This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Reorder some debugging statements
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
e9bc6d6b
KW
35 *
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
166f8a29
DM
44 */
45
98994639
HS
46#include "EXTERN.h"
47#define PERL_IN_LOCALE_C
f7416781 48#include "perl_langinfo.h"
98994639
HS
49#include "perl.h"
50
a4af207c
JH
51#include "reentr.h"
52
0dec74cd
KW
53#ifdef I_WCHAR
54# include <wchar.h>
55#endif
56
2fcc0ca9
KW
57/* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
5a4b0634
KW
60#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61# define debug_initialization 0
62# define DEBUG_INITIALIZATION_set(v)
63#else
2fcc0ca9 64static bool debug_initialization = FALSE;
5a4b0634 65# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
66#endif
67
48015184
KW
68
69/* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
71 * done generally. */
72#define GET_ERRNO saved_errno
73
291a84fb
KW
74/* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
77#define STRLENs(s) (sizeof("" s "") - 1)
78
63e5b0d7
KW
79/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
85 * cntrl: 84-97 9B-9F
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89#define isNAME_C_OR_POSIX(name) \
90 ( (name) != NULL \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
93
8ef6e574
KW
94#ifdef USE_LOCALE
95
47280b20
KW
96/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
98
99#define UTF8NESS_SEP "\v"
100#define UTF8NESS_PREFIX "\f"
101
102/* So, the string looks like:
98994639 103 *
47280b20 104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 105 *
47280b20
KW
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
108
109STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
111
112#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
114
115/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
118
98994639
HS
119STATIC char *
120S_stdize_locale(pTHX_ char *locs)
121{
47280b20
KW
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
124 *
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
130 *
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
134 * */
135
7452cf6a 136 const char * const s = strchr(locs, '=');
98994639
HS
137 bool okay = TRUE;
138
7918f24d
NC
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
140
8772537c
AL
141 if (s) {
142 const char * const t = strchr(s, '.');
98994639 143 okay = FALSE;
8772537c
AL
144 if (t) {
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
149 locs[len] = 0;
150 okay = TRUE;
98994639
HS
151 }
152 }
153 }
154
155 if (!okay)
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
157
158 return locs;
159}
160
e5f10d49
KW
161/* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
163 * order. */
164
165const int categories[] = {
166
167# ifdef USE_LOCALE_NUMERIC
168 LC_NUMERIC,
169# endif
170# ifdef USE_LOCALE_CTYPE
171 LC_CTYPE,
172# endif
173# ifdef USE_LOCALE_COLLATE
174 LC_COLLATE,
175# endif
176# ifdef USE_LOCALE_TIME
177 LC_TIME,
178# endif
179# ifdef USE_LOCALE_MESSAGES
180 LC_MESSAGES,
181# endif
182# ifdef USE_LOCALE_MONETARY
183 LC_MONETARY,
184# endif
9821811f
KW
185# ifdef USE_LOCALE_ADDRESS
186 LC_ADDRESS,
187# endif
188# ifdef USE_LOCALE_IDENTIFICATION
189 LC_IDENTIFICATION,
190# endif
191# ifdef USE_LOCALE_MEASUREMENT
192 LC_MEASUREMENT,
193# endif
194# ifdef USE_LOCALE_PAPER
195 LC_PAPER,
196# endif
197# ifdef USE_LOCALE_TELEPHONE
198 LC_TELEPHONE,
199# endif
e5f10d49
KW
200# ifdef LC_ALL
201 LC_ALL,
202# endif
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
206};
207
208/* The top-most real element is LC_ALL */
209
210const char * category_names[] = {
211
212# ifdef USE_LOCALE_NUMERIC
213 "LC_NUMERIC",
214# endif
215# ifdef USE_LOCALE_CTYPE
216 "LC_CTYPE",
217# endif
218# ifdef USE_LOCALE_COLLATE
219 "LC_COLLATE",
220# endif
221# ifdef USE_LOCALE_TIME
222 "LC_TIME",
223# endif
224# ifdef USE_LOCALE_MESSAGES
225 "LC_MESSAGES",
226# endif
227# ifdef USE_LOCALE_MONETARY
228 "LC_MONETARY",
229# endif
9821811f
KW
230# ifdef USE_LOCALE_ADDRESS
231 "LC_ADDRESS",
232# endif
233# ifdef USE_LOCALE_IDENTIFICATION
234 "LC_IDENTIFICATION",
235# endif
236# ifdef USE_LOCALE_MEASUREMENT
237 "LC_MEASUREMENT",
238# endif
239# ifdef USE_LOCALE_PAPER
240 "LC_PAPER",
241# endif
242# ifdef USE_LOCALE_TELEPHONE
243 "LC_TELEPHONE",
244# endif
e5f10d49
KW
245# ifdef LC_ALL
246 "LC_ALL",
247# endif
248 NULL /* Placeholder */
249 };
250
251# ifdef LC_ALL
252
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
256
257# else
258
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
262
263# endif
264
265/* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
948523db
KW
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
e5f10d49 272
b09aaf40
KW
273STATIC const char *
274S_category_name(const int category)
275{
276 unsigned int i;
277
278#ifdef LC_ALL
279
280 if (category == LC_ALL) {
281 return "LC_ALL";
282 }
283
284#endif
285
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
289 }
290 }
291
292 {
293 const char suffix[] = " (unknown)";
294 int temp = category;
295 Size_t length = sizeof(suffix) + 1;
296 char * unknown;
297 dTHX;
298
299 if (temp < 0) {
300 length++;
301 temp = - temp;
302 }
303
304 /* Calculate the number of digits */
305 while (temp >= 10) {
306 temp /= 10;
307 length++;
308 }
309
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
312 SAVEFREEPV(unknown);
313 return unknown;
314 }
315}
316
948523db
KW
317/* Now create LC_foo_INDEX #defines for just those categories on this system */
318# ifdef USE_LOCALE_NUMERIC
319# define LC_NUMERIC_INDEX 0
320# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
321# else
322# define _DUMMY_NUMERIC -1
323# endif
324# ifdef USE_LOCALE_CTYPE
325# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326# define _DUMMY_CTYPE LC_CTYPE_INDEX
327# else
328# define _DUMMY_CTYPE _DUMMY_NUMERIC
329# endif
330# ifdef USE_LOCALE_COLLATE
331# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332# define _DUMMY_COLLATE LC_COLLATE_INDEX
333# else
334# define _DUMMY_COLLATE _DUMMY_COLLATE
335# endif
336# ifdef USE_LOCALE_TIME
337# define LC_TIME_INDEX _DUMMY_COLLATE + 1
338# define _DUMMY_TIME LC_TIME_INDEX
339# else
340# define _DUMMY_TIME _DUMMY_COLLATE
341# endif
342# ifdef USE_LOCALE_MESSAGES
343# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
345# else
346# define _DUMMY_MESSAGES _DUMMY_TIME
347# endif
348# ifdef USE_LOCALE_MONETARY
349# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350# define _DUMMY_MONETARY LC_MONETARY_INDEX
351# else
352# define _DUMMY_MONETARY _DUMMY_MESSAGES
353# endif
9821811f
KW
354# ifdef USE_LOCALE_ADDRESS
355# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
357# else
358# define _DUMMY_ADDRESS _DUMMY_MONETARY
359# endif
360# ifdef USE_LOCALE_IDENTIFICATION
361# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
363# else
364# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
365# endif
366# ifdef USE_LOCALE_MEASUREMENT
367# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
369# else
370# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
371# endif
372# ifdef USE_LOCALE_PAPER
373# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374# define _DUMMY_PAPER LC_PAPER_INDEX
375# else
376# define _DUMMY_PAPER _DUMMY_MEASUREMENT
377# endif
378# ifdef USE_LOCALE_TELEPHONE
379# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
381# else
382# define _DUMMY_TELEPHONE _DUMMY_PAPER
383# endif
948523db 384# ifdef LC_ALL
9821811f 385# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
386# endif
387#endif /* ifdef USE_LOCALE */
8ef6e574 388
d2b24094 389/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
390#ifdef WIN32
391# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
392#else
393# define my_setlocale(cat, locale) setlocale(cat, locale)
394#endif
395
396#ifndef USE_POSIX_2008_LOCALE
397
398/* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
402
403#else /* Below uses POSIX 2008 */
404
405/* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411# define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
414
415/* A third array, parallel to the ones above to map from category to its
416 * equivalent mask */
417const int category_masks[] = {
418# ifdef USE_LOCALE_NUMERIC
419 LC_NUMERIC_MASK,
420# endif
421# ifdef USE_LOCALE_CTYPE
422 LC_CTYPE_MASK,
423# endif
424# ifdef USE_LOCALE_COLLATE
425 LC_COLLATE_MASK,
426# endif
427# ifdef USE_LOCALE_TIME
428 LC_TIME_MASK,
429# endif
430# ifdef USE_LOCALE_MESSAGES
431 LC_MESSAGES_MASK,
432# endif
433# ifdef USE_LOCALE_MONETARY
434 LC_MONETARY_MASK,
435# endif
436# ifdef USE_LOCALE_ADDRESS
437 LC_ADDRESS_MASK,
438# endif
439# ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
441# endif
442# ifdef USE_LOCALE_MEASUREMENT
443 LC_MEASUREMENT_MASK,
444# endif
445# ifdef USE_LOCALE_PAPER
446 LC_PAPER_MASK,
447# endif
448# ifdef USE_LOCALE_TELEPHONE
449 LC_TELEPHONE_MASK,
450# endif
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
455 * time */
456 LC_ALL_MASK
457 };
458
459STATIC const char *
460S_emulate_setlocale(const int category,
461 const char * locale,
462 unsigned int index,
463 const bool is_index_valid
464 )
465{
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
473 *
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
476 *
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
486 *
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
501 */
502
503 int mask;
504 locale_t old_obj;
505 locale_t new_obj;
506 dTHX;
507
508# ifdef DEBUGGING
509
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
512 }
513
514# endif
515
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
518 unsigned int i;
519
520# ifdef DEBUGGING
521
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
524 }
525
526# endif
527
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
530 index = i;
531 goto found_index;
532 }
533 }
534
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
539 category, locale);
540 return NULL;
541
542 found_index: ;
543
544# ifdef DEBUGGING
545
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
548 }
549
550# endif
551
552 }
553
554 mask = category_masks[index];
555
556# ifdef DEBUGGING
557
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
560 }
561
562# endif
563
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
567
568# ifdef DEBUGGING
569
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
572 }
573
574# endif
575
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
578 }
579
580# ifdef HAS_QUERYLOCALE
581
582 return (char *) querylocale(mask, cur_obj);
583
d2b24094 584# else
ecdda939
KW
585
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
588
e9bc6d6b
KW
589# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
590
591 {
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
595
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
598 /*
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
602 */
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
607
608# ifdef USE_C_BACKTRACE
609
610 dump_c_backtrace(Perl_debug_log, 20, 1);
611
612# endif
613
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
618 }
619
620 return temp_name;
621 }
622 }
623
624# endif
625
626 /* Without querylocale(), we have to use our record-keeping we've
627 * done. */
628
629 if (category != LC_ALL) {
630
631# ifdef DEBUGGING
632
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 }
636
637# endif
638
639 return PL_curlocales[index];
640 }
641 else { /* For LC_ALL */
642 unsigned int i;
643 Size_t names_len = 0;
644 char * all_string;
7c93a471 645 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
646
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
649
650# ifdef DEBUGGING
651
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
654 }
655
656# endif
657
658 return PL_curlocales[LC_ALL_INDEX];
659 }
660
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
e9bc6d6b
KW
666 for (i = 0; i < LC_ALL_INDEX; i++) {
667
668# ifdef DEBUGGING
669
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
672 }
673
674# endif
675
676 names_len += strlen(category_names[i])
677 + 1 /* '=' */
678 + strlen(PL_curlocales[i])
679 + 1; /* ';' */
7c93a471
KW
680
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
683 }
684 }
685
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 692 }
7c93a471 693
e9bc6d6b
KW
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
696 *all_string = '\0';
697
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
700
701# ifdef DEBUGGING
702
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
705 }
706
707# endif
708
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
713 }
714
715# ifdef DEBUGGING
716
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
719 }
720
721 #endif
722
723 return all_string;
724 }
725
726# ifdef EINVAL
727
728 SETERRNO(EINVAL, LIB_INVARG);
729
730# endif
731
732 return NULL;
733
d2b24094
KW
734# endif
735
e9bc6d6b
KW
736 }
737
f8115d60 738 /* Here, we are switching locales. */
e9bc6d6b
KW
739
740# ifndef HAS_QUERYLOCALE
741
742 if (strEQ(locale, "")) {
743
744 /* For non-querylocale() systems, we do the setting of "" ourselves to
745 * be sure that we really know what's going on. We follow the Linux
746 * documented behavior (but if that differs from the actual behavior,
747 * this won't work exactly as the OS implements). We go out and
748 * examine the environment based on our understanding of how the system
749 * works, and use that to figure things out */
750
751 const char * const lc_all = PerlEnv_getenv("LC_ALL");
752
753 /* Use any "LC_ALL" environment variable, as it overrides everything
754 * else. */
755 if (lc_all && strNE(lc_all, "")) {
756 locale = lc_all;
757 }
758 else {
759
760 /* Otherwise, we need to dig deeper. Unless overridden, the
761 * default is the LANG environment variable; if it doesn't exist,
762 * then "C" */
763
764 const char * default_name;
765
766 /* To minimize other threads messing with the environment, we copy
767 * the variable, making it a temporary. But this doesn't work upon
768 * program initialization before any scopes are created, and at
769 * this time, there's nothing else going on that would interfere.
770 * So skip the copy in that case */
771 if (PL_scopestack_ix == 0) {
772 default_name = PerlEnv_getenv("LANG");
773 }
774 else {
775 default_name = savepv(PerlEnv_getenv("LANG"));
776 }
777
778 if (! default_name || strEQ(default_name, "")) {
929a7f8c 779 default_name = "C";
e9bc6d6b
KW
780 }
781 else if (PL_scopestack_ix != 0) {
782 SAVEFREEPV(default_name);
783 }
784
785 if (category != LC_ALL) {
786 const char * const name = PerlEnv_getenv(category_names[index]);
787
788 /* Here we are setting a single category. Assume will have the
789 * default name */
790 locale = default_name;
791
792 /* But then look for an overriding environment variable */
793 if (name && strNE(name, "")) {
794 locale = name;
795 }
796 }
797 else {
798 bool did_override = FALSE;
799 unsigned int i;
800
801 /* Here, we are getting LC_ALL. Any categories that don't have
802 * a corresponding environment variable set should be set to
803 * LANG, or to "C" if there is no LANG. If no individual
804 * categories differ from this, we can just set LC_ALL. This
805 * is buggy on systems that have extra categories that we don't
806 * know about. If there is an environment variable that sets
807 * that category, we won't know to look for it, and so our use
808 * of LANG or "C" improperly overrides it. On the other hand,
809 * if we don't do what is done here, and there is no
810 * environment variable, the category's locale should be set to
811 * LANG or "C". So there is no good solution. khw thinks the
812 * best is to look at systems to see what categories they have,
813 * and include them, and then to assume that we know the
814 * complete set */
815
816 for (i = 0; i < LC_ALL_INDEX; i++) {
817 const char * const env_override
818 = savepv(PerlEnv_getenv(category_names[i]));
819 const char * this_locale = ( env_override
820 && strNE(env_override, ""))
821 ? env_override
822 : default_name;
6435f98d
KW
823 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
824 {
825 Safefree(env_override);
826 return NULL;
827 }
e9bc6d6b
KW
828
829 if (strNE(this_locale, default_name)) {
830 did_override = TRUE;
831 }
832
833 Safefree(env_override);
834 }
835
836 /* If all the categories are the same, we can set LC_ALL to
837 * that */
838 if (! did_override) {
839 locale = default_name;
840 }
841 else {
842
843 /* Here, LC_ALL is no longer valid, as some individual
844 * categories don't match it. We call ourselves
845 * recursively, as that will execute the code that
846 * generates the proper locale string for this situation.
847 * We don't do the remainder of this function, as that is
848 * to update our records, and we've just done that for the
849 * individual categories in the loop above, and doing so
850 * would cause LC_ALL to be done as well */
851 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
852 }
853 }
854 }
855 }
856 else if (strchr(locale, ';')) {
857
858 /* LC_ALL may actually incude a conglomeration of various categories.
859 * Without querylocale, this code uses the glibc (as of this writing)
860 * syntax for representing that, but that is not a stable API, and
861 * other platforms do it differently, so we have to handle all cases
862 * ourselves */
863
f0023d47 864 unsigned int i;
e9bc6d6b
KW
865 const char * s = locale;
866 const char * e = locale + strlen(locale);
867 const char * p = s;
868 const char * category_end;
869 const char * name_start;
870 const char * name_end;
871
f0023d47
KW
872 /* If the string that gives what to set doesn't include all categories,
873 * the omitted ones get set to "C". To get this behavior, first set
874 * all the individual categories to "C", and override the furnished
875 * ones below */
876 for (i = 0; i < LC_ALL_INDEX; i++) {
877 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
878 return NULL;
879 }
880 }
881
e9bc6d6b 882 while (s < e) {
e9bc6d6b
KW
883
884 /* Parse through the category */
885 while (isWORDCHAR(*p)) {
886 p++;
887 }
888 category_end = p;
889
890 if (*p++ != '=') {
891 Perl_croak(aTHX_
892 "panic: %s: %d: Unexpected character in locale name '%02X",
893 __FILE__, __LINE__, *(p-1));
894 }
895
896 /* Parse through the locale name */
897 name_start = p;
bee74f4b
KW
898 while (p < e && *p != ';') {
899 if (! isGRAPH(*p)) {
900 Perl_croak(aTHX_
901 "panic: %s: %d: Unexpected character in locale name '%02X",
902 __FILE__, __LINE__, *(p-1));
903 }
e9bc6d6b
KW
904 p++;
905 }
906 name_end = p;
907
bee74f4b
KW
908 /* Space past the semi-colon */
909 if (p < e) {
910 p++;
e9bc6d6b
KW
911 }
912
913 /* Find the index of the category name in our lists */
914 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 915 char * individ_locale;
e9bc6d6b
KW
916
917 /* Keep going if this isn't the index. The strnNE() avoids a
918 * Perl_form(), but would fail if ever a category name could be
919 * a substring of another one, like if there were a
920 * "LC_TIME_DATE" */
921 if strnNE(s, category_names[i], category_end - s) {
922 continue;
923 }
924
925 /* If this index is for the single category we're changing, we
926 * have found the locale to set it to. */
927 if (category == categories[i]) {
928 locale = Perl_form(aTHX_ "%.*s",
929 (int) (name_end - name_start),
930 name_start);
931 goto ready_to_set;
932 }
933
3c76a412 934 assert(category == LC_ALL);
bee74f4b
KW
935 individ_locale = Perl_form(aTHX_ "%.*s",
936 (int) (name_end - name_start), name_start);
6435f98d
KW
937 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
938 {
939 return NULL;
940 }
e9bc6d6b
KW
941 }
942
943 s = p;
944 }
945
946 /* Here we have set all the individual categories by recursive calls.
947 * These collectively should have fixed up LC_ALL, so can just query
948 * what that now is */
949 assert(category == LC_ALL);
950
951 return do_setlocale_c(LC_ALL, NULL);
952 }
953
954 ready_to_set: ;
955
f8115d60
KW
956 /* Here at the end of having to deal with the absence of querylocale().
957 * Some cases have already been fully handled by recursive calls to this
958 * function. But at this point, we haven't dealt with those, but are now
959 * prepared to, knowing what the locale name to set this category to is.
960 * This would have come for free if this system had had querylocale() */
961
e9bc6d6b
KW
962# endif /* end of ! querylocale */
963
f8115d60
KW
964 assert(PL_C_locale_obj);
965
966 /* Switching locales generally entails freeing the current one's space (at
967 * the C library's discretion). We need to stop using that locale before
968 * the switch. So switch to a known locale object that we don't otherwise
969 * mess with. This returns the locale object in effect at the time of the
970 * switch. */
971 old_obj = uselocale(PL_C_locale_obj);
972
973# ifdef DEBUGGING
974
975 if (DEBUG_Lv_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
977 }
978
979# endif
980
981 if (! old_obj) {
982
983# ifdef DEBUGGING
984
985 if (DEBUG_L_TEST || debug_initialization) {
986 dSAVE_ERRNO;
987 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
988 RESTORE_ERRNO;
989 }
990
991# endif
992
993 return NULL;
994 }
995
996# ifdef DEBUGGING
997
998 if (DEBUG_Lv_TEST || debug_initialization) {
999 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1000 }
1001
1002# endif
1003
1004 /* If we weren't in a thread safe locale, set so that newlocale() below
1005 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1006 The latter is defensive coding, so that, even if there is some bug, we
1007 will never end up trying to modify either of these, as if passed to
1008 newlocale(), they can be. */
1009 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1010 old_obj = (locale_t) 0;
1011 }
1012
e9bc6d6b
KW
1013 /* Ready to create a new locale by modification of the exising one */
1014 new_obj = newlocale(mask, locale, old_obj);
1015
1016 if (! new_obj) {
1017 dSAVE_ERRNO;
1018
1019# ifdef DEBUGGING
1020
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1023 }
1024
1025# endif
1026
1027 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1028
1029# ifdef DEBUGGING
1030
1031 if (DEBUG_L_TEST || debug_initialization) {
1032 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1033 }
1034
1035# endif
1036
1037 }
1038 RESTORE_ERRNO;
1039 return NULL;
1040 }
1041
1042# ifdef DEBUGGING
1043
1044 if (DEBUG_Lv_TEST || debug_initialization) {
d15b6e8b 1045 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p; should have freed %p\n", __FILE__, __LINE__, new_obj, old_obj);
e9bc6d6b
KW
1046 }
1047
1048# endif
1049
1050 /* And switch into it */
1051 if (! uselocale(new_obj)) {
1052 dSAVE_ERRNO;
1053
1054# ifdef DEBUGGING
1055
1056 if (DEBUG_L_TEST || debug_initialization) {
1057 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1058 }
1059
1060# endif
1061
1062 if (! uselocale(old_obj)) {
1063
1064# ifdef DEBUGGING
1065
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1068 }
1069
1070# endif
1071
1072 }
1073 freelocale(new_obj);
1074 RESTORE_ERRNO;
1075 return NULL;
1076 }
1077
1078# ifdef DEBUGGING
1079
1080 if (DEBUG_Lv_TEST || debug_initialization) {
1081 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1082 }
1083
1084# endif
1085
1086 /* We are done, except for updating our records (if the system doesn't keep
1087 * them) and in the case of locale "", we don't actually know what the
1088 * locale that got switched to is, as it came from the environment. So
1089 * have to find it */
1090
1091# ifdef HAS_QUERYLOCALE
1092
1093 if (strEQ(locale, "")) {
1094 locale = querylocale(mask, new_obj);
1095 }
1096
1097# else
1098
1099 /* Here, 'locale' is the return value */
1100
1101 /* Without querylocale(), we have to update our records */
1102
1103 if (category == LC_ALL) {
1104 unsigned int i;
1105
1106 /* For LC_ALL, we change all individual categories to correspond */
1107 /* PL_curlocales is a parallel array, so has same
1108 * length as 'categories' */
1109 for (i = 0; i <= LC_ALL_INDEX; i++) {
1110 Safefree(PL_curlocales[i]);
1111 PL_curlocales[i] = savepv(locale);
1112 }
1113 }
1114 else {
1115
1116 /* For a single category, if it's not the same as the one in LC_ALL, we
1117 * nullify LC_ALL */
1118
1119 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1120 Safefree(PL_curlocales[LC_ALL_INDEX]);
1121 PL_curlocales[LC_ALL_INDEX] = NULL;
1122 }
1123
1124 /* Then update the category's record */
1125 Safefree(PL_curlocales[index]);
1126 PL_curlocales[index] = savepv(locale);
1127 }
1128
1129# endif
1130
1131 return locale;
1132}
1133
1134#endif /* USE_POSIX_2008_LOCALE */
1135
1136#if 0 /* Code that was to emulate thread-safe locales on platforms that
1137 didn't natively support them */
1138
1139/* The way this would work is that we would keep a per-thread list of the
1140 * correct locale for that thread. Any operation that was locale-sensitive
1141 * would have to be changed so that it would look like this:
1142 *
1143 * LOCALE_LOCK;
1144 * setlocale to the correct locale for this operation
1145 * do operation
1146 * LOCALE_UNLOCK
1147 *
1148 * This leaves the global locale in the most recently used operation's, but it
1149 * was locked long enough to get the result. If that result is static, it
1150 * needs to be copied before the unlock.
1151 *
1152 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1153 * the setup, but are no-ops when not needed, and similarly,
1154 * END_LOCALE_DEPENDENT_OP for the tear-down
1155 *
1156 * But every call to a locale-sensitive function would have to be changed, and
1157 * if a module didn't cooperate by using the mutex, things would break.
1158 *
1159 * This code was abandoned before being completed or tested, and is left as-is
1160*/
1161
1162# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1163# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1164
1165STATIC char *
1166S_locking_setlocale(pTHX_
1167 const int category,
1168 const char * locale,
1169 int index,
1170 const bool is_index_valid
1171 )
1172{
1173 /* This function kind of performs a setlocale() on just the current thread;
1174 * thus it is kind of thread-safe. It does this by keeping a thread-level
1175 * array of the current locales for each category. Every time a locale is
1176 * switched to, it does the switch globally, but updates the thread's
1177 * array. A query as to what the current locale is just returns the
1178 * appropriate element from the array, and doesn't actually call the system
1179 * setlocale(). The saving into the array is done in an uninterruptible
1180 * section of code, so is unaffected by whatever any other threads might be
1181 * doing.
1182 *
1183 * All locale-sensitive operations must work by first starting a critical
1184 * section, then switching to the thread's locale as kept by this function,
1185 * and then doing the operation, then ending the critical section. Thus,
1186 * each gets done in the appropriate locale. simulating thread-safety.
1187 *
1188 * This function takes the same parameters, 'category' and 'locale', that
1189 * the regular setlocale() function does, but it also takes two additional
1190 * ones. This is because as described earlier. If we know on input the
1191 * index corresponding to the category into the array where we store the
1192 * current locales, we don't have to calculate it. If the caller knows at
1193 * compile time what the index is, it it can pass it, setting
1194 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1195 *
1196 */
1197
1198 /* If the input index might be incorrect, calculate the correct one */
1199 if (! is_index_valid) {
1200 unsigned int i;
1201
1202 if (DEBUG_Lv_TEST || debug_initialization) {
1203 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1204 }
1205
1206 for (i = 0; i <= LC_ALL_INDEX; i++) {
1207 if (category == categories[i]) {
1208 index = i;
1209 goto found_index;
1210 }
1211 }
1212
1213 /* Here, we don't know about this category, so can't handle it.
1214 * XXX best we can do is to unsafely set this
1215 * XXX warning */
1216
1217 return my_setlocale(category, locale);
1218
1219 found_index: ;
1220
1221 if (DEBUG_Lv_TEST || debug_initialization) {
1222 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1223 }
1224 }
1225
1226 /* For a query, just return what's in our records */
1227 if (new_locale == NULL) {
1228 return curlocales[index];
1229 }
1230
1231
1232 /* Otherwise, we need to do the switch, and save the result, all in a
1233 * critical section */
1234
1235 Safefree(curlocales[[index]]);
1236
1237 /* It might be that this is called from an already-locked section of code.
1238 * We would have to detect and skip the LOCK/UNLOCK if so */
1239 LOCALE_LOCK;
1240
1241 curlocales[index] = savepv(my_setlocale(category, new_locale));
1242
1243 if (strEQ(new_locale, "")) {
1244
1245#ifdef LC_ALL
1246
1247 /* The locale values come from the environment, and may not all be the
1248 * same, so for LC_ALL, we have to update all the others, while the
1249 * mutex is still locked */
1250
1251 if (category == LC_ALL) {
1252 unsigned int i;
1253 for (i = 0; i < LC_ALL_INDEX) {
1254 curlocales[i] = my_setlocale(categories[i], NULL);
1255 }
1256 }
1257 }
1258
1259#endif
1260
1261 LOCALE_UNLOCK;
1262
1263 return curlocales[index];
1264}
1265
1266#endif
837ce802 1267
a4f00dcc 1268STATIC void
86799d2d 1269S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1270{
86799d2d
KW
1271 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1272 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1273
86799d2d
KW
1274#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1275 || defined(HAS_NL_LANGINFO))
98994639 1276
87f8e8e7 1277 const char * radix = (use_locale)
4e6826bf 1278 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1279 /* FALSE => already in dest locale */
1280 : ".";
2213a3be 1281
87f8e8e7 1282 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1283
87f8e8e7
KW
1284 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1285 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1286 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1287 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1288 && _is_cur_LC_category_utf8(LC_NUMERIC))
1289 {
1290 SvUTF8_on(PL_numeric_radix_sv);
1291 }
86799d2d
KW
1292
1293# ifdef DEBUGGING
7d4bcc4a 1294
2fcc0ca9
KW
1295 if (DEBUG_L_TEST || debug_initialization) {
1296 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1297 SvPVX(PL_numeric_radix_sv),
1298 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1299 }
69014004 1300
86799d2d
KW
1301# endif
1302#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1303
98994639
HS
1304}
1305
398eeea9
KW
1306STATIC void
1307S_new_numeric(pTHX_ const char *newnum)
98994639 1308{
7d4bcc4a
KW
1309
1310#ifndef USE_LOCALE_NUMERIC
1311
1312 PERL_UNUSED_ARG(newnum);
1313
1314#else
0d071d52 1315
291a84fb 1316 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1317 * core Perl this and that 'newnum' is the name of the new locale.
1318 * It installs this locale as the current underlying default.
1319 *
1320 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1321 * set_numeric_underlying() and set_numeric_standard() functions, which
1322 * should probably not be called directly, but only via macros like
0d071d52
KW
1323 * SET_NUMERIC_STANDARD() in perl.h.
1324 *
1325 * The toggling is necessary mainly so that a non-dot radix decimal point
1326 * character can be output, while allowing internal calculations to use a
1327 * dot.
1328 *
1329 * This sets several interpreter-level variables:
bb304765 1330 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1331 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1332 * that the current locale is the program's underlying
1333 * locale
1334 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1335 * that the current locale is the C locale or
1336 * indistinguishable from the C locale. If non-zero, it
1337 * is in C; if > 1, it means it may not be toggled away
7738054c 1338 * from C.
4c68b815
KW
1339 * PL_numeric_underlying_is_standard A bool kept by this function
1340 * indicating that the underlying locale and the standard
1341 * C locale are indistinguishable for the purposes of
1342 * LC_NUMERIC. This happens when both of the above two
1343 * variables are true at the same time. (Toggling is a
1344 * no-op under these circumstances.) This variable is
1345 * used to avoid having to recalculate.
398eeea9 1346 */
0d071d52 1347
b03f34cf 1348 char *save_newnum;
98994639
HS
1349
1350 if (! newnum) {
43c5f42d
NC
1351 Safefree(PL_numeric_name);
1352 PL_numeric_name = NULL;
98994639 1353 PL_numeric_standard = TRUE;
892e6465 1354 PL_numeric_underlying = TRUE;
4c68b815 1355 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1356 return;
1357 }
1358
b03f34cf 1359 save_newnum = stdize_locale(savepv(newnum));
892e6465 1360 PL_numeric_underlying = TRUE;
4c68b815
KW
1361 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1362
69c5e0db
KW
1363#ifndef TS_W32_BROKEN_LOCALECONV
1364
4c68b815 1365 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1366 * them. But on broken Windows systems calling my_nl_langinfo() for
1367 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1368 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1369 if (! PL_numeric_standard) {
4e6826bf 1370 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1371 FALSE /* Don't toggle locale */ ))
4e6826bf 1372 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1373 }
abe1abcf 1374
69c5e0db
KW
1375#endif
1376
4c68b815 1377 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1378 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1379 Safefree(PL_numeric_name);
b03f34cf 1380 PL_numeric_name = save_newnum;
b03f34cf 1381 }
abe1abcf
KW
1382 else {
1383 Safefree(save_newnum);
1384 }
4c28b29c 1385
4c68b815
KW
1386 PL_numeric_underlying_is_standard = PL_numeric_standard;
1387
7e5377f7 1388# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1389
1390 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1391 PL_numeric_name,
1392 PL_underlying_numeric_obj);
1393
1394#endif
1395
4c68b815
KW
1396 if (DEBUG_L_TEST || debug_initialization) {
1397 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1398 }
1399
4c28b29c
KW
1400 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1401 * have to worry about the radix being a non-dot. (Core operations that
1402 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1403 if (PL_numeric_standard) {
1404 set_numeric_radix(0);
1405 }
1406 else {
1407 set_numeric_standard();
1408 }
4c28b29c 1409
98994639 1410#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1411
98994639
HS
1412}
1413
1414void
1415Perl_set_numeric_standard(pTHX)
1416{
7d4bcc4a 1417
98994639 1418#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1419
28c1bf33
KW
1420 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1421 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1422 * macro avoids calling this routine if toggling isn't necessary according
1423 * to our records (which could be wrong if some XS code has changed the
1424 * locale behind our back) */
0d071d52 1425
7d4bcc4a
KW
1426# ifdef DEBUGGING
1427
2fcc0ca9
KW
1428 if (DEBUG_L_TEST || debug_initialization) {
1429 PerlIO_printf(Perl_debug_log,
7bb8f702 1430 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1431 }
98994639 1432
7d4bcc4a 1433# endif
7bb8f702
KW
1434
1435 do_setlocale_c(LC_NUMERIC, "C");
1436 PL_numeric_standard = TRUE;
1437 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1438 set_numeric_radix(0);
1439
98994639 1440#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1441
98994639
HS
1442}
1443
1444void
5792c642 1445Perl_set_numeric_underlying(pTHX)
98994639 1446{
7d4bcc4a 1447
98994639 1448#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1449
28c1bf33 1450 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1451 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1452 * instead of calling this directly. The macro avoids calling this routine
1453 * if toggling isn't necessary according to our records (which could be
1454 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1455
7d4bcc4a
KW
1456# ifdef DEBUGGING
1457
2fcc0ca9
KW
1458 if (DEBUG_L_TEST || debug_initialization) {
1459 PerlIO_printf(Perl_debug_log,
7bb8f702 1460 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1461 PL_numeric_name);
1462 }
98994639 1463
7d4bcc4a 1464# endif
7bb8f702
KW
1465
1466 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1467 PL_numeric_standard = PL_numeric_underlying_is_standard;
1468 PL_numeric_underlying = TRUE;
1469 set_numeric_radix(! PL_numeric_standard);
1470
98994639 1471#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1472
98994639
HS
1473}
1474
1475/*
1476 * Set up for a new ctype locale.
1477 */
a4f00dcc
KW
1478STATIC void
1479S_new_ctype(pTHX_ const char *newctype)
98994639 1480{
7d4bcc4a
KW
1481
1482#ifndef USE_LOCALE_CTYPE
1483
1484 PERL_ARGS_ASSERT_NEW_CTYPE;
1485 PERL_UNUSED_ARG(newctype);
1486 PERL_UNUSED_CONTEXT;
1487
1488#else
0d071d52 1489
291a84fb 1490 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1491 * core Perl this and that 'newctype' is the name of the new locale.
1492 *
1493 * This function sets up the folding arrays for all 256 bytes, assuming
1494 * that tofold() is tolc() since fold case is not a concept in POSIX,
1495 *
1496 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1497 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1498 * this function should be called directly only from this file and from
0d071d52
KW
1499 * POSIX::setlocale() */
1500
27da23d5 1501 dVAR;
013f4e03 1502 unsigned int i;
98994639 1503
8b7358b9
KW
1504 /* Don't check for problems if we are suppressing the warnings */
1505 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1506
7918f24d
NC
1507 PERL_ARGS_ASSERT_NEW_CTYPE;
1508
215c5139
KW
1509 /* We will replace any bad locale warning with 1) nothing if the new one is
1510 * ok; or 2) a new warning for the bad new locale */
1511 if (PL_warn_locale) {
1512 SvREFCNT_dec_NN(PL_warn_locale);
1513 PL_warn_locale = NULL;
1514 }
1515
c1284011 1516 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1517
1518 /* A UTF-8 locale gets standard rules. But note that code still has to
1519 * handle this specially because of the three problematic code points */
1520 if (PL_in_utf8_CTYPE_locale) {
1521 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1522 }
8b7358b9
KW
1523
1524 /* We don't populate the other lists if a UTF-8 locale, but do check that
1525 * everything works as expected, unless checking turned off */
1526 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1527 /* Assume enough space for every character being bad. 4 spaces each
1528 * for the 94 printable characters that are output like "'x' "; and 5
1529 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1530 * NUL */
8b7358b9 1531 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1532 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1533 to start */
1534 unsigned int bad_count = 0; /* Count of bad characters */
1535
baa60164 1536 for (i = 0; i < 256; i++) {
8b7358b9 1537 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1538 if (isupper(i))
1539 PL_fold_locale[i] = (U8) tolower(i);
1540 else if (islower(i))
1541 PL_fold_locale[i] = (U8) toupper(i);
1542 else
1543 PL_fold_locale[i] = (U8) i;
8b7358b9 1544 }
8c6180a9
KW
1545
1546 /* If checking for locale problems, see if the native ASCII-range
1547 * printables plus \n and \t are in their expected categories in
1548 * the new locale. If not, this could mean big trouble, upending
1549 * Perl's and most programs' assumptions, like having a
1550 * metacharacter with special meaning become a \w. Fortunately,
1551 * it's very rare to find locales that aren't supersets of ASCII
1552 * nowadays. It isn't a problem for most controls to be changed
1553 * into something else; we check only \n and \t, though perhaps \r
1554 * could be an issue as well. */
7d4bcc4a 1555 if ( check_for_problems
8c6180a9
KW
1556 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1557 {
8b7358b9 1558 bool is_bad = FALSE;
6726b4f4 1559 char name[4] = { '\0' };
8b7358b9
KW
1560
1561 /* Convert the name into a string */
6726b4f4 1562 if (isGRAPH_A(i)) {
8b7358b9
KW
1563 name[0] = i;
1564 name[1] = '\0';
1565 }
1566 else if (i == '\n') {
6726b4f4
KW
1567 my_strlcpy(name, "\\n", sizeof(name));
1568 }
1569 else if (i == '\t') {
1570 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1571 }
1572 else {
6726b4f4
KW
1573 assert(i == ' ');
1574 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1575 }
1576
1577 /* Check each possibe class */
1578 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1579 is_bad = TRUE;
1580 DEBUG_L(PerlIO_printf(Perl_debug_log,
1581 "isalnum('%s') unexpectedly is %d\n",
1582 name, cBOOL(isalnum(i))));
1583 }
1584 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1585 is_bad = TRUE;
1586 DEBUG_L(PerlIO_printf(Perl_debug_log,
1587 "isalpha('%s') unexpectedly is %d\n",
1588 name, cBOOL(isalpha(i))));
1589 }
1590 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1591 is_bad = TRUE;
1592 DEBUG_L(PerlIO_printf(Perl_debug_log,
1593 "isdigit('%s') unexpectedly is %d\n",
1594 name, cBOOL(isdigit(i))));
1595 }
1596 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1597 is_bad = TRUE;
1598 DEBUG_L(PerlIO_printf(Perl_debug_log,
1599 "isgraph('%s') unexpectedly is %d\n",
1600 name, cBOOL(isgraph(i))));
1601 }
1602 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1603 is_bad = TRUE;
1604 DEBUG_L(PerlIO_printf(Perl_debug_log,
1605 "islower('%s') unexpectedly is %d\n",
1606 name, cBOOL(islower(i))));
1607 }
1608 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1609 is_bad = TRUE;
1610 DEBUG_L(PerlIO_printf(Perl_debug_log,
1611 "isprint('%s') unexpectedly is %d\n",
1612 name, cBOOL(isprint(i))));
1613 }
1614 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1615 is_bad = TRUE;
1616 DEBUG_L(PerlIO_printf(Perl_debug_log,
1617 "ispunct('%s') unexpectedly is %d\n",
1618 name, cBOOL(ispunct(i))));
1619 }
1620 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1621 is_bad = TRUE;
1622 DEBUG_L(PerlIO_printf(Perl_debug_log,
1623 "isspace('%s') unexpectedly is %d\n",
1624 name, cBOOL(isspace(i))));
1625 }
1626 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1627 is_bad = TRUE;
1628 DEBUG_L(PerlIO_printf(Perl_debug_log,
1629 "isupper('%s') unexpectedly is %d\n",
1630 name, cBOOL(isupper(i))));
1631 }
1632 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1633 is_bad = TRUE;
1634 DEBUG_L(PerlIO_printf(Perl_debug_log,
1635 "isxdigit('%s') unexpectedly is %d\n",
1636 name, cBOOL(isxdigit(i))));
1637 }
1638 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1639 is_bad = TRUE;
1640 DEBUG_L(PerlIO_printf(Perl_debug_log,
1641 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1642 name, tolower(i), (int) toLOWER_A(i)));
1643 }
1644 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1645 is_bad = TRUE;
1646 DEBUG_L(PerlIO_printf(Perl_debug_log,
1647 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1648 name, toupper(i), (int) toUPPER_A(i)));
1649 }
1650 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1651 is_bad = TRUE;
1652 DEBUG_L(PerlIO_printf(Perl_debug_log,
1653 "'\\n' (=%02X) is not a control\n", (int) i));
1654 }
1655
1656 /* Add to the list; Separate multiple entries with a blank */
1657 if (is_bad) {
1658 if (bad_count) {
1659 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1660 }
8b7358b9
KW
1661 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1662 bad_count++;
8c6180a9
KW
1663 }
1664 }
1665 }
1666
7d4bcc4a
KW
1667# ifdef MB_CUR_MAX
1668
8c6180a9 1669 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1670 * this locale requires more than one byte, there are going to be
8c6180a9 1671 * problems. */
9c8a6dc2
KW
1672 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1673 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1674 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1675
8b7358b9
KW
1676 if ( check_for_problems && MB_CUR_MAX > 1
1677 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1678
1679 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1680 * locale. Just assume that the implementation for them (plus
1681 * for POSIX) is correct and the > 1 value is spurious. (Since
1682 * these are specially handled to never be considered UTF-8
1683 * locales, as long as this is the only problem, everything
1684 * should work fine */
1685 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1686 {
8c6180a9
KW
1687 multi_byte_locale = TRUE;
1688 }
7d4bcc4a
KW
1689
1690# endif
8c6180a9 1691
8b7358b9
KW
1692 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1693 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1694 PL_warn_locale = Perl_newSVpvf(aTHX_
1695 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1696 " which have\nunexpected meanings: %s\nThe Perl program"
1697 " will use the expected meanings",
8b7358b9 1698 newctype, bad_chars_list);
8b7358b9
KW
1699 }
1700 else {
1701 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1702 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1703 newctype,
8c6180a9
KW
1704 (multi_byte_locale)
1705 ? " Some characters in it are not recognized by"
1706 " Perl."
1707 : "",
1708 (bad_count)
1709 ? "\nThe following characters (and maybe others)"
1710 " may not have the same meaning as the Perl"
1711 " program expects:\n"
1712 : "",
1713 (bad_count)
1714 ? bad_chars_list
1715 : ""
1716 );
8b7358b9
KW
1717 }
1718
b119c2be
KW
1719# ifdef HAS_NL_LANGINFO
1720
1721 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1722 /* parameter FALSE is a don't care here */
4e6826bf 1723 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1724
1725# endif
1726
8b7358b9
KW
1727 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1728
cc9eaeb0 1729 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1730 * output the message now. If not in that scope, we save the
1731 * message to be output at the first operation using this locale,
1732 * if that actually happens. Most programs don't use locales, so
1733 * they are immune to bad ones. */
cc9eaeb0 1734 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1735
780fcc9f
KW
1736 /* The '0' below suppresses a bogus gcc compiler warning */
1737 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1738
bddebb56
KW
1739 if (IN_LC(LC_CTYPE)) {
1740 SvREFCNT_dec_NN(PL_warn_locale);
1741 PL_warn_locale = NULL;
1742 }
780fcc9f 1743 }
baa60164 1744 }
31f05a37 1745 }
98994639
HS
1746
1747#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1748
98994639
HS
1749}
1750
98994639 1751void
2726666d
KW
1752Perl__warn_problematic_locale()
1753{
2726666d
KW
1754
1755#ifdef USE_LOCALE_CTYPE
1756
5f04a188
KW
1757 dTHX;
1758
1759 /* Internal-to-core function that outputs the message in PL_warn_locale,
1760 * and then NULLS it. Should be called only through the macro
1761 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1762
2726666d 1763 if (PL_warn_locale) {
2726666d
KW
1764 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1765 SvPVX(PL_warn_locale),
1766 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1767 SvREFCNT_dec_NN(PL_warn_locale);
1768 PL_warn_locale = NULL;
1769 }
1770
1771#endif
1772
1773}
1774
a4f00dcc
KW
1775STATIC void
1776S_new_collate(pTHX_ const char *newcoll)
98994639 1777{
7d4bcc4a
KW
1778
1779#ifndef USE_LOCALE_COLLATE
1780
1781 PERL_UNUSED_ARG(newcoll);
1782 PERL_UNUSED_CONTEXT;
1783
1784#else
0d071d52 1785
291a84fb 1786 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1787 * core Perl this and that 'newcoll' is the name of the new locale.
1788 *
d35fca5f
KW
1789 * The design of locale collation is that every locale change is given an
1790 * index 'PL_collation_ix'. The first time a string particpates in an
1791 * operation that requires collation while locale collation is active, it
1792 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1793 * magic includes the collation index, and the transformation of the string
1794 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1795 * instead of the string itself. If a string changes, the magic is
1796 * cleared. The next time the locale changes, the index is incremented,
1797 * and so we know during a comparison that the transformation is not
1798 * necessarily still valid, and so is recomputed. Note that if the locale
1799 * changes enough times, the index could wrap (a U32), and it is possible
1800 * that a transformation would improperly be considered valid, leading to
1801 * an unlikely bug */
0d071d52 1802
98994639
HS
1803 if (! newcoll) {
1804 if (PL_collation_name) {
1805 ++PL_collation_ix;
1806 Safefree(PL_collation_name);
1807 PL_collation_name = NULL;
1808 }
1809 PL_collation_standard = TRUE;
00bf60ca 1810 is_standard_collation:
98994639
HS
1811 PL_collxfrm_base = 0;
1812 PL_collxfrm_mult = 2;
165a1c52 1813 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1814 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1815 PL_strxfrm_max_cp = 0;
98994639
HS
1816 return;
1817 }
1818
d35fca5f 1819 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1820 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1821 ++PL_collation_ix;
1822 Safefree(PL_collation_name);
1823 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1824 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1825 if (PL_collation_standard) {
1826 goto is_standard_collation;
1827 }
98994639 1828
165a1c52 1829 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1830 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1831 PL_strxfrm_max_cp = 0;
165a1c52 1832
59c018b9
KW
1833 /* A locale collation definition includes primary, secondary, tertiary,
1834 * etc. weights for each character. To sort, the primary weights are
1835 * used, and only if they compare equal, then the secondary weights are
1836 * used, and only if they compare equal, then the tertiary, etc.
1837 *
1838 * strxfrm() works by taking the input string, say ABC, and creating an
1839 * output transformed string consisting of first the primary weights,
1840 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1841 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1842 * may not have weights at every level. In our example, let's say B
1843 * doesn't have a tertiary weight, and A doesn't have a secondary
1844 * weight. The constructed string is then going to be
1845 * A¹B¹C¹ B²C² A³C³ ....
1846 * This has the desired effect that strcmp() will look at the secondary
1847 * or tertiary weights only if the strings compare equal at all higher
1848 * priority weights. The spaces shown here, like in
c342d20e 1849 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1850 * are not just for readability. In the general case, these must
1851 * actually be bytes, which we will call here 'separator weights'; and
1852 * they must be smaller than any other weight value, but since these
1853 * are C strings, only the terminating one can be a NUL (some
1854 * implementations may include a non-NUL separator weight just before
1855 * the NUL). Implementations tend to reserve 01 for the separator
1856 * weights. They are needed so that a shorter string's secondary
1857 * weights won't be misconstrued as primary weights of a longer string,
1858 * etc. By making them smaller than any other weight, the shorter
1859 * string will sort first. (Actually, if all secondary weights are
1860 * smaller than all primary ones, there is no need for a separator
1861 * weight between those two levels, etc.)
1862 *
1863 * The length of the transformed string is roughly a linear function of
1864 * the input string. It's not exactly linear because some characters
1865 * don't have weights at all levels. When we call strxfrm() we have to
1866 * allocate some memory to hold the transformed string. The
1867 * calculations below try to find coefficients 'm' and 'b' for this
1868 * locale so that m*x + b equals how much space we need, given the size
1869 * of the input string in 'x'. If we calculate too small, we increase
1870 * the size as needed, and call strxfrm() again, but it is better to
1871 * get it right the first time to avoid wasted expensive string
1872 * transformations. */
1873
98994639 1874 {
79f120c8
KW
1875 /* We use the string below to find how long the tranformation of it
1876 * is. Almost all locales are supersets of ASCII, or at least the
1877 * ASCII letters. We use all of them, half upper half lower,
1878 * because if we used fewer, we might hit just the ones that are
1879 * outliers in a particular locale. Most of the strings being
1880 * collated will contain a preponderance of letters, and even if
1881 * they are above-ASCII, they are likely to have the same number of
1882 * weight levels as the ASCII ones. It turns out that digits tend
1883 * to have fewer levels, and some punctuation has more, but those
1884 * are relatively sparse in text, and khw believes this gives a
1885 * reasonable result, but it could be changed if experience so
1886 * dictates. */
1887 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1888 char * x_longer; /* Transformed 'longer' */
1889 Size_t x_len_longer; /* Length of 'x_longer' */
1890
1891 char * x_shorter; /* We also transform a substring of 'longer' */
1892 Size_t x_len_shorter;
1893
a4a439fb 1894 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1895 * are interested only in its length). It is used because it has
1896 * the intelligence to handle all cases, but to work, it needs some
1897 * values of 'm' and 'b' to get it started. For the purposes of
1898 * this calculation we use a very conservative estimate of 'm' and
1899 * 'b'. This assumes a weight can be multiple bytes, enough to
1900 * hold any UV on the platform, and there are 5 levels, 4 weight
1901 * bytes, and a trailing NUL. */
1902 PL_collxfrm_base = 5;
1903 PL_collxfrm_mult = 5 * sizeof(UV);
1904
1905 /* Find out how long the transformation really is */
a4a439fb
KW
1906 x_longer = _mem_collxfrm(longer,
1907 sizeof(longer) - 1,
1908 &x_len_longer,
1909
1910 /* We avoid converting to UTF-8 in the
1911 * called function by telling it the
1912 * string is in UTF-8 if the locale is a
1913 * UTF-8 one. Since the string passed
1914 * here is invariant under UTF-8, we can
1915 * claim it's UTF-8 even though it isn't.
1916 * */
1917 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1918 Safefree(x_longer);
1919
1920 /* Find out how long the transformation of a substring of 'longer'
1921 * is. Together the lengths of these transformations are
1922 * sufficient to calculate 'm' and 'b'. The substring is all of
1923 * 'longer' except the first character. This minimizes the chances
1924 * of being swayed by outliers */
a4a439fb 1925 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1926 sizeof(longer) - 2,
a4a439fb
KW
1927 &x_len_shorter,
1928 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1929 Safefree(x_shorter);
1930
1931 /* If the results are nonsensical for this simple test, the whole
1932 * locale definition is suspect. Mark it so that locale collation
1933 * is not active at all for it. XXX Should we warn? */
1934 if ( x_len_shorter == 0
1935 || x_len_longer == 0
1936 || x_len_shorter >= x_len_longer)
1937 {
1938 PL_collxfrm_mult = 0;
1939 PL_collxfrm_base = 0;
1940 }
1941 else {
1942 SSize_t base; /* Temporary */
1943
1944 /* We have both: m * strlen(longer) + b = x_len_longer
1945 * m * strlen(shorter) + b = x_len_shorter;
1946 * subtracting yields:
1947 * m * (strlen(longer) - strlen(shorter))
1948 * = x_len_longer - x_len_shorter
1949 * But we have set things up so that 'shorter' is 1 byte smaller
1950 * than 'longer'. Hence:
1951 * m = x_len_longer - x_len_shorter
1952 *
1953 * But if something went wrong, make sure the multiplier is at
1954 * least 1.
1955 */
1956 if (x_len_longer > x_len_shorter) {
1957 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1958 }
1959 else {
1960 PL_collxfrm_mult = 1;
1961 }
1962
1963 /* mx + b = len
1964 * so: b = len - mx
1965 * but in case something has gone wrong, make sure it is
1966 * non-negative */
1967 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1968 if (base < 0) {
1969 base = 0;
1970 }
1971
1972 /* Add 1 for the trailing NUL */
1973 PL_collxfrm_base = base + 1;
1974 }
58eebef2 1975
7d4bcc4a
KW
1976# ifdef DEBUGGING
1977
58eebef2
KW
1978 if (DEBUG_L_TEST || debug_initialization) {
1979 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1980 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1981 "x_len_longer=%zu,"
1982 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1983 __FILE__, __LINE__,
1984 PL_in_utf8_COLLATE_locale,
1985 x_len_shorter, x_len_longer,
1986 PL_collxfrm_mult, PL_collxfrm_base);
1987 }
7d4bcc4a
KW
1988# endif
1989
98994639
HS
1990 }
1991 }
1992
1993#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1994
98994639
HS
1995}
1996
d2b24094 1997#ifdef WIN32
b8cc575c 1998
a4f00dcc 1999STATIC char *
b8cc575c 2000S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2001{
2002 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2003 * difference between the two unless the input locale is "", which normally
2004 * means on Windows to get the machine default, which is set via the
2005 * computer's "Regional and Language Options" (or its current equivalent).
2006 * In POSIX, it instead means to find the locale from the user's
2007 * environment. This routine changes the Windows behavior to first look in
2008 * the environment, and, if anything is found, use that instead of going to
2009 * the machine default. If there is no environment override, the machine
2010 * default is used, by calling the real setlocale() with "".
2011 *
2012 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2013 * use the particular category's variable if set; otherwise to use the LANG
2014 * variable. */
b385bb4d 2015
175c4cf9 2016 bool override_LC_ALL = FALSE;
89f7b9aa 2017 char * result;
e5f10d49 2018 unsigned int i;
89f7b9aa 2019
b385bb4d 2020 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2021
2022# ifdef LC_ALL
2023
b385bb4d
KW
2024 locale = PerlEnv_getenv("LC_ALL");
2025 if (! locale) {
e5f10d49
KW
2026 if (category == LC_ALL) {
2027 override_LC_ALL = TRUE;
2028 }
2029 else {
7d4bcc4a
KW
2030
2031# endif
7d4bcc4a 2032
e5f10d49
KW
2033 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2034 if (category == categories[i]) {
2035 locale = PerlEnv_getenv(category_names[i]);
2036 goto found_locale;
2037 }
2038 }
7d4bcc4a 2039
b385bb4d 2040 locale = PerlEnv_getenv("LANG");
481465ea 2041 if (! locale) {
b385bb4d
KW
2042 locale = "";
2043 }
e5f10d49
KW
2044
2045 found_locale: ;
7d4bcc4a
KW
2046
2047# ifdef LC_ALL
2048
e5f10d49 2049 }
b385bb4d 2050 }
7d4bcc4a
KW
2051
2052# endif
2053
b385bb4d
KW
2054 }
2055
89f7b9aa 2056 result = setlocale(category, locale);
48015184
KW
2057 DEBUG_L(STMT_START {
2058 dSAVE_ERRNO;
2059 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2060 setlocale_debug_string(category, locale, result));
2061 RESTORE_ERRNO;
2062 } STMT_END);
89f7b9aa 2063
481465ea 2064 if (! override_LC_ALL) {
89f7b9aa
KW
2065 return result;
2066 }
2067
dfd77d7a 2068 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2069 * LANG variable or the system default if there is no LANG. But these have
2070 * lower priority than the other LC_foo variables, so override it for each
2071 * one that is set. (If they are set to "", it means to use the same thing
2072 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2073
948523db 2074 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2075 result = PerlEnv_getenv(category_names[i]);
2076 if (result && strNE(result, "")) {
2077 setlocale(categories[i], result);
2078 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2079 __FILE__, __LINE__,
2080 setlocale_debug_string(categories[i], result, "not captured")));
2081 }
89f7b9aa 2082 }
7d4bcc4a 2083
bbc98134 2084 result = setlocale(LC_ALL, NULL);
48015184
KW
2085 DEBUG_L(STMT_START {
2086 dSAVE_ERRNO;
2087 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2088 __FILE__, __LINE__,
48015184
KW
2089 setlocale_debug_string(LC_ALL, NULL, result));
2090 RESTORE_ERRNO;
2091 } STMT_END);
89f7b9aa 2092
bbc98134 2093 return result;
b385bb4d
KW
2094}
2095
2096#endif
2097
9aac5db8
KW
2098/*
2099
2100=head1 Locale-related functions and macros
2101
2102=for apidoc Perl_setlocale
2103
2104This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2105taking the same parameters, and returning the same information, except that it
9487427b
KW
2106returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2107instead return C<C> if the underlying locale has a non-dot decimal point
2108character, or a non-empty thousands separator for displaying floating point
2109numbers. This is because perl keeps that locale category such that it has a
2110dot and empty separator, changing the locale briefly during the operations
2111where the underlying one is required. C<Perl_setlocale> knows about this, and
2112compensates; regular C<setlocale> doesn't.
9aac5db8 2113
e9bc6d6b 2114Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2115declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2116C<const> (presumably because its API was specified long ago, and can't be
2117updated; it is illegal to change the information C<setlocale> returns; doing
2118so leads to segfaults.)
9aac5db8 2119
e9bc6d6b
KW
2120Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2121C<setlocale> can be completely ineffective on some platforms under some
2122configurations.
2123
9aac5db8 2124C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2125where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2126the system C<setlocale()> is ineffective, returning the wrong information, and
2127failing to actually change the locale. C<Perl_setlocale>, however works
2128properly in all circumstances.
9aac5db8
KW
2129
2130The return points to a per-thread static buffer, which is overwritten the next
2131time C<Perl_setlocale> is called from the same thread.
2132
2133=cut
2134
2135*/
2136
2137const char *
2138Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2139{
2140 /* This wraps POSIX::setlocale() */
2141
9aac5db8
KW
2142 const char * retval;
2143 const char * newlocale;
2144 dSAVEDERRNO;
2145 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2146 dTHX;
2147
a4f00dcc
KW
2148#ifdef USE_LOCALE_NUMERIC
2149
291a84fb
KW
2150 /* A NULL locale means only query what the current one is. We have the
2151 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2152 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2153 * the correct results. All other categories don't require special
2154 * handling */
a4f00dcc
KW
2155 if (locale == NULL) {
2156 if (category == LC_NUMERIC) {
9aac5db8
KW
2157
2158 /* We don't have to copy this return value, as it is a per-thread
2159 * variable, and won't change until a future setlocale */
2160 return PL_numeric_name;
a4f00dcc
KW
2161 }
2162
7d4bcc4a 2163# ifdef LC_ALL
a4f00dcc 2164
9aac5db8
KW
2165 else if (category == LC_ALL) {
2166 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2167 }
2168
7d4bcc4a 2169# endif
a4f00dcc
KW
2170
2171 }
2172
2173#endif
2174
33e5a354
KW
2175 retval = save_to_buffer(do_setlocale_r(category, locale),
2176 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2177 SAVE_ERRNO;
2178
2179#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2180
2181 if (locale == NULL && category == LC_ALL) {
2182 RESTORE_LC_NUMERIC();
2183 }
2184
2185#endif
a4f00dcc
KW
2186
2187 DEBUG_L(PerlIO_printf(Perl_debug_log,
2188 "%s:%d: %s\n", __FILE__, __LINE__,
2189 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2190
9aac5db8
KW
2191 RESTORE_ERRNO;
2192
2193 if (! retval) {
a4f00dcc
KW
2194 return NULL;
2195 }
2196
9aac5db8 2197 /* If locale == NULL, we are just querying the state */
a4f00dcc 2198 if (locale == NULL) {
a4f00dcc
KW
2199 return retval;
2200 }
a4f00dcc 2201
1159483a
KW
2202 /* Now that have switched locales, we have to update our records to
2203 * correspond. */
a4f00dcc 2204
1159483a 2205 switch (category) {
a4f00dcc 2206
1159483a 2207#ifdef USE_LOCALE_CTYPE
a4f00dcc 2208
1159483a
KW
2209 case LC_CTYPE:
2210 new_ctype(retval);
2211 break;
a4f00dcc 2212
1159483a 2213#endif
a4f00dcc
KW
2214#ifdef USE_LOCALE_COLLATE
2215
1159483a
KW
2216 case LC_COLLATE:
2217 new_collate(retval);
2218 break;
a4f00dcc 2219
1159483a 2220#endif
a4f00dcc
KW
2221#ifdef USE_LOCALE_NUMERIC
2222
1159483a
KW
2223 case LC_NUMERIC:
2224 new_numeric(retval);
2225 break;
a4f00dcc 2226
1159483a
KW
2227#endif
2228#ifdef LC_ALL
a4f00dcc 2229
1159483a 2230 case LC_ALL:
a4f00dcc 2231
1159483a
KW
2232 /* LC_ALL updates all the things we care about. The values may not
2233 * be the same as 'retval', as the locale "" may have set things
2234 * individually */
a4f00dcc 2235
1159483a 2236# ifdef USE_LOCALE_CTYPE
a4f00dcc 2237
b0bde642 2238 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2239 new_ctype(newlocale);
b0bde642 2240 Safefree(newlocale);
a4f00dcc 2241
1159483a
KW
2242# endif /* USE_LOCALE_CTYPE */
2243# ifdef USE_LOCALE_COLLATE
2244
b0bde642 2245 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2246 new_collate(newlocale);
b0bde642 2247 Safefree(newlocale);
a4f00dcc 2248
7d4bcc4a 2249# endif
1159483a 2250# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2251
b0bde642 2252 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2253 new_numeric(newlocale);
b0bde642 2254 Safefree(newlocale);
a4f00dcc 2255
1159483a
KW
2256# endif /* USE_LOCALE_NUMERIC */
2257#endif /* LC_ALL */
a4f00dcc 2258
1159483a
KW
2259 default:
2260 break;
a4f00dcc
KW
2261 }
2262
2263 return retval;
2264
f7416781
KW
2265}
2266
2267PERL_STATIC_INLINE const char *
2268S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2269{
2270 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2271 * growing it if necessary */
2272
0b6697c5 2273 Size_t string_size;
f7416781
KW
2274
2275 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2276
0b6697c5
KW
2277 if (! string) {
2278 return NULL;
2279 }
2280
2281 string_size = strlen(string) + offset + 1;
2282
f7416781
KW
2283 if (*buf_size == 0) {
2284 Newx(*buf, string_size, char);
2285 *buf_size = string_size;
2286 }
2287 else if (string_size > *buf_size) {
2288 Renew(*buf, string_size, char);
2289 *buf_size = string_size;
2290 }
2291
2292 Copy(string, *buf + offset, string_size - offset, char);
2293 return *buf;
2294}
2295
2296/*
2297
f7416781
KW
2298=for apidoc Perl_langinfo
2299
ce181bf6 2300This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2301taking the same C<item> parameter values, and returning the same information.
2302But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2303of Perl's locale handling from your code, and can be used on systems that lack
2304a native C<nl_langinfo>.
2305
2306Expanding on these:
2307
2308=over
2309
2310=item *
2311
ce181bf6
KW
2312The reason it isn't quite a drop-in replacement is actually an advantage. The
2313only difference is that it returns S<C<const char *>>, whereas plain
2314C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2315forbidden to write into the buffer. By declaring this C<const>, the compiler
2316enforces this restriction, so if it is violated, you know at compilation time,
2317rather than getting segfaults at runtime.
2318
2319=item *
2320
69e2275e 2321It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2322without you having to write extra code. The reason for the extra code would be
2323because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2324kept set by Perl so that the radix is a dot, and the separator is the empty
2325string, no matter what the underlying locale is supposed to be, and so to get
2326the expected results, you have to temporarily toggle into the underlying
2327locale, and later toggle back. (You could use plain C<nl_langinfo> and
2328C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2329the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2330(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2331(decimal point) character to be a dot.)
ce181bf6
KW
2332
2333=item *
2334
2335The system function it replaces can have its static return buffer trashed,
2336not only by a subesequent call to that function, but by a C<freelocale>,
2337C<setlocale>, or other locale change. The returned buffer of this function is
2338not changed until the next call to it, so the buffer is never in a trashed
2339state.
f7416781
KW
2340
2341=item *
2342
ce181bf6
KW
2343Its return buffer is per-thread, so it also is never overwritten by a call to
2344this function from another thread; unlike the function it replaces.
2345
2346=item *
2347
2348But most importantly, it works on systems that don't have C<nl_langinfo>, such
2349as Windows, hence makes your code more portable. Of the fifty-some possible
2350items specified by the POSIX 2008 standard,
f7416781 2351L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2352only one is completely unimplemented, though on non-Windows platforms, another
2353significant one is also not implemented). It uses various techniques to
2354recover the other items, including calling C<L<localeconv(3)>>, and
2355C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2356available. Later C<strftime()> versions have additional capabilities; C<""> is
2357returned for those not available on your system.
ce181bf6
KW
2358
2359It is important to note that when called with an item that is recovered by
2360using C<localeconv>, the buffer from any previous explicit call to
2361C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2362contents if you need to access them after a call to this function. (But note
2363that you might not want to be using C<localeconv()> directly anyway, because of
2364issues like the ones listed in the second item of this list (above) for
2365C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2366call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2367unpack).
9f0bc911 2368
6201c220
KW
2369The details for those items which may deviate from what this emulation returns
2370and what a native C<nl_langinfo()> would return are specified in
2371L<I18N::Langinfo>.
f7416781 2372
ce181bf6
KW
2373=back
2374
f7416781
KW
2375When using C<Perl_langinfo> on systems that don't have a native
2376C<nl_langinfo()>, you must
2377
2378 #include "perl_langinfo.h"
2379
2380before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2381C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2382C<langinfo.h> would try to import into the namespace for code that doesn't need
2383it.)
f7416781 2384
f7416781
KW
2385The original impetus for C<Perl_langinfo()> was so that code that needs to
2386find out the current currency symbol, floating point radix character, or digit
2387grouping separator can use, on all systems, the simpler and more
2388thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2389pain to make thread-friendly. For other fields returned by C<localeconv>, it
2390is better to use the methods given in L<perlcall> to call
2391L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2392
2393=cut
2394
2395*/
2396
2397const char *
2398#ifdef HAS_NL_LANGINFO
2399Perl_langinfo(const nl_item item)
2400#else
2401Perl_langinfo(const int item)
2402#endif
2403{
f61748ac
KW
2404 return my_nl_langinfo(item, TRUE);
2405}
2406
59a15af0 2407STATIC const char *
f61748ac
KW
2408#ifdef HAS_NL_LANGINFO
2409S_my_nl_langinfo(const nl_item item, bool toggle)
2410#else
2411S_my_nl_langinfo(const int item, bool toggle)
2412#endif
2413{
ae74815b 2414 dTHX;
0b6697c5 2415 const char * retval;
f7416781 2416
5a854ab3
KW
2417 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2418 * two items, and only if not already there */
4e6826bf 2419 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3
KW
2420 || PL_numeric_underlying))
2421 {
2422 toggle = FALSE;
2423 }
2424
ab340fff 2425#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2426# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2427 || ! defined(HAS_POSIX_2008_LOCALE) \
2428 || ! defined(DUPLOCALE)
f7416781 2429
ab340fff 2430 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2431 * for those items dependent on it. This must be copied to a buffer before
2432 * switching back, as some systems destroy the buffer when setlocale() is
2433 * called */
f7416781 2434
038d3702
KW
2435 {
2436 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2437
b2fee59c 2438 if (toggle) {
038d3702 2439 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2440 }
f7416781 2441
5acc4454
KW
2442 LOCALE_LOCK; /* Prevent interference from another thread executing
2443 this code section (the only call to nl_langinfo in
2444 the core) */
2445
ee90eb2d 2446
628ff75a
KW
2447 /* Copy to a per-thread buffer, which is also one that won't be
2448 * destroyed by a subsequent setlocale(), such as the
2449 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2450 retval = save_to_buffer(nl_langinfo(item),
2451 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2452
5acc4454
KW
2453 LOCALE_UNLOCK;
2454
b2fee59c 2455 if (toggle) {
038d3702 2456 RESTORE_LC_NUMERIC();
b2fee59c 2457 }
f7416781
KW
2458 }
2459
ab340fff
KW
2460# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2461
5a854ab3 2462 {
b2fee59c
KW
2463 bool do_free = FALSE;
2464 locale_t cur = uselocale((locale_t) 0);
ab340fff 2465
b2fee59c
KW
2466 if (cur == LC_GLOBAL_LOCALE) {
2467 cur = duplocale(LC_GLOBAL_LOCALE);
2468 do_free = TRUE;
2469 }
ab340fff 2470
b2fee59c 2471 if (toggle) {
e1aa2579
KW
2472 if (PL_underlying_numeric_obj) {
2473 cur = PL_underlying_numeric_obj;
2474 }
2475 else {
2476 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2477 do_free = TRUE;
2478 }
b2fee59c 2479 }
ab340fff 2480
628ff75a
KW
2481 /* We have to save it to a buffer, because the freelocale() just below
2482 * can invalidate the internal one */
0b6697c5
KW
2483 retval = save_to_buffer(nl_langinfo_l(item, cur),
2484 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2485
b2fee59c
KW
2486 if (do_free) {
2487 freelocale(cur);
2488 }
5a854ab3 2489 }
ab340fff 2490
c1566110
KW
2491# endif
2492
0b6697c5 2493 if (strEQ(retval, "")) {
4e6826bf 2494 if (item == YESSTR) {
c1566110
KW
2495 return "yes";
2496 }
4e6826bf 2497 if (item == NOSTR) {
c1566110
KW
2498 return "no";
2499 }
2500 }
2501
0b6697c5 2502 return retval;
ab340fff 2503
f7416781 2504#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2505
2506 {
2507
f7416781
KW
2508# ifdef HAS_LOCALECONV
2509
43dd6b15 2510 const struct lconv* lc;
628ff75a 2511 const char * temp;
038d3702 2512 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2513
69c5e0db
KW
2514# ifdef TS_W32_BROKEN_LOCALECONV
2515
2516 const char * save_global;
2517 const char * save_thread;
2518 int needed_size;
2519 char * ptr;
2520 char * e;
2521 char * item_start;
2522
2523# endif
f7416781
KW
2524# endif
2525# ifdef HAS_STRFTIME
2526
43dd6b15
KW
2527 struct tm tm;
2528 bool return_format = FALSE; /* Return the %format, not the value */
2529 const char * format;
f7416781
KW
2530
2531# endif
2532
43dd6b15
KW
2533 /* We copy the results to a per-thread buffer, even if not
2534 * multi-threaded. This is in part to simplify this code, and partly
2535 * because we need a buffer anyway for strftime(), and partly because a
2536 * call of localeconv() could otherwise wipe out the buffer, and the
2537 * programmer would not be expecting this, as this is a nl_langinfo()
2538 * substitute after all, so s/he might be thinking their localeconv()
2539 * is safe until another localeconv() call. */
f7416781 2540
43dd6b15
KW
2541 switch (item) {
2542 Size_t len;
f7416781 2543
8d72e74e 2544 /* This is unimplemented */
4e6826bf 2545 case ERA: /* For use with strftime() %E modifier */
f7416781 2546
43dd6b15
KW
2547 default:
2548 return "";
f7416781 2549
43dd6b15 2550 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2551 case YESEXPR: return "^[+1yY]";
2552 case YESSTR: return "yes";
2553 case NOEXPR: return "^[-0nN]";
2554 case NOSTR: return "no";
2555
8d72e74e
KW
2556 case CODESET:
2557
2558# ifndef WIN32
2559
2560 /* On non-windows, this is unimplemented, in part because of
2561 * inconsistencies between vendors. The Darwin native
2562 * nl_langinfo() implementation simply looks at everything past
2563 * any dot in the name, but that doesn't work for other
2564 * vendors. Many Linux locales that don't have UTF-8 in their
2565 * names really are UTF-8, for example; z/OS locales that do
2566 * have UTF-8 in their names, aren't really UTF-8 */
2567 return "";
2568
2569# else
2570
2571 { /* But on Windows, the name does seem to be consistent, so
2572 use that. */
2573 const char * p;
2574 const char * first;
2575 Size_t offset = 0;
2576 const char * name = my_setlocale(LC_CTYPE, NULL);
2577
2578 if (isNAME_C_OR_POSIX(name)) {
2579 return "ANSI_X3.4-1968";
2580 }
2581
2582 /* Find the dot in the locale name */
2583 first = (const char *) strchr(name, '.');
2584 if (! first) {
2585 first = name;
2586 goto has_nondigit;
2587 }
2588
2589 /* Look at everything past the dot */
2590 first++;
2591 p = first;
2592
2593 while (*p) {
2594 if (! isDIGIT(*p)) {
2595 goto has_nondigit;
2596 }
2597
2598 p++;
2599 }
2600
2601 /* Here everything past the dot is a digit. Treat it as a
2602 * code page */
18503cbd 2603 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2604 &PL_langinfo_bufsize, 0);
8d72e74e 2605 offset = STRLENs("CP");
f7416781 2606
8d72e74e
KW
2607 has_nondigit:
2608
2609 retval = save_to_buffer(first, &PL_langinfo_buf,
2610 &PL_langinfo_bufsize, offset);
2611 }
2612
2613 break;
2614
2615# endif
f7416781
KW
2616# ifdef HAS_LOCALECONV
2617
4e6826bf 2618 case CRNCYSTR:
f7416781 2619
43dd6b15
KW
2620 /* We don't bother with localeconv_l() because any system that
2621 * has it is likely to also have nl_langinfo() */
291a84fb 2622
69c5e0db
KW
2623 LOCALE_LOCK_V; /* Prevent interference with other threads
2624 using localeconv() */
2625
2626# ifdef TS_W32_BROKEN_LOCALECONV
2627
2628 /* This is a workaround for a Windows bug prior to VS 15.
2629 * What we do here is, while locked, switch to the global
2630 * locale so localeconv() works; then switch back just before
2631 * the unlock. This can screw things up if some thread is
2632 * already using the global locale while assuming no other is.
2633 * A different workaround would be to call GetCurrencyFormat on
2634 * a known value, and parse it; patches welcome
2635 *
2636 * We have to use LC_ALL instead of LC_MONETARY because of
2637 * another bug in Windows */
2638
2639 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2640 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2641 save_global= savepv(my_setlocale(LC_ALL, NULL));
2642 my_setlocale(LC_ALL, save_thread);
2643
2644# endif
5acc4454 2645
43dd6b15
KW
2646 lc = localeconv();
2647 if ( ! lc
2648 || ! lc->currency_symbol
2649 || strEQ("", lc->currency_symbol))
2650 {
69c5e0db 2651 LOCALE_UNLOCK_V;
43dd6b15
KW
2652 return "";
2653 }
f7416781 2654
43dd6b15 2655 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2656 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2657 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2658 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2659 { /* khw couldn't figure out how the localedef specifications
2660 would show that the $ should replace the radix; this is
2661 just a guess as to how it might work.*/
a34edee3 2662 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2663 }
2664 else if (lc->p_cs_precedes) {
a34edee3 2665 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2666 }
2667 else {
a34edee3 2668 PL_langinfo_buf[0] = '+';
43dd6b15 2669 }
f7416781 2670
69c5e0db
KW
2671# ifdef TS_W32_BROKEN_LOCALECONV
2672
2673 my_setlocale(LC_ALL, save_global);
2674 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2675 my_setlocale(LC_ALL, save_thread);
2676 Safefree(save_global);
2677 Safefree(save_thread);
2678
2679# endif
2680
2681 LOCALE_UNLOCK_V;
43dd6b15 2682 break;
f7416781 2683
69c5e0db
KW
2684# ifdef TS_W32_BROKEN_LOCALECONV
2685
4e6826bf 2686 case RADIXCHAR:
69c5e0db
KW
2687
2688 /* For this, we output a known simple floating point number to
2689 * a buffer, and parse it, looking for the radix */
2690
2691 if (toggle) {
2692 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2693 }
2694
2695 if (PL_langinfo_bufsize < 10) {
2696 PL_langinfo_bufsize = 10;
2697 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2698 }
2699
2700 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2701 "%.1f", 1.5);
2702 if (needed_size >= (int) PL_langinfo_bufsize) {
2703 PL_langinfo_bufsize = needed_size + 1;
2704 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2705 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2706 "%.1f", 1.5);
2707 assert(needed_size < (int) PL_langinfo_bufsize);
2708 }
2709
2710 ptr = PL_langinfo_buf;
2711 e = PL_langinfo_buf + PL_langinfo_bufsize;
2712
2713 /* Find the '1' */
2714 while (ptr < e && *ptr != '1') {
2715 ptr++;
2716 }
2717 ptr++;
2718
2719 /* Find the '5' */
2720 item_start = ptr;
2721 while (ptr < e && *ptr != '5') {
2722 ptr++;
2723 }
2724
2725 /* Everything in between is the radix string */
2726 if (ptr >= e) {
2727 PL_langinfo_buf[0] = '?';
2728 PL_langinfo_buf[1] = '\0';
2729 }
2730 else {
2731 *ptr = '\0';
2732 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2733 }
2734
2735 if (toggle) {
2736 RESTORE_LC_NUMERIC();
2737 }
2738
2739 retval = PL_langinfo_buf;
2740 break;
2741
2742# else
2743
2744 case RADIXCHAR: /* No special handling needed */
2745
2746# endif
2747
4e6826bf 2748 case THOUSEP:
f7416781 2749
43dd6b15 2750 if (toggle) {
038d3702 2751 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2752 }
f7416781 2753
69c5e0db
KW
2754 LOCALE_LOCK_V; /* Prevent interference with other threads
2755 using localeconv() */
2756
2757# ifdef TS_W32_BROKEN_LOCALECONV
2758
2759 /* This should only be for the thousands separator. A
2760 * different work around would be to use GetNumberFormat on a
2761 * known value and parse the result to find the separator */
2762 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2763 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2764 save_global = savepv(my_setlocale(LC_ALL, NULL));
2765 my_setlocale(LC_ALL, save_thread);
2766# if 0
2767 /* This is the start of code that for broken Windows replaces
2768 * the above and below code, and instead calls
2769 * GetNumberFormat() and then would parse that to find the
2770 * thousands separator. It needs to handle UTF-16 vs -8
2771 * issues. */
2772
2773 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2774 DEBUG_L(PerlIO_printf(Perl_debug_log,
2775 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2776 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2777
2778# endif
2779# endif
5acc4454 2780
43dd6b15
KW
2781 lc = localeconv();
2782 if (! lc) {
628ff75a 2783 temp = "";
33394adc 2784 }
43dd6b15 2785 else {
4e6826bf 2786 temp = (item == RADIXCHAR)
43dd6b15
KW
2787 ? lc->decimal_point
2788 : lc->thousands_sep;
628ff75a
KW
2789 if (! temp) {
2790 temp = "";
43dd6b15
KW
2791 }
2792 }
f7416781 2793
0b6697c5
KW
2794 retval = save_to_buffer(temp, &PL_langinfo_buf,
2795 &PL_langinfo_bufsize, 0);
f7416781 2796
69c5e0db
KW
2797# ifdef TS_W32_BROKEN_LOCALECONV
2798
2799 my_setlocale(LC_ALL, save_global);
2800 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2801 my_setlocale(LC_ALL, save_thread);
2802 Safefree(save_global);
2803 Safefree(save_thread);
2804
2805# endif
2806
2807 LOCALE_UNLOCK_V;
5acc4454 2808
43dd6b15 2809 if (toggle) {
038d3702 2810 RESTORE_LC_NUMERIC();
43dd6b15 2811 }
f7416781 2812
43dd6b15 2813 break;
f7416781
KW
2814
2815# endif
2816# ifdef HAS_STRFTIME
2817
43dd6b15
KW
2818 /* These are defined by C89, so we assume that strftime supports
2819 * them, and so are returned unconditionally; they may not be what
2820 * the locale actually says, but should give good enough results
2821 * for someone using them as formats (as opposed to trying to parse
2822 * them to figure out what the locale says). The other format
2823 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2824 case D_FMT: return "%x";
2825 case T_FMT: return "%X";
2826 case D_T_FMT: return "%c";
43dd6b15
KW
2827
2828 /* These formats are only available in later strfmtime's */
4e6826bf 2829 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2830
2831 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2832 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2833 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2834 case ALT_DIGITS:
2835 case AM_STR: case PM_STR:
2836 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2837 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2838 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2839 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2840 case DAY_5: case DAY_6: case DAY_7:
2841 case MON_1: case MON_2: case MON_3: case MON_4:
2842 case MON_5: case MON_6: case MON_7: case MON_8:
2843 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2844
2845 LOCALE_LOCK;
2846
2847 init_tm(&tm); /* Precaution against core dumps */
2848 tm.tm_sec = 30;
2849 tm.tm_min = 30;
2850 tm.tm_hour = 6;
2851 tm.tm_year = 2017 - 1900;
2852 tm.tm_wday = 0;
2853 tm.tm_mon = 0;
2854 switch (item) {
2855 default:
2856 LOCALE_UNLOCK;
2857 Perl_croak(aTHX_
2858 "panic: %s: %d: switch case: %d problem",
2859 __FILE__, __LINE__, item);
2860 NOT_REACHED; /* NOTREACHED */
2861
4e6826bf
KW
2862 case PM_STR: tm.tm_hour = 18;
2863 case AM_STR:
43dd6b15
KW
2864 format = "%p";
2865 break;
2866
4e6826bf
KW
2867 case ABDAY_7: tm.tm_wday++;
2868 case ABDAY_6: tm.tm_wday++;
2869 case ABDAY_5: tm.tm_wday++;
2870 case ABDAY_4: tm.tm_wday++;
2871 case ABDAY_3: tm.tm_wday++;
2872 case ABDAY_2: tm.tm_wday++;
2873 case ABDAY_1:
43dd6b15
KW
2874 format = "%a";
2875 break;
2876
4e6826bf
KW
2877 case DAY_7: tm.tm_wday++;
2878 case DAY_6: tm.tm_wday++;
2879 case DAY_5: tm.tm_wday++;
2880 case DAY_4: tm.tm_wday++;
2881 case DAY_3: tm.tm_wday++;
2882 case DAY_2: tm.tm_wday++;
2883 case DAY_1:
43dd6b15
KW
2884 format = "%A";
2885 break;
2886
4e6826bf
KW
2887 case ABMON_12: tm.tm_mon++;
2888 case ABMON_11: tm.tm_mon++;
2889 case ABMON_10: tm.tm_mon++;
2890 case ABMON_9: tm.tm_mon++;
2891 case ABMON_8: tm.tm_mon++;
2892 case ABMON_7: tm.tm_mon++;
2893 case ABMON_6: tm.tm_mon++;
2894 case ABMON_5: tm.tm_mon++;
2895 case ABMON_4: tm.tm_mon++;
2896 case ABMON_3: tm.tm_mon++;
2897 case ABMON_2: tm.tm_mon++;
2898 case ABMON_1:
43dd6b15
KW
2899 format = "%b";
2900 break;
2901
4e6826bf
KW
2902 case MON_12: tm.tm_mon++;
2903 case MON_11: tm.tm_mon++;
2904 case MON_10: tm.tm_mon++;
2905 case MON_9: tm.tm_mon++;
2906 case MON_8: tm.tm_mon++;
2907 case MON_7: tm.tm_mon++;
2908 case MON_6: tm.tm_mon++;
2909 case MON_5: tm.tm_mon++;
2910 case MON_4: tm.tm_mon++;
2911 case MON_3: tm.tm_mon++;
2912 case MON_2: tm.tm_mon++;
2913 case MON_1:
43dd6b15
KW
2914 format = "%B";
2915 break;
2916
4e6826bf 2917 case T_FMT_AMPM:
43dd6b15
KW
2918 format = "%r";
2919 return_format = TRUE;
2920 break;
2921
4e6826bf 2922 case ERA_D_FMT:
43dd6b15
KW
2923 format = "%Ex";
2924 return_format = TRUE;
2925 break;
2926
4e6826bf 2927 case ERA_T_FMT:
43dd6b15
KW
2928 format = "%EX";
2929 return_format = TRUE;
2930 break;
2931
4e6826bf 2932 case ERA_D_T_FMT:
43dd6b15
KW
2933 format = "%Ec";
2934 return_format = TRUE;
2935 break;
2936
4e6826bf 2937 case ALT_DIGITS:
43dd6b15
KW
2938 tm.tm_wday = 0;
2939 format = "%Ow"; /* Find the alternate digit for 0 */
2940 break;
2941 }
f7416781 2942
43dd6b15
KW
2943 /* We can't use my_strftime() because it doesn't look at
2944 * tm_wday */
2945 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2946 format, &tm))
2947 {
2948 /* A zero return means one of:
2949 * a) there wasn't enough space in PL_langinfo_buf
2950 * b) the format, like a plain %p, returns empty
2951 * c) it was an illegal format, though some
2952 * implementations of strftime will just return the
2953 * illegal format as a plain character sequence.
2954 *
2955 * To quickly test for case 'b)', try again but precede
2956 * the format with a plain character. If that result is
2957 * still empty, the problem is either 'a)' or 'c)' */
2958
2959 Size_t format_size = strlen(format) + 1;
2960 Size_t mod_size = format_size + 1;
2961 char * mod_format;
2962 char * temp_result;
2963
2964 Newx(mod_format, mod_size, char);
2965 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2966 *mod_format = ' ';
43dd6b15
KW
2967 my_strlcpy(mod_format + 1, format, mod_size);
2968 len = strftime(temp_result,
2969 PL_langinfo_bufsize,
2970 mod_format, &tm);
2971 Safefree(mod_format);
2972 Safefree(temp_result);
2973
2974 /* If 'len' is non-zero, it means that we had a case like
2975 * %p which means the current locale doesn't use a.m. or
2976 * p.m., and that is valid */
2977 if (len == 0) {
2978
2979 /* Here, still didn't work. If we get well beyond a
2980 * reasonable size, bail out to prevent an infinite
2981 * loop. */
2982
2983 if (PL_langinfo_bufsize > 100 * format_size) {
2984 *PL_langinfo_buf = '\0';
2985 }
2986 else {
2987 /* Double the buffer size to retry; Add 1 in case
2988 * original was 0, so we aren't stuck at 0. */
2989 PL_langinfo_bufsize *= 2;
2990 PL_langinfo_bufsize++;
2991 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2992 continue;
2993 }
2994 }
f7416781 2995
f7416781 2996 break;
43dd6b15 2997 }
f7416781 2998
43dd6b15
KW
2999 /* Here, we got a result.
3000 *
3001 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3002 * alternate format for wday 0. If the value is the same as
3003 * the normal 0, there isn't an alternate, so clear the buffer.
3004 * */
4e6826bf 3005 if ( item == ALT_DIGITS
43dd6b15
KW
3006 && strEQ(PL_langinfo_buf, "0"))
3007 {
3008 *PL_langinfo_buf = '\0';
3009 }
f7416781 3010
43dd6b15
KW
3011 /* ALT_DIGITS is problematic. Experiments on it showed that
3012 * strftime() did not always work properly when going from
3013 * alt-9 to alt-10. Only a few locales have this item defined,
3014 * and in all of them on Linux that khw was able to find,
3015 * nl_langinfo() merely returned the alt-0 character, possibly
3016 * doubled. Most Unicode digits are in blocks of 10
3017 * consecutive code points, so that is sufficient information
3018 * for those scripts, as we can infer alt-1, alt-2, .... But
3019 * for a Japanese locale, a CJK ideographic 0 is returned, and
3020 * the CJK digits are not in code point order, so you can't
3021 * really infer anything. The localedef for this locale did
3022 * specify the succeeding digits, so that strftime() works
3023 * properly on them, without needing to infer anything. But
3024 * the nl_langinfo() return did not give sufficient information
3025 * for the caller to understand what's going on. So until
3026 * there is evidence that it should work differently, this
3027 * returns the alt-0 string for ALT_DIGITS.
3028 *
3029 * wday was chosen because its range is all a single digit.
3030 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3031
43dd6b15 3032 LOCALE_UNLOCK;
f7416781 3033
d1b150d9
KW
3034 retval = PL_langinfo_buf;
3035
43dd6b15
KW
3036 /* If to return the format, not the value, overwrite the buffer
3037 * with it. But some strftime()s will keep the original format
3038 * if illegal, so change those to "" */
3039 if (return_format) {
3040 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3041 *PL_langinfo_buf = '\0';
3042 }
43dd6b15 3043 else {
0b6697c5
KW
3044 retval = save_to_buffer(format, &PL_langinfo_buf,
3045 &PL_langinfo_bufsize, 0);
f7416781
KW
3046 }
3047 }
3048
3049 break;
f7416781
KW
3050
3051# endif
3052
43dd6b15 3053 }
f7416781
KW
3054 }
3055
0b6697c5 3056 return retval;
f7416781
KW
3057
3058#endif
3059
a4f00dcc 3060}
b385bb4d 3061
98994639
HS
3062/*
3063 * Initialize locale awareness.
3064 */
3065int
3066Perl_init_i18nl10n(pTHX_ int printwarn)
3067{
0e92a118
KW
3068 /* printwarn is
3069 *
3070 * 0 if not to output warning when setup locale is bad
3071 * 1 if to output warning based on value of PERL_BADLANG
3072 * >1 if to output regardless of PERL_BADLANG
3073 *
3074 * returns
98994639 3075 * 1 = set ok or not applicable,
0e92a118
KW
3076 * 0 = fallback to a locale of lower priority
3077 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3078 *
3079 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3080 * set, debugging information is output.
3081 *
3082 * This looks more complicated than it is, mainly due to the #ifdefs.
3083 *
3084 * We try to set LC_ALL to the value determined by the environment. If
3085 * there is no LC_ALL on this platform, we try the individual categories we
3086 * know about. If this works, we are done.
3087 *
3088 * But if it doesn't work, we have to do something else. We search the
3089 * environment variables ourselves instead of relying on the system to do
3090 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3091 * think there is one), and the ultimate fallback "C". This is all done in
3092 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3093 * more complex. The 'trial_locales' array is initialized with just one
3094 * element; it causes the behavior described in the paragraph above this to
3095 * happen. If that fails, we add elements to 'trial_locales', and do extra
3096 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3097 *
3098 * On Ultrix, the locale MUST come from the environment, so there is
3099 * preliminary code to set it. I (khw) am not sure that it is necessary,
3100 * and that this couldn't be folded into the loop, but barring any real
3101 * platforms to test on, it's staying as-is
3102 *
3103 * A slight complication is that in embedded Perls, the locale may already
3104 * be set-up, and we don't want to get it from the normal environment
3105 * variables. This is handled by having a special environment variable
3106 * indicate we're in this situation. We simply set setlocale's 2nd
3107 * parameter to be a NULL instead of "". That indicates to setlocale that
3108 * it is not to change anything, but to return the current value,
3109 * effectively initializing perl's db to what the locale already is.
3110 *
3111 * We play the same trick with NULL if a LC_ALL succeeds. We call
3112 * setlocale() on the individual categores with NULL to get their existing
3113 * values for our db, instead of trying to change them.
3114 * */
98994639 3115
0e92a118
KW
3116 int ok = 1;
3117
7d4bcc4a
KW
3118#ifndef USE_LOCALE
3119
3120 PERL_UNUSED_ARG(printwarn);
3121
3122#else /* USE_LOCALE */
7d4bcc4a
KW
3123# ifdef __GLIBC__
3124
175c4cf9 3125 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3126
3127# endif
65ebb059 3128
ccd65d51
KW
3129 /* NULL uses the existing already set up locale */
3130 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3131 ? NULL
3132 : "";
c3fcd832
KW
3133 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3134 unsigned int trial_locales_count;
175c4cf9
KW
3135 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3136 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3137 bool setlocale_failure = FALSE;
65ebb059 3138 unsigned int i;
175c4cf9
KW
3139
3140 /* A later getenv() could zap this, so only use here */
3141 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3142
3143 const bool locwarn = (printwarn > 1
e5f10d49
KW
3144 || ( printwarn
3145 && ( ! bad_lang_use_once
22ff3130 3146 || (
e5f10d49
KW
3147 /* disallow with "" or "0" */
3148 *bad_lang_use_once
3149 && strNE("0", bad_lang_use_once)))));
ea92aad8 3150
291a84fb 3151 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3152 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3153
3154 /* current locale for given category; should have been copied so aren't
3155 * volatile */
8de4332b 3156 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3157
7d4bcc4a
KW
3158# ifdef WIN32
3159
6bce99ee
JH
3160 /* In some systems you can find out the system default locale
3161 * and use that as the fallback locale. */
7d4bcc4a
KW
3162# define SYSTEM_DEFAULT_LOCALE
3163# endif
3164# ifdef SYSTEM_DEFAULT_LOCALE
3165
65ebb059 3166 const char *system_default_locale = NULL;
98994639 3167
7d4bcc4a 3168# endif
948523db
KW
3169
3170# ifndef DEBUGGING
3171# define DEBUG_LOCALE_INIT(a,b,c)
3172# else
7d4bcc4a 3173
8298454c 3174 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3175
3176# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3177 STMT_START { \
3178 if (debug_initialization) { \
3179 PerlIO_printf(Perl_debug_log, \
3180 "%s:%d: %s\n", \
3181 __FILE__, __LINE__, \
a4f00dcc 3182 setlocale_debug_string(category, \
2fcc0ca9
KW
3183 locale, \
3184 result)); \
3185 } \
3186 } STMT_END
2fcc0ca9 3187
948523db
KW
3188/* Make sure the parallel arrays are properly set up */
3189# ifdef USE_LOCALE_NUMERIC
3190 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3191 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3192# ifdef USE_POSIX_2008_LOCALE
3193 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3194# endif
948523db
KW
3195# endif
3196# ifdef USE_LOCALE_CTYPE
3197 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3198 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3199# ifdef USE_POSIX_2008_LOCALE
3200 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3201# endif
948523db
KW
3202# endif
3203# ifdef USE_LOCALE_COLLATE
3204 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3205 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3206# ifdef USE_POSIX_2008_LOCALE
3207 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3208# endif
948523db
KW
3209# endif
3210# ifdef USE_LOCALE_TIME
3211 assert(categories[LC_TIME_INDEX] == LC_TIME);
3212 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3213# ifdef USE_POSIX_2008_LOCALE
3214 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3215# endif
948523db
KW
3216# endif
3217# ifdef USE_LOCALE_MESSAGES
3218 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3219 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3220# ifdef USE_POSIX_2008_LOCALE
3221 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3222# endif
948523db
KW
3223# endif
3224# ifdef USE_LOCALE_MONETARY
3225 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3226 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3227# ifdef USE_POSIX_2008_LOCALE
3228 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3229# endif
948523db 3230# endif
9821811f
KW
3231# ifdef USE_LOCALE_ADDRESS
3232 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3233 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3234# ifdef USE_POSIX_2008_LOCALE
3235 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3236# endif
9821811f
KW
3237# endif
3238# ifdef USE_LOCALE_IDENTIFICATION
3239 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3240 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3241# ifdef USE_POSIX_2008_LOCALE
3242 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3243# endif
9821811f
KW
3244# endif
3245# ifdef USE_LOCALE_MEASUREMENT
3246 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3247 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3248# ifdef USE_POSIX_2008_LOCALE
3249 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3250# endif
9821811f
KW
3251# endif
3252# ifdef USE_LOCALE_PAPER
3253 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3254 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3255# ifdef USE_POSIX_2008_LOCALE
3256 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3257# endif
9821811f
KW
3258# endif
3259# ifdef USE_LOCALE_TELEPHONE
3260 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3261 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3262# ifdef USE_POSIX_2008_LOCALE
3263 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3264# endif
9821811f 3265# endif
948523db
KW
3266# ifdef LC_ALL
3267 assert(categories[LC_ALL_INDEX] == LC_ALL);
3268 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3269 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3270# ifdef USE_POSIX_2008_LOCALE
3271 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3272# endif
948523db
KW
3273# endif
3274# endif /* DEBUGGING */
47280b20
KW
3275
3276 /* Initialize the cache of the program's UTF-8ness for the always known
3277 * locales C and POSIX */
3278 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3279 sizeof(PL_locale_utf8ness));
3280
e9bc6d6b
KW
3281# ifdef USE_THREAD_SAFE_LOCALE
3282# ifdef WIN32
3283
3284 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3285
3286# endif
3287# endif
39e69e77 3288# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3289
3290 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3291 if (! PL_C_locale_obj) {
3292 Perl_croak_nocontext(
3293 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3294 }
3295 if (DEBUG_Lv_TEST || debug_initialization) {
3296 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3297 }
3298
3299# endif
3300
3ca88433
KW
3301 PL_numeric_radix_sv = newSVpvs(".");
3302
e9bc6d6b
KW
3303# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3304
3305 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3306 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3307
3308# endif
b56c4436 3309# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3310
3311 /*
3312 * Ultrix setlocale(..., "") fails if there are no environment
3313 * variables from which to get a locale name.
3314 */
3315
b56c4436
KW
3316# ifndef LC_ALL
3317# error Ultrix without LC_ALL not implemented
3318# else
7d4bcc4a 3319
b56c4436
KW
3320 {
3321 bool done = FALSE;
ea92aad8
KW
3322 if (lang) {
3323 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3324 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3325 if (sl_result[LC_ALL_INDEX])
3326 done = TRUE;
3327 else
e5f10d49 3328 setlocale_failure = TRUE;
ea92aad8
KW
3329 }
3330 if (! setlocale_failure) {
3331 const char * locale_param;
3332 for (i = 0; i < LC_ALL_INDEX; i++) {
3333 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3334 ? setlocale_init
3335 : NULL;
3336 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3337 if (! sl_result[i]) {
3338 setlocale_failure = TRUE;
3339 }
3340 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3341 }
c835d6be 3342 }
7d4bcc4a
KW
3343 }
3344
3345# endif /* LC_ALL */
e5f10d49 3346# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3347
65ebb059 3348 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3349 * the list. Normally the loop is executed just once. But if setting the
3350 * locale fails, inside the loop we add fallback trials to the array and so
3351 * will execute the loop multiple times */
c3fcd832
KW
3352 trial_locales[0] = setlocale_init;
3353 trial_locales_count = 1;
7d4bcc4a 3354
65ebb059
KW
3355 for (i= 0; i < trial_locales_count; i++) {
3356 const char * trial_locale = trial_locales[i];
3357
3358 if (i > 0) {
3359
3360 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3361 * when i==0, but I (khw) don't think that behavior makes much
3362 * sense */
3363 setlocale_failure = FALSE;
3364
7d4bcc4a 3365# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3366# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3367
65ebb059
KW
3368 /* On Windows machines, an entry of "" after the 0th means to use
3369 * the system default locale, which we now proceed to get. */
3370 if (strEQ(trial_locale, "")) {
3371 unsigned int j;
3372
3373 /* Note that this may change the locale, but we are going to do
3374 * that anyway just below */
837ce802 3375 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3376 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3377
7d4bcc4a 3378 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3379 * try */
3380 if (! system_default_locale) {
3381 goto next_iteration;
3382 }
3383 for (j = 0; j < trial_locales_count; j++) {
3384 if (strEQ(system_default_locale, trial_locales[j])) {
3385 goto next_iteration;
3386 }
3387 }
3388
3389 trial_locale = system_default_locale;
3390 }
ec0202b5
KW
3391# else
3392# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3393# endif
7d4bcc4a 3394# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3395
3396 } /* For i > 0 */
65ebb059 3397
7d4bcc4a
KW
3398# ifdef LC_ALL
3399
948523db
KW
3400 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3401 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3402 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3403 setlocale_failure = TRUE;
7cd8b568
KW
3404 }
3405 else {
3406 /* Since LC_ALL succeeded, it should have changed all the other
3407 * categories it can to its value; so we massage things so that the
3408 * setlocales below just return their category's current values.
3409 * This adequately handles the case in NetBSD where LC_COLLATE may
3410 * not be defined for a locale, and setting it individually will
7d4bcc4a 3411 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3412 * the POSIX locale. */
3413 trial_locale = NULL;
3414 }
7d4bcc4a
KW
3415
3416# endif /* LC_ALL */
98994639 3417
e5f10d49
KW
3418 if (! setlocale_failure) {
3419 unsigned int j;
3420 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3421 curlocales[j]
3422 = savepv(do_setlocale_r(categories[j], trial_locale));
3423 if (! curlocales[j]) {
3424 setlocale_failure = TRUE;
3425 }
3426 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3427 }
c835d6be 3428
e5f10d49
KW
3429 if (! setlocale_failure) { /* All succeeded */
3430 break; /* Exit trial_locales loop */
49c85077 3431 }
65ebb059 3432 }
98994639 3433
49c85077
KW
3434 /* Here, something failed; will need to try a fallback. */
3435 ok = 0;
65ebb059 3436
49c85077
KW
3437 if (i == 0) {
3438 unsigned int j;
98994639 3439
65ebb059 3440 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3441
3442# ifdef LC_ALL
98994639 3443
49c85077
KW
3444 PerlIO_printf(Perl_error_log,
3445 "perl: warning: Setting locale failed.\n");
98994639 3446
7d4bcc4a 3447# else /* !LC_ALL */
98994639 3448
49c85077
KW
3449 PerlIO_printf(Perl_error_log,
3450 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3451
e5f10d49
KW
3452 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3453 if (! curlocales[j]) {
3454 PerlIO_printf(Perl_error_log, category_names[j]);
3455 }
3456 else {
3457 Safefree(curlocales[j]);
3458 }
3459 }
7d4bcc4a 3460
7d4bcc4a 3461# endif /* LC_ALL */
98994639 3462
49c85077
KW
3463 PerlIO_printf(Perl_error_log,
3464 "perl: warning: Please check that your locale settings:\n");
98994639 3465
7d4bcc4a
KW
3466# ifdef __GLIBC__
3467
49c85077
KW
3468 PerlIO_printf(Perl_error_log,
3469 "\tLANGUAGE = %c%s%c,\n",
3470 language ? '"' : '(',
3471 language ? language : "unset",
3472 language ? '"' : ')');
7d4bcc4a 3473# endif
98994639 3474
49c85077
KW
3475 PerlIO_printf(Perl_error_log,
3476 "\tLC_ALL = %c%s%c,\n",
3477 lc_all ? '"' : '(',
3478 lc_all ? lc_all : "unset",
3479 lc_all ? '"' : ')');
98994639 3480
7d4bcc4a
KW
3481# if defined(USE_ENVIRON_ARRAY)
3482
49c85077 3483 {
cd999af9 3484 char **e;
d5e32b93
KW
3485
3486 /* Look through the environment for any variables of the
3487 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3488 * already handled above. These are assumed to be locale
3489 * settings. Output them and their values. */
cd999af9 3490 for (e = environ; *e; e++) {
d5e32b93
KW
3491 const STRLEN prefix_len = sizeof("LC_") - 1;
3492 STRLEN uppers_len;
3493
cd999af9 3494 if ( strBEGINs(*e, "LC_")
c8b388b0 3495 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3496 && (uppers_len = strspn(*e + prefix_len,
3497 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3498 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3499 {
3500 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3501 (int) (prefix_len + uppers_len), *e,
3502 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3503 }
3504 }
49c85077 3505 }
7d4bcc4a
KW
3506
3507# else
3508
49c85077
KW
3509 PerlIO_printf(Perl_error_log,
3510 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3511
3512# endif
98994639 3513
49c85077
KW
3514 PerlIO_printf(Perl_error_log,
3515 "\tLANG = %c%s%c\n",
3516 lang ? '"' : '(',
3517 lang ? lang : "unset",
3518 lang ? '"' : ')');
98994639 3519
49c85077
KW
3520 PerlIO_printf(Perl_error_log,
3521 " are supported and installed on your system.\n");
3522 }
98994639 3523
65ebb059 3524 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3525 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3526 * usually change the upper bound of the loop we are in.
3527 *
3528 * Since the system's default way of setting the locale has not
3529 * found one that works, We use Perl's defined ordering: LC_ALL,
3530 * LANG, and the C locale. We don't try the same locale twice, so
3531 * don't add to the list if already there. (On POSIX systems, the
3532 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3533 * but there's no harm done by doing it explicitly.
3534 *
3535 * Note that this tries the LC_ALL environment variable even on
3536 * systems which have no LC_ALL locale setting. This may or may
3537 * not have been originally intentional, but there's no real need
3538 * to change the behavior. */
65ebb059
KW
3539 if (lc_all) {
3540 for (j = 0; j < trial_locales_count; j++) {
3541 if (strEQ(lc_all, trial_locales[j])) {
3542 goto done_lc_all;
3543 }
3544 }
3545 trial_locales[trial_locales_count++] = lc_all;
3546 }
3547 done_lc_all:
98994639 3548
65ebb059
KW
3549 if (lang) {
3550 for (j = 0; j < trial_locales_count; j++) {
3551 if (strEQ(lang, trial_locales[j])) {
3552 goto done_lang;
3553 }
3554 }
3555 trial_locales[trial_locales_count++] = lang;
3556 }
3557 done_lang:
3558
7d4bcc4a
KW
3559# if defined(WIN32) && defined(LC_ALL)
3560
65ebb059
KW
3561 /* For Windows, we also try the system default locale before "C".
3562 * (If there exists a Windows without LC_ALL we skip this because
3563 * it gets too complicated. For those, the "C" is the next
3564 * fallback possibility). The "" is the same as the 0th element of
3565 * the array, but the code at the loop above knows to treat it
3566 * differently when not the 0th */
3567 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3568
3569# endif
65ebb059
KW
3570
3571 for (j = 0; j < trial_locales_count; j++) {
3572 if (strEQ("C", trial_locales[j])) {
3573 goto done_C;
3574 }
3575 }
3576 trial_locales[trial_locales_count++] = "C";
98994639 3577
65ebb059
KW
3578 done_C: ;
3579 } /* end of first time through the loop */
98994639 3580
7d4bcc4a
KW
3581# ifdef WIN32
3582
65ebb059 3583 next_iteration: ;
7d4bcc4a
KW
3584
3585# endif
65ebb059
KW
3586
3587 } /* end of looping through the trial locales */
3588
3589 if (ok < 1) { /* If we tried to fallback */
3590 const char* msg;
3591 if (! setlocale_failure) { /* fallback succeeded */
3592 msg = "Falling back to";
3593 }
3594 else { /* fallback failed */
e5f10d49 3595 unsigned int j;
98994639 3596
65ebb059
KW
3597 /* We dropped off the end of the loop, so have to decrement i to
3598 * get back to the value the last time through */
3599 i--;
98994639 3600
65ebb059
KW
3601 ok = -1;
3602 msg = "Failed to fall back to";
3603
3604 /* To continue, we should use whatever values we've got */
7d4bcc4a 3605
e5f10d49
KW
3606 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3607 Safefree(curlocales[j]);
3608 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3609 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3610 }
65ebb059
KW
3611 }
3612
3613 if (locwarn) {
3614 const char * description;
3615 const char * name = "";
3616 if (strEQ(trial_locales[i], "C")) {
3617 description = "the standard locale";
3618 name = "C";
3619 }
7d4bcc4a
KW
3620
3621# ifdef SYSTEM_DEFAULT_LOCALE
3622
65ebb059
KW
3623 else if (strEQ(trial_locales[i], "")) {
3624 description = "the system default locale";
3625 if (system_default_locale) {
3626 name = system_default_locale;
3627 }
3628 }
7d4bcc4a
KW
3629
3630# endif /* SYSTEM_DEFAULT_LOCALE */
3631
65ebb059
KW
3632 else {
3633 description = "a fallback locale";
3634 name = trial_locales[i];
3635 }
3636 if (name && strNE(name, "")) {
3637 PerlIO_printf(Perl_error_log,
3638 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3639 }
3640 else {
3641 PerlIO_printf(Perl_error_log,
3642 "perl: warning: %s %s.\n", msg, description);
3643 }
3644 }
3645 } /* End of tried to fallback */
98994639 3646
e5f10d49
KW
3647 /* Done with finding the locales; update our records */
3648
7d4bcc4a
KW
3649# ifdef USE_LOCALE_CTYPE
3650
948523db 3651 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3652
e5f10d49 3653# endif
7d4bcc4a
KW
3654# ifdef USE_LOCALE_COLLATE
3655
948523db 3656 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3657
e5f10d49 3658# endif
7d4bcc4a
KW
3659# ifdef USE_LOCALE_NUMERIC
3660
948523db 3661 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3662
3663# endif
3664
948523db 3665 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3666
e9bc6d6b 3667# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3668
3669 /* This caches whether each category's locale is UTF-8 or not. This
3670 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3671 * initialization time before any threads have started, but not later
3672 * unless thread-safe operations are used.
47280b20
KW
3673 * Caching means that if the program heeds our dictate not to change
3674 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3675 * it may get queried without having to change locales. If the
3676 * environment is such that all categories have the same locale, this
3677 * isn't needed, as the code will not change the locale; but this
3678 * handles the uncommon case where the environment has disparate
3679 * locales for the categories */
47280b20
KW
3680 (void) _is_cur_LC_category_utf8(categories[i]);
3681
3682# endif
3683
e5f10d49
KW
3684 Safefree(curlocales[i]);
3685 }
b310b053 3686
7d4bcc4a
KW
3687# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3688
49c85077 3689 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3690 * locale is UTF-8. The call to new_ctype() just above has already
3691 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3692 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3693 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3694 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3695 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3696
a05d7ebb 3697 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3698 This is an alternative to using the -C command line switch
3699 (the -C if present will override this). */
3700 {
dd374669 3701 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3702 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3703 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3704 PL_utf8cache = -1;
b310b053
JH
3705 }
3706
7d4bcc4a 3707# endif
7d4bcc4a
KW
3708# ifdef __GLIBC__
3709
175c4cf9 3710 Safefree(language);
7d4bcc4a
KW
3711
3712# endif
175c4cf9
KW
3713
3714 Safefree(lc_all);
3715 Safefree(lang);
3716
e3305790 3717#endif /* USE_LOCALE */
2fcc0ca9 3718#ifdef DEBUGGING
7d4bcc4a 3719
2fcc0ca9 3720 /* So won't continue to output stuff */
27cdc72e 3721 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3722
2fcc0ca9
KW
3723#endif
3724
98994639
HS
3725 return ok;
3726}
3727
98994639
HS
3728#ifdef USE_LOCALE_COLLATE
3729
a4a439fb 3730char *
a4a439fb
KW
3731Perl__mem_collxfrm(pTHX_ const char *input_string,
3732 STRLEN len, /* Length of 'input_string' */
3733 STRLEN *xlen, /* Set to length of returned string
3734 (not including the collation index
3735 prefix) */
3736 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3737 )
98994639 3738{
a4a439fb
KW
3739
3740 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3741 * differences. First, it handles embedded NULs. Second, it allocates a bit
3742 * more memory than needed for the transformed data itself. The real
55e5378d 3743 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3744 * the length of that, and doesn't include the collation index size.
3745 * Please see sv_collxfrm() to see how this is used. */
3746
55e5378d
KW
3747#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3748
6696cfa7
KW
3749 char * s = (char *) input_string;
3750 STRLEN s_strlen = strlen(input_string);
79f120c8 3751 char *xbuf = NULL;
55e5378d 3752 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3753 STRLEN length_in_chars;
c664130f 3754 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3755
a4a439fb
KW
3756 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3757
3758 /* Must be NUL-terminated */
3759 assert(*(input_string + len) == '\0');
7918f24d 3760
79f120c8
KW
3761 /* If this locale has defective collation, skip */
3762 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3763 DEBUG_L(PerlIO_printf(Perl_debug_log,
3764 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3765 goto bad;
3766 }
3767
6696cfa7
KW
3768 /* Replace any embedded NULs with the control that sorts before any others.
3769 * This will give as good as possible results on strings that don't
3770 * otherwise contain that character, but otherwise there may be
3771 * less-than-perfect results with that character and NUL. This is
fdc080f3 3772 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3773 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3774 NUL */
6696cfa7
KW
3775 char * e = s + len;
3776 char * sans_nuls;
fdc080f3 3777 STRLEN sans_nuls_len;
94762aa0 3778 int try_non_controls;
afc4976f
KW
3779 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3780 making sure 2nd byte is NUL.
3781 */
3782 STRLEN this_replacement_len;
3783
1e4c9676
KW
3784 /* If we don't know what non-NUL control character sorts lowest for
3785 * this locale, find it */
f28f4d2a 3786 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3787 int j;
afc4976f 3788 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3789 includes the collation index
3790 prefixed. */
3791
91c0e2e0 3792 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3793
3794 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3795 * NUL, in which case we instead look for any character. Controls
3796 * are preferred because collation order is, in general, context
3797 * sensitive, with adjoining characters affecting the order, and
3798 * controls are less likely to have such interactions, allowing the
3799 * NUL-replacement to stand on its own. (Another way to look at it
3800 * is to imagine what would happen if the NUL were replaced by a
3801 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3802 for (try_non_controls = 0;
3803 try_non_controls < 2;
3804 try_non_controls++)
3805 {
d4ff9586
KW
3806 /* Look through all legal code points (NUL isn't) */
3807 for (j = 1; j < 256; j++) {
3808 char * x; /* j's xfrm plus collation index */
3809 STRLEN x_len; /* length of 'x' */
3810 STRLEN trial_len = 1;
736a4fed 3811 char cur_source[] = { '\0', '\0' };
d4ff9586 3812
736a4fed
KW
3813 /* Skip non-controls the first time through the loop. The
3814 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3815 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3816 ? ! isCNTRL_L1(j)
3817 : ! isCNTRL_LC(j))
3818 {
d4ff9586 3819 continue;
6696cfa7 3820 }
6696cfa7 3821
736a4fed
KW
3822 /* Create a 1-char string of the current code point */
3823 cur_source[0] = (char) j;
3824
d4ff9586
KW
3825 /* Then transform it */
3826 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3827 0 /* The string is not in UTF-8 */);
6696cfa7 3828
1e4c9676 3829 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3830 * */
3831 if (! x) {
3832 continue;
3833 }
6696cfa7 3834
d4ff9586
KW
3835 /* If this character's transformation is lower than
3836 * the current lowest, this one becomes the lowest */
3837 if ( cur_min_x == NULL
3838 || strLT(x + COLLXFRM_HDR_LEN,
3839 cur_min_x + COLLXFRM_HDR_LEN))
3840 {
f28f4d2a 3841 PL_strxfrm_NUL_replacement = j;
d4ff9586 3842 cur_min_x = x;
d4ff9586
KW
3843 }
3844 else {
3845 Safefree(x);
3846 }
1e4c9676 3847 } /* end of loop through all 255 characters */
6696cfa7 3848
1e4c9676 3849 /* Stop looking if found */
94762aa0
KW
3850 if (cur_min_x) {
3851 break;
3852 }
3853
3854 /* Unlikely, but possible, if there aren't any controls that
3855 * work in the locale, repeat the loop, looking for any
3856 * character that works */
3857 DEBUG_L(PerlIO_printf(Perl_debug_log,
3858 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3859 } /* End of loop to try first the controls, then any char */
6696cfa7 3860
94762aa0
KW
3861 if (! cur_min_x) {
3862 DEBUG_L(PerlIO_printf(Perl_d