This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add z/OS locale categories
[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
36dbc955
KW
203# ifdef USE_LOCALE_SYNTAX
204 LC_SYNTAX,
205# endif
206# ifdef USE_LOCALE_TOD
207 LC_TOD,
208# endif
e5f10d49
KW
209# ifdef LC_ALL
210 LC_ALL,
211# endif
212 -1 /* Placeholder because C doesn't allow a
213 trailing comma, and it would get complicated
214 with all the #ifdef's */
215};
216
217/* The top-most real element is LC_ALL */
218
4ef8bdf9 219const char * const category_names[] = {
e5f10d49
KW
220
221# ifdef USE_LOCALE_NUMERIC
222 "LC_NUMERIC",
223# endif
224# ifdef USE_LOCALE_CTYPE
225 "LC_CTYPE",
226# endif
227# ifdef USE_LOCALE_COLLATE
228 "LC_COLLATE",
229# endif
230# ifdef USE_LOCALE_TIME
231 "LC_TIME",
232# endif
233# ifdef USE_LOCALE_MESSAGES
234 "LC_MESSAGES",
235# endif
236# ifdef USE_LOCALE_MONETARY
237 "LC_MONETARY",
238# endif
9821811f
KW
239# ifdef USE_LOCALE_ADDRESS
240 "LC_ADDRESS",
241# endif
242# ifdef USE_LOCALE_IDENTIFICATION
243 "LC_IDENTIFICATION",
244# endif
245# ifdef USE_LOCALE_MEASUREMENT
246 "LC_MEASUREMENT",
247# endif
248# ifdef USE_LOCALE_PAPER
249 "LC_PAPER",
250# endif
251# ifdef USE_LOCALE_TELEPHONE
252 "LC_TELEPHONE",
253# endif
36dbc955
KW
254# ifdef USE_LOCALE_SYNTAX
255 "LC_SYNTAX",
256# endif
257# ifdef USE_LOCALE_TOD
258 "LC_TOD",
259# endif
e5f10d49
KW
260# ifdef LC_ALL
261 "LC_ALL",
262# endif
263 NULL /* Placeholder */
264 };
265
266# ifdef LC_ALL
267
268 /* On systems with LC_ALL, it is kept in the highest index position. (-2
269 * to account for the final unused placeholder element.) */
270# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
271
272# else
273
274 /* On systems without LC_ALL, we pretend it is there, one beyond the real
275 * top element, hence in the unused placeholder element. */
276# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
277
278# endif
279
280/* Pretending there is an LC_ALL element just above allows us to avoid most
281 * special cases. Most loops through these arrays in the code below are
282 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
283 * on either type of system. But the code must be written to not access the
948523db
KW
284 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
285 * checked for at compile time by using the #define LC_ALL_INDEX which is only
286 * defined if we do have LC_ALL. */
e5f10d49 287
b09aaf40
KW
288STATIC const char *
289S_category_name(const int category)
290{
291 unsigned int i;
292
293#ifdef LC_ALL
294
295 if (category == LC_ALL) {
296 return "LC_ALL";
297 }
298
299#endif
300
301 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
302 if (category == categories[i]) {
303 return category_names[i];
304 }
305 }
306
307 {
308 const char suffix[] = " (unknown)";
309 int temp = category;
310 Size_t length = sizeof(suffix) + 1;
311 char * unknown;
312 dTHX;
313
314 if (temp < 0) {
315 length++;
316 temp = - temp;
317 }
318
319 /* Calculate the number of digits */
320 while (temp >= 10) {
321 temp /= 10;
322 length++;
323 }
324
325 Newx(unknown, length, char);
326 my_snprintf(unknown, length, "%d%s", category, suffix);
327 SAVEFREEPV(unknown);
328 return unknown;
329 }
330}
331
948523db
KW
332/* Now create LC_foo_INDEX #defines for just those categories on this system */
333# ifdef USE_LOCALE_NUMERIC
334# define LC_NUMERIC_INDEX 0
335# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
336# else
337# define _DUMMY_NUMERIC -1
338# endif
339# ifdef USE_LOCALE_CTYPE
340# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
341# define _DUMMY_CTYPE LC_CTYPE_INDEX
342# else
343# define _DUMMY_CTYPE _DUMMY_NUMERIC
344# endif
345# ifdef USE_LOCALE_COLLATE
346# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
347# define _DUMMY_COLLATE LC_COLLATE_INDEX
348# else
8203b3c3 349# define _DUMMY_COLLATE _DUMMY_CTYPE
948523db
KW
350# endif
351# ifdef USE_LOCALE_TIME
352# define LC_TIME_INDEX _DUMMY_COLLATE + 1
353# define _DUMMY_TIME LC_TIME_INDEX
354# else
355# define _DUMMY_TIME _DUMMY_COLLATE
356# endif
357# ifdef USE_LOCALE_MESSAGES
358# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
359# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
360# else
361# define _DUMMY_MESSAGES _DUMMY_TIME
362# endif
363# ifdef USE_LOCALE_MONETARY
364# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
365# define _DUMMY_MONETARY LC_MONETARY_INDEX
366# else
367# define _DUMMY_MONETARY _DUMMY_MESSAGES
368# endif
9821811f
KW
369# ifdef USE_LOCALE_ADDRESS
370# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
371# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
372# else
373# define _DUMMY_ADDRESS _DUMMY_MONETARY
374# endif
375# ifdef USE_LOCALE_IDENTIFICATION
376# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
377# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
378# else
379# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
380# endif
381# ifdef USE_LOCALE_MEASUREMENT
382# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
383# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
384# else
385# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
386# endif
387# ifdef USE_LOCALE_PAPER
388# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
389# define _DUMMY_PAPER LC_PAPER_INDEX
390# else
391# define _DUMMY_PAPER _DUMMY_MEASUREMENT
392# endif
393# ifdef USE_LOCALE_TELEPHONE
394# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
395# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
396# else
397# define _DUMMY_TELEPHONE _DUMMY_PAPER
398# endif
36dbc955
KW
399# ifdef USE_LOCALE_SYNTAX
400# define LC_SYNTAX_INDEX _DUMMY_TELEPHONE + 1
401# define _DUMMY_SYNTAX LC_SYNTAX_INDEX
402# else
403# define _DUMMY_SYNTAX _DUMMY_TELEPHONE
404# endif
405# ifdef USE_LOCALE_TOD
406# define LC_TOD_INDEX _DUMMY_SYNTAX + 1
407# define _DUMMY_TOD LC_TOD_INDEX
408# else
409# define _DUMMY_TOD _DUMMY_SYNTAX
410# endif
948523db 411# ifdef LC_ALL
36dbc955 412# define LC_ALL_INDEX _DUMMY_TOD + 1
948523db
KW
413# endif
414#endif /* ifdef USE_LOCALE */
8ef6e574 415
d2b24094 416/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
417#ifdef WIN32
418# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
419#else
420# define my_setlocale(cat, locale) setlocale(cat, locale)
421#endif
422
423#ifndef USE_POSIX_2008_LOCALE
424
425/* "do_setlocale_c" is intended to be called when the category is a constant
426 * known at compile time; "do_setlocale_r", not known until run time */
427# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
428# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
45cef8fb 429# define FIX_GLIBC_LC_MESSAGES_BUG(i)
e9bc6d6b
KW
430
431#else /* Below uses POSIX 2008 */
432
433/* We emulate setlocale with our own function. LC_foo is not valid for the
434 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
435 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
436 * proper offset into the array 'category_masks[]'. At runtime, we have to
437 * search through the array (as the actual numbers may not be small contiguous
438 * positive integers which would lend themselves to array lookup). */
439# define do_setlocale_c(cat, locale) \
440 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
441# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
442
45cef8fb
KW
443# if ! defined(__GLIBC__) || ! defined(USE_LOCALE_MESSAGES)
444
445# define FIX_GLIBC_LC_MESSAGES_BUG(i)
446
447# else /* Invalidate glibc cache of loaded translations, see [perl #134264] */
448
449# include <libintl.h>
450# define FIX_GLIBC_LC_MESSAGES_BUG(i) \
451 STMT_START { \
452 if ((i) == LC_MESSAGES_INDEX) { \
453 textdomain(textdomain(NULL)); \
454 } \
455 } STMT_END
456
457# endif
458
e9bc6d6b
KW
459/* A third array, parallel to the ones above to map from category to its
460 * equivalent mask */
461const int category_masks[] = {
462# ifdef USE_LOCALE_NUMERIC
463 LC_NUMERIC_MASK,
464# endif
465# ifdef USE_LOCALE_CTYPE
466 LC_CTYPE_MASK,
467# endif
468# ifdef USE_LOCALE_COLLATE
469 LC_COLLATE_MASK,
470# endif
471# ifdef USE_LOCALE_TIME
472 LC_TIME_MASK,
473# endif
474# ifdef USE_LOCALE_MESSAGES
475 LC_MESSAGES_MASK,
476# endif
477# ifdef USE_LOCALE_MONETARY
478 LC_MONETARY_MASK,
479# endif
480# ifdef USE_LOCALE_ADDRESS
481 LC_ADDRESS_MASK,
482# endif
483# ifdef USE_LOCALE_IDENTIFICATION
484 LC_IDENTIFICATION_MASK,
485# endif
486# ifdef USE_LOCALE_MEASUREMENT
487 LC_MEASUREMENT_MASK,
488# endif
489# ifdef USE_LOCALE_PAPER
490 LC_PAPER_MASK,
491# endif
492# ifdef USE_LOCALE_TELEPHONE
493 LC_TELEPHONE_MASK,
494# endif
36dbc955
KW
495# ifdef USE_LOCALE_SYNTAX
496 LC_SYNTAX_MASK,
497# endif
498# ifdef USE_LOCALE_TOD
499 LC_TOD_MASK,
500# endif
e9bc6d6b
KW
501 /* LC_ALL can't be turned off by a Configure
502 * option, and in Posix 2008, should always be
503 * here, so compile it in unconditionally.
504 * This could catch some glitches at compile
505 * time */
506 LC_ALL_MASK
507 };
508
509STATIC const char *
510S_emulate_setlocale(const int category,
511 const char * locale,
512 unsigned int index,
513 const bool is_index_valid
514 )
515{
516 /* This function effectively performs a setlocale() on just the current
517 * thread; thus it is thread-safe. It does this by using the POSIX 2008
518 * locale functions to emulate the behavior of setlocale(). Similar to
519 * regular setlocale(), the return from this function points to memory that
520 * can be overwritten by other system calls, so needs to be copied
521 * immediately if you need to retain it. The difference here is that
522 * system calls besides another setlocale() can overwrite it.
523 *
524 * By doing this, most locale-sensitive functions become thread-safe. The
525 * exceptions are mostly those that return a pointer to static memory.
526 *
527 * This function takes the same parameters, 'category' and 'locale', that
528 * the regular setlocale() function does, but it also takes two additional
529 * ones. This is because the 2008 functions don't use a category; instead
530 * they use a corresponding mask. Because this function operates in both
531 * worlds, it may need one or the other or both. This function can
532 * calculate the mask from the input category, but to avoid this
533 * calculation, if the caller knows at compile time what the mask is, it
534 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
535 * parameter is ignored.
536 *
537 * POSIX 2008, for some sick reason, chose not to provide a method to find
538 * the category name of a locale. Some vendors have created a
539 * querylocale() function to do just that. This function is a lot simpler
540 * to implement on systems that have this. Otherwise, we have to keep
541 * track of what the locale has been set to, so that we can return its
542 * name to emulate setlocale(). It's also possible for C code in some
543 * library to change the locale without us knowing it, though as of
544 * September 2017, there are no occurrences in CPAN of uselocale(). Some
545 * libraries do use setlocale(), but that changes the global locale, and
546 * threads using per-thread locales will just ignore those changes.
547 * Another problem is that without querylocale(), we have to guess at what
548 * was meant by setting a locale of "". We handle this by not actually
549 * ever setting to "" (unless querylocale exists), but to emulate what we
550 * think should happen for "".
551 */
552
553 int mask;
554 locale_t old_obj;
555 locale_t new_obj;
556 dTHX;
557
558# ifdef DEBUGGING
559
560 if (DEBUG_Lv_TEST || debug_initialization) {
561 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);
562 }
563
564# endif
565
566 /* If the input mask might be incorrect, calculate the correct one */
567 if (! is_index_valid) {
568 unsigned int i;
569
570# ifdef DEBUGGING
571
572 if (DEBUG_Lv_TEST || debug_initialization) {
573 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
574 }
575
576# endif
577
578 for (i = 0; i <= LC_ALL_INDEX; i++) {
579 if (category == categories[i]) {
580 index = i;
581 goto found_index;
582 }
583 }
584
585 /* Here, we don't know about this category, so can't handle it.
586 * Fallback to the early POSIX usages */
587 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
588 "Unknown locale category %d; can't set it to %s\n",
589 category, locale);
590 return NULL;
591
592 found_index: ;
593
594# ifdef DEBUGGING
595
596 if (DEBUG_Lv_TEST || debug_initialization) {
597 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
598 }
599
600# endif
601
602 }
603
604 mask = category_masks[index];
605
606# ifdef DEBUGGING
607
608 if (DEBUG_Lv_TEST || debug_initialization) {
609 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
610 }
611
612# endif
613
614 /* If just querying what the existing locale is ... */
615 if (locale == NULL) {
616 locale_t cur_obj = uselocale((locale_t) 0);
617
618# ifdef DEBUGGING
619
620 if (DEBUG_Lv_TEST || debug_initialization) {
621 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
622 }
623
624# endif
625
626 if (cur_obj == LC_GLOBAL_LOCALE) {
627 return my_setlocale(category, NULL);
628 }
629
630# ifdef HAS_QUERYLOCALE
631
632 return (char *) querylocale(mask, cur_obj);
633
d2b24094 634# else
ecdda939
KW
635
636 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
637 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
638
8e13243a
KW
639# if defined(_NL_LOCALE_NAME) \
640 && defined(DEBUGGING) \
641 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
642 /* On systems that accept any locale name, the real underlying locale
643 * is often returned by this internal function, so we can't use it */
e9bc6d6b
KW
644 {
645 /* Internal glibc for querylocale(), but doesn't handle
646 * empty-string ("") locale properly; who knows what other
3a732259 647 * glitches. Check for it now, under debug. */
e9bc6d6b
KW
648
649 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
650 uselocale((locale_t) 0));
651 /*
652 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
653 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
654 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
655 */
656 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
657 if ( strNE(PL_curlocales[index], temp_name)
658 && ! ( isNAME_C_OR_POSIX(temp_name)
659 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
660
661# ifdef USE_C_BACKTRACE
662
663 dump_c_backtrace(Perl_debug_log, 20, 1);
664
665# endif
666
667 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
668 " (%s) and what internal glibc thinks"
669 " (%s)\n", category_names[index],
670 PL_curlocales[index], temp_name);
671 }
672
673 return temp_name;
674 }
675 }
676
677# endif
678
679 /* Without querylocale(), we have to use our record-keeping we've
680 * done. */
681
682 if (category != LC_ALL) {
683
684# ifdef DEBUGGING
685
686 if (DEBUG_Lv_TEST || debug_initialization) {
687 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
688 }
689
690# endif
691
692 return PL_curlocales[index];
693 }
694 else { /* For LC_ALL */
695 unsigned int i;
696 Size_t names_len = 0;
697 char * all_string;
7c93a471 698 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
699
700 /* If we have a valid LC_ALL value, just return it */
701 if (PL_curlocales[LC_ALL_INDEX]) {
702
703# ifdef DEBUGGING
704
705 if (DEBUG_Lv_TEST || debug_initialization) {
706 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
707 }
708
709# endif
710
711 return PL_curlocales[LC_ALL_INDEX];
712 }
713
714 /* Otherwise, we need to construct a string of name=value pairs.
715 * We use the glibc syntax, like
716 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
717 * First calculate the needed size. Along the way, check if all
718 * the locale names are the same */
e9bc6d6b
KW
719 for (i = 0; i < LC_ALL_INDEX; i++) {
720
721# ifdef DEBUGGING
722
723 if (DEBUG_Lv_TEST || debug_initialization) {
724 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]);
725 }
726
727# endif
728
729 names_len += strlen(category_names[i])
730 + 1 /* '=' */
731 + strlen(PL_curlocales[i])
732 + 1; /* ';' */
7c93a471
KW
733
734 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
735 are_all_categories_the_same_locale = FALSE;
736 }
737 }
738
739 /* If they are the same, we don't actually have to construct the
740 * string; we just make the entry in LC_ALL_INDEX valid, and be
741 * that single name */
742 if (are_all_categories_the_same_locale) {
743 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
744 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 745 }
7c93a471 746
e9bc6d6b
KW
747 names_len++; /* Trailing '\0' */
748 SAVEFREEPV(Newx(all_string, names_len, char));
749 *all_string = '\0';
750
751 /* Then fill in the string */
752 for (i = 0; i < LC_ALL_INDEX; i++) {
753
754# ifdef DEBUGGING
755
756 if (DEBUG_Lv_TEST || debug_initialization) {
757 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]);
758 }
759
760# endif
761
762 my_strlcat(all_string, category_names[i], names_len);
763 my_strlcat(all_string, "=", names_len);
764 my_strlcat(all_string, PL_curlocales[i], names_len);
765 my_strlcat(all_string, ";", names_len);
766 }
767
768# ifdef DEBUGGING
769
770 if (DEBUG_L_TEST || debug_initialization) {
771 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
772 }
773
774 #endif
775
776 return all_string;
777 }
778
779# ifdef EINVAL
780
781 SETERRNO(EINVAL, LIB_INVARG);
782
783# endif
784
785 return NULL;
786
d2b24094
KW
787# endif
788
6aba5c5e 789 } /* End of this being setlocale(LC_foo, NULL) */
e9bc6d6b 790
f8115d60 791 /* Here, we are switching locales. */
e9bc6d6b
KW
792
793# ifndef HAS_QUERYLOCALE
794
795 if (strEQ(locale, "")) {
796
797 /* For non-querylocale() systems, we do the setting of "" ourselves to
798 * be sure that we really know what's going on. We follow the Linux
799 * documented behavior (but if that differs from the actual behavior,
800 * this won't work exactly as the OS implements). We go out and
801 * examine the environment based on our understanding of how the system
802 * works, and use that to figure things out */
803
804 const char * const lc_all = PerlEnv_getenv("LC_ALL");
805
806 /* Use any "LC_ALL" environment variable, as it overrides everything
807 * else. */
808 if (lc_all && strNE(lc_all, "")) {
809 locale = lc_all;
810 }
811 else {
812
813 /* Otherwise, we need to dig deeper. Unless overridden, the
814 * default is the LANG environment variable; if it doesn't exist,
815 * then "C" */
816
817 const char * default_name;
818
207cc8df 819 default_name = PerlEnv_getenv("LANG");
e9bc6d6b
KW
820
821 if (! default_name || strEQ(default_name, "")) {
929a7f8c 822 default_name = "C";
e9bc6d6b 823 }
e9bc6d6b
KW
824
825 if (category != LC_ALL) {
826 const char * const name = PerlEnv_getenv(category_names[index]);
827
828 /* Here we are setting a single category. Assume will have the
829 * default name */
830 locale = default_name;
831
832 /* But then look for an overriding environment variable */
833 if (name && strNE(name, "")) {
834 locale = name;
835 }
836 }
837 else {
838 bool did_override = FALSE;
839 unsigned int i;
840
841 /* Here, we are getting LC_ALL. Any categories that don't have
842 * a corresponding environment variable set should be set to
843 * LANG, or to "C" if there is no LANG. If no individual
844 * categories differ from this, we can just set LC_ALL. This
845 * is buggy on systems that have extra categories that we don't
846 * know about. If there is an environment variable that sets
847 * that category, we won't know to look for it, and so our use
848 * of LANG or "C" improperly overrides it. On the other hand,
849 * if we don't do what is done here, and there is no
850 * environment variable, the category's locale should be set to
851 * LANG or "C". So there is no good solution. khw thinks the
852 * best is to look at systems to see what categories they have,
853 * and include them, and then to assume that we know the
854 * complete set */
855
856 for (i = 0; i < LC_ALL_INDEX; i++) {
857 const char * const env_override
24f3e849 858 = PerlEnv_getenv(category_names[i]);
e9bc6d6b
KW
859 const char * this_locale = ( env_override
860 && strNE(env_override, ""))
861 ? env_override
862 : default_name;
6435f98d
KW
863 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
864 {
6435f98d
KW
865 return NULL;
866 }
e9bc6d6b
KW
867
868 if (strNE(this_locale, default_name)) {
869 did_override = TRUE;
870 }
e9bc6d6b
KW
871 }
872
873 /* If all the categories are the same, we can set LC_ALL to
874 * that */
875 if (! did_override) {
876 locale = default_name;
877 }
878 else {
879
880 /* Here, LC_ALL is no longer valid, as some individual
881 * categories don't match it. We call ourselves
882 * recursively, as that will execute the code that
883 * generates the proper locale string for this situation.
884 * We don't do the remainder of this function, as that is
885 * to update our records, and we've just done that for the
886 * individual categories in the loop above, and doing so
887 * would cause LC_ALL to be done as well */
888 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
889 }
890 }
891 }
6aba5c5e 892 } /* End of this being setlocale(LC_foo, "") */
e9bc6d6b
KW
893 else if (strchr(locale, ';')) {
894
895 /* LC_ALL may actually incude a conglomeration of various categories.
896 * Without querylocale, this code uses the glibc (as of this writing)
897 * syntax for representing that, but that is not a stable API, and
898 * other platforms do it differently, so we have to handle all cases
899 * ourselves */
900
f0023d47 901 unsigned int i;
e9bc6d6b
KW
902 const char * s = locale;
903 const char * e = locale + strlen(locale);
904 const char * p = s;
905 const char * category_end;
906 const char * name_start;
907 const char * name_end;
908
f0023d47
KW
909 /* If the string that gives what to set doesn't include all categories,
910 * the omitted ones get set to "C". To get this behavior, first set
911 * all the individual categories to "C", and override the furnished
912 * ones below */
913 for (i = 0; i < LC_ALL_INDEX; i++) {
914 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
915 return NULL;
916 }
917 }
918
e9bc6d6b 919 while (s < e) {
e9bc6d6b
KW
920
921 /* Parse through the category */
922 while (isWORDCHAR(*p)) {
923 p++;
924 }
925 category_end = p;
926
927 if (*p++ != '=') {
928 Perl_croak(aTHX_
929 "panic: %s: %d: Unexpected character in locale name '%02X",
930 __FILE__, __LINE__, *(p-1));
931 }
932
933 /* Parse through the locale name */
934 name_start = p;
bee74f4b
KW
935 while (p < e && *p != ';') {
936 if (! isGRAPH(*p)) {
937 Perl_croak(aTHX_
938 "panic: %s: %d: Unexpected character in locale name '%02X",
939 __FILE__, __LINE__, *(p-1));
940 }
e9bc6d6b
KW
941 p++;
942 }
943 name_end = p;
944
bee74f4b
KW
945 /* Space past the semi-colon */
946 if (p < e) {
947 p++;
e9bc6d6b
KW
948 }
949
950 /* Find the index of the category name in our lists */
951 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 952 char * individ_locale;
e9bc6d6b
KW
953
954 /* Keep going if this isn't the index. The strnNE() avoids a
955 * Perl_form(), but would fail if ever a category name could be
956 * a substring of another one, like if there were a
957 * "LC_TIME_DATE" */
958 if strnNE(s, category_names[i], category_end - s) {
959 continue;
960 }
961
962 /* If this index is for the single category we're changing, we
963 * have found the locale to set it to. */
964 if (category == categories[i]) {
965 locale = Perl_form(aTHX_ "%.*s",
966 (int) (name_end - name_start),
967 name_start);
968 goto ready_to_set;
969 }
970
3c76a412 971 assert(category == LC_ALL);
bee74f4b
KW
972 individ_locale = Perl_form(aTHX_ "%.*s",
973 (int) (name_end - name_start), name_start);
6435f98d
KW
974 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
975 {
976 return NULL;
977 }
e9bc6d6b
KW
978 }
979
980 s = p;
981 }
982
983 /* Here we have set all the individual categories by recursive calls.
984 * These collectively should have fixed up LC_ALL, so can just query
985 * what that now is */
986 assert(category == LC_ALL);
987
988 return do_setlocale_c(LC_ALL, NULL);
6aba5c5e
KW
989 } /* End of this being setlocale(LC_ALL,
990 "LC_CTYPE=foo;LC_NUMERIC=bar;...") */
e9bc6d6b
KW
991
992 ready_to_set: ;
993
f8115d60
KW
994 /* Here at the end of having to deal with the absence of querylocale().
995 * Some cases have already been fully handled by recursive calls to this
996 * function. But at this point, we haven't dealt with those, but are now
997 * prepared to, knowing what the locale name to set this category to is.
998 * This would have come for free if this system had had querylocale() */
999
e9bc6d6b
KW
1000# endif /* end of ! querylocale */
1001
f8115d60
KW
1002 assert(PL_C_locale_obj);
1003
1004 /* Switching locales generally entails freeing the current one's space (at
1005 * the C library's discretion). We need to stop using that locale before
1006 * the switch. So switch to a known locale object that we don't otherwise
1007 * mess with. This returns the locale object in effect at the time of the
1008 * switch. */
1009 old_obj = uselocale(PL_C_locale_obj);
1010
1011# ifdef DEBUGGING
1012
1013 if (DEBUG_Lv_TEST || debug_initialization) {
1014 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
1015 }
1016
1017# endif
1018
1019 if (! old_obj) {
1020
1021# ifdef DEBUGGING
1022
1023 if (DEBUG_L_TEST || debug_initialization) {
1024 dSAVE_ERRNO;
1025 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1026 RESTORE_ERRNO;
1027 }
1028
1029# endif
1030
1031 return NULL;
1032 }
1033
1034# ifdef DEBUGGING
1035
1036 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1037 PerlIO_printf(Perl_debug_log,
1038 "%s:%d: emulate_setlocale now using %p\n",
1039 __FILE__, __LINE__, PL_C_locale_obj);
f8115d60
KW
1040 }
1041
1042# endif
1043
6aba5c5e
KW
1044 /* If this call is to switch to the LC_ALL C locale, it already exists, and
1045 * in fact, we already have switched to it (in preparation for what
1046 * normally is to come). But since we're already there, continue to use
70bd6bc8
KW
1047 * it instead of trying to create a new locale */
1048 if (mask == LC_ALL_MASK && isNAME_C_OR_POSIX(locale)) {
1049
1050# ifdef DEBUGGING
1051
1052 if (DEBUG_Lv_TEST || debug_initialization) {
1053 PerlIO_printf(Perl_debug_log,
1054 "%s:%d: will stay in C object\n", __FILE__, __LINE__);
1055 }
1056
1057# endif
1058
1059 new_obj = PL_C_locale_obj;
1060
1061 /* We already had switched to the C locale in preparation for freeing
b22e9937 1062 * 'old_obj' */
70bd6bc8
KW
1063 if (old_obj != LC_GLOBAL_LOCALE && old_obj != PL_C_locale_obj) {
1064 freelocale(old_obj);
1065 }
1066 }
1067 else {
b22e9937
KW
1068 /* If we weren't in a thread safe locale, set so that newlocale() below
1069 * which uses 'old_obj', uses an empty one. Same for our reserved C
1070 * object. The latter is defensive coding, so that, even if there is
1071 * some bug, we will never end up trying to modify either of these, as
1072 * if passed to newlocale(), they can be. */
1073 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1074 old_obj = (locale_t) 0;
1075 }
f8115d60 1076
b22e9937
KW
1077 /* Ready to create a new locale by modification of the exising one */
1078 new_obj = newlocale(mask, locale, old_obj);
e9bc6d6b 1079
b22e9937
KW
1080 if (! new_obj) {
1081 dSAVE_ERRNO;
e9bc6d6b
KW
1082
1083# ifdef DEBUGGING
1084
b22e9937
KW
1085 if (DEBUG_L_TEST || debug_initialization) {
1086 PerlIO_printf(Perl_debug_log,
1087 "%s:%d: emulate_setlocale creating new object"
1088 " failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1089 }
e9bc6d6b
KW
1090
1091# endif
1092
b22e9937 1093 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1094
1095# ifdef DEBUGGING
1096
b22e9937
KW
1097 if (DEBUG_L_TEST || debug_initialization) {
1098 PerlIO_printf(Perl_debug_log,
1099 "%s:%d: switching back failed: %d\n",
1100 __FILE__, __LINE__, GET_ERRNO);
1101 }
e9bc6d6b
KW
1102
1103# endif
1104
b22e9937
KW
1105 }
1106 RESTORE_ERRNO;
1107 return NULL;
e9bc6d6b 1108 }
e9bc6d6b
KW
1109
1110# ifdef DEBUGGING
1111
b22e9937
KW
1112 if (DEBUG_Lv_TEST || debug_initialization) {
1113 PerlIO_printf(Perl_debug_log,
1114 "%s:%d: emulate_setlocale created %p",
1115 __FILE__, __LINE__, new_obj);
1116 if (old_obj) {
1117 PerlIO_printf(Perl_debug_log,
1118 "; should have freed %p", old_obj);
1119 }
1120 PerlIO_printf(Perl_debug_log, "\n");
19ee3daf 1121 }
e9bc6d6b
KW
1122
1123# endif
1124
b22e9937
KW
1125 /* And switch into it */
1126 if (! uselocale(new_obj)) {
1127 dSAVE_ERRNO;
e9bc6d6b
KW
1128
1129# ifdef DEBUGGING
1130
b22e9937
KW
1131 if (DEBUG_L_TEST || debug_initialization) {
1132 PerlIO_printf(Perl_debug_log,
1133 "%s:%d: emulate_setlocale switching to new object"
1134 " failed\n", __FILE__, __LINE__);
1135 }
e9bc6d6b
KW
1136
1137# endif
1138
b22e9937 1139 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1140
1141# ifdef DEBUGGING
1142
b22e9937
KW
1143 if (DEBUG_L_TEST || debug_initialization) {
1144 PerlIO_printf(Perl_debug_log,
1145 "%s:%d: switching back failed: %d\n",
1146 __FILE__, __LINE__, GET_ERRNO);
1147 }
e9bc6d6b
KW
1148
1149# endif
1150
b22e9937
KW
1151 }
1152 freelocale(new_obj);
1153 RESTORE_ERRNO;
1154 return NULL;
e9bc6d6b 1155 }
70bd6bc8 1156 }
e9bc6d6b
KW
1157
1158# ifdef DEBUGGING
1159
1160 if (DEBUG_Lv_TEST || debug_initialization) {
b22e9937
KW
1161 PerlIO_printf(Perl_debug_log,
1162 "%s:%d: emulate_setlocale now using %p\n",
1163 __FILE__, __LINE__, new_obj);
e9bc6d6b
KW
1164 }
1165
1166# endif
1167
1168 /* We are done, except for updating our records (if the system doesn't keep
1169 * them) and in the case of locale "", we don't actually know what the
1170 * locale that got switched to is, as it came from the environment. So
1171 * have to find it */
1172
1173# ifdef HAS_QUERYLOCALE
1174
1175 if (strEQ(locale, "")) {
1176 locale = querylocale(mask, new_obj);
1177 }
1178
1179# else
1180
1181 /* Here, 'locale' is the return value */
1182
1183 /* Without querylocale(), we have to update our records */
1184
1185 if (category == LC_ALL) {
1186 unsigned int i;
1187
1188 /* For LC_ALL, we change all individual categories to correspond */
1189 /* PL_curlocales is a parallel array, so has same
1190 * length as 'categories' */
1191 for (i = 0; i <= LC_ALL_INDEX; i++) {
1192 Safefree(PL_curlocales[i]);
1193 PL_curlocales[i] = savepv(locale);
1194 }
45cef8fb
KW
1195
1196 FIX_GLIBC_LC_MESSAGES_BUG(LC_MESSAGES_INDEX);
e9bc6d6b
KW
1197 }
1198 else {
1199
1200 /* For a single category, if it's not the same as the one in LC_ALL, we
1201 * nullify LC_ALL */
1202
1203 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1204 Safefree(PL_curlocales[LC_ALL_INDEX]);
1205 PL_curlocales[LC_ALL_INDEX] = NULL;
1206 }
1207
1208 /* Then update the category's record */
1209 Safefree(PL_curlocales[index]);
1210 PL_curlocales[index] = savepv(locale);
45cef8fb
KW
1211
1212 FIX_GLIBC_LC_MESSAGES_BUG(index);
e9bc6d6b
KW
1213 }
1214
1215# endif
1216
1217 return locale;
1218}
1219
1220#endif /* USE_POSIX_2008_LOCALE */
1221
1222#if 0 /* Code that was to emulate thread-safe locales on platforms that
1223 didn't natively support them */
1224
1225/* The way this would work is that we would keep a per-thread list of the
1226 * correct locale for that thread. Any operation that was locale-sensitive
1227 * would have to be changed so that it would look like this:
1228 *
1229 * LOCALE_LOCK;
1230 * setlocale to the correct locale for this operation
1231 * do operation
1232 * LOCALE_UNLOCK
1233 *
1234 * This leaves the global locale in the most recently used operation's, but it
1235 * was locked long enough to get the result. If that result is static, it
1236 * needs to be copied before the unlock.
1237 *
1238 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1239 * the setup, but are no-ops when not needed, and similarly,
1240 * END_LOCALE_DEPENDENT_OP for the tear-down
1241 *
1242 * But every call to a locale-sensitive function would have to be changed, and
1243 * if a module didn't cooperate by using the mutex, things would break.
1244 *
1245 * This code was abandoned before being completed or tested, and is left as-is
1246*/
1247
1248# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1249# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1250
1251STATIC char *
1252S_locking_setlocale(pTHX_
1253 const int category,
1254 const char * locale,
1255 int index,
1256 const bool is_index_valid
1257 )
1258{
1259 /* This function kind of performs a setlocale() on just the current thread;
1260 * thus it is kind of thread-safe. It does this by keeping a thread-level
1261 * array of the current locales for each category. Every time a locale is
1262 * switched to, it does the switch globally, but updates the thread's
1263 * array. A query as to what the current locale is just returns the
1264 * appropriate element from the array, and doesn't actually call the system
1265 * setlocale(). The saving into the array is done in an uninterruptible
1266 * section of code, so is unaffected by whatever any other threads might be
1267 * doing.
1268 *
1269 * All locale-sensitive operations must work by first starting a critical
1270 * section, then switching to the thread's locale as kept by this function,
1271 * and then doing the operation, then ending the critical section. Thus,
1272 * each gets done in the appropriate locale. simulating thread-safety.
1273 *
1274 * This function takes the same parameters, 'category' and 'locale', that
1275 * the regular setlocale() function does, but it also takes two additional
1276 * ones. This is because as described earlier. If we know on input the
1277 * index corresponding to the category into the array where we store the
1278 * current locales, we don't have to calculate it. If the caller knows at
06cc5386 1279 * compile time what the index is, it can pass it, setting
e9bc6d6b
KW
1280 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1281 *
1282 */
1283
1284 /* If the input index might be incorrect, calculate the correct one */
1285 if (! is_index_valid) {
1286 unsigned int i;
1287
1288 if (DEBUG_Lv_TEST || debug_initialization) {
1289 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1290 }
1291
1292 for (i = 0; i <= LC_ALL_INDEX; i++) {
1293 if (category == categories[i]) {
1294 index = i;
1295 goto found_index;
1296 }
1297 }
1298
1299 /* Here, we don't know about this category, so can't handle it.
1300 * XXX best we can do is to unsafely set this
1301 * XXX warning */
1302
1303 return my_setlocale(category, locale);
1304
1305 found_index: ;
1306
1307 if (DEBUG_Lv_TEST || debug_initialization) {
1308 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1309 }
1310 }
1311
1312 /* For a query, just return what's in our records */
1313 if (new_locale == NULL) {
1314 return curlocales[index];
1315 }
1316
1317
1318 /* Otherwise, we need to do the switch, and save the result, all in a
1319 * critical section */
1320
1321 Safefree(curlocales[[index]]);
1322
1323 /* It might be that this is called from an already-locked section of code.
1324 * We would have to detect and skip the LOCK/UNLOCK if so */
1325 LOCALE_LOCK;
1326
1327 curlocales[index] = savepv(my_setlocale(category, new_locale));
1328
1329 if (strEQ(new_locale, "")) {
1330
1331#ifdef LC_ALL
1332
1333 /* The locale values come from the environment, and may not all be the
1334 * same, so for LC_ALL, we have to update all the others, while the
1335 * mutex is still locked */
1336
1337 if (category == LC_ALL) {
1338 unsigned int i;
1339 for (i = 0; i < LC_ALL_INDEX) {
1340 curlocales[i] = my_setlocale(categories[i], NULL);
1341 }
1342 }
1343 }
1344
1345#endif
1346
1347 LOCALE_UNLOCK;
1348
1349 return curlocales[index];
1350}
1351
1352#endif
d36adde0 1353#ifdef USE_LOCALE
837ce802 1354
a4f00dcc 1355STATIC void
86799d2d 1356S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1357{
86799d2d
KW
1358 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1359 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1360
86799d2d
KW
1361#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1362 || defined(HAS_NL_LANGINFO))
98994639 1363
87f8e8e7 1364 const char * radix = (use_locale)
4e6826bf 1365 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1366 /* FALSE => already in dest locale */
1367 : ".";
2213a3be 1368
87f8e8e7 1369 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1370
87f8e8e7
KW
1371 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1372 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1373 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1374 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1375 && _is_cur_LC_category_utf8(LC_NUMERIC))
1376 {
1377 SvUTF8_on(PL_numeric_radix_sv);
1378 }
86799d2d
KW
1379
1380# ifdef DEBUGGING
7d4bcc4a 1381
2fcc0ca9
KW
1382 if (DEBUG_L_TEST || debug_initialization) {
1383 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1384 SvPVX(PL_numeric_radix_sv),
1385 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1386 }
69014004 1387
86799d2d 1388# endif
d36adde0
KW
1389#else
1390
1391 PERL_UNUSED_ARG(use_locale);
1392
86799d2d 1393#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1394
98994639
HS
1395}
1396
398eeea9
KW
1397STATIC void
1398S_new_numeric(pTHX_ const char *newnum)
98994639 1399{
7d4bcc4a
KW
1400
1401#ifndef USE_LOCALE_NUMERIC
1402
1403 PERL_UNUSED_ARG(newnum);
1404
1405#else
0d071d52 1406
291a84fb 1407 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1408 * core Perl this and that 'newnum' is the name of the new locale.
1409 * It installs this locale as the current underlying default.
1410 *
1411 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1412 * set_numeric_underlying() and set_numeric_standard() functions, which
1413 * should probably not be called directly, but only via macros like
0d071d52
KW
1414 * SET_NUMERIC_STANDARD() in perl.h.
1415 *
1416 * The toggling is necessary mainly so that a non-dot radix decimal point
1417 * character can be output, while allowing internal calculations to use a
1418 * dot.
1419 *
1420 * This sets several interpreter-level variables:
bb304765 1421 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1422 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1423 * that the current locale is the program's underlying
1424 * locale
1425 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1426 * that the current locale is the C locale or
1427 * indistinguishable from the C locale. If non-zero, it
1428 * is in C; if > 1, it means it may not be toggled away
7738054c 1429 * from C.
4c68b815
KW
1430 * PL_numeric_underlying_is_standard A bool kept by this function
1431 * indicating that the underlying locale and the standard
1432 * C locale are indistinguishable for the purposes of
1433 * LC_NUMERIC. This happens when both of the above two
1434 * variables are true at the same time. (Toggling is a
1435 * no-op under these circumstances.) This variable is
1436 * used to avoid having to recalculate.
398eeea9 1437 */
0d071d52 1438
b03f34cf 1439 char *save_newnum;
98994639
HS
1440
1441 if (! newnum) {
43c5f42d
NC
1442 Safefree(PL_numeric_name);
1443 PL_numeric_name = NULL;
98994639 1444 PL_numeric_standard = TRUE;
892e6465 1445 PL_numeric_underlying = TRUE;
4c68b815 1446 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1447 return;
1448 }
1449
b03f34cf 1450 save_newnum = stdize_locale(savepv(newnum));
892e6465 1451 PL_numeric_underlying = TRUE;
4c68b815
KW
1452 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1453
69c5e0db
KW
1454#ifndef TS_W32_BROKEN_LOCALECONV
1455
4c68b815 1456 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1457 * them. But on broken Windows systems calling my_nl_langinfo() for
1458 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1459 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1460 if (! PL_numeric_standard) {
4e6826bf 1461 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1462 FALSE /* Don't toggle locale */ ))
4e6826bf 1463 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1464 }
abe1abcf 1465
69c5e0db
KW
1466#endif
1467
4c68b815 1468 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1469 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1470 Safefree(PL_numeric_name);
b03f34cf 1471 PL_numeric_name = save_newnum;
b03f34cf 1472 }
abe1abcf
KW
1473 else {
1474 Safefree(save_newnum);
1475 }
4c28b29c 1476
4c68b815
KW
1477 PL_numeric_underlying_is_standard = PL_numeric_standard;
1478
7e5377f7 1479# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1480
1481 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1482 PL_numeric_name,
1483 PL_underlying_numeric_obj);
1484
1485#endif
1486
4c68b815
KW
1487 if (DEBUG_L_TEST || debug_initialization) {
1488 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1489 }
1490
4c28b29c
KW
1491 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1492 * have to worry about the radix being a non-dot. (Core operations that
1493 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1494 if (PL_numeric_standard) {
1495 set_numeric_radix(0);
1496 }
1497 else {
1498 set_numeric_standard();
1499 }
4c28b29c 1500
98994639 1501#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1502
98994639
HS
1503}
1504
1505void
1506Perl_set_numeric_standard(pTHX)
1507{
7d4bcc4a 1508
98994639 1509#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1510
28c1bf33
KW
1511 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1512 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1513 * macro avoids calling this routine if toggling isn't necessary according
1514 * to our records (which could be wrong if some XS code has changed the
1515 * locale behind our back) */
0d071d52 1516
7d4bcc4a
KW
1517# ifdef DEBUGGING
1518
2fcc0ca9
KW
1519 if (DEBUG_L_TEST || debug_initialization) {
1520 PerlIO_printf(Perl_debug_log,
7bb8f702 1521 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1522 }
98994639 1523
7d4bcc4a 1524# endif
7bb8f702
KW
1525
1526 do_setlocale_c(LC_NUMERIC, "C");
1527 PL_numeric_standard = TRUE;
1528 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1529 set_numeric_radix(0);
1530
98994639 1531#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1532
98994639
HS
1533}
1534
1535void
5792c642 1536Perl_set_numeric_underlying(pTHX)
98994639 1537{
7d4bcc4a 1538
98994639 1539#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1540
28c1bf33 1541 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1542 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1543 * instead of calling this directly. The macro avoids calling this routine
1544 * if toggling isn't necessary according to our records (which could be
1545 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1546
7d4bcc4a
KW
1547# ifdef DEBUGGING
1548
2fcc0ca9
KW
1549 if (DEBUG_L_TEST || debug_initialization) {
1550 PerlIO_printf(Perl_debug_log,
7bb8f702 1551 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1552 PL_numeric_name);
1553 }
98994639 1554
7d4bcc4a 1555# endif
7bb8f702
KW
1556
1557 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1558 PL_numeric_standard = PL_numeric_underlying_is_standard;
1559 PL_numeric_underlying = TRUE;
1560 set_numeric_radix(! PL_numeric_standard);
1561
98994639 1562#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1563
98994639
HS
1564}
1565
1566/*
1567 * Set up for a new ctype locale.
1568 */
a4f00dcc
KW
1569STATIC void
1570S_new_ctype(pTHX_ const char *newctype)
98994639 1571{
7d4bcc4a
KW
1572
1573#ifndef USE_LOCALE_CTYPE
1574
7d4bcc4a
KW
1575 PERL_UNUSED_ARG(newctype);
1576 PERL_UNUSED_CONTEXT;
1577
1578#else
0d071d52 1579
291a84fb 1580 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1581 * core Perl this and that 'newctype' is the name of the new locale.
1582 *
1583 * This function sets up the folding arrays for all 256 bytes, assuming
1584 * that tofold() is tolc() since fold case is not a concept in POSIX,
1585 *
1586 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1587 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1588 * this function should be called directly only from this file and from
0d071d52
KW
1589 * POSIX::setlocale() */
1590
27da23d5 1591 dVAR;
013f4e03 1592 unsigned int i;
98994639 1593
8b7358b9
KW
1594 /* Don't check for problems if we are suppressing the warnings */
1595 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
30d8090d 1596 bool maybe_utf8_turkic = FALSE;
8b7358b9 1597
7918f24d
NC
1598 PERL_ARGS_ASSERT_NEW_CTYPE;
1599
215c5139
KW
1600 /* We will replace any bad locale warning with 1) nothing if the new one is
1601 * ok; or 2) a new warning for the bad new locale */
1602 if (PL_warn_locale) {
1603 SvREFCNT_dec_NN(PL_warn_locale);
1604 PL_warn_locale = NULL;
1605 }
1606
c1284011 1607 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1608
1609 /* A UTF-8 locale gets standard rules. But note that code still has to
1610 * handle this specially because of the three problematic code points */
1611 if (PL_in_utf8_CTYPE_locale) {
1612 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
30d8090d
KW
1613
1614 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1615 * Turkic. Make sure these two are the only anomalies. (We don't use
1616 * towupper and towlower because they aren't in C89.) */
5b64f24c
KW
1617
1618#if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1619
1620 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1621
1622#else
1623
30d8090d 1624 if (toupper('i') == 'i' && tolower('I') == 'I') {
5b64f24c
KW
1625
1626#endif
30d8090d
KW
1627 check_for_problems = TRUE;
1628 maybe_utf8_turkic = TRUE;
1629 }
31f05a37 1630 }
8b7358b9
KW
1631
1632 /* We don't populate the other lists if a UTF-8 locale, but do check that
1633 * everything works as expected, unless checking turned off */
1634 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1635 /* Assume enough space for every character being bad. 4 spaces each
1636 * for the 94 printable characters that are output like "'x' "; and 5
1637 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1638 * NUL */
8b7358b9 1639 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1640 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1641 to start */
1642 unsigned int bad_count = 0; /* Count of bad characters */
1643
baa60164 1644 for (i = 0; i < 256; i++) {
8b7358b9 1645 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1646 if (isupper(i))
1647 PL_fold_locale[i] = (U8) tolower(i);
1648 else if (islower(i))
1649 PL_fold_locale[i] = (U8) toupper(i);
1650 else
1651 PL_fold_locale[i] = (U8) i;
8b7358b9 1652 }
8c6180a9
KW
1653
1654 /* If checking for locale problems, see if the native ASCII-range
1655 * printables plus \n and \t are in their expected categories in
1656 * the new locale. If not, this could mean big trouble, upending
1657 * Perl's and most programs' assumptions, like having a
1658 * metacharacter with special meaning become a \w. Fortunately,
1659 * it's very rare to find locales that aren't supersets of ASCII
1660 * nowadays. It isn't a problem for most controls to be changed
1661 * into something else; we check only \n and \t, though perhaps \r
1662 * could be an issue as well. */
7d4bcc4a 1663 if ( check_for_problems
8c6180a9
KW
1664 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1665 {
8b7358b9 1666 bool is_bad = FALSE;
6726b4f4 1667 char name[4] = { '\0' };
8b7358b9
KW
1668
1669 /* Convert the name into a string */
6726b4f4 1670 if (isGRAPH_A(i)) {
8b7358b9
KW
1671 name[0] = i;
1672 name[1] = '\0';
1673 }
1674 else if (i == '\n') {
6726b4f4
KW
1675 my_strlcpy(name, "\\n", sizeof(name));
1676 }
1677 else if (i == '\t') {
1678 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1679 }
1680 else {
6726b4f4
KW
1681 assert(i == ' ');
1682 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1683 }
1684
1685 /* Check each possibe class */
1686 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1687 is_bad = TRUE;
1688 DEBUG_L(PerlIO_printf(Perl_debug_log,
1689 "isalnum('%s') unexpectedly is %d\n",
1690 name, cBOOL(isalnum(i))));
1691 }
1692 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1693 is_bad = TRUE;
1694 DEBUG_L(PerlIO_printf(Perl_debug_log,
1695 "isalpha('%s') unexpectedly is %d\n",
1696 name, cBOOL(isalpha(i))));
1697 }
1698 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1699 is_bad = TRUE;
1700 DEBUG_L(PerlIO_printf(Perl_debug_log,
1701 "isdigit('%s') unexpectedly is %d\n",
1702 name, cBOOL(isdigit(i))));
1703 }
1704 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1705 is_bad = TRUE;
1706 DEBUG_L(PerlIO_printf(Perl_debug_log,
1707 "isgraph('%s') unexpectedly is %d\n",
1708 name, cBOOL(isgraph(i))));
1709 }
1710 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1711 is_bad = TRUE;
1712 DEBUG_L(PerlIO_printf(Perl_debug_log,
1713 "islower('%s') unexpectedly is %d\n",
1714 name, cBOOL(islower(i))));
1715 }
1716 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1717 is_bad = TRUE;
1718 DEBUG_L(PerlIO_printf(Perl_debug_log,
1719 "isprint('%s') unexpectedly is %d\n",
1720 name, cBOOL(isprint(i))));
1721 }
1722 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1723 is_bad = TRUE;
1724 DEBUG_L(PerlIO_printf(Perl_debug_log,
1725 "ispunct('%s') unexpectedly is %d\n",
1726 name, cBOOL(ispunct(i))));
1727 }
1728 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1729 is_bad = TRUE;
1730 DEBUG_L(PerlIO_printf(Perl_debug_log,
1731 "isspace('%s') unexpectedly is %d\n",
1732 name, cBOOL(isspace(i))));
1733 }
1734 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1735 is_bad = TRUE;
1736 DEBUG_L(PerlIO_printf(Perl_debug_log,
1737 "isupper('%s') unexpectedly is %d\n",
1738 name, cBOOL(isupper(i))));
1739 }
1740 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1741 is_bad = TRUE;
1742 DEBUG_L(PerlIO_printf(Perl_debug_log,
1743 "isxdigit('%s') unexpectedly is %d\n",
1744 name, cBOOL(isxdigit(i))));
1745 }
65c15174 1746 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
8b7358b9
KW
1747 is_bad = TRUE;
1748 DEBUG_L(PerlIO_printf(Perl_debug_log,
1749 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1750 name, tolower(i), (int) toLOWER_A(i)));
1751 }
65c15174 1752 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
8b7358b9
KW
1753 is_bad = TRUE;
1754 DEBUG_L(PerlIO_printf(Perl_debug_log,
1755 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1756 name, toupper(i), (int) toUPPER_A(i)));
1757 }
1758 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1759 is_bad = TRUE;
1760 DEBUG_L(PerlIO_printf(Perl_debug_log,
1761 "'\\n' (=%02X) is not a control\n", (int) i));
1762 }
1763
1764 /* Add to the list; Separate multiple entries with a blank */
1765 if (is_bad) {
1766 if (bad_count) {
1767 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1768 }
8b7358b9
KW
1769 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1770 bad_count++;
8c6180a9
KW
1771 }
1772 }
1773 }
1774
30d8090d
KW
1775 if (bad_count == 2 && maybe_utf8_turkic) {
1776 bad_count = 0;
1777 *bad_chars_list = '\0';
1778 PL_fold_locale['I'] = 'I';
1779 PL_fold_locale['i'] = 'i';
1780 PL_in_utf8_turkic_locale = TRUE;
1781 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1782 __FILE__, __LINE__, newctype));
1783 }
1784 else {
b8df1494 1785 PL_in_utf8_turkic_locale = FALSE;
30d8090d 1786 }
b8df1494 1787
7d4bcc4a
KW
1788# ifdef MB_CUR_MAX
1789
8c6180a9 1790 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1791 * this locale requires more than one byte, there are going to be
8c6180a9 1792 * problems. */
9c8a6dc2
KW
1793 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1794 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1795 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1796
8b7358b9
KW
1797 if ( check_for_problems && MB_CUR_MAX > 1
1798 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1799
1800 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1801 * locale. Just assume that the implementation for them (plus
1802 * for POSIX) is correct and the > 1 value is spurious. (Since
1803 * these are specially handled to never be considered UTF-8
1804 * locales, as long as this is the only problem, everything
1805 * should work fine */
1806 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1807 {
8c6180a9
KW
1808 multi_byte_locale = TRUE;
1809 }
7d4bcc4a
KW
1810
1811# endif
8c6180a9 1812
30d8090d
KW
1813 /* If we found problems and we want them output, do so */
1814 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1815 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1816 {
8b7358b9
KW
1817 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1818 PL_warn_locale = Perl_newSVpvf(aTHX_
1819 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1820 " which have\nunexpected meanings: %s\nThe Perl program"
1821 " will use the expected meanings",
8b7358b9 1822 newctype, bad_chars_list);
8b7358b9
KW
1823 }
1824 else {
1825 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1826 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1827 newctype,
8c6180a9
KW
1828 (multi_byte_locale)
1829 ? " Some characters in it are not recognized by"
1830 " Perl."
1831 : "",
1832 (bad_count)
1833 ? "\nThe following characters (and maybe others)"
1834 " may not have the same meaning as the Perl"
1835 " program expects:\n"
1836 : "",
1837 (bad_count)
1838 ? bad_chars_list
1839 : ""
1840 );
8b7358b9
KW
1841 }
1842
b119c2be
KW
1843# ifdef HAS_NL_LANGINFO
1844
1845 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1846 /* parameter FALSE is a don't care here */
4e6826bf 1847 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1848
1849# endif
1850
8b7358b9
KW
1851 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1852
cc9eaeb0 1853 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1854 * output the message now. If not in that scope, we save the
1855 * message to be output at the first operation using this locale,
1856 * if that actually happens. Most programs don't use locales, so
1857 * they are immune to bad ones. */
cc9eaeb0 1858 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1859
780fcc9f
KW
1860 /* The '0' below suppresses a bogus gcc compiler warning */
1861 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1862
bddebb56
KW
1863 if (IN_LC(LC_CTYPE)) {
1864 SvREFCNT_dec_NN(PL_warn_locale);
1865 PL_warn_locale = NULL;
1866 }
780fcc9f 1867 }
baa60164 1868 }
31f05a37 1869 }
98994639
HS
1870
1871#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1872
98994639
HS
1873}
1874
98994639 1875void
2726666d
KW
1876Perl__warn_problematic_locale()
1877{
2726666d
KW
1878
1879#ifdef USE_LOCALE_CTYPE
1880
5f04a188
KW
1881 dTHX;
1882
1883 /* Internal-to-core function that outputs the message in PL_warn_locale,
1884 * and then NULLS it. Should be called only through the macro
1885 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1886
2726666d 1887 if (PL_warn_locale) {
2726666d
KW
1888 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1889 SvPVX(PL_warn_locale),
1890 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1891 SvREFCNT_dec_NN(PL_warn_locale);
1892 PL_warn_locale = NULL;
1893 }
1894
1895#endif
1896
1897}
1898
a4f00dcc
KW
1899STATIC void
1900S_new_collate(pTHX_ const char *newcoll)
98994639 1901{
7d4bcc4a
KW
1902
1903#ifndef USE_LOCALE_COLLATE
1904
1905 PERL_UNUSED_ARG(newcoll);
1906 PERL_UNUSED_CONTEXT;
1907
1908#else
0d071d52 1909
291a84fb 1910 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1911 * core Perl this and that 'newcoll' is the name of the new locale.
1912 *
d35fca5f
KW
1913 * The design of locale collation is that every locale change is given an
1914 * index 'PL_collation_ix'. The first time a string particpates in an
1915 * operation that requires collation while locale collation is active, it
1916 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1917 * magic includes the collation index, and the transformation of the string
1918 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1919 * instead of the string itself. If a string changes, the magic is
1920 * cleared. The next time the locale changes, the index is incremented,
1921 * and so we know during a comparison that the transformation is not
1922 * necessarily still valid, and so is recomputed. Note that if the locale
1923 * changes enough times, the index could wrap (a U32), and it is possible
1924 * that a transformation would improperly be considered valid, leading to
1925 * an unlikely bug */
0d071d52 1926
98994639
HS
1927 if (! newcoll) {
1928 if (PL_collation_name) {
1929 ++PL_collation_ix;
1930 Safefree(PL_collation_name);
1931 PL_collation_name = NULL;
1932 }
1933 PL_collation_standard = TRUE;
00bf60ca 1934 is_standard_collation:
98994639
HS
1935 PL_collxfrm_base = 0;
1936 PL_collxfrm_mult = 2;
165a1c52 1937 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1938 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1939 PL_strxfrm_max_cp = 0;
98994639
HS
1940 return;
1941 }
1942
d35fca5f 1943 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1944 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1945 ++PL_collation_ix;
1946 Safefree(PL_collation_name);
1947 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1948 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1949 if (PL_collation_standard) {
1950 goto is_standard_collation;
1951 }
98994639 1952
165a1c52 1953 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1954 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1955 PL_strxfrm_max_cp = 0;
165a1c52 1956
59c018b9
KW
1957 /* A locale collation definition includes primary, secondary, tertiary,
1958 * etc. weights for each character. To sort, the primary weights are
1959 * used, and only if they compare equal, then the secondary weights are
1960 * used, and only if they compare equal, then the tertiary, etc.
1961 *
1962 * strxfrm() works by taking the input string, say ABC, and creating an
1963 * output transformed string consisting of first the primary weights,
1964 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1965 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1966 * may not have weights at every level. In our example, let's say B
1967 * doesn't have a tertiary weight, and A doesn't have a secondary
1968 * weight. The constructed string is then going to be
1969 * A¹B¹C¹ B²C² A³C³ ....
1970 * This has the desired effect that strcmp() will look at the secondary
1971 * or tertiary weights only if the strings compare equal at all higher
1972 * priority weights. The spaces shown here, like in
c342d20e 1973 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1974 * are not just for readability. In the general case, these must
1975 * actually be bytes, which we will call here 'separator weights'; and
1976 * they must be smaller than any other weight value, but since these
1977 * are C strings, only the terminating one can be a NUL (some
1978 * implementations may include a non-NUL separator weight just before
1979 * the NUL). Implementations tend to reserve 01 for the separator
1980 * weights. They are needed so that a shorter string's secondary
1981 * weights won't be misconstrued as primary weights of a longer string,
1982 * etc. By making them smaller than any other weight, the shorter
1983 * string will sort first. (Actually, if all secondary weights are
1984 * smaller than all primary ones, there is no need for a separator
1985 * weight between those two levels, etc.)
1986 *
1987 * The length of the transformed string is roughly a linear function of
1988 * the input string. It's not exactly linear because some characters
1989 * don't have weights at all levels. When we call strxfrm() we have to
1990 * allocate some memory to hold the transformed string. The
1991 * calculations below try to find coefficients 'm' and 'b' for this
1992 * locale so that m*x + b equals how much space we need, given the size
1993 * of the input string in 'x'. If we calculate too small, we increase
1994 * the size as needed, and call strxfrm() again, but it is better to
1995 * get it right the first time to avoid wasted expensive string
1996 * transformations. */
1997
98994639 1998 {
79f120c8
KW
1999 /* We use the string below to find how long the tranformation of it
2000 * is. Almost all locales are supersets of ASCII, or at least the
2001 * ASCII letters. We use all of them, half upper half lower,
2002 * because if we used fewer, we might hit just the ones that are
2003 * outliers in a particular locale. Most of the strings being
2004 * collated will contain a preponderance of letters, and even if
2005 * they are above-ASCII, they are likely to have the same number of
2006 * weight levels as the ASCII ones. It turns out that digits tend
2007 * to have fewer levels, and some punctuation has more, but those
2008 * are relatively sparse in text, and khw believes this gives a
2009 * reasonable result, but it could be changed if experience so
2010 * dictates. */
2011 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
2012 char * x_longer; /* Transformed 'longer' */
2013 Size_t x_len_longer; /* Length of 'x_longer' */
2014
2015 char * x_shorter; /* We also transform a substring of 'longer' */
2016 Size_t x_len_shorter;
2017
a4a439fb 2018 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
2019 * are interested only in its length). It is used because it has
2020 * the intelligence to handle all cases, but to work, it needs some
2021 * values of 'm' and 'b' to get it started. For the purposes of
2022 * this calculation we use a very conservative estimate of 'm' and
2023 * 'b'. This assumes a weight can be multiple bytes, enough to
2024 * hold any UV on the platform, and there are 5 levels, 4 weight
2025 * bytes, and a trailing NUL. */
2026 PL_collxfrm_base = 5;
2027 PL_collxfrm_mult = 5 * sizeof(UV);
2028
2029 /* Find out how long the transformation really is */
a4a439fb
KW
2030 x_longer = _mem_collxfrm(longer,
2031 sizeof(longer) - 1,
2032 &x_len_longer,
2033
2034 /* We avoid converting to UTF-8 in the
2035 * called function by telling it the
2036 * string is in UTF-8 if the locale is a
2037 * UTF-8 one. Since the string passed
2038 * here is invariant under UTF-8, we can
2039 * claim it's UTF-8 even though it isn't.
2040 * */
2041 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2042 Safefree(x_longer);
2043
2044 /* Find out how long the transformation of a substring of 'longer'
2045 * is. Together the lengths of these transformations are
2046 * sufficient to calculate 'm' and 'b'. The substring is all of
2047 * 'longer' except the first character. This minimizes the chances
2048 * of being swayed by outliers */
a4a439fb 2049 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 2050 sizeof(longer) - 2,
a4a439fb
KW
2051 &x_len_shorter,
2052 PL_in_utf8_COLLATE_locale);
79f120c8
KW
2053 Safefree(x_shorter);
2054
2055 /* If the results are nonsensical for this simple test, the whole
2056 * locale definition is suspect. Mark it so that locale collation
2057 * is not active at all for it. XXX Should we warn? */
2058 if ( x_len_shorter == 0
2059 || x_len_longer == 0
2060 || x_len_shorter >= x_len_longer)
2061 {
2062 PL_collxfrm_mult = 0;
2063 PL_collxfrm_base = 0;
2064 }
2065 else {
2066 SSize_t base; /* Temporary */
2067
2068 /* We have both: m * strlen(longer) + b = x_len_longer
2069 * m * strlen(shorter) + b = x_len_shorter;
2070 * subtracting yields:
2071 * m * (strlen(longer) - strlen(shorter))
2072 * = x_len_longer - x_len_shorter
2073 * But we have set things up so that 'shorter' is 1 byte smaller
2074 * than 'longer'. Hence:
2075 * m = x_len_longer - x_len_shorter
2076 *
2077 * But if something went wrong, make sure the multiplier is at
2078 * least 1.
2079 */
2080 if (x_len_longer > x_len_shorter) {
2081 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2082 }
2083 else {
2084 PL_collxfrm_mult = 1;
2085 }
2086
2087 /* mx + b = len
2088 * so: b = len - mx
2089 * but in case something has gone wrong, make sure it is
2090 * non-negative */
2091 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2092 if (base < 0) {
2093 base = 0;
2094 }
2095
2096 /* Add 1 for the trailing NUL */
2097 PL_collxfrm_base = base + 1;
2098 }
58eebef2 2099
7d4bcc4a
KW
2100# ifdef DEBUGGING
2101
58eebef2
KW
2102 if (DEBUG_L_TEST || debug_initialization) {
2103 PerlIO_printf(Perl_debug_log,
b07929e4
KW
2104 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2105 "x_len_longer=%zu,"
2106 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
2107 __FILE__, __LINE__,
2108 PL_in_utf8_COLLATE_locale,
2109 x_len_shorter, x_len_longer,
2110 PL_collxfrm_mult, PL_collxfrm_base);
2111 }
7d4bcc4a
KW
2112# endif
2113
98994639
HS
2114 }
2115 }
2116
2117#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 2118
98994639
HS
2119}
2120
d36adde0
KW
2121#endif
2122
d2b24094 2123#ifdef WIN32
b8cc575c 2124
6c332036
TC
2125#define USE_WSETLOCALE
2126
2127#ifdef USE_WSETLOCALE
2128
2129STATIC char *
2130S_wrap_wsetlocale(pTHX_ int category, const char *locale) {
2131 wchar_t *wlocale;
2132 wchar_t *wresult;
2133 char *result;
2134
2135 if (locale) {
2136 int req_size =
2137 MultiByteToWideChar(CP_UTF8, 0, locale, -1, NULL, 0);
2138
2139 if (!req_size) {
2140 errno = EINVAL;
2141 return NULL;
2142 }
2143
2144 Newx(wlocale, req_size, wchar_t);
2145 if (!MultiByteToWideChar(CP_UTF8, 0, locale, -1, wlocale, req_size)) {
2146 Safefree(wlocale);
2147 errno = EINVAL;
2148 return NULL;
2149 }
2150 }
2151 else {
2152 wlocale = NULL;
2153 }
2154 wresult = _wsetlocale(category, wlocale);
2155 Safefree(wlocale);
2156 if (wresult) {
2157 int req_size =
2158 WideCharToMultiByte(CP_UTF8, 0, wresult, -1, NULL, 0, NULL, NULL);
2159 Newx(result, req_size, char);
2160 SAVEFREEPV(result); /* is there something better we can do here? */
2161 if (!WideCharToMultiByte(CP_UTF8, 0, wresult, -1,
2162 result, req_size, NULL, NULL)) {
2163 errno = EINVAL;
2164 return NULL;
2165 }
2166 }
2167 else {
2168 result = NULL;
2169 }
2170
2171 return result;
2172}
2173
2174#endif
2175
a4f00dcc 2176STATIC char *
b8cc575c 2177S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2178{
2179 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2180 * difference between the two unless the input locale is "", which normally
2181 * means on Windows to get the machine default, which is set via the
2182 * computer's "Regional and Language Options" (or its current equivalent).
2183 * In POSIX, it instead means to find the locale from the user's
2184 * environment. This routine changes the Windows behavior to first look in
2185 * the environment, and, if anything is found, use that instead of going to
2186 * the machine default. If there is no environment override, the machine
2187 * default is used, by calling the real setlocale() with "".
2188 *
2189 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2190 * use the particular category's variable if set; otherwise to use the LANG
2191 * variable. */
b385bb4d 2192
175c4cf9 2193 bool override_LC_ALL = FALSE;
89f7b9aa 2194 char * result;
e5f10d49 2195 unsigned int i;
89f7b9aa 2196
b385bb4d 2197 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2198
2199# ifdef LC_ALL
2200
b385bb4d
KW
2201 locale = PerlEnv_getenv("LC_ALL");
2202 if (! locale) {
e5f10d49
KW
2203 if (category == LC_ALL) {
2204 override_LC_ALL = TRUE;
2205 }
2206 else {
7d4bcc4a
KW
2207
2208# endif
7d4bcc4a 2209
e5f10d49
KW
2210 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2211 if (category == categories[i]) {
2212 locale = PerlEnv_getenv(category_names[i]);
2213 goto found_locale;
2214 }
2215 }
7d4bcc4a 2216
b385bb4d 2217 locale = PerlEnv_getenv("LANG");
481465ea 2218 if (! locale) {
b385bb4d
KW
2219 locale = "";
2220 }
e5f10d49
KW
2221
2222 found_locale: ;
7d4bcc4a
KW
2223
2224# ifdef LC_ALL
2225
e5f10d49 2226 }
b385bb4d 2227 }
7d4bcc4a
KW
2228
2229# endif
2230
b385bb4d
KW
2231 }
2232
6c332036
TC
2233#ifdef USE_WSETLOCALE
2234 result = S_wrap_wsetlocale(aTHX_ category, locale);
2235#else
89f7b9aa 2236 result = setlocale(category, locale);
6c332036 2237#endif
48015184
KW
2238 DEBUG_L(STMT_START {
2239 dSAVE_ERRNO;
2240 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2241 setlocale_debug_string(category, locale, result));
2242 RESTORE_ERRNO;
2243 } STMT_END);
89f7b9aa 2244
481465ea 2245 if (! override_LC_ALL) {
89f7b9aa
KW
2246 return result;
2247 }
2248
dfd77d7a 2249 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2250 * LANG variable or the system default if there is no LANG. But these have
2251 * lower priority than the other LC_foo variables, so override it for each
2252 * one that is set. (If they are set to "", it means to use the same thing
2253 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2254
948523db 2255 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2256 result = PerlEnv_getenv(category_names[i]);
2257 if (result && strNE(result, "")) {
6c332036 2258#ifdef USE_WSETLOCALE
17a1e9ac 2259 S_wrap_wsetlocale(aTHX_ categories[i], result);
6c332036 2260#else
17a1e9ac 2261 setlocale(categories[i], result);
6c332036 2262#endif
e5f10d49
KW
2263 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2264 __FILE__, __LINE__,
2265 setlocale_debug_string(categories[i], result, "not captured")));
2266 }
89f7b9aa 2267 }
7d4bcc4a 2268
bbc98134 2269 result = setlocale(LC_ALL, NULL);
48015184
KW
2270 DEBUG_L(STMT_START {
2271 dSAVE_ERRNO;
2272 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2273 __FILE__, __LINE__,
48015184
KW
2274 setlocale_debug_string(LC_ALL, NULL, result));
2275 RESTORE_ERRNO;
2276 } STMT_END);
89f7b9aa 2277
bbc98134 2278 return result;
b385bb4d
KW
2279}
2280
2281#endif
2282
9aac5db8
KW
2283/*
2284
2285=head1 Locale-related functions and macros
2286
2287=for apidoc Perl_setlocale
2288
2289This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2290taking the same parameters, and returning the same information, except that it
9487427b
KW
2291returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2292instead return C<C> if the underlying locale has a non-dot decimal point
2293character, or a non-empty thousands separator for displaying floating point
2294numbers. This is because perl keeps that locale category such that it has a
2295dot and empty separator, changing the locale briefly during the operations
2296where the underlying one is required. C<Perl_setlocale> knows about this, and
2297compensates; regular C<setlocale> doesn't.
9aac5db8 2298
e9bc6d6b 2299Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2300declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2301C<const> (presumably because its API was specified long ago, and can't be
2302updated; it is illegal to change the information C<setlocale> returns; doing
2303so leads to segfaults.)
9aac5db8 2304
e9bc6d6b
KW
2305Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2306C<setlocale> can be completely ineffective on some platforms under some
2307configurations.
2308
9aac5db8 2309C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2310where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2311the system C<setlocale()> is ineffective, returning the wrong information, and
2312failing to actually change the locale. C<Perl_setlocale>, however works
2313properly in all circumstances.
9aac5db8
KW
2314
2315The return points to a per-thread static buffer, which is overwritten the next
2316time C<Perl_setlocale> is called from the same thread.
2317
2318=cut
2319
2320*/
2321
2322const char *
2323Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2324{
2325 /* This wraps POSIX::setlocale() */
2326
52129632 2327#ifndef USE_LOCALE
d36adde0
KW
2328
2329 PERL_UNUSED_ARG(category);
2330 PERL_UNUSED_ARG(locale);
2331
2332 return "C";
2333
2334#else
2335
9aac5db8
KW
2336 const char * retval;
2337 const char * newlocale;
2338 dSAVEDERRNO;
a4f00dcc 2339 dTHX;
d36adde0 2340 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2341
a4f00dcc
KW
2342#ifdef USE_LOCALE_NUMERIC
2343
291a84fb
KW
2344 /* A NULL locale means only query what the current one is. We have the
2345 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2346 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2347 * the correct results. All other categories don't require special
2348 * handling */
a4f00dcc
KW
2349 if (locale == NULL) {
2350 if (category == LC_NUMERIC) {
9aac5db8
KW
2351
2352 /* We don't have to copy this return value, as it is a per-thread
2353 * variable, and won't change until a future setlocale */
2354 return PL_numeric_name;
a4f00dcc
KW
2355 }
2356
7d4bcc4a 2357# ifdef LC_ALL
a4f00dcc 2358
9aac5db8
KW
2359 else if (category == LC_ALL) {
2360 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2361 }
2362
7d4bcc4a 2363# endif
a4f00dcc
KW
2364
2365 }
2366
2367#endif
2368
33e5a354
KW
2369 retval = save_to_buffer(do_setlocale_r(category, locale),
2370 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2371 SAVE_ERRNO;
2372
2373#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2374
2375 if (locale == NULL && category == LC_ALL) {
2376 RESTORE_LC_NUMERIC();
2377 }
2378
2379#endif
a4f00dcc
KW
2380
2381 DEBUG_L(PerlIO_printf(Perl_debug_log,
2382 "%s:%d: %s\n", __FILE__, __LINE__,
2383 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2384
9aac5db8
KW
2385 RESTORE_ERRNO;
2386
2387 if (! retval) {
a4f00dcc
KW
2388 return NULL;
2389 }
2390
9aac5db8 2391 /* If locale == NULL, we are just querying the state */
a4f00dcc 2392 if (locale == NULL) {
a4f00dcc
KW
2393 return retval;
2394 }
a4f00dcc 2395
1159483a
KW
2396 /* Now that have switched locales, we have to update our records to
2397 * correspond. */
a4f00dcc 2398
1159483a 2399 switch (category) {
a4f00dcc 2400
1159483a 2401#ifdef USE_LOCALE_CTYPE
a4f00dcc 2402
1159483a
KW
2403 case LC_CTYPE:
2404 new_ctype(retval);
2405 break;
a4f00dcc 2406
1159483a 2407#endif
a4f00dcc
KW
2408#ifdef USE_LOCALE_COLLATE
2409
1159483a
KW
2410 case LC_COLLATE:
2411 new_collate(retval);
2412 break;
a4f00dcc 2413
1159483a 2414#endif
a4f00dcc
KW
2415#ifdef USE_LOCALE_NUMERIC
2416
1159483a
KW
2417 case LC_NUMERIC:
2418 new_numeric(retval);
2419 break;
a4f00dcc 2420
1159483a
KW
2421#endif
2422#ifdef LC_ALL
a4f00dcc 2423
1159483a 2424 case LC_ALL:
a4f00dcc 2425
1159483a
KW
2426 /* LC_ALL updates all the things we care about. The values may not
2427 * be the same as 'retval', as the locale "" may have set things
2428 * individually */
a4f00dcc 2429
1159483a 2430# ifdef USE_LOCALE_CTYPE
a4f00dcc 2431
b0bde642 2432 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2433 new_ctype(newlocale);
b0bde642 2434 Safefree(newlocale);
a4f00dcc 2435
1159483a
KW
2436# endif /* USE_LOCALE_CTYPE */
2437# ifdef USE_LOCALE_COLLATE
2438
b0bde642 2439 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2440 new_collate(newlocale);
b0bde642 2441 Safefree(newlocale);
a4f00dcc 2442
7d4bcc4a 2443# endif
1159483a 2444# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2445
b0bde642 2446 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2447 new_numeric(newlocale);
b0bde642 2448 Safefree(newlocale);
a4f00dcc 2449
1159483a
KW
2450# endif /* USE_LOCALE_NUMERIC */
2451#endif /* LC_ALL */
a4f00dcc 2452
1159483a
KW
2453 default:
2454 break;
a4f00dcc
KW
2455 }
2456
2457 return retval;
2458
d36adde0
KW
2459#endif
2460
f7416781
KW
2461}
2462
2463PERL_STATIC_INLINE const char *
2464S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2465{
2466 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2467 * growing it if necessary */
2468
0b6697c5 2469 Size_t string_size;
f7416781
KW
2470
2471 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2472
0b6697c5
KW
2473 if (! string) {
2474 return NULL;
2475 }
2476
2477 string_size = strlen(string) + offset + 1;
2478
f7416781
KW
2479 if (*buf_size == 0) {
2480 Newx(*buf, string_size, char);
2481 *buf_size = string_size;
2482 }
2483 else if (string_size > *buf_size) {
2484 Renew(*buf, string_size, char);
2485 *buf_size = string_size;
2486 }
2487
2488 Copy(string, *buf + offset, string_size - offset, char);
2489 return *buf;
2490}
2491
2492/*
2493
f7416781
KW
2494=for apidoc Perl_langinfo
2495
ce181bf6 2496This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2497taking the same C<item> parameter values, and returning the same information.
2498But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2499of Perl's locale handling from your code, and can be used on systems that lack
2500a native C<nl_langinfo>.
2501
2502Expanding on these:
2503
2504=over
2505
2506=item *
2507
ce181bf6
KW
2508The reason it isn't quite a drop-in replacement is actually an advantage. The
2509only difference is that it returns S<C<const char *>>, whereas plain
2510C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2511forbidden to write into the buffer. By declaring this C<const>, the compiler
2512enforces this restriction, so if it is violated, you know at compilation time,
2513rather than getting segfaults at runtime.
2514
2515=item *
2516
69e2275e 2517It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2518without you having to write extra code. The reason for the extra code would be
2519because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2520kept set by Perl so that the radix is a dot, and the separator is the empty
2521string, no matter what the underlying locale is supposed to be, and so to get
2522the expected results, you have to temporarily toggle into the underlying
2523locale, and later toggle back. (You could use plain C<nl_langinfo> and
2524C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2525the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2526(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2527(decimal point) character to be a dot.)
ce181bf6
KW
2528
2529=item *
2530
2531The system function it replaces can have its static return buffer trashed,
2532not only by a subesequent call to that function, but by a C<freelocale>,
2533C<setlocale>, or other locale change. The returned buffer of this function is
2534not changed until the next call to it, so the buffer is never in a trashed
2535state.
f7416781
KW
2536
2537=item *
2538
ce181bf6
KW
2539Its return buffer is per-thread, so it also is never overwritten by a call to
2540this function from another thread; unlike the function it replaces.
2541
2542=item *
2543
2544But most importantly, it works on systems that don't have C<nl_langinfo>, such
2545as Windows, hence makes your code more portable. Of the fifty-some possible
2546items specified by the POSIX 2008 standard,
f7416781 2547L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2548only one is completely unimplemented, though on non-Windows platforms, another
2549significant one is also not implemented). It uses various techniques to
2550recover the other items, including calling C<L<localeconv(3)>>, and
2551C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2552available. Later C<strftime()> versions have additional capabilities; C<""> is
2553returned for those not available on your system.
ce181bf6
KW
2554
2555It is important to note that when called with an item that is recovered by
2556using C<localeconv>, the buffer from any previous explicit call to
2557C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2558contents if you need to access them after a call to this function. (But note
2559that you might not want to be using C<localeconv()> directly anyway, because of
2560issues like the ones listed in the second item of this list (above) for
2561C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2562call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2563unpack).
9f0bc911 2564
6201c220
KW
2565The details for those items which may deviate from what this emulation returns
2566and what a native C<nl_langinfo()> would return are specified in
2567L<I18N::Langinfo>.
f7416781 2568
ce181bf6
KW
2569=back
2570
f7416781
KW
2571When using C<Perl_langinfo> on systems that don't have a native
2572C<nl_langinfo()>, you must
2573
2574 #include "perl_langinfo.h"
2575
2576before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2577C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2578C<langinfo.h> would try to import into the namespace for code that doesn't need
2579it.)
f7416781 2580
f7416781
KW
2581The original impetus for C<Perl_langinfo()> was so that code that needs to
2582find out the current currency symbol, floating point radix character, or digit
2583grouping separator can use, on all systems, the simpler and more
2584thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2585pain to make thread-friendly. For other fields returned by C<localeconv>, it
2586is better to use the methods given in L<perlcall> to call
2587L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2588
2589=cut
2590
2591*/
2592
2593const char *
2594#ifdef HAS_NL_LANGINFO
2595Perl_langinfo(const nl_item item)
2596#else
2597Perl_langinfo(const int item)
2598#endif
2599{
f61748ac
KW
2600 return my_nl_langinfo(item, TRUE);
2601}
2602
59a15af0 2603STATIC const char *
f61748ac
KW
2604#ifdef HAS_NL_LANGINFO
2605S_my_nl_langinfo(const nl_item item, bool toggle)
2606#else
2607S_my_nl_langinfo(const int item, bool toggle)
2608#endif
2609{
ae74815b 2610 dTHX;
0b6697c5 2611 const char * retval;
f7416781 2612
d36adde0
KW
2613#ifdef USE_LOCALE_NUMERIC
2614
5a854ab3
KW
2615 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2616 * two items, and only if not already there */
4e6826bf 2617 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2618 || PL_numeric_underlying))
d36adde0
KW
2619
2620#endif /* No toggling needed if not using LC_NUMERIC */
2621
5a854ab3 2622 toggle = FALSE;
5a854ab3 2623
ab340fff 2624#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2625# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2626 || ! defined(HAS_POSIX_2008_LOCALE) \
2627 || ! defined(DUPLOCALE)
f7416781 2628
ab340fff 2629 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2630 * for those items dependent on it. This must be copied to a buffer before
2631 * switching back, as some systems destroy the buffer when setlocale() is
2632 * called */
f7416781 2633
038d3702
KW
2634 {
2635 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2636
b2fee59c 2637 if (toggle) {
038d3702 2638 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2639 }
f7416781 2640
5acc4454
KW
2641 LOCALE_LOCK; /* Prevent interference from another thread executing
2642 this code section (the only call to nl_langinfo in
2643 the core) */
2644
ee90eb2d 2645
628ff75a
KW
2646 /* Copy to a per-thread buffer, which is also one that won't be
2647 * destroyed by a subsequent setlocale(), such as the
2648 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2649 retval = save_to_buffer(nl_langinfo(item),
2650 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2651
5acc4454
KW
2652 LOCALE_UNLOCK;
2653
b2fee59c 2654 if (toggle) {
038d3702 2655 RESTORE_LC_NUMERIC();
b2fee59c 2656 }
f7416781
KW
2657 }
2658
ab340fff
KW
2659# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2660
5a854ab3 2661 {
b2fee59c
KW
2662 bool do_free = FALSE;
2663 locale_t cur = uselocale((locale_t) 0);
ab340fff 2664
b2fee59c
KW
2665 if (cur == LC_GLOBAL_LOCALE) {
2666 cur = duplocale(LC_GLOBAL_LOCALE);
2667 do_free = TRUE;
2668 }
ab340fff 2669
d36adde0
KW
2670# ifdef USE_LOCALE_NUMERIC
2671
b2fee59c 2672 if (toggle) {
e1aa2579
KW
2673 if (PL_underlying_numeric_obj) {
2674 cur = PL_underlying_numeric_obj;
2675 }
2676 else {
2677 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2678 do_free = TRUE;
2679 }
b2fee59c 2680 }
ab340fff 2681
d36adde0
KW
2682# endif
2683
628ff75a
KW
2684 /* We have to save it to a buffer, because the freelocale() just below
2685 * can invalidate the internal one */
0b6697c5
KW
2686 retval = save_to_buffer(nl_langinfo_l(item, cur),
2687 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2688
b2fee59c
KW
2689 if (do_free) {
2690 freelocale(cur);
2691 }
5a854ab3 2692 }
ab340fff 2693
c1566110
KW
2694# endif
2695
0b6697c5 2696 if (strEQ(retval, "")) {
4e6826bf 2697 if (item == YESSTR) {
c1566110
KW
2698 return "yes";
2699 }
4e6826bf 2700 if (item == NOSTR) {
c1566110
KW
2701 return "no";
2702 }
2703 }
2704
0b6697c5 2705 return retval;
ab340fff 2706
f7416781 2707#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2708
2709 {
2710
f7416781
KW
2711# ifdef HAS_LOCALECONV
2712
43dd6b15 2713 const struct lconv* lc;
628ff75a 2714 const char * temp;
038d3702 2715 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2716
69c5e0db
KW
2717# ifdef TS_W32_BROKEN_LOCALECONV
2718
2719 const char * save_global;
2720 const char * save_thread;
2721 int needed_size;
2722 char * ptr;
2723 char * e;
2724 char * item_start;
2725
2726# endif
f7416781
KW
2727# endif
2728# ifdef HAS_STRFTIME
2729
43dd6b15
KW
2730 struct tm tm;
2731 bool return_format = FALSE; /* Return the %format, not the value */
2732 const char * format;
f7416781
KW
2733
2734# endif
2735
43dd6b15
KW
2736 /* We copy the results to a per-thread buffer, even if not
2737 * multi-threaded. This is in part to simplify this code, and partly
2738 * because we need a buffer anyway for strftime(), and partly because a
2739 * call of localeconv() could otherwise wipe out the buffer, and the
2740 * programmer would not be expecting this, as this is a nl_langinfo()
2741 * substitute after all, so s/he might be thinking their localeconv()
2742 * is safe until another localeconv() call. */
f7416781 2743
43dd6b15
KW
2744 switch (item) {
2745 Size_t len;
f7416781 2746
8d72e74e 2747 /* This is unimplemented */
4e6826bf 2748 case ERA: /* For use with strftime() %E modifier */
f7416781 2749
43dd6b15
KW
2750 default:
2751 return "";
f7416781 2752
43dd6b15 2753 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2754 case YESEXPR: return "^[+1yY]";
2755 case YESSTR: return "yes";
2756 case NOEXPR: return "^[-0nN]";
2757 case NOSTR: return "no";
2758
8d72e74e
KW
2759 case CODESET:
2760
2761# ifndef WIN32
2762
2763 /* On non-windows, this is unimplemented, in part because of
2764 * inconsistencies between vendors. The Darwin native
2765 * nl_langinfo() implementation simply looks at everything past
2766 * any dot in the name, but that doesn't work for other
2767 * vendors. Many Linux locales that don't have UTF-8 in their
2768 * names really are UTF-8, for example; z/OS locales that do
2769 * have UTF-8 in their names, aren't really UTF-8 */
2770 return "";
2771
2772# else
2773
2774 { /* But on Windows, the name does seem to be consistent, so
2775 use that. */
2776 const char * p;
2777 const char * first;
2778 Size_t offset = 0;
2779 const char * name = my_setlocale(LC_CTYPE, NULL);
2780
2781 if (isNAME_C_OR_POSIX(name)) {
2782 return "ANSI_X3.4-1968";
2783 }
2784
2785 /* Find the dot in the locale name */
2786 first = (const char *) strchr(name, '.');
2787 if (! first) {
2788 first = name;
2789 goto has_nondigit;
2790 }
2791
2792 /* Look at everything past the dot */
2793 first++;
2794 p = first;
2795
2796 while (*p) {
2797 if (! isDIGIT(*p)) {
2798 goto has_nondigit;
2799 }
2800
2801 p++;
2802 }
2803
2804 /* Here everything past the dot is a digit. Treat it as a
2805 * code page */
18503cbd 2806 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2807 &PL_langinfo_bufsize, 0);
8d72e74e 2808 offset = STRLENs("CP");
f7416781 2809
8d72e74e
KW
2810 has_nondigit:
2811
2812 retval = save_to_buffer(first, &PL_langinfo_buf,
2813 &PL_langinfo_bufsize, offset);
2814 }
2815
2816 break;
2817
2818# endif
f7416781
KW
2819# ifdef HAS_LOCALECONV
2820
4e6826bf 2821 case CRNCYSTR:
f7416781 2822
43dd6b15
KW
2823 /* We don't bother with localeconv_l() because any system that
2824 * has it is likely to also have nl_langinfo() */
291a84fb 2825
69c5e0db
KW
2826 LOCALE_LOCK_V; /* Prevent interference with other threads
2827 using localeconv() */
2828
2829# ifdef TS_W32_BROKEN_LOCALECONV
2830
2831 /* This is a workaround for a Windows bug prior to VS 15.
2832 * What we do here is, while locked, switch to the global
2833 * locale so localeconv() works; then switch back just before
2834 * the unlock. This can screw things up if some thread is
2835 * already using the global locale while assuming no other is.
2836 * A different workaround would be to call GetCurrencyFormat on
2837 * a known value, and parse it; patches welcome
2838 *
2839 * We have to use LC_ALL instead of LC_MONETARY because of
2840 * another bug in Windows */
2841
2842 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2843 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2844 save_global= savepv(my_setlocale(LC_ALL, NULL));
2845 my_setlocale(LC_ALL, save_thread);
2846
2847# endif
5acc4454 2848
43dd6b15
KW
2849 lc = localeconv();
2850 if ( ! lc
2851 || ! lc->currency_symbol
2852 || strEQ("", lc->currency_symbol))
2853 {
69c5e0db 2854 LOCALE_UNLOCK_V;
43dd6b15
KW
2855 return "";
2856 }
f7416781 2857
43dd6b15 2858 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2859 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2860 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2861 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2862 { /* khw couldn't figure out how the localedef specifications
2863 would show that the $ should replace the radix; this is
2864 just a guess as to how it might work.*/
a34edee3 2865 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2866 }
2867 else if (lc->p_cs_precedes) {
a34edee3 2868 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2869 }
2870 else {
a34edee3 2871 PL_langinfo_buf[0] = '+';
43dd6b15 2872 }
f7416781 2873
69c5e0db
KW
2874# ifdef TS_W32_BROKEN_LOCALECONV
2875
2876 my_setlocale(LC_ALL, save_global);
2877 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2878 my_setlocale(LC_ALL, save_thread);
2879 Safefree(save_global);
2880 Safefree(save_thread);
2881
2882# endif
2883
2884 LOCALE_UNLOCK_V;
43dd6b15 2885 break;
f7416781 2886
69c5e0db
KW
2887# ifdef TS_W32_BROKEN_LOCALECONV
2888
4e6826bf 2889 case RADIXCHAR:
69c5e0db
KW
2890
2891 /* For this, we output a known simple floating point number to
2892 * a buffer, and parse it, looking for the radix */
2893
2894 if (toggle) {
2895 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2896 }
2897
2898 if (PL_langinfo_bufsize < 10) {
2899 PL_langinfo_bufsize = 10;
2900 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2901 }
2902
2903 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2904 "%.1f", 1.5);
2905 if (needed_size >= (int) PL_langinfo_bufsize) {
2906 PL_langinfo_bufsize = needed_size + 1;
2907 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2908 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2909 "%.1f", 1.5);
2910 assert(needed_size < (int) PL_langinfo_bufsize);
2911 }
2912
2913 ptr = PL_langinfo_buf;
2914 e = PL_langinfo_buf + PL_langinfo_bufsize;
2915
2916 /* Find the '1' */
2917 while (ptr < e && *ptr != '1') {
2918 ptr++;
2919 }
2920 ptr++;
2921
2922 /* Find the '5' */
2923 item_start = ptr;
2924 while (ptr < e && *ptr != '5') {
2925 ptr++;
2926 }
2927
2928 /* Everything in between is the radix string */
2929 if (ptr >= e) {
2930 PL_langinfo_buf[0] = '?';
2931 PL_langinfo_buf[1] = '\0';
2932 }
2933 else {
2934 *ptr = '\0';
2935 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2936 }
2937
2938 if (toggle) {
2939 RESTORE_LC_NUMERIC();
2940 }
2941
2942 retval = PL_langinfo_buf;
2943 break;
2944
2945# else
2946
2947 case RADIXCHAR: /* No special handling needed */
2948
2949# endif
2950
4e6826bf 2951 case THOUSEP:
f7416781 2952
43dd6b15 2953 if (toggle) {
038d3702 2954 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2955 }
f7416781 2956
69c5e0db
KW
2957 LOCALE_LOCK_V; /* Prevent interference with other threads
2958 using localeconv() */
2959
2960# ifdef TS_W32_BROKEN_LOCALECONV
2961
2962 /* This should only be for the thousands separator. A
2963 * different work around would be to use GetNumberFormat on a
2964 * known value and parse the result to find the separator */
2965 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2966 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2967 save_global = savepv(my_setlocale(LC_ALL, NULL));
2968 my_setlocale(LC_ALL, save_thread);
2969# if 0
2970 /* This is the start of code that for broken Windows replaces
2971 * the above and below code, and instead calls
2972 * GetNumberFormat() and then would parse that to find the
2973 * thousands separator. It needs to handle UTF-16 vs -8
2974 * issues. */
2975
2976 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2977 DEBUG_L(PerlIO_printf(Perl_debug_log,
2978 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2979 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2980
2981# endif
2982# endif
5acc4454 2983
43dd6b15
KW
2984 lc = localeconv();
2985 if (! lc) {
628ff75a 2986 temp = "";
33394adc 2987 }
43dd6b15 2988 else {
4e6826bf 2989 temp = (item == RADIXCHAR)
43dd6b15
KW
2990 ? lc->decimal_point
2991 : lc->thousands_sep;
628ff75a
KW
2992 if (! temp) {
2993 temp = "";
43dd6b15
KW
2994 }
2995 }
f7416781 2996
0b6697c5
KW
2997 retval = save_to_buffer(temp, &PL_langinfo_buf,
2998 &PL_langinfo_bufsize, 0);
f7416781 2999
69c5e0db
KW
3000# ifdef TS_W32_BROKEN_LOCALECONV
3001
3002 my_setlocale(LC_ALL, save_global);
3003 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3004 my_setlocale(LC_ALL, save_thread);
3005 Safefree(save_global);
3006 Safefree(save_thread);
3007
3008# endif
3009
3010 LOCALE_UNLOCK_V;
5acc4454 3011
43dd6b15 3012 if (toggle) {
038d3702 3013 RESTORE_LC_NUMERIC();
43dd6b15 3014 }
f7416781 3015
43dd6b15 3016 break;
f7416781
KW
3017
3018# endif
3019# ifdef HAS_STRFTIME
3020
43dd6b15
KW
3021 /* These are defined by C89, so we assume that strftime supports
3022 * them, and so are returned unconditionally; they may not be what
3023 * the locale actually says, but should give good enough results
3024 * for someone using them as formats (as opposed to trying to parse
3025 * them to figure out what the locale says). The other format
3026 * items are actually tested to verify they work on the platform */
4e6826bf
KW
3027 case D_FMT: return "%x";
3028 case T_FMT: return "%X";
3029 case D_T_FMT: return "%c";
43dd6b15
KW
3030
3031 /* These formats are only available in later strfmtime's */
4e6826bf 3032 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
3033
3034 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
3035 case ABDAY_1: case ABDAY_2: case ABDAY_3:
3036 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
3037 case ALT_DIGITS:
3038 case AM_STR: case PM_STR:
3039 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
3040 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
3041 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
3042 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
3043 case DAY_5: case DAY_6: case DAY_7:
3044 case MON_1: case MON_2: case MON_3: case MON_4:
3045 case MON_5: case MON_6: case MON_7: case MON_8:
3046 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
3047
3048 LOCALE_LOCK;
3049
3050 init_tm(&tm); /* Precaution against core dumps */
3051 tm.tm_sec = 30;
3052 tm.tm_min = 30;
3053 tm.tm_hour = 6;
3054 tm.tm_year = 2017 - 1900;
3055 tm.tm_wday = 0;
3056 tm.tm_mon = 0;
3057 switch (item) {
3058 default:
3059 LOCALE_UNLOCK;
3060 Perl_croak(aTHX_
3061 "panic: %s: %d: switch case: %d problem",
3062 __FILE__, __LINE__, item);
3063 NOT_REACHED; /* NOTREACHED */
3064
4e6826bf
KW
3065 case PM_STR: tm.tm_hour = 18;
3066 case AM_STR:
43dd6b15
KW
3067 format = "%p";
3068 break;
3069
4e6826bf
KW
3070 case ABDAY_7: tm.tm_wday++;
3071 case ABDAY_6: tm.tm_wday++;
3072 case ABDAY_5: tm.tm_wday++;
3073 case ABDAY_4: tm.tm_wday++;
3074 case ABDAY_3: tm.tm_wday++;
3075 case ABDAY_2: tm.tm_wday++;
3076 case ABDAY_1:
43dd6b15
KW
3077 format = "%a";
3078 break;
3079
4e6826bf
KW
3080 case DAY_7: tm.tm_wday++;
3081 case DAY_6: tm.tm_wday++;
3082 case DAY_5: tm.tm_wday++;
3083 case DAY_4: tm.tm_wday++;
3084 case DAY_3: tm.tm_wday++;
3085 case DAY_2: tm.tm_wday++;
3086 case DAY_1:
43dd6b15
KW
3087 format = "%A";
3088 break;
3089
4e6826bf
KW
3090 case ABMON_12: tm.tm_mon++;
3091 case ABMON_11: tm.tm_mon++;
3092 case ABMON_10: tm.tm_mon++;
3093 case ABMON_9: tm.tm_mon++;
3094 case ABMON_8: tm.tm_mon++;
3095 case ABMON_7: tm.tm_mon++;
3096 case ABMON_6: tm.tm_mon++;
3097 case ABMON_5: tm.tm_mon++;
3098 case ABMON_4: tm.tm_mon++;
3099 case ABMON_3: tm.tm_mon++;
3100 case ABMON_2: tm.tm_mon++;
3101 case ABMON_1:
43dd6b15
KW
3102 format = "%b";
3103 break;
3104
4e6826bf
KW
3105 case MON_12: tm.tm_mon++;
3106 case MON_11: tm.tm_mon++;
3107 case MON_10: tm.tm_mon++;
3108 case MON_9: tm.tm_mon++;
3109 case MON_8: tm.tm_mon++;
3110 case MON_7: tm.tm_mon++;
3111 case MON_6: tm.tm_mon++;
3112 case MON_5: tm.tm_mon++;
3113 case MON_4: tm.tm_mon++;
3114 case MON_3: tm.tm_mon++;
3115 case MON_2: tm.tm_mon++;
3116 case MON_1:
43dd6b15
KW
3117 format = "%B";
3118 break;
3119
4e6826bf 3120 case T_FMT_AMPM:
43dd6b15
KW
3121 format = "%r";
3122 return_format = TRUE;
3123 break;
3124
4e6826bf 3125 case ERA_D_FMT:
43dd6b15
KW
3126 format = "%Ex";
3127 return_format = TRUE;
3128 break;
3129
4e6826bf 3130 case ERA_T_FMT:
43dd6b15
KW
3131 format = "%EX";
3132 return_format = TRUE;
3133 break;
3134
4e6826bf 3135 case ERA_D_T_FMT:
43dd6b15
KW
3136 format = "%Ec";
3137 return_format = TRUE;
3138 break;
3139
4e6826bf 3140 case ALT_DIGITS:
43dd6b15
KW
3141 tm.tm_wday = 0;
3142 format = "%Ow"; /* Find the alternate digit for 0 */
3143 break;
3144 }
f7416781 3145
43dd6b15
KW
3146 /* We can't use my_strftime() because it doesn't look at
3147 * tm_wday */
3148 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3149 format, &tm))
3150 {
3151 /* A zero return means one of:
3152 * a) there wasn't enough space in PL_langinfo_buf
3153 * b) the format, like a plain %p, returns empty
3154 * c) it was an illegal format, though some
3155 * implementations of strftime will just return the
3156 * illegal format as a plain character sequence.
3157 *
3158 * To quickly test for case 'b)', try again but precede
3159 * the format with a plain character. If that result is
3160 * still empty, the problem is either 'a)' or 'c)' */
3161
3162 Size_t format_size = strlen(format) + 1;
3163 Size_t mod_size = format_size + 1;
3164 char * mod_format;
3165 char * temp_result;
3166
3167 Newx(mod_format, mod_size, char);
3168 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 3169 *mod_format = ' ';
43dd6b15
KW
3170 my_strlcpy(mod_format + 1, format, mod_size);
3171 len = strftime(temp_result,
3172 PL_langinfo_bufsize,
3173 mod_format, &tm);
3174 Safefree(mod_format);
3175 Safefree(temp_result);
3176
3177 /* If 'len' is non-zero, it means that we had a case like
3178 * %p which means the current locale doesn't use a.m. or
3179 * p.m., and that is valid */
3180 if (len == 0) {
3181
3182 /* Here, still didn't work. If we get well beyond a
3183 * reasonable size, bail out to prevent an infinite
3184 * loop. */
3185
3186 if (PL_langinfo_bufsize > 100 * format_size) {
3187 *PL_langinfo_buf = '\0';
3188 }
3189 else {
3190 /* Double the buffer size to retry; Add 1 in case
3191 * original was 0, so we aren't stuck at 0. */
3192 PL_langinfo_bufsize *= 2;
3193 PL_langinfo_bufsize++;
3194 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3195 continue;
3196 }
3197 }
f7416781 3198
f7416781 3199 break;
43dd6b15 3200 }
f7416781 3201
43dd6b15
KW
3202 /* Here, we got a result.
3203 *
3204 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3205 * alternate format for wday 0. If the value is the same as
3206 * the normal 0, there isn't an alternate, so clear the buffer.
3207 * */
4e6826bf 3208 if ( item == ALT_DIGITS
43dd6b15
KW
3209 && strEQ(PL_langinfo_buf, "0"))
3210 {
3211 *PL_langinfo_buf = '\0';
3212 }
f7416781 3213
43dd6b15
KW
3214 /* ALT_DIGITS is problematic. Experiments on it showed that
3215 * strftime() did not always work properly when going from
3216 * alt-9 to alt-10. Only a few locales have this item defined,
3217 * and in all of them on Linux that khw was able to find,
3218 * nl_langinfo() merely returned the alt-0 character, possibly
3219 * doubled. Most Unicode digits are in blocks of 10
3220 * consecutive code points, so that is sufficient information
3221 * for those scripts, as we can infer alt-1, alt-2, .... But
3222 * for a Japanese locale, a CJK ideographic 0 is returned, and
3223 * the CJK digits are not in code point order, so you can't
3224 * really infer anything. The localedef for this locale did
3225 * specify the succeeding digits, so that strftime() works
3226 * properly on them, without needing to infer anything. But
3227 * the nl_langinfo() return did not give sufficient information
3228 * for the caller to understand what's going on. So until
3229 * there is evidence that it should work differently, this
3230 * returns the alt-0 string for ALT_DIGITS.
3231 *
3232 * wday was chosen because its range is all a single digit.
3233 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3234
43dd6b15 3235 LOCALE_UNLOCK;
f7416781 3236
d1b150d9
KW
3237 retval = PL_langinfo_buf;
3238
43dd6b15
KW
3239 /* If to return the format, not the value, overwrite the buffer
3240 * with it. But some strftime()s will keep the original format
3241 * if illegal, so change those to "" */
3242 if (return_format) {
3243 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3244 *PL_langinfo_buf = '\0';
3245 }
43dd6b15 3246 else {
0b6697c5
KW
3247 retval = save_to_buffer(format, &PL_langinfo_buf,
3248 &PL_langinfo_bufsize, 0);
f7416781
KW
3249 }
3250 }
3251
3252 break;
f7416781
KW
3253
3254# endif
3255
43dd6b15 3256 }
f7416781
KW
3257 }
3258
0b6697c5 3259 return retval;
f7416781
KW
3260
3261#endif
3262
a4f00dcc 3263}
b385bb4d 3264
98994639
HS
3265/*
3266 * Initialize locale awareness.
3267 */
3268int
3269Perl_init_i18nl10n(pTHX_ int printwarn)
3270{
0e92a118
KW
3271 /* printwarn is
3272 *
3273 * 0 if not to output warning when setup locale is bad
3274 * 1 if to output warning based on value of PERL_BADLANG
3275 * >1 if to output regardless of PERL_BADLANG
3276 *
3277 * returns
98994639 3278 * 1 = set ok or not applicable,
0e92a118
KW
3279 * 0 = fallback to a locale of lower priority
3280 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3281 *
3282 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3283 * set, debugging information is output.
3284 *
3285 * This looks more complicated than it is, mainly due to the #ifdefs.
3286 *
3287 * We try to set LC_ALL to the value determined by the environment. If
3288 * there is no LC_ALL on this platform, we try the individual categories we
3289 * know about. If this works, we are done.
3290 *
3291 * But if it doesn't work, we have to do something else. We search the
3292 * environment variables ourselves instead of relying on the system to do
3293 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3294 * think there is one), and the ultimate fallback "C". This is all done in
3295 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3296 * more complex. The 'trial_locales' array is initialized with just one
3297 * element; it causes the behavior described in the paragraph above this to
3298 * happen. If that fails, we add elements to 'trial_locales', and do extra
3299 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3300 *
3301 * On Ultrix, the locale MUST come from the environment, so there is
3302 * preliminary code to set it. I (khw) am not sure that it is necessary,
3303 * and that this couldn't be folded into the loop, but barring any real
3304 * platforms to test on, it's staying as-is
3305 *
3306 * A slight complication is that in embedded Perls, the locale may already
3307 * be set-up, and we don't want to get it from the normal environment
3308 * variables. This is handled by having a special environment variable
3309 * indicate we're in this situation. We simply set setlocale's 2nd
3310 * parameter to be a NULL instead of "". That indicates to setlocale that
3311 * it is not to change anything, but to return the current value,
3312 * effectively initializing perl's db to what the locale already is.
3313 *
3314 * We play the same trick with NULL if a LC_ALL succeeds. We call
3315 * setlocale() on the individual categores with NULL to get their existing
3316 * values for our db, instead of trying to change them.
3317 * */
98994639 3318
1565c085
DM
3319 dVAR;
3320
0e92a118
KW
3321 int ok = 1;
3322
7d4bcc4a
KW
3323#ifndef USE_LOCALE
3324
3325 PERL_UNUSED_ARG(printwarn);
3326
3327#else /* USE_LOCALE */
7d4bcc4a
KW
3328# ifdef __GLIBC__
3329
24f3e849 3330 const char * const language = PerlEnv_getenv("LANGUAGE");
7d4bcc4a
KW
3331
3332# endif
65ebb059 3333
ccd65d51
KW
3334 /* NULL uses the existing already set up locale */
3335 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3336 ? NULL
3337 : "";
c3fcd832
KW
3338 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3339 unsigned int trial_locales_count;
24f3e849
KW
3340 const char * const lc_all = PerlEnv_getenv("LC_ALL");
3341 const char * const lang = PerlEnv_getenv("LANG");
98994639 3342 bool setlocale_failure = FALSE;
65ebb059 3343 unsigned int i;
175c4cf9
KW
3344
3345 /* A later getenv() could zap this, so only use here */
3346 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3347
3348 const bool locwarn = (printwarn > 1
e5f10d49
KW
3349 || ( printwarn
3350 && ( ! bad_lang_use_once
22ff3130 3351 || (
e5f10d49
KW
3352 /* disallow with "" or "0" */
3353 *bad_lang_use_once
3354 && strNE("0", bad_lang_use_once)))));
ea92aad8 3355
291a84fb 3356 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3357 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3358
3359 /* current locale for given category; should have been copied so aren't
3360 * volatile */
8de4332b 3361 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3362
7d4bcc4a
KW
3363# ifdef WIN32
3364
6bce99ee
JH
3365 /* In some systems you can find out the system default locale
3366 * and use that as the fallback locale. */
7d4bcc4a
KW
3367# define SYSTEM_DEFAULT_LOCALE
3368# endif
3369# ifdef SYSTEM_DEFAULT_LOCALE
3370
65ebb059 3371 const char *system_default_locale = NULL;
98994639 3372
7d4bcc4a 3373# endif
948523db
KW
3374
3375# ifndef DEBUGGING
3376# define DEBUG_LOCALE_INIT(a,b,c)
3377# else
7d4bcc4a 3378
8298454c 3379 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3380
3381# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3382 STMT_START { \
3383 if (debug_initialization) { \
3384 PerlIO_printf(Perl_debug_log, \
3385 "%s:%d: %s\n", \
3386 __FILE__, __LINE__, \
a4f00dcc 3387 setlocale_debug_string(category, \
2fcc0ca9
KW
3388 locale, \
3389 result)); \
3390 } \
3391 } STMT_END
2fcc0ca9 3392
948523db
KW
3393/* Make sure the parallel arrays are properly set up */
3394# ifdef USE_LOCALE_NUMERIC
3395 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3396 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3397# ifdef USE_POSIX_2008_LOCALE
3398 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3399# endif
948523db
KW
3400# endif
3401# ifdef USE_LOCALE_CTYPE
3402 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3403 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3404# ifdef USE_POSIX_2008_LOCALE
3405 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3406# endif
948523db
KW
3407# endif
3408# ifdef USE_LOCALE_COLLATE
3409 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3410 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3411# ifdef USE_POSIX_2008_LOCALE
3412 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3413# endif
948523db
KW
3414# endif
3415# ifdef USE_LOCALE_TIME
3416 assert(categories[LC_TIME_INDEX] == LC_TIME);
3417 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3418# ifdef USE_POSIX_2008_LOCALE
3419 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3420# endif
948523db
KW
3421# endif
3422# ifdef USE_LOCALE_MESSAGES
3423 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3424 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3425# ifdef USE_POSIX_2008_LOCALE
3426 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3427# endif
948523db
KW
3428# endif
3429# ifdef USE_LOCALE_MONETARY
3430 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3431 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3432# ifdef USE_POSIX_2008_LOCALE
3433 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3434# endif
948523db 3435# endif
9821811f
KW
3436# ifdef USE_LOCALE_ADDRESS
3437 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3438 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3439# ifdef USE_POSIX_2008_LOCALE
3440 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3441# endif
9821811f
KW
3442# endif
3443# ifdef USE_LOCALE_IDENTIFICATION
3444 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3445 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3446# ifdef USE_POSIX_2008_LOCALE
3447 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3448# endif
9821811f
KW
3449# endif
3450# ifdef USE_LOCALE_MEASUREMENT
3451 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3452 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3453# ifdef USE_POSIX_2008_LOCALE
3454 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3455# endif
9821811f
KW
3456# endif
3457# ifdef USE_LOCALE_PAPER
3458 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3459 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3460# ifdef USE_POSIX_2008_LOCALE
3461 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3462# endif
9821811f
KW
3463# endif
3464# ifdef USE_LOCALE_TELEPHONE
3465 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3466 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3467# ifdef USE_POSIX_2008_LOCALE
3468 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3469# endif
9821811f 3470# endif
36dbc955
KW
3471# ifdef USE_LOCALE_SYNTAX
3472 assert(categories[LC_SYNTAX_INDEX] == LC_SYNTAX);
3473 assert(strEQ(category_names[LC_SYNTAX_INDEX], "LC_SYNTAX"));
3474# ifdef USE_POSIX_2008_LOCALE
3475 assert(category_masks[LC_SYNTAX_INDEX] == LC_SYNTAX_MASK);
3476# endif
3477# endif
3478# ifdef USE_LOCALE_TOD
3479 assert(categories[LC_TOD_INDEX] == LC_TOD);
3480 assert(strEQ(category_names[LC_TOD_INDEX], "LC_TOD"));
3481# ifdef USE_POSIX_2008_LOCALE
3482 assert(category_masks[LC_TOD_INDEX] == LC_TOD_MASK);
3483# endif
3484# endif
948523db
KW
3485# ifdef LC_ALL
3486 assert(categories[LC_ALL_INDEX] == LC_ALL);
3487 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3488 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3489# ifdef USE_POSIX_2008_LOCALE
3490 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3491# endif
948523db
KW
3492# endif
3493# endif /* DEBUGGING */
47280b20 3494
5a6637f0
KW
3495 /* Initialize the per-thread mbrFOO() state variables. See POSIX.xs for
3496 * why these particular incantations are used. */
d2c9cb53
KW
3497#ifdef HAS_MBRLEN
3498 memzero(&PL_mbrlen_ps, sizeof(PL_mbrlen_ps));
3499#endif
5a6637f0
KW
3500#ifdef HAS_MBRTOWC
3501 memzero(&PL_mbrtowc_ps, sizeof(PL_mbrtowc_ps));
3502#endif
3503#ifdef HAS_WCTOMBR
3504 wcrtomb(NULL, L'\0', &PL_wcrtomb_ps);
3505#endif
d2c9cb53 3506
47280b20
KW
3507 /* Initialize the cache of the program's UTF-8ness for the always known
3508 * locales C and POSIX */
3509 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3510 sizeof(PL_locale_utf8ness));
3511
e9bc6d6b
KW
3512# ifdef USE_THREAD_SAFE_LOCALE
3513# ifdef WIN32
3514
3515 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3516
3517# endif
3518# endif
39e69e77 3519# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3520
3521 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3522 if (! PL_C_locale_obj) {
3523 Perl_croak_nocontext(
3524 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3525 }
3526 if (DEBUG_Lv_TEST || debug_initialization) {
3527 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3528 }
3529
3530# endif
3531
78f7c09f
KW
3532# ifdef USE_LOCALE_NUMERIC
3533
3ca88433
KW
3534 PL_numeric_radix_sv = newSVpvs(".");
3535
78f7c09f
KW
3536# endif
3537
e9bc6d6b
KW
3538# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3539
3540 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3541 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3542
3543# endif
b56c4436 3544# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3545
3546 /*
3547 * Ultrix setlocale(..., "") fails if there are no environment
3548 * variables from which to get a locale name.
3549 */
3550
b56c4436
KW
3551# ifndef LC_ALL
3552# error Ultrix without LC_ALL not implemented
3553# else
7d4bcc4a 3554
b56c4436
KW
3555 {
3556 bool done = FALSE;
ea92aad8
KW
3557 if (lang) {
3558 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3559 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3560 if (sl_result[LC_ALL_INDEX])
3561 done = TRUE;
3562 else
e5f10d49 3563 setlocale_failure = TRUE;
ea92aad8
KW
3564 }
3565 if (! setlocale_failure) {
3566 const char * locale_param;
3567 for (i = 0; i < LC_ALL_INDEX; i++) {
3568 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3569 ? setlocale_init
3570 : NULL;
3571 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3572 if (! sl_result[i]) {
3573 setlocale_failure = TRUE;
3574 }
3575 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3576 }
c835d6be 3577 }
7d4bcc4a
KW
3578 }
3579
3580# endif /* LC_ALL */
e5f10d49 3581# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3582
65ebb059 3583 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3584 * the list. Normally the loop is executed just once. But if setting the
3585 * locale fails, inside the loop we add fallback trials to the array and so
3586 * will execute the loop multiple times */
c3fcd832
KW
3587 trial_locales[0] = setlocale_init;
3588 trial_locales_count = 1;
7d4bcc4a 3589
65ebb059
KW
3590 for (i= 0; i < trial_locales_count; i++) {
3591 const char * trial_locale = trial_locales[i];
3592
3593 if (i > 0) {
3594
3595 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3596 * when i==0, but I (khw) don't think that behavior makes much
3597 * sense */
3598 setlocale_failure = FALSE;
3599
7d4bcc4a 3600# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3601# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3602
65ebb059
KW
3603 /* On Windows machines, an entry of "" after the 0th means to use
3604 * the system default locale, which we now proceed to get. */
3605 if (strEQ(trial_locale, "")) {
3606 unsigned int j;
3607
3608 /* Note that this may change the locale, but we are going to do
3609 * that anyway just below */
837ce802 3610 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3611 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3612
7d4bcc4a 3613 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3614 * try */
3615 if (! system_default_locale) {
3616 goto next_iteration;
3617 }
3618 for (j = 0; j < trial_locales_count; j++) {
3619 if (strEQ(system_default_locale, trial_locales[j])) {
3620 goto next_iteration;
3621 }
3622 }
3623
3624 trial_locale = system_default_locale;
3625 }
ec0202b5
KW
3626# else
3627# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3628# endif
7d4bcc4a 3629# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3630
3631 } /* For i > 0 */
65ebb059 3632
7d4bcc4a
KW
3633# ifdef LC_ALL
3634
948523db
KW
3635 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3636 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3637 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3638 setlocale_failure = TRUE;
7cd8b568
KW
3639 }
3640 else {
3641 /* Since LC_ALL succeeded, it should have changed all the other
3642 * categories it can to its value; so we massage things so that the
3643 * setlocales below just return their category's current values.
3644 * This adequately handles the case in NetBSD where LC_COLLATE may
3645 * not be defined for a locale, and setting it individually will
7d4bcc4a 3646 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3647 * the POSIX locale. */
3648 trial_locale = NULL;
3649 }
7d4bcc4a
KW
3650
3651# endif /* LC_ALL */
98994639 3652
e5f10d49
KW
3653 if (! setlocale_failure) {
3654 unsigned int j;
3655 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3656 curlocales[j]
3657 = savepv(do_setlocale_r(categories[j], trial_locale));
3658 if (! curlocales[j]) {
3659 setlocale_failure = TRUE;
3660 }
3661 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3662 }
c835d6be 3663
e5f10d49
KW
3664 if (! setlocale_failure) { /* All succeeded */
3665 break; /* Exit trial_locales loop */
49c85077 3666 }
65ebb059 3667 }
98994639 3668
49c85077
KW
3669 /* Here, something failed; will need to try a fallback. */
3670 ok = 0;
65ebb059 3671
49c85077
KW
3672 if (i == 0) {
3673 unsigned int j;
98994639 3674
65ebb059 3675 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3676
3677# ifdef LC_ALL
98994639 3678
49c85077
KW
3679 PerlIO_printf(Perl_error_log,
3680 "perl: warning: Setting locale failed.\n");
98994639 3681
7d4bcc4a 3682# else /* !LC_ALL */
98994639 3683
49c85077
KW
3684 PerlIO_printf(Perl_error_log,
3685 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3686
e5f10d49
KW
3687 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3688 if (! curlocales[j]) {
3689 PerlIO_printf(Perl_error_log, category_names[j]);
3690 }
3691 else {
3692 Safefree(curlocales[j]);
3693 }
3694 }
7d4bcc4a 3695
7d4bcc4a 3696# endif /* LC_ALL */
98994639 3697
49c85077
KW
3698 PerlIO_printf(Perl_error_log,
3699 "perl: warning: Please check that your locale settings:\n");
98994639 3700
7d4bcc4a
KW
3701# ifdef __GLIBC__
3702
49c85077
KW
3703 PerlIO_printf(Perl_error_log,
3704 "\tLANGUAGE = %c%s%c,\n",
3705 language ? '"' : '(',
3706 language ? language : "unset",
3707 language ? '"' : ')');
7d4bcc4a 3708# endif
98994639 3709
49c85077
KW
3710 PerlIO_printf(Perl_error_log,
3711 "\tLC_ALL = %c%s%c,\n",
3712 lc_all ? '"' : '(',
3713 lc_all ? lc_all : "unset",
3714 lc_all ? '"' : ')');
98994639 3715
7d4bcc4a
KW
3716# if defined(USE_ENVIRON_ARRAY)
3717
49c85077 3718 {
cd999af9 3719 char **e;
d5e32b93
KW
3720
3721 /* Look through the environment for any variables of the
3722 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3723 * already handled above. These are assumed to be locale
3724 * settings. Output them and their values. */
cd999af9 3725 for (e = environ; *e; e++) {
d5e32b93
KW
3726 const STRLEN prefix_len = sizeof("LC_") - 1;
3727 STRLEN uppers_len;
3728
cd999af9 3729 if ( strBEGINs(*e, "LC_")
c8b388b0 3730 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3731 && (uppers_len = strspn(*e + prefix_len,
3732 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3733 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3734 {
3735 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3736 (int) (prefix_len + uppers_len), *e,
3737 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3738 }
3739 }
49c85077 3740 }
7d4bcc4a
KW
3741
3742# else
3743
49c85077
KW
3744 PerlIO_printf(Perl_error_log,
3745 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3746
3747# endif
98994639 3748
49c85077
KW
3749 PerlIO_printf(Perl_error_log,
3750 "\tLANG = %c%s%c\n",
3751 lang ? '"' : '(',
3752 lang ? lang : "unset",
3753 lang ? '"' : ')');
98994639 3754
49c85077
KW
3755 PerlIO_printf(Perl_error_log,
3756 " are supported and installed on your system.\n");
3757 }
98994639 3758
65ebb059 3759 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3760 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3761 * usually change the upper bound of the loop we are in.
3762 *
3763 * Since the system's default way of setting the locale has not
3764 * found one that works, We use Perl's defined ordering: LC_ALL,
3765 * LANG, and the C locale. We don't try the same locale twice, so
3766 * don't add to the list if already there. (On POSIX systems, the
3767 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3768 * but there's no harm done by doing it explicitly.
3769 *
3770 * Note that this tries the LC_ALL environment variable even on
3771 * systems which have no LC_ALL locale setting. This may or may
3772 * not have been originally intentional, but there's no real need
3773 * to change the behavior. */
65ebb059
KW
3774 if (lc_all) {
3775 for (j = 0; j < trial_locales_count; j++) {
3776 if (strEQ(lc_all, trial_locales[j])) {
3777 goto done_lc_all;
3778 }
3779 }
3780 trial_locales[trial_locales_count++] = lc_all;
3781 }
3782 done_lc_all:
98994639 3783
65ebb059
KW
3784 if (lang) {
3785 for (j = 0; j < trial_locales_count; j++) {
3786 if (strEQ(lang, trial_locales[j])) {
3787 goto done_lang;
3788 }
3789 }
3790 trial_locales[trial_locales_count++] = lang;
3791 }
3792 done_lang:
3793
7d4bcc4a
KW
3794# if defined(WIN32) && defined(LC_ALL)
3795
65ebb059
KW
3796 /* For Windows, we also try the system default locale before "C".
3797 * (If there exists a Windows without LC_ALL we skip this because
3798 * it gets too complicated. For those, the "C" is the next
3799 * fallback possibility). The "" is the same as the 0th element of
3800 * the array, but the code at the loop above knows to treat it
3801 * differently when not the 0th */
3802 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3803
3804# endif
65ebb059
KW
3805
3806 for (j = 0; j < trial_locales_count; j++) {
3807 if (strEQ("C", trial_locales[j])) {
3808 goto done_C;
3809 }
3810 }
3811 trial_locales[trial_locales_count++] = "C";
98994639 3812
65ebb059
KW
3813 done_C: ;
3814 } /* end of first time through the loop */
98994639 3815
7d4bcc4a
KW
3816# ifdef WIN32
3817
65ebb059 3818 next_iteration: ;
7d4bcc4a
KW
3819
3820# endif
65ebb059
KW
3821
3822 } /* end of looping through the trial locales */
3823
3824 if (ok < 1) { /* If we tried to fallback */
3825 const char* msg;
3826 if (! setlocale_failure) { /* fallback succeeded */
3827 msg = "Falling back to";
3828 }
3829 else { /* fallback failed */
e5f10d49 3830 unsigned int j;
98994639 3831
65ebb059
KW
3832 /* We dropped off the end of the loop, so have to decrement i to
3833 * get back to the value the last time through */
3834 i--;
98994639 3835
65ebb059
KW
3836 ok = -1;
3837 msg = "Failed to fall back to";
3838
3839 /* To continue, we should use whatever values we've got */
7d4bcc4a 3840
e5f10d49
KW
3841 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3842 Safefree(curlocales[j]);
3843 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3844 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3845 }
65ebb059
KW
3846 }
3847
3848 if (locwarn) {
3849 const char * description;
3850 const char * name = "";
3851 if (strEQ(trial_locales[i], "C")) {
3852 description = "the standard locale";
3853 name = "C";
3854 }
7d4bcc4a
KW
3855
3856# ifdef SYSTEM_DEFAULT_LOCALE
3857
65ebb059
KW
3858 else if (strEQ(trial_locales[i], "")) {
3859 description = "the system default locale";
3860 if (system_default_locale) {
3861 name = system_default_locale;
3862 }
3863 }