL_UNUSED_CONTEXT;
1560
1561#else
0d071d52 1562
291a84fb 1563 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1564 * core Perl this and that 'newctype' is the name of the new locale.
1565 *
1566 * This function sets up the folding arrays for all 256 bytes, assuming
1567 * that tofold() is tolc() since fold case is not a concept in POSIX,
1568 *
1569 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1570 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1571 * this function should be called directly only from this file and from
0d071d52
KW
1572 * POSIX::setlocale() */
1573
27da23d5 1574 dVAR;
013f4e03 1575 unsigned int i;
98994639 1576
8b7358b9
KW
1577 /* Don't check for problems if we are suppressing the warnings */
1578 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
30d8090d 1579 bool maybe_utf8_turkic = FALSE;
8b7358b9 1580
7918f24d
NC
1581 PERL_ARGS_ASSERT_NEW_CTYPE;
1582
215c5139
KW
1583 /* We will replace any bad locale warning with 1) nothing if the new one is
1584 * ok; or 2) a new warning for the bad new locale */
1585 if (PL_warn_locale) {
1586 SvREFCNT_dec_NN(PL_warn_locale);
1587 PL_warn_locale = NULL;
1588 }
1589
c1284011 1590 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1591
1592 /* A UTF-8 locale gets standard rules. But note that code still has to
1593 * handle this specially because of the three problematic code points */
1594 if (PL_in_utf8_CTYPE_locale) {
1595 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
30d8090d
KW
1596
1597 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1598 * Turkic. Make sure these two are the only anomalies. (We don't use
1599 * towupper and towlower because they aren't in C89.) */
5b64f24c
KW
1600
1601#if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1602
1603 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1604
1605#else
1606
30d8090d 1607 if (toupper('i') == 'i' && tolower('I') == 'I') {
5b64f24c
KW
1608
1609#endif
30d8090d
KW
1610 check_for_problems = TRUE;
1611 maybe_utf8_turkic = TRUE;
1612 }
31f05a37 1613 }
8b7358b9
KW
1614
1615 /* We don't populate the other lists if a UTF-8 locale, but do check that
1616 * everything works as expected, unless checking turned off */
1617 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1618 /* Assume enough space for every character being bad. 4 spaces each
1619 * for the 94 printable characters that are output like "'x' "; and 5
1620 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1621 * NUL */
8b7358b9 1622 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1623 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1624 to start */
1625 unsigned int bad_count = 0; /* Count of bad characters */
1626
baa60164 1627 for (i = 0; i < 256; i++) {
8b7358b9 1628 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1629 if (isupper(i))
1630 PL_fold_locale[i] = (U8) tolower(i);
1631 else if (islower(i))
1632 PL_fold_locale[i] = (U8) toupper(i);
1633 else
1634 PL_fold_locale[i] = (U8) i;
8b7358b9 1635 }
8c6180a9
KW
1636
1637 /* If checking for locale problems, see if the native ASCII-range
1638 * printables plus \n and \t are in their expected categories in
1639 * the new locale. If not, this could mean big trouble, upending
1640 * Perl's and most programs' assumptions, like having a
1641 * metacharacter with special meaning become a \w. Fortunately,
1642 * it's very rare to find locales that aren't supersets of ASCII
1643 * nowadays. It isn't a problem for most controls to be changed
1644 * into something else; we check only \n and \t, though perhaps \r
1645 * could be an issue as well. */
7d4bcc4a 1646 if ( check_for_problems
8c6180a9
KW
1647 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1648 {
8b7358b9 1649 bool is_bad = FALSE;
6726b4f4 1650 char name[4] = { '\0' };
8b7358b9
KW
1651
1652 /* Convert the name into a string */
6726b4f4 1653 if (isGRAPH_A(i)) {
8b7358b9
KW
1654 name[0] = i;
1655 name[1] = '\0';
1656 }
1657 else if (i == '\n') {
6726b4f4
KW
1658 my_strlcpy(name, "\\n", sizeof(name));
1659 }
1660 else if (i == '\t') {
1661 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1662 }
1663 else {
6726b4f4
KW
1664 assert(i == ' ');
1665 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1666 }
1667
1668 /* Check each possibe class */
1669 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1670 is_bad = TRUE;
1671 DEBUG_L(PerlIO_printf(Perl_debug_log,
1672 "isalnum('%s') unexpectedly is %d\n",
1673 name, cBOOL(isalnum(i))));
1674 }
1675 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1676 is_bad = TRUE;
1677 DEBUG_L(PerlIO_printf(Perl_debug_log,
1678 "isalpha('%s') unexpectedly is %d\n",
1679 name, cBOOL(isalpha(i))));
1680 }
1681 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1682 is_bad = TRUE;
1683 DEBUG_L(PerlIO_printf(Perl_debug_log,
1684 "isdigit('%s') unexpectedly is %d\n",
1685 name, cBOOL(isdigit(i))));
1686 }
1687 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1688 is_bad = TRUE;
1689 DEBUG_L(PerlIO_printf(Perl_debug_log,
1690 "isgraph('%s') unexpectedly is %d\n",
1691 name, cBOOL(isgraph(i))));
1692 }
1693 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1694 is_bad = TRUE;
1695 DEBUG_L(PerlIO_printf(Perl_debug_log,
1696 "islower('%s') unexpectedly is %d\n",
1697 name, cBOOL(islower(i))));
1698 }
1699 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1700 is_bad = TRUE;
1701 DEBUG_L(PerlIO_printf(Perl_debug_log,
1702 "isprint('%s') unexpectedly is %d\n",
1703 name, cBOOL(isprint(i))));
1704 }
1705 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1706 is_bad = TRUE;
1707 DEBUG_L(PerlIO_printf(Perl_debug_log,
1708 "ispunct('%s') unexpectedly is %d\n",
1709 name, cBOOL(ispunct(i))));
1710 }
1711 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1712 is_bad = TRUE;
1713 DEBUG_L(PerlIO_printf(Perl_debug_log,
1714 "isspace('%s') unexpectedly is %d\n",
1715 name, cBOOL(isspace(i))));
1716 }
1717 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1718 is_bad = TRUE;
1719 DEBUG_L(PerlIO_printf(Perl_debug_log,
1720 "isupper('%s') unexpectedly is %d\n",
1721 name, cBOOL(isupper(i))));
1722 }
1723 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1724 is_bad = TRUE;
1725 DEBUG_L(PerlIO_printf(Perl_debug_log,
1726 "isxdigit('%s') unexpectedly is %d\n",
1727 name, cBOOL(isxdigit(i))));
1728 }
65c15174 1729 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
8b7358b9
KW
1730 is_bad = TRUE;
1731 DEBUG_L(PerlIO_printf(Perl_debug_log,
1732 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1733 name, tolower(i), (int) toLOWER_A(i)));
1734 }
65c15174 1735 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
8b7358b9
KW
1736 is_bad = TRUE;
1737 DEBUG_L(PerlIO_printf(Perl_debug_log,
1738 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1739 name, toupper(i), (int) toUPPER_A(i)));
1740 }
1741 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1742 is_bad = TRUE;
1743 DEBUG_L(PerlIO_printf(Perl_debug_log,
1744 "'\\n' (=%02X) is not a control\n", (int) i));
1745 }
1746
1747 /* Add to the list; Separate multiple entries with a blank */
1748 if (is_bad) {
1749 if (bad_count) {
1750 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1751 }
8b7358b9
KW
1752 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1753 bad_count++;
8c6180a9
KW
1754 }
1755 }
1756 }
1757
30d8090d
KW
1758 if (bad_count == 2 && maybe_utf8_turkic) {
1759 bad_count = 0;
1760 *bad_chars_list = '\0';
1761 PL_fold_locale['I'] = 'I';
1762 PL_fold_locale['i'] = 'i';
1763 PL_in_utf8_turkic_locale = TRUE;
1764 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1765 __FILE__, __LINE__, newctype));
1766 }
1767 else {
b8df1494 1768 PL_in_utf8_turkic_locale = FALSE;
30d8090d 1769 }
b8df1494 1770
7d4bcc4a
KW
1771# ifdef MB_CUR_MAX
1772
8c6180a9 1773 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1774 * this locale requires more than one byte, there are going to be
8c6180a9 1775 * problems. */
9c8a6dc2
KW
1776 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1777 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1778 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1779
8b7358b9
KW
1780 if ( check_for_problems && MB_CUR_MAX > 1
1781 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1782
1783 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1784 * locale. Just assume that the implementation for them (plus
1785 * for POSIX) is correct and the > 1 value is spurious. (Since
1786 * these are specially handled to never be considered UTF-8
1787 * locales, as long as this is the only problem, everything
1788 * should work fine */
1789 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1790 {
8c6180a9
KW
1791 multi_byte_locale = TRUE;
1792 }
7d4bcc4a
KW
1793
1794# endif
8c6180a9 1795
30d8090d
KW
1796 /* If we found problems and we want them output, do so */
1797 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1798 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1799 {
8b7358b9
KW
1800 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1801 PL_warn_locale = Perl_newSVpvf(aTHX_
1802 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1803 " which have\nunexpected meanings: %s\nThe Perl program"
1804 " will use the expected meanings",
8b7358b9 1805 newctype, bad_chars_list);
8b7358b9
KW
1806 }
1807 else {
1808 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1809 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1810 newctype,
8c6180a9
KW
1811 (multi_byte_locale)
1812 ? " Some characters in it are not recognized by"
1813 " Perl."
1814 : "",
1815 (bad_count)
1816 ? "\nThe following characters (and maybe others)"
1817 " may not have the same meaning as the Perl"
1818 " program expects:\n"
1819 : "",
1820 (bad_count)
1821 ? bad_chars_list
1822 : ""
1823 );
8b7358b9
KW
1824 }
1825
b119c2be
KW
1826# ifdef HAS_NL_LANGINFO
1827
1828 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1829 /* parameter FALSE is a don't care here */
4e6826bf 1830 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1831
1832# endif
1833
8b7358b9
KW
1834 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1835
cc9eaeb0 1836 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1837 * output the message now. If not in that scope, we save the
1838 * message to be output at the first operation using this locale,
1839 * if that actually happens. Most programs don't use locales, so
1840 * they are immune to bad ones. */
cc9eaeb0 1841 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1842
780fcc9f
KW
1843 /* The '0' below suppresses a bogus gcc compiler warning */
1844 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1845
bddebb56
KW
1846 if (IN_LC(LC_CTYPE)) {
1847 SvREFCNT_dec_NN(PL_warn_locale);
1848 PL_warn_locale = NULL;
1849 }
780fcc9f 1850 }
baa60164 1851 }
31f05a37 1852 }
98994639
HS
1853
1854#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1855
98994639
HS
1856}
1857
98994639 1858void
2726666d
KW
1859Perl__warn_problematic_locale()
1860{
2726666d
KW
1861
1862#ifdef USE_LOCALE_CTYPE
1863
5f04a188
KW
1864 dTHX;
1865
1866 /* Internal-to-core function that outputs the message in PL_warn_locale,
1867 * and then NULLS it. Should be called only through the macro
1868 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1869
2726666d 1870 if (PL_warn_locale) {
2726666d
KW
1871 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1872 SvPVX(PL_warn_locale),
1873 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1874 SvREFCNT_dec_NN(PL_warn_locale);
1875 PL_warn_locale = NULL;
1876 }
1877
1878#endif
1879
1880}
1881
a4f00dcc
KW
1882STATIC void
1883S_new_collate(pTHX_ const char *newcoll)
98994639 1884{
7d4bcc4a
KW
1885
1886#ifndef USE_LOCALE_COLLATE
1887
1888 PERL_UNUSED_ARG(newcoll);
1889 PERL_UNUSED_CONTEXT;
1890
1891#else
0d071d52 1892
291a84fb 1893 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1894 * core Perl this and that 'newcoll' is the name of the new locale.
1895 *
d35fca5f
KW
1896 * The design of locale collation is that every locale change is given an
1897 * index 'PL_collation_ix'. The first time a string particpates in an
1898 * operation that requires collation while locale collation is active, it
1899 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1900 * magic includes the collation index, and the transformation of the string
1901 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1902 * instead of the string itself. If a string changes, the magic is
1903 * cleared. The next time the locale changes, the index is incremented,
1904 * and so we know during a comparison that the transformation is not
1905 * necessarily still valid, and so is recomputed. Note that if the locale
1906 * changes enough times, the index could wrap (a U32), and it is possible
1907 * that a transformation would improperly be considered valid, leading to
1908 * an unlikely bug */
0d071d52 1909
98994639
HS
1910 if (! newcoll) {
1911 if (PL_collation_name) {
1912 ++PL_collation_ix;
1913 Safefree(PL_collation_name);
1914 PL_collation_name = NULL;
1915 }
1916 PL_collation_standard = TRUE;
00bf60ca 1917 is_standard_collation:
98994639
HS
1918 PL_collxfrm_base = 0;
1919 PL_collxfrm_mult = 2;
165a1c52 1920 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1921 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1922 PL_strxfrm_max_cp = 0;
98994639
HS
1923 return;
1924 }
1925
d35fca5f 1926 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1927 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1928 ++PL_collation_ix;
1929 Safefree(PL_collation_name);
1930 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1931 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1932 if (PL_collation_standard) {
1933 goto is_standard_collation;
1934 }
98994639 1935
165a1c52 1936 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1937 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1938 PL_strxfrm_max_cp = 0;
165a1c52 1939
59c018b9
KW
1940 /* A locale collation definition includes primary, secondary, tertiary,
1941 * etc. weights for each character. To sort, the primary weights are
1942 * used, and only if they compare equal, then the secondary weights are
1943 * used, and only if they compare equal, then the tertiary, etc.
1944 *
1945 * strxfrm() works by taking the input string, say ABC, and creating an
1946 * output transformed string consisting of first the primary weights,
1947 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1948 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1949 * may not have weights at every level. In our example, let's say B
1950 * doesn't have a tertiary weight, and A doesn't have a secondary
1951 * weight. The constructed string is then going to be
1952 * A¹B¹C¹ B²C² A³C³ ....
1953 * This has the desired effect that strcmp() will look at the secondary
1954 * or tertiary weights only if the strings compare equal at all higher
1955 * priority weights. The spaces shown here, like in
c342d20e 1956 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1957 * are not just for readability. In the general case, these must
1958 * actually be bytes, which we will call here 'separator weights'; and
1959 * they must be smaller than any other weight value, but since these
1960 * are C strings, only the terminating one can be a NUL (some
1961 * implementations may include a non-NUL separator weight just before
1962 * the NUL). Implementations tend to reserve 01 for the separator
1963 * weights. They are needed so that a shorter string's secondary
1964 * weights won't be misconstrued as primary weights of a longer string,
1965 * etc. By making them smaller than any other weight, the shorter
1966 * string will sort first. (Actually, if all secondary weights are
1967 * smaller than all primary ones, there is no need for a separator
1968 * weight between those two levels, etc.)
1969 *
1970 * The length of the transformed string is roughly a linear function of
1971 * the input string. It's not exactly linear because some characters
1972 * don't have weights at all levels. When we call strxfrm() we have to
1973 * allocate some memory to hold the transformed string. The
1974 * calculations below try to find coefficients 'm' and 'b' for this
1975 * locale so that m*x + b equals how much space we need, given the size
1976 * of the input string in 'x'. If we calculate too small, we increase
1977 * the size as needed, and call strxfrm() again, but it is better to
1978 * get it right the first time to avoid wasted expensive string
1979 * transformations. */
1980
98994639 1981 {
79f120c8
KW
1982 /* We use the string below to find how long the tranformation of it
1983 * is. Almost all locales are supersets of ASCII, or at least the
1984 * ASCII letters. We use all of them, half upper half lower,
1985 * because if we used fewer, we might hit just the ones that are
1986 * outliers in a particular locale. Most of the strings being
1987 * collated will contain a preponderance of letters, and even if
1988 * they are above-ASCII, they are likely to have the same number of
1989 * weight levels as the ASCII ones. It turns out that digits tend
1990 * to have fewer levels, and some punctuation has more, but those
1991 * are relatively sparse in text, and khw believes this gives a
1992 * reasonable result, but it could be changed if experience so
1993 * dictates. */
1994 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1995 char * x_longer; /* Transformed 'longer' */
1996 Size_t x_len_longer; /* Length of 'x_longer' */
1997
1998 char * x_shorter; /* We also transform a substring of 'longer' */
1999 Size_t x_len_shorter;
2000
a4a439fb 2001 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
2002 * are interested only in its length). It is used because it has
2003 * the intelligence to handle all cases, but to work, it needs some
2004 * values of 'm' and 'b' to get it started. For the purposes of
2005 * this calculation we use a very conservative estimate of 'm' and
2006 * 'b'. This assumes a weight can be multiple bytes, enough to
2007 * hold any UV on the platform, and there are 5 levels, 4 weight
2008 * bytes, and a trailing NUL. */
2009 PL_collxfrm_base = 5;
2010 PL_collxfrm_mult = 5 * sizeof(UV);
2011
2012 /* Find out how long the transformation really is */
a4a439fb
KW
2013 x_longer = _mem_collxfrm(longer,
2014 sizeof(longer) - 1,
2015 &x_len_longer,
2016
2017 /* We avoid converting to UTF-8 in the
2018 * called function by telling it the
2019 * string is in UTF-8 if the locale is a
2020 * UTF-8 one. Since the string passed
2021 * here is invariant under UTF-8, we can
2022 * claim it's UTF-8 even though it isn't.
2023 * */
2024 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2025 Safefree(x_longer);
2026
2027 /* Find out how long the transformation of a substring of 'longer'
2028 * is. Together the lengths of these transformations are
2029 * sufficient to calculate 'm' and 'b'. The substring is all of
2030 * 'longer' except the first character. This minimizes the chances
2031 * of being swayed by outliers */
a4a439fb 2032 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 2033 sizeof(longer) - 2,
a4a439fb
KW
2034 &x_len_shorter,
2035 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2036 Safefree(x_shorter);
2037
2038 /* If the results are nonsensical for this simple test, the whole
2039 * locale definition is suspect. Mark it so that locale collation
2040 * is not active at all for it. XXX Should we warn? */
2041 if ( x_len_shorter == 0
2042 || x_len_longer == 0
2043 || x_len_shorter >= x_len_longer)
2044 {
2045 PL_collxfrm_mult = 0;
2046 PL_collxfrm_base = 0;
2047 }
2048 else {
2049 SSize_t base; /* Temporary */
2050
2051 /* We have both: m * strlen(longer) + b = x_len_longer
2052 * m * strlen(shorter) + b = x_len_shorter;
2053 * subtracting yields:
2054 * m * (strlen(longer) - strlen(shorter))
2055 * = x_len_longer - x_len_shorter
2056 * But we have set things up so that 'shorter' is 1 byte smaller
2057 * than 'longer'. Hence:
2058 * m = x_len_longer - x_len_shorter
2059 *
2060 * But if something went wrong, make sure the multiplier is at
2061 * least 1.
2062 */
2063 if (x_len_longer > x_len_shorter) {
2064 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2065 }
2066 else {
2067 PL_collxfrm_mult = 1;
2068 }
2069
2070 /* mx + b = len
2071 * so: b = len - mx
2072 * but in case something has gone wrong, make sure it is
2073 * non-negative */
2074 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2075 if (base < 0) {
2076 base = 0;
2077 }
2078
2079 /* Add 1 for the trailing NUL */
2080 PL_collxfrm_base = base + 1;
2081 }
58eebef2 2082
7d4bcc4a
KW
2083# ifdef DEBUGGING
2084
58eebef2
KW
2085 if (DEBUG_L_TEST || debug_initialization) {
2086 PerlIO_printf(Perl_debug_log,
b07929e4
KW
2087 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2088 "x_len_longer=%zu,"
2089 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
2090 __FILE__, __LINE__,
2091 PL_in_utf8_COLLATE_locale,
2092 x_len_shorter, x_len_longer,
2093 PL_collxfrm_mult, PL_collxfrm_base);
2094 }
7d4bcc4a
KW
2095# endif
2096
98994639
HS
2097 }
2098 }
2099
2100#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 2101
98994639
HS
2102}
2103
d36adde0
KW
2104#endif
2105
d2b24094 2106#ifdef WIN32
b8cc575c 2107
6c332036
TC
2108#define USE_WSETLOCALE
2109
2110#ifdef USE_WSETLOCALE
2111
2112STATIC char *
2113S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2114 wchar_t *wlocale;
2115 wchar_t *wresult;
2116 char *result;
2117
2118 if (locale) {
2119 int req_size =
2120 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2121
2122 if (!req_size) {
2123 errno = EINVAL;
2124 return NULL;
2125 }
2126
2127 Newx(wlocale, req_size, wchar_t);
2128 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2129 Safefree(wlocale);
2130 errno = EINVAL;
2131 return NULL;
2132 }
2133 }
2134 else {
2135 wlocale = NULL;
2136 }
2137 wresult = _wsetlocale(category, wlocale);
2138 Safefree(wlocale);
2139 if (wresult) {
2140 int req_size =
2141 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2142 Newx(result, req_size, char);
2143 SAVEFREEPV(result); /* is there something better we can do here? */
2144 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2145 result, req_size, NULL, NULL)) {
2146 errno = EINVAL;
2147 return NULL;
2148 }
2149 }
2150 else {
2151 result = NULL;
2152 }
2153
2154 return result;
2155}
2156
2157#endif
2158
a4f00dcc 2159STATIC char *
b8cc575c 2160S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2161{
2162 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2163 * difference between the two unless the input locale is "", which normally
2164 * means on Windows to get the machine default, which is set via the
2165 * computer's "Regional and Language Options" (or its current equivalent).
2166 * In POSIX, it instead means to find the locale from the user's
2167 * environment. This routine changes the Windows behavior to first look in
2168 * the environment, and, if anything is found, use that instead of going to
2169 * the machine default. If there is no environment override, the machine
2170 * default is used, by calling the real setlocale() with "".
2171 *
2172 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2173 * use the particular category's variable if set; otherwise to use the LANG
2174 * variable. */
b385bb4d 2175
175c4cf9 2176 bool override_LC_ALL = FALSE;
89f7b9aa 2177 char * result;
e5f10d49 2178 unsigned int i;
89f7b9aa 2179
b385bb4d 2180 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2181
2182# ifdef LC_ALL
2183
b385bb4d
KW
2184 locale = PerlEnv_getenv("LC_ALL");
2185 if (! locale) {
e5f10d49
KW
2186 if (category == LC_ALL) {
2187 override_LC_ALL = TRUE;
2188 }
2189 else {
7d4bcc4a
KW
2190
2191# endif
7d4bcc4a 2192
e5f10d49
KW
2193 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2194 if (category == categories[i]) {
2195 locale = PerlEnv_getenv(category_names[i]);
2196 goto found_locale;
2197 }
2198 }
7d4bcc4a 2199
b385bb4d 2200 locale = PerlEnv_getenv("LANG");
481465ea 2201 if (! locale) {
b385bb4d
KW
2202 locale = "";
2203 }
e5f10d49
KW
2204
2205 found_locale: ;
7d4bcc4a
KW
2206
2207# ifdef LC_ALL
2208
e5f10d49 2209 }
b385bb4d 2210 }
7d4bcc4a
KW
2211
2212# endif
2213
b385bb4d
KW
2214 }
2215
6c332036
TC
2216#ifdef USE_WSETLOCALE
2217 result = S_wrap_wsetlocale(aTHX_ category, locale);
2218#else
89f7b9aa 2219 result = setlocale(category, locale);
6c332036 2220#endif
48015184
KW
2221 DEBUG_L(STMT_START {
2222 dSAVE_ERRNO;
2223 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2224 setlocale_debug_string(category, locale, result));
2225 RESTORE_ERRNO;
2226 } STMT_END);
89f7b9aa 2227
481465ea 2228 if (! override_LC_ALL) {
89f7b9aa
KW
2229 return result;
2230 }
2231
dfd77d7a 2232 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2233 * LANG variable or the system default if there is no LANG. But these have
2234 * lower priority than the other LC_foo variables, so override it for each
2235 * one that is set. (If they are set to "", it means to use the same thing
2236 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2237
948523db 2238 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2239 result = PerlEnv_getenv(category_names[i]);
2240 if (result && strNE(result, "")) {
6c332036 2241#ifdef USE_WSETLOCALE
17a1e9ac 2242 S_wrap_wsetlocale(aTHX_ categories[i], result);
6c332036 2243#else
17a1e9ac 2244 setlocale(categories[i], result);
6c332036 2245#endif
e5f10d49
KW
2246 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2247 __FILE__, __LINE__,
2248 setlocale_debug_string(categories[i], result, "not captured")));
2249 }
89f7b9aa 2250 }
7d4bcc4a 2251
bbc98134 2252 result = setlocale(LC_ALL, NULL);
48015184
KW
2253 DEBUG_L(STMT_START {
2254 dSAVE_ERRNO;
2255 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2256 __FILE__, __LINE__,
48015184
KW
2257 setlocale_debug_string(LC_ALL, NULL, result));
2258 RESTORE_ERRNO;
2259 } STMT_END);
89f7b9aa 2260
bbc98134 2261 return result;
b385bb4d
KW
2262}
2263
2264#endif
2265
9aac5db8
KW
2266/*
2267
2268=head1 Locale-related functions and macros
2269
2270=for apidoc Perl_setlocale
2271
2272This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2273taking the same parameters, and returning the same information, except that it
9487427b
KW
2274returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2275instead return C<C> if the underlying locale has a non-dot decimal point
2276character, or a non-empty thousands separator for displaying floating point
2277numbers. This is because perl keeps that locale category such that it has a
2278dot and empty separator, changing the locale briefly during the operations
2279where the underlying one is required. C<Perl_setlocale> knows about this, and
2280compensates; regular C<setlocale> doesn't.
9aac5db8 2281
e9bc6d6b 2282Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2283declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2284C<const> (presumably because its API was specified long ago, and can't be
2285updated; it is illegal to change the information C<setlocale> returns; doing
2286so leads to segfaults.)
9aac5db8 2287
e9bc6d6b
KW
2288Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2289C<setlocale> can be completely ineffective on some platforms under some
2290configurations.
2291
9aac5db8 2292C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2293where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2294the system C<setlocale()> is ineffective, returning the wrong information, and
2295failing to actually change the locale. C<Perl_setlocale>, however works
2296properly in all circumstances.
9aac5db8
KW
2297
2298The return points to a per-thread static buffer, which is overwritten the next
2299time C<Perl_setlocale> is called from the same thread.
2300
2301=cut
2302
2303*/
2304
2305const char *
2306Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2307{
2308 /* This wraps POSIX::setlocale() */
2309
52129632 2310#ifndef USE_LOCALE
d36adde0
KW
2311
2312 PERL_UNUSED_ARG(category);
2313 PERL_UNUSED_ARG(locale);
2314
2315 return "C";
2316
2317#else
2318
9aac5db8
KW
2319 const char * retval;
2320 const char * newlocale;
2321 dSAVEDERRNO;
a4f00dcc 2322 dTHX;
d36adde0 2323 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2324
a4f00dcc
KW
2325#ifdef USE_LOCALE_NUMERIC
2326
291a84fb
KW
2327 /* A NULL locale means only query what the current one is. We have the
2328 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2329 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2330 * the correct results. All other categories don't require special
2331 * handling */
a4f00dcc
KW
2332 if (locale == NULL) {
2333 if (category == LC_NUMERIC) {
9aac5db8
KW
2334
2335 /* We don't have to copy this return value, as it is a per-thread
2336 * variable, and won't change until a future setlocale */
2337 return PL_numeric_name;
a4f00dcc
KW
2338 }
2339
7d4bcc4a 2340# ifdef LC_ALL
a4f00dcc 2341
9aac5db8
KW
2342 else if (category == LC_ALL) {
2343 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2344 }
2345
7d4bcc4a 2346# endif
a4f00dcc
KW
2347
2348 }
2349
2350#endif
2351
33e5a354
KW
2352 retval = save_to_buffer(do_setlocale_r(category, locale),
2353 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2354 SAVE_ERRNO;
2355
2356#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2357
2358 if (locale == NULL && category == LC_ALL) {
2359 RESTORE_LC_NUMERIC();
2360 }
2361
2362#endif
a4f00dcc
KW
2363
2364 DEBUG_L(PerlIO_printf(Perl_debug_log,
2365 "%s:%d: %s\n", __FILE__, __LINE__,
2366 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2367
9aac5db8
KW
2368 RESTORE_ERRNO;
2369
2370 if (! retval) {
a4f00dcc
KW
2371 return NULL;
2372 }
2373
9aac5db8 2374 /* If locale == NULL, we are just querying the state */
a4f00dcc 2375 if (locale == NULL) {
a4f00dcc
KW
2376 return retval;
2377 }
a4f00dcc 2378
1159483a
KW
2379 /* Now that have switched locales, we have to update our records to
2380 * correspond. */
a4f00dcc 2381
1159483a 2382 switch (category) {
a4f00dcc 2383
1159483a 2384#ifdef USE_LOCALE_CTYPE
a4f00dcc 2385
1159483a
KW
2386 case LC_CTYPE:
2387 new_ctype(retval);
2388 break;
a4f00dcc 2389
1159483a 2390#endif
a4f00dcc
KW
2391#ifdef USE_LOCALE_COLLATE
2392
1159483a
KW
2393 case LC_COLLATE:
2394 new_collate(retval);
2395 break;
a4f00dcc 2396
1159483a 2397#endif
a4f00dcc
KW
2398#ifdef USE_LOCALE_NUMERIC
2399
1159483a
KW
2400 case LC_NUMERIC:
2401 new_numeric(retval);
2402 break;
a4f00dcc 2403
1159483a
KW
2404#endif
2405#ifdef LC_ALL
a4f00dcc 2406
1159483a 2407 case LC_ALL:
a4f00dcc 2408
1159483a
KW
2409 /* LC_ALL updates all the things we care about. The values may not
2410 * be the same as 'retval', as the locale "" may have set things
2411 * individually */
a4f00dcc 2412
1159483a 2413# ifdef USE_LOCALE_CTYPE
a4f00dcc 2414
b0bde642 2415 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2416 new_ctype(newlocale);
b0bde642 2417 Safefree(newlocale);
a4f00dcc 2418
1159483a
KW
2419# endif /* USE_LOCALE_CTYPE */
2420# ifdef USE_LOCALE_COLLATE
2421
b0bde642 2422 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2423 new_collate(newlocale);
b0bde642 2424 Safefree(newlocale);
a4f00dcc 2425
7d4bcc4a 2426# endif
1159483a 2427# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2428
b0bde642 2429 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2430 new_numeric(newlocale);
b0bde642 2431 Safefree(newlocale);
a4f00dcc 2432
1159483a
KW
2433# endif /* USE_LOCALE_NUMERIC */
2434#endif /* LC_ALL */
a4f00dcc 2435
1159483a
KW
2436 default:
2437 break;
a4f00dcc
KW
2438 }
2439
2440 return retval;
2441
d36adde0
KW
2442#endif
2443
f7416781
KW
2444}
2445
2446PERL_STATIC_INLINE const char *
2447S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2448{
2449 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2450 * growing it if necessary */
2451
0b6697c5 2452 Size_t string_size;
f7416781
KW
2453
2454 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2455
0b6697c5
KW
2456 if (! string) {
2457 return NULL;
2458 }
2459
2460 string_size = strlen(string) + offset + 1;
2461
f7416781
KW
2462 if (*buf_size == 0) {
2463 Newx(*buf, string_size, char);
2464 *buf_size = string_size;
2465 }
2466 else if (string_size > *buf_size) {
2467 Renew(*buf, string_size, char);
2468 *buf_size = string_size;
2469 }
2470
2471 Copy(string, *buf + offset, string_size - offset, char);
2472 return *buf;
2473}
2474
2475/*
2476
f7416781
KW
2477=for apidoc Perl_langinfo
2478
ce181bf6 2479This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2480taking the same C<item> parameter values, and returning the same information.
2481But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2482of Perl's locale handling from your code, and can be used on systems that lack
2483a native C<nl_langinfo>.
2484
2485Expanding on these:
2486
2487=over
2488
2489=item *
2490
ce181bf6
KW
2491The reason it isn't quite a drop-in replacement is actually an advantage. The
2492only difference is that it returns S<C<const char *>>, whereas plain
2493C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2494forbidden to write into the buffer. By declaring this C<const>, the compiler
2495enforces this restriction, so if it is violated, you know at compilation time,
2496rather than getting segfaults at runtime.
2497
2498=item *
2499
69e2275e 2500It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2501without you having to write extra code. The reason for the extra code would be
2502because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2503kept set by Perl so that the radix is a dot, and the separator is the empty
2504string, no matter what the underlying locale is supposed to be, and so to get
2505the expected results, you have to temporarily toggle into the underlying
2506locale, and later toggle back. (You could use plain C<nl_langinfo> and
2507C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2508the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2509(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2510(decimal point) character to be a dot.)
ce181bf6
KW
2511
2512=item *
2513
2514The system function it replaces can have its static return buffer trashed,
2515not only by a subesequent call to that function, but by a C<freelocale>,
2516C<setlocale>, or other locale change. The returned buffer of this function is
2517not changed until the next call to it, so the buffer is never in a trashed
2518state.
f7416781
KW
2519
2520=item *
2521
ce181bf6
KW
2522Its return buffer is per-thread, so it also is never overwritten by a call to
2523this function from another thread; unlike the function it replaces.
2524
2525=item *
2526
2527But most importantly, it works on systems that don't have C<nl_langinfo>, such
2528as Windows, hence makes your code more portable. Of the fifty-some possible
2529items specified by the POSIX 2008 standard,
f7416781 2530L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2531only one is completely unimplemented, though on non-Windows platforms, another
2532significant one is also not implemented). It uses various techniques to
2533recover the other items, including calling C<L<localeconv(3)>>, and
2534C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2535available. Later C<strftime()> versions have additional capabilities; C<""> is
2536returned for those not available on your system.
ce181bf6
KW
2537
2538It is important to note that when called with an item that is recovered by
2539using C<localeconv>, the buffer from any previous explicit call to
2540C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2541contents if you need to access them after a call to this function. (But note
2542that you might not want to be using C<localeconv()> directly anyway, because of
2543issues like the ones listed in the second item of this list (above) for
2544C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2545call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2546unpack).
9f0bc911 2547
6201c220
KW
2548The details for those items which may deviate from what this emulation returns
2549and what a native C<nl_langinfo()> would return are specified in
2550L<I18N::Langinfo>.
f7416781 2551
ce181bf6
KW
2552=back
2553
f7416781
KW
2554When using C<Perl_langinfo> on systems that don't have a native
2555C<nl_langinfo()>, you must
2556
2557 #include "perl_langinfo.h"
2558
2559before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2560C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2561C<langinfo.h> would try to import into the namespace for code that doesn't need
2562it.)
f7416781 2563
f7416781
KW
2564The original impetus for C<Perl_langinfo()> was so that code that needs to
2565find out the current currency symbol, floating point radix character, or digit
2566grouping separator can use, on all systems, the simpler and more
2567thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2568pain to make thread-friendly. For other fields returned by C<localeconv>, it
2569is better to use the methods given in L<perlcall> to call
2570L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2571
2572=cut
2573
2574*/
2575
2576const char *
2577#ifdef HAS_NL_LANGINFO
2578Perl_langinfo(const nl_item item)
2579#else
2580Perl_langinfo(const int item)
2581#endif
2582{
f61748ac
KW
2583 return my_nl_langinfo(item, TRUE);
2584}
2585
59a15af0 2586STATIC const char *
f61748ac
KW
2587#ifdef HAS_NL_LANGINFO
2588S_my_nl_langinfo(const nl_item item, bool toggle)
2589#else
2590S_my_nl_langinfo(const int item, bool toggle)
2591#endif
2592{
ae74815b 2593 dTHX;
0b6697c5 2594 const char * retval;
f7416781 2595
d36adde0
KW
2596#ifdef USE_LOCALE_NUMERIC
2597
5a854ab3
KW
2598 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2599 * two items, and only if not already there */
4e6826bf 2600 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2601 || PL_numeric_underlying))
d36adde0
KW
2602
2603#endif /* No toggling needed if not using LC_NUMERIC */
2604
5a854ab3 2605 toggle = FALSE;
5a854ab3 2606
ab340fff 2607#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2608# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2609 || ! defined(HAS_POSIX_2008_LOCALE) \
2610 || ! defined(DUPLOCALE)
f7416781 2611
ab340fff 2612 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2613 * for those items dependent on it. This must be copied to a buffer before
2614 * switching back, as some systems destroy the buffer when setlocale() is
2615 * called */
f7416781 2616
038d3702
KW
2617 {
2618 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2619
b2fee59c 2620 if (toggle) {
038d3702 2621 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2622 }
f7416781 2623
5acc4454
KW
2624 LOCALE_LOCK; /* Prevent interference from another thread executing
2625 this code section (the only call to nl_langinfo in
2626 the core) */
2627
ee90eb2d 2628
628ff75a
KW
2629 /* Copy to a per-thread buffer, which is also one that won't be
2630 * destroyed by a subsequent setlocale(), such as the
2631 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2632 retval = save_to_buffer(nl_langinfo(item),
2633 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2634
5acc4454
KW
2635 LOCALE_UNLOCK;
2636
b2fee59c 2637 if (toggle) {
038d3702 2638 RESTORE_LC_NUMERIC();
b2fee59c 2639 }
f7416781
KW
2640 }
2641
ab340fff
KW
2642# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2643
5a854ab3 2644 {
b2fee59c
KW
2645 bool do_free = FALSE;
2646 locale_t cur = uselocale((locale_t) 0);
ab340fff 2647
b2fee59c
KW
2648 if (cur == LC_GLOBAL_LOCALE) {
2649 cur = duplocale(LC_GLOBAL_LOCALE);
2650 do_free = TRUE;
2651 }
ab340fff 2652
d36adde0
KW
2653# ifdef USE_LOCALE_NUMERIC
2654
b2fee59c 2655 if (toggle) {
e1aa2579
KW
2656 if (PL_underlying_numeric_obj) {
2657 cur = PL_underlying_numeric_obj;
2658 }
2659 else {
2660 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2661 do_free = TRUE;
2662 }
b2fee59c 2663 }
ab340fff 2664
d36adde0
KW
2665# endif
2666
628ff75a
KW
2667 /* We have to save it to a buffer, because the freelocale() just below
2668 * can invalidate the internal one */
0b6697c5
KW
2669 retval = save_to_buffer(nl_langinfo_l(item, cur),
2670 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2671
b2fee59c
KW
2672 if (do_free) {
2673 freelocale(cur);
2674 }
5a854ab3 2675 }
ab340fff 2676
c1566110
KW
2677# endif
2678
0b6697c5 2679 if (strEQ(retval, "")) {
4e6826bf 2680 if (item == YESSTR) {
c1566110
KW
2681 return "yes";
2682 }
4e6826bf 2683 if (item == NOSTR) {
c1566110
KW
2684 return "no";
2685 }
2686 }
2687
0b6697c5 2688 return retval;
ab340fff 2689
f7416781 2690#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2691
2692 {
2693
f7416781
KW
2694# ifdef HAS_LOCALECONV
2695
43dd6b15 2696 const struct lconv* lc;
628ff75a 2697 const char * temp;
038d3702 2698 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2699
69c5e0db
KW
2700# ifdef TS_W32_BROKEN_LOCALECONV
2701
2702 const char * save_global;
2703 const char * save_thread;
2704 int needed_size;
2705 char * ptr;
2706 char * e;
2707 char * item_start;
2708
2709# endif
f7416781
KW
2710# endif
2711# ifdef HAS_STRFTIME
2712
43dd6b15
KW
2713 struct tm tm;
2714 bool return_format = FALSE; /* Return the %format, not the value */
2715 const char * format;
f7416781
KW
2716
2717# endif
2718
43dd6b15
KW
2719 /* We copy the results to a per-thread buffer, even if not
2720 * multi-threaded. This is in part to simplify this code, and partly
2721 * because we need a buffer anyway for strftime(), and partly because a
2722 * call of localeconv() could otherwise wipe out the buffer, and the
2723 * programmer would not be expecting this, as this is a nl_langinfo()
2724 * substitute after all, so s/he might be thinking their localeconv()
2725 * is safe until another localeconv() call. */
f7416781 2726
43dd6b15
KW
2727 switch (item) {
2728 Size_t len;
f7416781 2729
8d72e74e 2730 /* This is unimplemented */
4e6826bf 2731 case ERA: /* For use with strftime() %E modifier */
f7416781 2732
43dd6b15
KW
2733 default:
2734 return "";
f7416781 2735
43dd6b15 2736 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2737 case YESEXPR: return "^[+1yY]";
2738 case YESSTR: return "yes";
2739 case NOEXPR: return "^[-0nN]";
2740 case NOSTR: return "no";
2741
8d72e74e
KW
2742 case CODESET:
2743
2744# ifndef WIN32
2745
2746 /* On non-windows, this is unimplemented, in part because of
2747 * inconsistencies between vendors. The Darwin native
2748 * nl_langinfo() implementation simply looks at everything past
2749 * any dot in the name, but that doesn't work for other
2750 * vendors. Many Linux locales that don't have UTF-8 in their
2751 * names really are UTF-8, for example; z/OS locales that do
2752 * have UTF-8 in their names, aren't really UTF-8 */
2753 return "";
2754
2755# else
2756
2757 { /* But on Windows, the name does seem to be consistent, so
2758 use that. */
2759 const char * p;
2760 const char * first;
2761 Size_t offset = 0;
2762 const char * name = my_setlocale(LC_CTYPE, NULL);
2763
2764 if (isNAME_C_OR_POSIX(name)) {
2765 return "ANSI_X3.4-1968";
2766 }
2767
2768 /* Find the dot in the locale name */
2769 first = (const char *) strchr(name, '.');
2770 if (! first) {
2771 first = name;
2772 goto has_nondigit;
2773 }
2774
2775 /* Look at everything past the dot */
2776 first++;
2777 p = first;
2778
2779 while (*p) {
2780 if (! isDIGIT(*p)) {
2781 goto has_nondigit;
2782 }
2783
2784 p++;
2785 }
2786
2787 /* Here everything past the dot is a digit. Treat it as a
2788 * code page */
18503cbd 2789 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2790 &PL_langinfo_bufsize, 0);
8d72e74e 2791 offset = STRLENs("CP");
f7416781 2792
8d72e74e
KW
2793 has_nondigit:
2794
2795 retval = save_to_buffer(first, &PL_langinfo_buf,
2796 &PL_langinfo_bufsize, offset);
2797 }
2798
2799 break;
2800
2801# endif
f7416781
KW
2802# ifdef HAS_LOCALECONV
2803
4e6826bf 2804 case CRNCYSTR:
f7416781 2805
43dd6b15
KW
2806 /* We don't bother with localeconv_l() because any system that
2807 * has it is likely to also have nl_langinfo() */
291a84fb 2808
69c5e0db
KW
2809 LOCALE_LOCK_V; /* Prevent interference with other threads
2810 using localeconv() */
2811
2812# ifdef TS_W32_BROKEN_LOCALECONV
2813
2814 /* This is a workaround for a Windows bug prior to VS 15.
2815 * What we do here is, while locked, switch to the global
2816 * locale so localeconv() works; then switch back just before
2817 * the unlock. This can screw things up if some thread is
2818 * already using the global locale while assuming no other is.
2819 * A different workaround would be to call GetCurrencyFormat on
2820 * a known value, and parse it; patches welcome
2821 *
2822 * We have to use LC_ALL instead of LC_MONETARY because of
2823 * another bug in Windows */
2824
2825 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2826 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2827 save_global= savepv(my_setlocale(LC_ALL, NULL));
2828 my_setlocale(LC_ALL, save_thread);
2829
2830# endif
5acc4454 2831
43dd6b15
KW
2832 lc = localeconv();
2833 if ( ! lc
2834 || ! lc->currency_symbol
2835 || strEQ("", lc->currency_symbol))
2836 {
69c5e0db 2837 LOCALE_UNLOCK_V;
43dd6b15
KW
2838 return "";
2839 }
f7416781 2840
43dd6b15 2841 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2842 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2843 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2844 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2845 { /* khw couldn't figure out how the localedef specifications
2846 would show that the $ should replace the radix; this is
2847 just a guess as to how it might work.*/
a34edee3 2848 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2849 }
2850 else if (lc->p_cs_precedes) {
a34edee3 2851 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2852 }
2853 else {
a34edee3 2854 PL_langinfo_buf[0] = '+';
43dd6b15 2855 }
f7416781 2856
69c5e0db
KW
2857# ifdef TS_W32_BROKEN_LOCALECONV
2858
2859 my_setlocale(LC_ALL, save_global);
2860 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2861 my_setlocale(LC_ALL, save_thread);
2862 Safefree(save_global);
2863 Safefree(save_thread);
2864
2865# endif
2866
2867 LOCALE_UNLOCK_V;
43dd6b15 2868 break;
f7416781 2869
69c5e0db
KW
2870# ifdef TS_W32_BROKEN_LOCALECONV
2871
4e6826bf 2872 case RADIXCHAR:
69c5e0db
KW
2873
2874 /* For this, we output a known simple floating point number to
2875 * a buffer, and parse it, looking for the radix */
2876
2877 if (toggle) {
2878 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2879 }
2880
2881 if (PL_langinfo_bufsize < 10) {
2882 PL_langinfo_bufsize = 10;
2883 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2884 }
2885
2886 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2887 "%.1f", 1.5);
2888 if (needed_size >= (int) PL_langinfo_bufsize) {
2889 PL_langinfo_bufsize = needed_size + 1;
2890 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2891 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2892 "%.1f", 1.5);
2893 assert(needed_size < (int) PL_langinfo_bufsize);
2894 }
2895
2896 ptr = PL_langinfo_buf;
2897 e = PL_langinfo_buf + PL_langinfo_bufsize;
2898
2899 /* Find the '1' */
2900 while (ptr < e && *ptr != '1') {
2901 ptr++;
2902 }
2903 ptr++;
2904
2905 /* Find the '5' */
2906 item_start = ptr;
2907 while (ptr < e && *ptr != '5') {
2908 ptr++;
2909 }
2910
2911 /* Everything in between is the radix string */
2912 if (ptr >= e) {
2913 PL_langinfo_buf[0] = '?';
2914 PL_langinfo_buf[1] = '\0';
2915 }
2916 else {
2917 *ptr = '\0';
2918 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2919 }
2920
2921 if (toggle) {
2922 RESTORE_LC_NUMERIC();
2923 }
2924
2925 retval = PL_langinfo_buf;
2926 break;
2927
2928# else
2929
2930 case RADIXCHAR: /* No special handling needed */
2931
2932# endif
2933
4e6826bf 2934 case THOUSEP:
f7416781 2935
43dd6b15 2936 if (toggle) {
038d3702 2937 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2938 }
f7416781 2939
69c5e0db
KW
2940 LOCALE_LOCK_V; /* Prevent interference with other threads
2941 using localeconv() */
2942
2943# ifdef TS_W32_BROKEN_LOCALECONV
2944
2945 /* This should only be for the thousands separator. A
2946 * different work around would be to use GetNumberFormat on a
2947 * known value and parse the result to find the separator */
2948 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2949 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2950 save_global = savepv(my_setlocale(LC_ALL, NULL));
2951 my_setlocale(LC_ALL, save_thread);
2952# if 0
2953 /* This is the start of code that for broken Windows replaces
2954 * the above and below code, and instead calls
2955 * GetNumberFormat() and then would parse that to find the
2956 * thousands separator. It needs to handle UTF-16 vs -8
2957 * issues. */
2958
2959 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2960 DEBUG_L(PerlIO_printf(Perl_debug_log,
2961 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2962 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2963
2964# endif
2965# endif
5acc4454 2966
43dd6b15
KW
2967 lc = localeconv();
2968 if (! lc) {
628ff75a 2969 temp = "";
33394adc 2970 }
43dd6b15 2971 else {
4e6826bf 2972 temp = (item == RADIXCHAR)
43dd6b15
KW
2973 ? lc->decimal_point
2974 : lc->thousands_sep;
628ff75a
KW
2975 if (! temp) {
2976 temp = "";
43dd6b15
KW
2977 }
2978 }
f7416781 2979
0b6697c5
KW
2980 retval = save_to_buffer(temp, &PL_langinfo_buf,
2981 &PL_langinfo_bufsize, 0);
f7416781 2982
69c5e0db
KW
2983# ifdef TS_W32_BROKEN_LOCALECONV
2984
2985 my_setlocale(LC_ALL, save_global);
2986 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2987 my_setlocale(LC_ALL, save_thread);
2988 Safefree(save_global);
2989 Safefree(save_thread);
2990
2991# endif
2992
2993 LOCALE_UNLOCK_V;
5acc4454 2994
43dd6b15 2995 if (toggle) {
038d3702 2996 RESTORE_LC_NUMERIC();
43dd6b15 2997 }
f7416781 2998
43dd6b15 2999 break;
f7416781
KW
3000
3001# endif
3002# ifdef HAS_STRFTIME
3003
43dd6b15
KW
3004 /* These are defined by C89, so we assume that strftime supports
3005 * them, and so are returned unconditionally; they may not be what
3006 * the locale actually says, but should give good enough results
3007 * for someone using them as formats (as opposed to trying to parse
3008 * them to figure out what the locale says). The other format
3009 * items are actually tested to verify they work on the platform */
4e6826bf
KW
3010 case D_FMT: return "%x";
3011 case T_FMT: return "%X";
3012 case D_T_FMT: return "%c";
43dd6b15
KW
3013
3014 /* These formats are only available in later strfmtime's */
4e6826bf 3015 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
3016
3017 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
3018 case ABDAY_1: case ABDAY_2: case ABDAY_3:
3019 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
3020 case ALT_DIGITS:
3021 case AM_STR: case PM_STR:
3022 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
3023 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
3024 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
3025 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
3026 case DAY_5: case DAY_6: case DAY_7:
3027 case MON_1: case MON_2: case MON_3: case MON_4:
3028 case MON_5: case MON_6: case MON_7: case MON_8:
3029 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
3030
3031 LOCALE_LOCK;
3032
3033 init_tm(&tm); /* Precaution against core dumps */
3034 tm.tm_sec = 30;
3035 tm.tm_min = 30;
3036 tm.tm_hour = 6;
3037 tm.tm_year = 2017 - 1900;
3038 tm.tm_wday = 0;
3039 tm.tm_mon = 0;
3040 switch (item) {
3041 default:
3042 LOCALE_UNLOCK;
3043 Perl_croak(aTHX_
3044 "panic: %s: %d: switch case: %d problem",
3045 __FILE__, __LINE__, item);
3046 NOT_REACHED; /* NOTREACHED */
3047
4e6826bf
KW
3048 case PM_STR: tm.tm_hour = 18;
3049 case AM_STR:
43dd6b15
KW
3050 format = "%p";
3051 break;
3052
4e6826bf
KW
3053 case ABDAY_7: tm.tm_wday++;
3054 case ABDAY_6: tm.tm_wday++;
3055 case ABDAY_5: tm.tm_wday++;
3056 case ABDAY_4: tm.tm_wday++;
3057 case ABDAY_3: tm.tm_wday++;
3058 case ABDAY_2: tm.tm_wday++;
3059 case ABDAY_1:
43dd6b15
KW
3060 format = "%a";
3061 break;
3062
4e6826bf
KW
3063 case DAY_7: tm.tm_wday++;
3064 case DAY_6: tm.tm_wday++;
3065 case DAY_5: tm.tm_wday++;
3066 case DAY_4: tm.tm_wday++;
3067 case DAY_3: tm.tm_wday++;
3068 case DAY_2: tm.tm_wday++;
3069 case DAY_1:
43dd6b15
KW
3070 format = "%A";
3071 break;
3072
4e6826bf
KW
3073 case ABMON_12: tm.tm_mon++;
3074 case ABMON_11: tm.tm_mon++;
3075 case ABMON_10: tm.tm_mon++;
3076 case ABMON_9: tm.tm_mon++;
3077 case ABMON_8: tm.tm_mon++;
3078 case ABMON_7: tm.tm_mon++;
3079 case ABMON_6: tm.tm_mon++;
3080 case ABMON_5: tm.tm_mon++;
3081 case ABMON_4: tm.tm_mon++;
3082 case ABMON_3: tm.tm_mon++;
3083 case ABMON_2: tm.tm_mon++;
3084 case ABMON_1:
43dd6b15
KW
3085 format = "%b";
3086 break;
3087
4e6826bf
KW
3088 case MON_12: tm.tm_mon++;
3089 case MON_11: tm.tm_mon++;
3090 case MON_10: tm.tm_mon++;
3091 case MON_9: tm.tm_mon++;
3092 case MON_8: tm.tm_mon++;
3093 case MON_7: tm.tm_mon++;
3094 case MON_6: tm.tm_mon++;
3095 case MON_5: tm.tm_mon++;
3096 case MON_4: tm.tm_mon++;
3097 case MON_3: tm.tm_mon++;
3098 case MON_2: tm.tm_mon++;
3099 case MON_1:
43dd6b15
KW
3100 format = "%B";
3101 break;
3102
4e6826bf 3103 case T_FMT_AMPM:
43dd6b15
KW
3104 format = "%r";
3105 return_format = TRUE;
3106 break;
3107
4e6826bf 3108 case ERA_D_FMT:
43dd6b15
KW
3109 format = "%Ex";
3110 return_format = TRUE;
3111 break;
3112
4e6826bf 3113 case ERA_T_FMT:
43dd6b15
KW
3114 format = "%EX";
3115 return_format = TRUE;
3116 break;
3117
4e6826bf 3118 case ERA_D_T_FMT:
43dd6b15
KW
3119 format = "%Ec";
3120 return_format = TRUE;
3121 break;
3122
4e6826bf 3123 case ALT_DIGITS:
43dd6b15
KW
3124 tm.tm_wday = 0;
3125 format = "%Ow"; /* Find the alternate digit for 0 */
3126 break;
3127 }
f7416781 3128
43dd6b15
KW
3129 /* We can't use my_strftime() because it doesn't look at
3130 * tm_wday */
3131 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3132 format, &tm))
3133 {
3134 /* A zero return means one of:
3135 * a) there wasn't enough space in PL_langinfo_buf
3136 * b) the format, like a plain %p, returns empty
3137 * c) it was an illegal format, though some
3138 * implementations of strftime will just return the
3139 * illegal format as a plain character sequence.
3140 *
3141 * To quickly test for case 'b)', try again but precede
3142 * the format with a plain character. If that result is
3143 * still empty, the problem is either 'a)' or 'c)' */
3144
3145 Size_t format_size = strlen(format) + 1;
3146 Size_t mod_size = format_size + 1;
3147 char * mod_format;
3148 char * temp_result;
3149
3150 Newx(mod_format, mod_size, char);
3151 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 3152 *mod_format = ' ';
43dd6b15
KW
3153 my_strlcpy(mod_format + 1, format, mod_size);
3154 len = strftime(temp_result,
3155 PL_langinfo_bufsize,
3156 mod_format, &tm);
3157 Safefree(mod_format);
3158 Safefree(temp_result);
3159
3160 /* If 'len' is non-zero, it means that we had a case like
3161 * %p which means the current locale doesn't use a.m. or
3162 * p.m., and that is valid */
3163 if (len == 0) {
3164
3165 /* Here, still didn't work. If we get well beyond a
3166 * reasonable size, bail out to prevent an infinite
3167 * loop. */
3168
3169 if (PL_langinfo_bufsize > 100 * format_size) {
3170 *PL_langinfo_buf = '\0';
3171 }
3172 else {
3173 /* Double the buffer size to retry; Add 1 in case
3174 * original was 0, so we aren't stuck at 0. */
3175 PL_langinfo_bufsize *= 2;
3176 PL_langinfo_bufsize++;
3177 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3178 continue;
3179 }
3180 }
f7416781 3181
f7416781 3182 break;
43dd6b15 3183 }
f7416781 3184
43dd6b15
KW
3185 /* Here, we got a result.
3186 *
3187 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3188 * alternate format for wday 0. If the value is the same as
3189 * the normal 0, there isn't an alternate, so clear the buffer.
3190 * */
4e6826bf 3191 if ( item == ALT_DIGITS
43dd6b15
KW
3192 && strEQ(PL_langinfo_buf, "0"))
3193 {
3194 *PL_langinfo_buf = '\0';
3195 }
f7416781 3196
43dd6b15
KW
3197 /* ALT_DIGITS is problematic. Experiments on it showed that
3198 * strftime() did not always work properly when going from
3199 * alt-9 to alt-10. Only a few locales have this item defined,
3200 * and in all of them on Linux that khw was able to find,
3201 * nl_langinfo() merely returned the alt-0 character, possibly
3202 * doubled. Most Unicode digits are in blocks of 10
3203 * consecutive code points, so that is sufficient information
3204 * for those scripts, as we can infer alt-1, alt-2, .... But
3205 * for a Japanese locale, a CJK ideographic 0 is returned, and
3206 * the CJK digits are not in code point order, so you can't
3207 * really infer anything. The localedef for this locale did
3208 * specify the succeeding digits, so that strftime() works
3209 * properly on them, without needing to infer anything. But
3210 * the nl_langinfo() return did not give sufficient information
3211 * for the caller to understand what's going on. So until
3212 * there is evidence that it should work differently, this
3213 * returns the alt-0 string for ALT_DIGITS.
3214 *
3215 * wday was chosen because its range is all a single digit.
3216 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3217
43dd6b15 3218 LOCALE_UNLOCK;
f7416781 3219
d1b150d9
KW
3220 retval = PL_langinfo_buf;
3221
43dd6b15
KW
3222 /* If to return the format, not the value, overwrite the buffer
3223 * with it. But some strftime()s will keep the original format
3224 * if illegal, so change those to "" */
3225 if (return_format) {
3226 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3227 *PL_langinfo_buf = '\0';
3228 }
43dd6b15 3229 else {
0b6697c5
KW
3230 retval = save_to_buffer(format, &PL_langinfo_buf,
3231 &PL_langinfo_bufsize, 0);
f7416781
KW
3232 }
3233 }
3234
3235 break;
f7416781
KW
3236
3237# endif
3238
43dd6b15 3239 }
f7416781
KW
3240 }
3241
0b6697c5 3242 return retval;
f7416781
KW
3243
3244#endif
3245
a4f00dcc 3246}
b385bb4d 3247
98994639
HS
3248/*
3249 * Initialize locale awareness.
3250 */
3251int
3252Perl_init_i18nl10n(pTHX_ int printwarn)
3253{
0e92a118
KW
3254 /* printwarn is
3255 *
3256 * 0 if not to output warning when setup locale is bad
3257 * 1 if to output warning based on value of PERL_BADLANG
3258 * >1 if to output regardless of PERL_BADLANG
3259 *
3260 * returns
98994639 3261 * 1 = set ok or not applicable,
0e92a118
KW
3262 * 0 = fallback to a locale of lower priority
3263 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3264 *
3265 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3266 * set, debugging information is output.
3267 *
3268 * This looks more complicated than it is, mainly due to the #ifdefs.
3269 *
3270 * We try to set LC_ALL to the value determined by the environment. If
3271 * there is no LC_ALL on this platform, we try the individual categories we
3272 * know about. If this works, we are done.
3273 *
3274 * But if it doesn't work, we have to do something else. We search the
3275 * environment variables ourselves instead of relying on the system to do
3276 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3277 * think there is one), and the ultimate fallback "C". This is all done in
3278 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3279 * more complex. The 'trial_locales' array is initialized with just one
3280 * element; it causes the behavior described in the paragraph above this to
3281 * happen. If that fails, we add elements to 'trial_locales', and do extra
3282 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3283 *
3284 * On Ultrix, the locale MUST come from the environment, so there is
3285 * preliminary code to set it. I (khw) am not sure that it is necessary,
3286 * and that this couldn't be folded into the loop, but barring any real
3287 * platforms to test on, it's staying as-is
3288 *
3289 * A slight complication is that in embedded Perls, the locale may already
3290 * be set-up, and we don't want to get it from the normal environment
3291 * variables. This is handled by having a special environment variable
3292 * indicate we're in this situation. We simply set setlocale's 2nd
3293 * parameter to be a NULL instead of "". That indicates to setlocale that
3294 * it is not to change anything, but to return the current value,
3295 * effectively initializing perl's db to what the locale already is.
3296 *
3297 * We play the same trick with NULL if a LC_ALL succeeds. We call
3298 * setlocale() on the individual categores with NULL to get their existing
3299 * values for our db, instead of trying to change them.
3300 * */
98994639 3301
1565c085
DM
3302 dVAR;
3303
0e92a118
KW
3304 int ok = 1;
3305
7d4bcc4a
KW
3306#ifndef USE_LOCALE
3307
3308 PERL_UNUSED_ARG(printwarn);
3309
3310#else /* USE_LOCALE */
7d4bcc4a
KW
3311# ifdef __GLIBC__
3312
175c4cf9 3313 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3314
3315# endif
65ebb059 3316
ccd65d51
KW
3317 /* NULL uses the existing already set up locale */
3318 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3319 ? NULL
3320 : "";
c3fcd832
KW
3321 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3322 unsigned int trial_locales_count;
175c4cf9
KW
3323 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3324 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3325 bool setlocale_failure = FALSE;
65ebb059 3326 unsigned int i;
175c4cf9
KW
3327
3328 /* A later getenv() could zap this, so only use here */
3329 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3330
3331 const bool locwarn = (printwarn > 1
e5f10d49
KW
3332 || ( printwarn
3333 && ( ! bad_lang_use_once
22ff3130 3334 || (
e5f10d49
KW
3335 /* disallow with "" or "0" */
3336 *bad_lang_use_once
3337 && strNE("0", bad_lang_use_once)))));
ea92aad8 3338
291a84fb 3339 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3340 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3341
3342 /* current locale for given category; should have been copied so aren't
3343 * volatile */
8de4332b 3344 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3345
7d4bcc4a
KW
3346# ifdef WIN32
3347
6bce99ee
JH
3348 /* In some systems you can find out the system default locale
3349 * and use that as the fallback locale. */
7d4bcc4a
KW
3350# define SYSTEM_DEFAULT_LOCALE
3351# endif
3352# ifdef SYSTEM_DEFAULT_LOCALE
3353
65ebb059 3354 const char *system_default_locale = NULL;
98994639 3355
7d4bcc4a 3356# endif
948523db
KW
3357
3358# ifndef DEBUGGING
3359# define DEBUG_LOCALE_INIT(a,b,c)
3360# else
7d4bcc4a 3361
8298454c 3362 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3363
3364# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3365 STMT_START { \
3366 if (debug_initialization) { \
3367 PerlIO_printf(Perl_debug_log, \
3368 "%s:%d: %s\n", \
3369 __FILE__, __LINE__, \
a4f00dcc 3370 setlocale_debug_string(category, \
2fcc0ca9
KW
3371 locale, \
3372 result)); \
3373 } \
3374 } STMT_END
2fcc0ca9 3375
948523db
KW
3376/* Make sure the parallel arrays are properly set up */
3377# ifdef USE_LOCALE_NUMERIC
3378 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3379 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3380# ifdef USE_POSIX_2008_LOCALE
3381 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3382# endif
948523db
KW
3383# endif
3384# ifdef USE_LOCALE_CTYPE
3385 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3386 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3387# ifdef USE_POSIX_2008_LOCALE
3388 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3389# endif
948523db
KW
3390# endif
3391# ifdef USE_LOCALE_COLLATE
3392 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3393 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3394# ifdef USE_POSIX_2008_LOCALE
3395 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3396# endif
948523db
KW
3397# endif
3398# ifdef USE_LOCALE_TIME
3399 assert(categories[LC_TIME_INDEX] == LC_TIME);
3400 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3401# ifdef USE_POSIX_2008_LOCALE
3402 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3403# endif
948523db
KW
3404# endif
3405# ifdef USE_LOCALE_MESSAGES
3406 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3407 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3408# ifdef USE_POSIX_2008_LOCALE
3409 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3410# endif
948523db
KW
3411# endif
3412# ifdef USE_LOCALE_MONETARY
3413 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3414 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3415# ifdef USE_POSIX_2008_LOCALE
3416 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3417# endif
948523db 3418# endif
9821811f
KW
3419# ifdef USE_LOCALE_ADDRESS
3420 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3421 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3422# ifdef USE_POSIX_2008_LOCALE
3423 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3424# endif
9821811f
KW
3425# endif
3426# ifdef USE_LOCALE_IDENTIFICATION
3427 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3428 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3429# ifdef USE_POSIX_2008_LOCALE
3430 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3431# endif
9821811f
KW
3432# endif
3433# ifdef USE_LOCALE_MEASUREMENT
3434 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3435 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3436# ifdef USE_POSIX_2008_LOCALE
3437 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3438# endif
9821811f
KW
3439# endif
3440# ifdef USE_LOCALE_PAPER
3441 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3442 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3443# ifdef USE_POSIX_2008_LOCALE
3444 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3445# endif
9821811f
KW
3446# endif
3447# ifdef USE_LOCALE_TELEPHONE
3448 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3449 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3450# ifdef USE_POSIX_2008_LOCALE
3451 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3452# endif
9821811f 3453# endif
948523db
KW
3454# ifdef LC_ALL
3455 assert(categories[LC_ALL_INDEX] == LC_ALL);
3456 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3457 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3458# ifdef USE_POSIX_2008_LOCALE
3459 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3460# endif
948523db
KW
3461# endif
3462# endif /* DEBUGGING */
47280b20 3463
5a6637f0
KW
3464 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
3465 * why these particular incantations are used. */
d2c9cb53
KW
3466#ifdef HAS_MBRLEN
3467 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
3468#endif
5a6637f0
KW
3469#ifdef HAS_MBRTOWC
3470 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
3471#endif
3472#ifdef HAS_WCTOMBR
3473 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
3474#endif
d2c9cb53 3475
47280b20
KW
3476 /* Initialize the cache of the program's UTF-8ness for the always known
3477 * locales C and POSIX */
3478 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3479 sizeof(PL_locale_utf8ness));
3480
e9bc6d6b
KW
3481# ifdef USE_THREAD_SAFE_LOCALE
3482# ifdef WIN32
3483
3484 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3485
3486# endif
3487# endif
39e69e77 3488# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3489
3490 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3491 if (! PL_C_locale_obj) {
3492 Perl_croak_nocontext(
3493 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3494 }
3495 if (DEBUG_Lv_TEST || debug_initialization) {
3496 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3497 }
3498
3499# endif
3500
78f7c09f
KW
3501# ifdef USE_LOCALE_NUMERIC
3502
3ca88433
KW
3503 PL_numeric_radix_sv = newSVpvs(".");
3504
78f7c09f
KW
3505# endif
3506
e9bc6d6b
KW
3507# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3508
3509 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3510 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3511
3512# endif
b56c4436 3513# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3514
3515 /*
3516 * Ultrix setlocale(..., "") fails if there are no environment
3517 * variables from which to get a locale name.
3518 */
3519
b56c4436
KW
3520# ifndef LC_ALL
3521# error Ultrix without LC_ALL not implemented
3522# else
7d4bcc4a 3523
b56c4436
KW
3524 {
3525 bool done = FALSE;
ea92aad8
KW
3526 if (lang) {
3527 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3528 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3529 if (sl_result[LC_ALL_INDEX])
3530 done = TRUE;
3531 else
e5f10d49 3532 setlocale_failure = TRUE;
ea92aad8
KW
3533 }
3534 if (! setlocale_failure) {
3535 const char * locale_param;
3536 for (i = 0; i < LC_ALL_INDEX; i++) {
3537 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3538 ? setlocale_init
3539 : NULL;
3540 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3541 if (! sl_result[i]) {
3542 setlocale_failure = TRUE;
3543 }
3544 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3545 }
c835d6be 3546 }
7d4bcc4a
KW
3547 }
3548
3549# endif /* LC_ALL */
e5f10d49 3550# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3551
65ebb059 3552 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3553 * the list. Normally the loop is executed just once. But if setting the
3554 * locale fails, inside the loop we add fallback trials to the array and so
3555 * will execute the loop multiple times */
c3fcd832
KW
3556 trial_locales[0] = setlocale_init;
3557 trial_locales_count = 1;
7d4bcc4a 3558
65ebb059
KW
3559 for (i= 0; i < trial_locales_count; i++) {
3560 const char * trial_locale = trial_locales[i];
3561
3562 if (i > 0) {
3563
3564 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3565 * when i==0, but I (khw) don't think that behavior makes much
3566 * sense */
3567 setlocale_failure = FALSE;
3568
7d4bcc4a 3569# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3570# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3571
65ebb059
KW
3572 /* On Windows machines, an entry of "" after the 0th means to use
3573 * the system default locale, which we now proceed to get. */
3574 if (strEQ(trial_locale, "")) {
3575 unsigned int j;
3576
3577 /* Note that this may change the locale, but we are going to do
3578 * that anyway just below */
837ce802 3579 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3580 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3581
7d4bcc4a 3582 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3583 * try */
3584 if (! system_default_locale) {
3585 goto next_iteration;
3586 }
3587 for (j = 0; j < trial_locales_count; j++) {
3588 if (strEQ(system_default_locale, trial_locales[j])) {
3589 goto next_iteration;
3590 }
3591 }
3592
3593 trial_locale = system_default_locale;
3594 }
ec0202b5
KW
3595# else
3596# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3597# endif
7d4bcc4a 3598# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3599
3600 } /* For i > 0 */
65ebb059 3601
7d4bcc4a
KW
3602# ifdef LC_ALL
3603
948523db
KW
3604 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3605 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3606 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3607 setlocale_failure = TRUE;
7cd8b568
KW
3608 }
3609 else {
3610 /* Since LC_ALL succeeded, it should have changed all the other
3611 * categories it can to its value; so we massage things so that the
3612 * setlocales below just return their category's current values.
3613 * This adequately handles the case in NetBSD where LC_COLLATE may
3614 * not be defined for a locale, and setting it individually will
7d4bcc4a 3615 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3616 * the POSIX locale. */
3617 trial_locale = NULL;
3618 }
7d4bcc4a
KW
3619
3620# endif /* LC_ALL */
98994639 3621
e5f10d49
KW
3622 if (! setlocale_failure) {
3623 unsigned int j;
3624 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3625 curlocales[j]
3626 = savepv(do_setlocale_r(categories[j], trial_locale));
3627 if (! curlocales[j]) {
3628 setlocale_failure = TRUE;
3629 }
3630 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3631 }
c835d6be 3632
e5f10d49
KW
3633 if (! setlocale_failure) { /* All succeeded */
3634 break; /* Exit trial_locales loop */
49c85077 3635 }
65ebb059 3636 }
98994639 3637
49c85077
KW
3638 /* Here, something failed; will need to try a fallback. */
3639 ok = 0;
65ebb059 3640
49c85077
KW
3641 if (i == 0) {
3642 unsigned int j;
98994639 3643
65ebb059 3644 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3645
3646# ifdef LC_ALL
98994639 3647
49c85077
KW
3648 PerlIO_printf(Perl_error_log,
3649 "perl: warning: Setting locale failed.\n");
98994639 3650
7d4bcc4a 3651# else /* !LC_ALL */
98994639 3652
49c85077
KW
3653 PerlIO_printf(Perl_error_log,
3654 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3655
e5f10d49
KW
3656 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3657 if (! curlocales[j]) {
3658 PerlIO_printf(Perl_error_log, category_names[j]);
3659 }
3660 else {
3661 Safefree(curlocales[j]);
3662 }
3663 }
7d4bcc4a 3664
7d4bcc4a 3665# endif /* LC_ALL */
98994639 3666
49c85077
KW
3667 PerlIO_printf(Perl_error_log,
3668 "perl: warning: Please check that your locale settings:\n");
98994639 3669
7d4bcc4a
KW
3670# ifdef __GLIBC__
3671
49c85077
KW
3672 PerlIO_printf(Perl_error_log,
3673 "\tLANGUAGE = %c%s%c,\n",
3674 language ? '"' : '(',
3675 language ? language : "unset",
3676 language ? '"' : ')');
7d4bcc4a 3677# endif
98994639 3678
49c85077
KW
3679 PerlIO_printf(Perl_error_log,
3680 "\tLC_ALL = %c%s%c,\n",
3681 lc_all ? '"' : '(',
3682 lc_all ? lc_all : "unset",
3683 lc_all ? '"' : ')');
98994639 3684
7d4bcc4a
KW
3685# if defined(USE_ENVIRON_ARRAY)
3686
49c85077 3687 {
cd999af9 3688 char **e;
d5e32b93
KW
3689
3690 /* Look through the environment for any variables of the
3691 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3692 * already handled above. These are assumed to be locale
3693 * settings. Output them and their values. */
cd999af9 3694 for (e = environ; *e; e++) {
d5e32b93
KW
3695 const STRLEN prefix_len = sizeof("LC_") - 1;
3696 STRLEN uppers_len;
3697
cd999af9 3698 if ( strBEGINs(*e, "LC_")
c8b388b0 3699 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3700 && (uppers_len = strspn(*e + prefix_len,
3701 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3702 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3703 {
3704 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3705 (int) (prefix_len + uppers_len), *e,
3706 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3707 }
3708 }
49c85077 3709 }
7d4bcc4a
KW
3710
3711# else
3712
49c85077
KW
3713 PerlIO_printf(Perl_error_log,
3714 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3715
3716# endif
98994639 3717
49c85077
KW
3718 PerlIO_printf(Perl_error_log,
3719 "\tLANG = %c%s%c\n",
3720 lang ? '"' : '(',
3721 lang ? lang : "unset",
3722 lang ? '"' : ')');
98994639 3723
49c85077
KW
3724 PerlIO_printf(Perl_error_log,
3725 " are supported and installed on your system.\n");
3726 }
98994639 3727
65ebb059 3728 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3729 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3730 * usually change the upper bound of the loop we are in.
3731 *
3732 * Since the system's default way of setting the locale has not
3733 * found one that works, We use Perl's defined ordering: LC_ALL,
3734 * LANG, and the C locale. We don't try the same locale twice, so
3735 * don't add to the list if already there. (On POSIX systems, the
3736 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3737 * but there's no harm done by doing it explicitly.
3738 *
3739 * Note that this tries the LC_ALL environment variable even on
3740 * systems which have no LC_ALL locale setting. This may or may
3741 * not have been originally intentional, but there's no real need
3742 * to change the behavior. */
65ebb059
KW
3743 if (lc_all) {
3744 for (j = 0; j < trial_locales_count; j++) {
3745 if (strEQ(lc_all, trial_locales[j])) {
3746 goto done_lc_all;
3747 }
3748 }
3749 trial_locales[trial_locales_count++] = lc_all;
3750 }
3751 done_lc_all:
98994639 3752
65ebb059
KW
3753 if (lang) {
3754 for (j = 0; j < trial_locales_count; j++) {
3755 if (strEQ(lang, trial_locales[j])) {
3756 goto done_lang;
3757 }
3758 }
3759 trial_locales[trial_locales_count++] = lang;
3760 }
3761 done_lang:
3762
7d4bcc4a
KW
3763# if defined(WIN32) && defined(LC_ALL)
3764
65ebb059
KW
3765 /* For Windows, we also try the system default locale before "C".
3766 * (If there exists a Windows without LC_ALL we skip this because
3767 * it gets too complicated. For those, the "C" is the next
3768 * fallback possibility). The "" is the same as the 0th element of
3769 * the array, but the code at the loop above knows to treat it
3770 * differently when not the 0th */
3771 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3772
3773# endif
65ebb059
KW
3774
3775 for (j = 0; j < trial_locales_count; j++) {
3776 if (strEQ("C", trial_locales[j])) {
3777 goto done_C;
3778 }
3779 }
3780 trial_locales[trial_locales_count++] = "C";
98994639 3781
65ebb059
KW
3782 done_C: ;
3783 } /* end of first time through the loop */
98994639 3784
7d4bcc4a
KW
3785# ifdef WIN32
3786
65ebb059 3787 next_iteration: ;
7d4bcc4a
KW
3788
3789# endif
65ebb059
KW
3790
3791 } /* end of looping through the trial locales */
3792
3793 if (ok < 1) { /* If we tried to fallback */
3794 const char* msg;
3795 if (! setlocale_failure) { /* fallback succeeded */
3796 msg = "Falling back to";
3797 }
3798 else { /* fallback failed */
e5f10d49 3799 unsigned int j;
98994639 3800
65ebb059
KW
3801 /* We dropped off the end of the loop, so have to decrement i to
3802 * get back to the value the last time through */
3803 i--;
98994639 3804
65ebb059
KW
3805 ok = -1;
3806 msg = "Failed to fall back to";
3807
3808 /* To continue, we should use whatever values we've got */
7d4bcc4a 3809
e5f10d49
KW
3810 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3811 Safefree(curlocales[j]);
3812 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3813 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3814 }
65ebb059
KW
3815 }
3816
3817 if (locwarn) {
3818 const char * description;
3819 const char * name = "";
3820 if (strEQ(trial_locales[i], "C")) {
3821 description = "the standard locale";
3822 name = "C";
3823 }
7d4bcc4a
KW
3824
3825# ifdef SYSTEM_DEFAULT_LOCALE
3826
65ebb059
KW
3827 else if (strEQ(trial_locales[i], "")) {
3828 description = "the system default locale";
3829 if (system_default_locale) {
3830 name = system_default_locale;
3831 }
3832 }
7d4bcc4a
KW
3833
3834# endif /* SYSTEM_DEFAULT_LOCALE */
3835
65ebb059
KW
3836 else {
3837 description = "a fallback locale";
3838 name = trial_locales[i];
3839 }
3840 if (name && strNE(name, "")) {
3841 PerlIO_printf(Perl_error_log,
3842 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3843 }
3844 else {
3845 PerlIO_printf(Perl_error_log,
3846 "perl: warning: %s %s.\n", msg, description);
3847 }
3848 }
3849 } /* End of tried to fallback */
98994639 3850
e5f10d49
KW
3851 /* Done with finding the locales; update our records */
3852
7d4bcc4a
KW
3853# ifdef USE_LOCALE_CTYPE
3854
948523db 3855 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3856
e5f10d49 3857# endif
7d4bcc4a
KW
3858# ifdef USE_LOCALE_COLLATE
3859
948523db 3860 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3861
e5f10d49