This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Save needed value before destroying it
[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
738 assert(PL_C_locale_obj);
739
740 /* Otherwise, we are switching locales. This will generally entail freeing
741 * the current one's space (at the C library's discretion). We need to
742 * stop using that locale before the switch. So switch to a known locale
743 * object that we don't otherwise mess with. This returns the locale
744 * object in effect at the time of the switch. */
745 old_obj = uselocale(PL_C_locale_obj);
746
747# ifdef DEBUGGING
748
749 if (DEBUG_Lv_TEST || debug_initialization) {
750 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
751 }
752
753# endif
754
755 if (! old_obj) {
756
757# ifdef DEBUGGING
758
759 if (DEBUG_L_TEST || debug_initialization) {
760 dSAVE_ERRNO;
761 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
762 RESTORE_ERRNO;
763 }
764
765# endif
766
767 return NULL;
768 }
769
770# ifdef DEBUGGING
771
772 if (DEBUG_Lv_TEST || debug_initialization) {
773 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
774 }
775
776# endif
777
778 /* If we weren't in a thread safe locale, set so that newlocale() below
779 which uses 'old_obj', uses an empty one. Same for our reserved C object.
780 The latter is defensive coding, so that, even if there is some bug, we
781 will never end up trying to modify either of these, as if passed to
782 newlocale(), they can be. */
783 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
784 old_obj = (locale_t) 0;
785 }
786
787 /* Create the new locale (it may actually modify the current one). */
788
789# ifndef HAS_QUERYLOCALE
790
791 if (strEQ(locale, "")) {
792
793 /* For non-querylocale() systems, we do the setting of "" ourselves to
794 * be sure that we really know what's going on. We follow the Linux
795 * documented behavior (but if that differs from the actual behavior,
796 * this won't work exactly as the OS implements). We go out and
797 * examine the environment based on our understanding of how the system
798 * works, and use that to figure things out */
799
800 const char * const lc_all = PerlEnv_getenv("LC_ALL");
801
802 /* Use any "LC_ALL" environment variable, as it overrides everything
803 * else. */
804 if (lc_all && strNE(lc_all, "")) {
805 locale = lc_all;
806 }
807 else {
808
809 /* Otherwise, we need to dig deeper. Unless overridden, the
810 * default is the LANG environment variable; if it doesn't exist,
811 * then "C" */
812
813 const char * default_name;
814
815 /* To minimize other threads messing with the environment, we copy
816 * the variable, making it a temporary. But this doesn't work upon
817 * program initialization before any scopes are created, and at
818 * this time, there's nothing else going on that would interfere.
819 * So skip the copy in that case */
820 if (PL_scopestack_ix == 0) {
821 default_name = PerlEnv_getenv("LANG");
822 }
823 else {
824 default_name = savepv(PerlEnv_getenv("LANG"));
825 }
826
827 if (! default_name || strEQ(default_name, "")) {
828 default_name = "C";
829 }
830 else if (PL_scopestack_ix != 0) {
831 SAVEFREEPV(default_name);
832 }
833
834 if (category != LC_ALL) {
835 const char * const name = PerlEnv_getenv(category_names[index]);
836
837 /* Here we are setting a single category. Assume will have the
838 * default name */
839 locale = default_name;
840
841 /* But then look for an overriding environment variable */
842 if (name && strNE(name, "")) {
843 locale = name;
844 }
845 }
846 else {
847 bool did_override = FALSE;
848 unsigned int i;
849
850 /* Here, we are getting LC_ALL. Any categories that don't have
851 * a corresponding environment variable set should be set to
852 * LANG, or to "C" if there is no LANG. If no individual
853 * categories differ from this, we can just set LC_ALL. This
854 * is buggy on systems that have extra categories that we don't
855 * know about. If there is an environment variable that sets
856 * that category, we won't know to look for it, and so our use
857 * of LANG or "C" improperly overrides it. On the other hand,
858 * if we don't do what is done here, and there is no
859 * environment variable, the category's locale should be set to
860 * LANG or "C". So there is no good solution. khw thinks the
861 * best is to look at systems to see what categories they have,
862 * and include them, and then to assume that we know the
863 * complete set */
864
865 for (i = 0; i < LC_ALL_INDEX; i++) {
866 const char * const env_override
867 = savepv(PerlEnv_getenv(category_names[i]));
868 const char * this_locale = ( env_override
869 && strNE(env_override, ""))
870 ? env_override
871 : default_name;
872 emulate_setlocale(categories[i], this_locale, i, TRUE);
873
874 if (strNE(this_locale, default_name)) {
875 did_override = TRUE;
876 }
877
878 Safefree(env_override);
879 }
880
881 /* If all the categories are the same, we can set LC_ALL to
882 * that */
883 if (! did_override) {
884 locale = default_name;
885 }
886 else {
887
888 /* Here, LC_ALL is no longer valid, as some individual
889 * categories don't match it. We call ourselves
890 * recursively, as that will execute the code that
891 * generates the proper locale string for this situation.
892 * We don't do the remainder of this function, as that is
893 * to update our records, and we've just done that for the
894 * individual categories in the loop above, and doing so
895 * would cause LC_ALL to be done as well */
896 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
897 }
898 }
899 }
900 }
901 else if (strchr(locale, ';')) {
902
903 /* LC_ALL may actually incude a conglomeration of various categories.
904 * Without querylocale, this code uses the glibc (as of this writing)
905 * syntax for representing that, but that is not a stable API, and
906 * other platforms do it differently, so we have to handle all cases
907 * ourselves */
908
909 const char * s = locale;
910 const char * e = locale + strlen(locale);
911 const char * p = s;
912 const char * category_end;
913 const char * name_start;
914 const char * name_end;
915
916 while (s < e) {
917 unsigned int i;
918
919 /* Parse through the category */
920 while (isWORDCHAR(*p)) {
921 p++;
922 }
923 category_end = p;
924
925 if (*p++ != '=') {
926 Perl_croak(aTHX_
927 "panic: %s: %d: Unexpected character in locale name '%02X",
928 __FILE__, __LINE__, *(p-1));
929 }
930
931 /* Parse through the locale name */
932 name_start = p;
933 while (isGRAPH(*p) && *p != ';') {
934 p++;
935 }
936 name_end = p;
937
938 if (*p++ != ';') {
939 Perl_croak(aTHX_
940 "panic: %s: %d: Unexpected character in locale name '%02X",
941 __FILE__, __LINE__, *(p-1));
942 }
943
944 /* Find the index of the category name in our lists */
945 for (i = 0; i < LC_ALL_INDEX; i++) {
946
947 /* Keep going if this isn't the index. The strnNE() avoids a
948 * Perl_form(), but would fail if ever a category name could be
949 * a substring of another one, like if there were a
950 * "LC_TIME_DATE" */
951 if strnNE(s, category_names[i], category_end - s) {
952 continue;
953 }
954
955 /* If this index is for the single category we're changing, we
956 * have found the locale to set it to. */
957 if (category == categories[i]) {
958 locale = Perl_form(aTHX_ "%.*s",
959 (int) (name_end - name_start),
960 name_start);
961 goto ready_to_set;
962 }
963
964 if (category == LC_ALL) {
965 char * individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
966 emulate_setlocale(categories[i], individ_locale, i, TRUE);
967 Safefree(individ_locale);
968 }
969 }
970
971 s = p;
972 }
973
974 /* Here we have set all the individual categories by recursive calls.
975 * These collectively should have fixed up LC_ALL, so can just query
976 * what that now is */
977 assert(category == LC_ALL);
978
979 return do_setlocale_c(LC_ALL, NULL);
980 }
981
982 ready_to_set: ;
983
984# endif /* end of ! querylocale */
985
986 /* Ready to create a new locale by modification of the exising one */
987 new_obj = newlocale(mask, locale, old_obj);
988
989 if (! new_obj) {
990 dSAVE_ERRNO;
991
992# ifdef DEBUGGING
993
994 if (DEBUG_L_TEST || debug_initialization) {
995 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
996 }
997
998# endif
999
1000 if (! uselocale(old_obj)) {
1001 SAVE_ERRNO;
1002
1003# ifdef DEBUGGING
1004
1005 if (DEBUG_L_TEST || debug_initialization) {
1006 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1007 }
1008
1009# endif
1010
1011 }
1012 RESTORE_ERRNO;
1013 return NULL;
1014 }
1015
1016# ifdef DEBUGGING
1017
1018 if (DEBUG_Lv_TEST || debug_initialization) {
1019 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1020 }
1021
1022# endif
1023
1024 /* And switch into it */
1025 if (! uselocale(new_obj)) {
1026 dSAVE_ERRNO;
1027
1028# ifdef DEBUGGING
1029
1030 if (DEBUG_L_TEST || debug_initialization) {
1031 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1032 }
1033
1034# endif
1035
1036 if (! uselocale(old_obj)) {
1037
1038# ifdef DEBUGGING
1039
1040 if (DEBUG_L_TEST || debug_initialization) {
1041 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1042 }
1043
1044# endif
1045
1046 }
1047 freelocale(new_obj);
1048 RESTORE_ERRNO;
1049 return NULL;
1050 }
1051
1052# ifdef DEBUGGING
1053
1054 if (DEBUG_Lv_TEST || debug_initialization) {
1055 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1056 }
1057
1058# endif
1059
1060 /* We are done, except for updating our records (if the system doesn't keep
1061 * them) and in the case of locale "", we don't actually know what the
1062 * locale that got switched to is, as it came from the environment. So
1063 * have to find it */
1064
1065# ifdef HAS_QUERYLOCALE
1066
1067 if (strEQ(locale, "")) {
1068 locale = querylocale(mask, new_obj);
1069 }
1070
1071# else
1072
1073 /* Here, 'locale' is the return value */
1074
1075 /* Without querylocale(), we have to update our records */
1076
1077 if (category == LC_ALL) {
1078 unsigned int i;
1079
1080 /* For LC_ALL, we change all individual categories to correspond */
1081 /* PL_curlocales is a parallel array, so has same
1082 * length as 'categories' */
1083 for (i = 0; i <= LC_ALL_INDEX; i++) {
1084 Safefree(PL_curlocales[i]);
1085 PL_curlocales[i] = savepv(locale);
1086 }
1087 }
1088 else {
1089
1090 /* For a single category, if it's not the same as the one in LC_ALL, we
1091 * nullify LC_ALL */
1092
1093 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1094 Safefree(PL_curlocales[LC_ALL_INDEX]);
1095 PL_curlocales[LC_ALL_INDEX] = NULL;
1096 }
1097
1098 /* Then update the category's record */
1099 Safefree(PL_curlocales[index]);
1100 PL_curlocales[index] = savepv(locale);
1101 }
1102
1103# endif
1104
1105 return locale;
1106}
1107
1108#endif /* USE_POSIX_2008_LOCALE */
1109
1110#if 0 /* Code that was to emulate thread-safe locales on platforms that
1111 didn't natively support them */
1112
1113/* The way this would work is that we would keep a per-thread list of the
1114 * correct locale for that thread. Any operation that was locale-sensitive
1115 * would have to be changed so that it would look like this:
1116 *
1117 * LOCALE_LOCK;
1118 * setlocale to the correct locale for this operation
1119 * do operation
1120 * LOCALE_UNLOCK
1121 *
1122 * This leaves the global locale in the most recently used operation's, but it
1123 * was locked long enough to get the result. If that result is static, it
1124 * needs to be copied before the unlock.
1125 *
1126 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1127 * the setup, but are no-ops when not needed, and similarly,
1128 * END_LOCALE_DEPENDENT_OP for the tear-down
1129 *
1130 * But every call to a locale-sensitive function would have to be changed, and
1131 * if a module didn't cooperate by using the mutex, things would break.
1132 *
1133 * This code was abandoned before being completed or tested, and is left as-is
1134*/
1135
1136# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1137# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1138
1139STATIC char *
1140S_locking_setlocale(pTHX_
1141 const int category,
1142 const char * locale,
1143 int index,
1144 const bool is_index_valid
1145 )
1146{
1147 /* This function kind of performs a setlocale() on just the current thread;
1148 * thus it is kind of thread-safe. It does this by keeping a thread-level
1149 * array of the current locales for each category. Every time a locale is
1150 * switched to, it does the switch globally, but updates the thread's
1151 * array. A query as to what the current locale is just returns the
1152 * appropriate element from the array, and doesn't actually call the system
1153 * setlocale(). The saving into the array is done in an uninterruptible
1154 * section of code, so is unaffected by whatever any other threads might be
1155 * doing.
1156 *
1157 * All locale-sensitive operations must work by first starting a critical
1158 * section, then switching to the thread's locale as kept by this function,
1159 * and then doing the operation, then ending the critical section. Thus,
1160 * each gets done in the appropriate locale. simulating thread-safety.
1161 *
1162 * This function takes the same parameters, 'category' and 'locale', that
1163 * the regular setlocale() function does, but it also takes two additional
1164 * ones. This is because as described earlier. If we know on input the
1165 * index corresponding to the category into the array where we store the
1166 * current locales, we don't have to calculate it. If the caller knows at
1167 * compile time what the index is, it it can pass it, setting
1168 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1169 *
1170 */
1171
1172 /* If the input index might be incorrect, calculate the correct one */
1173 if (! is_index_valid) {
1174 unsigned int i;
1175
1176 if (DEBUG_Lv_TEST || debug_initialization) {
1177 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1178 }
1179
1180 for (i = 0; i <= LC_ALL_INDEX; i++) {
1181 if (category == categories[i]) {
1182 index = i;
1183 goto found_index;
1184 }
1185 }
1186
1187 /* Here, we don't know about this category, so can't handle it.
1188 * XXX best we can do is to unsafely set this
1189 * XXX warning */
1190
1191 return my_setlocale(category, locale);
1192
1193 found_index: ;
1194
1195 if (DEBUG_Lv_TEST || debug_initialization) {
1196 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1197 }
1198 }
1199
1200 /* For a query, just return what's in our records */
1201 if (new_locale == NULL) {
1202 return curlocales[index];
1203 }
1204
1205
1206 /* Otherwise, we need to do the switch, and save the result, all in a
1207 * critical section */
1208
1209 Safefree(curlocales[[index]]);
1210
1211 /* It might be that this is called from an already-locked section of code.
1212 * We would have to detect and skip the LOCK/UNLOCK if so */
1213 LOCALE_LOCK;
1214
1215 curlocales[index] = savepv(my_setlocale(category, new_locale));
1216
1217 if (strEQ(new_locale, "")) {
1218
1219#ifdef LC_ALL
1220
1221 /* The locale values come from the environment, and may not all be the
1222 * same, so for LC_ALL, we have to update all the others, while the
1223 * mutex is still locked */
1224
1225 if (category == LC_ALL) {
1226 unsigned int i;
1227 for (i = 0; i < LC_ALL_INDEX) {
1228 curlocales[i] = my_setlocale(categories[i], NULL);
1229 }
1230 }
1231 }
1232
1233#endif
1234
1235 LOCALE_UNLOCK;
1236
1237 return curlocales[index];
1238}
1239
1240#endif
837ce802 1241
a4f00dcc 1242STATIC void
86799d2d 1243S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1244{
86799d2d
KW
1245 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1246 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1247
86799d2d
KW
1248#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1249 || defined(HAS_NL_LANGINFO))
98994639 1250
87f8e8e7
KW
1251 const char * radix = (use_locale)
1252 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1253 /* FALSE => already in dest locale */
1254 : ".";
2213a3be 1255
87f8e8e7 1256 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1257
87f8e8e7
KW
1258 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1259 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1260 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1261 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1262 && _is_cur_LC_category_utf8(LC_NUMERIC))
1263 {
1264 SvUTF8_on(PL_numeric_radix_sv);
1265 }
86799d2d
KW
1266
1267# ifdef DEBUGGING
7d4bcc4a 1268
2fcc0ca9
KW
1269 if (DEBUG_L_TEST || debug_initialization) {
1270 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1271 SvPVX(PL_numeric_radix_sv),
1272 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1273 }
69014004 1274
86799d2d
KW
1275# endif
1276#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1277
98994639
HS
1278}
1279
398eeea9
KW
1280STATIC void
1281S_new_numeric(pTHX_ const char *newnum)
98994639 1282{
7d4bcc4a
KW
1283
1284#ifndef USE_LOCALE_NUMERIC
1285
1286 PERL_UNUSED_ARG(newnum);
1287
1288#else
0d071d52 1289
291a84fb 1290 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1291 * core Perl this and that 'newnum' is the name of the new locale.
1292 * It installs this locale as the current underlying default.
1293 *
1294 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1295 * set_numeric_underlying() and set_numeric_standard() functions, which
1296 * should probably not be called directly, but only via macros like
0d071d52
KW
1297 * SET_NUMERIC_STANDARD() in perl.h.
1298 *
1299 * The toggling is necessary mainly so that a non-dot radix decimal point
1300 * character can be output, while allowing internal calculations to use a
1301 * dot.
1302 *
1303 * This sets several interpreter-level variables:
bb304765 1304 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1305 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1306 * that the current locale is the program's underlying
1307 * locale
1308 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1309 * that the current locale is the C locale or
1310 * indistinguishable from the C locale. If non-zero, it
1311 * is in C; if > 1, it means it may not be toggled away
7738054c 1312 * from C.
4c68b815
KW
1313 * PL_numeric_underlying_is_standard A bool kept by this function
1314 * indicating that the underlying locale and the standard
1315 * C locale are indistinguishable for the purposes of
1316 * LC_NUMERIC. This happens when both of the above two
1317 * variables are true at the same time. (Toggling is a
1318 * no-op under these circumstances.) This variable is
1319 * used to avoid having to recalculate.
398eeea9 1320 */
0d071d52 1321
b03f34cf 1322 char *save_newnum;
98994639
HS
1323
1324 if (! newnum) {
43c5f42d
NC
1325 Safefree(PL_numeric_name);
1326 PL_numeric_name = NULL;
98994639 1327 PL_numeric_standard = TRUE;
892e6465 1328 PL_numeric_underlying = TRUE;
4c68b815 1329 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1330 return;
1331 }
1332
b03f34cf 1333 save_newnum = stdize_locale(savepv(newnum));
892e6465 1334 PL_numeric_underlying = TRUE;
4c68b815
KW
1335 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1336
1337 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1338 * them */
1339 if (! PL_numeric_standard) {
1340 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1341 FALSE /* Don't toggle locale */ ))
1342 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1343 FALSE)));
1344 }
abe1abcf 1345
4c68b815 1346 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1347 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1348 Safefree(PL_numeric_name);
b03f34cf 1349 PL_numeric_name = save_newnum;
b03f34cf 1350 }
abe1abcf
KW
1351 else {
1352 Safefree(save_newnum);
1353 }
4c28b29c 1354
4c68b815
KW
1355 PL_numeric_underlying_is_standard = PL_numeric_standard;
1356
7e5377f7 1357# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1358
1359 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1360 PL_numeric_name,
1361 PL_underlying_numeric_obj);
1362
1363#endif
1364
4c68b815
KW
1365 if (DEBUG_L_TEST || debug_initialization) {
1366 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1367 }
1368
4c28b29c
KW
1369 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1370 * have to worry about the radix being a non-dot. (Core operations that
1371 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1372 if (PL_numeric_standard) {
1373 set_numeric_radix(0);
1374 }
1375 else {
1376 set_numeric_standard();
1377 }
4c28b29c 1378
98994639 1379#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1380
98994639
HS
1381}
1382
1383void
1384Perl_set_numeric_standard(pTHX)
1385{
7d4bcc4a 1386
98994639 1387#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1388
28c1bf33
KW
1389 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1390 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1391 * macro avoids calling this routine if toggling isn't necessary according
1392 * to our records (which could be wrong if some XS code has changed the
1393 * locale behind our back) */
0d071d52 1394
837ce802 1395 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1396 PL_numeric_standard = TRUE;
4c68b815 1397 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1398 set_numeric_radix(0);
7d4bcc4a
KW
1399
1400# ifdef DEBUGGING
1401
2fcc0ca9
KW
1402 if (DEBUG_L_TEST || debug_initialization) {
1403 PerlIO_printf(Perl_debug_log,
58e4a467 1404 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1405 }
98994639 1406
7d4bcc4a 1407# endif
98994639 1408#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1409
98994639
HS
1410}
1411
1412void
5792c642 1413Perl_set_numeric_underlying(pTHX)
98994639 1414{
7d4bcc4a 1415
98994639 1416#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1417
28c1bf33 1418 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1419 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1420 * instead of calling this directly. The macro avoids calling this routine
1421 * if toggling isn't necessary according to our records (which could be
1422 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1423
837ce802 1424 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1425 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1426 PL_numeric_underlying = TRUE;
3c62fd46 1427 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1428
1429# ifdef DEBUGGING
1430
2fcc0ca9
KW
1431 if (DEBUG_L_TEST || debug_initialization) {
1432 PerlIO_printf(Perl_debug_log,
58e4a467 1433 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1434 PL_numeric_name);
1435 }
98994639 1436
7d4bcc4a 1437# endif
98994639 1438#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1439
98994639
HS
1440}
1441
1442/*
1443 * Set up for a new ctype locale.
1444 */
a4f00dcc
KW
1445STATIC void
1446S_new_ctype(pTHX_ const char *newctype)
98994639 1447{
7d4bcc4a
KW
1448
1449#ifndef USE_LOCALE_CTYPE
1450
1451 PERL_ARGS_ASSERT_NEW_CTYPE;
1452 PERL_UNUSED_ARG(newctype);
1453 PERL_UNUSED_CONTEXT;
1454
1455#else
0d071d52 1456
291a84fb 1457 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1458 * core Perl this and that 'newctype' is the name of the new locale.
1459 *
1460 * This function sets up the folding arrays for all 256 bytes, assuming
1461 * that tofold() is tolc() since fold case is not a concept in POSIX,
1462 *
1463 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1464 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1465 * this function should be called directly only from this file and from
0d071d52
KW
1466 * POSIX::setlocale() */
1467
27da23d5 1468 dVAR;
68067e4e 1469 UV i;
98994639 1470
8b7358b9
KW
1471 /* Don't check for problems if we are suppressing the warnings */
1472 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1473
7918f24d
NC
1474 PERL_ARGS_ASSERT_NEW_CTYPE;
1475
215c5139
KW
1476 /* We will replace any bad locale warning with 1) nothing if the new one is
1477 * ok; or 2) a new warning for the bad new locale */
1478 if (PL_warn_locale) {
1479 SvREFCNT_dec_NN(PL_warn_locale);
1480 PL_warn_locale = NULL;
1481 }
1482
c1284011 1483 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1484
1485 /* A UTF-8 locale gets standard rules. But note that code still has to
1486 * handle this specially because of the three problematic code points */
1487 if (PL_in_utf8_CTYPE_locale) {
1488 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1489 }
8b7358b9
KW
1490
1491 /* We don't populate the other lists if a UTF-8 locale, but do check that
1492 * everything works as expected, unless checking turned off */
1493 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1494 /* Assume enough space for every character being bad. 4 spaces each
1495 * for the 94 printable characters that are output like "'x' "; and 5
1496 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1497 * NUL */
8b7358b9 1498 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1499 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1500 to start */
1501 unsigned int bad_count = 0; /* Count of bad characters */
1502
baa60164 1503 for (i = 0; i < 256; i++) {
8b7358b9 1504 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1505 if (isupper(i))
1506 PL_fold_locale[i] = (U8) tolower(i);
1507 else if (islower(i))
1508 PL_fold_locale[i] = (U8) toupper(i);
1509 else
1510 PL_fold_locale[i] = (U8) i;
8b7358b9 1511 }
8c6180a9
KW
1512
1513 /* If checking for locale problems, see if the native ASCII-range
1514 * printables plus \n and \t are in their expected categories in
1515 * the new locale. If not, this could mean big trouble, upending
1516 * Perl's and most programs' assumptions, like having a
1517 * metacharacter with special meaning become a \w. Fortunately,
1518 * it's very rare to find locales that aren't supersets of ASCII
1519 * nowadays. It isn't a problem for most controls to be changed
1520 * into something else; we check only \n and \t, though perhaps \r
1521 * could be an issue as well. */
7d4bcc4a 1522 if ( check_for_problems
8c6180a9
KW
1523 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1524 {
8b7358b9
KW
1525 bool is_bad = FALSE;
1526 char name[3] = { '\0' };
1527
1528 /* Convert the name into a string */
1529 if (isPRINT_A(i)) {
1530 name[0] = i;
1531 name[1] = '\0';
1532 }
1533 else if (i == '\n') {
1534 my_strlcpy(name, "\n", sizeof(name));
1535 }
1536 else {
1537 my_strlcpy(name, "\t", sizeof(name));
1538 }
1539
1540 /* Check each possibe class */
1541 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1542 is_bad = TRUE;
1543 DEBUG_L(PerlIO_printf(Perl_debug_log,
1544 "isalnum('%s') unexpectedly is %d\n",
1545 name, cBOOL(isalnum(i))));
1546 }
1547 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1548 is_bad = TRUE;
1549 DEBUG_L(PerlIO_printf(Perl_debug_log,
1550 "isalpha('%s') unexpectedly is %d\n",
1551 name, cBOOL(isalpha(i))));
1552 }
1553 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1554 is_bad = TRUE;
1555 DEBUG_L(PerlIO_printf(Perl_debug_log,
1556 "isdigit('%s') unexpectedly is %d\n",
1557 name, cBOOL(isdigit(i))));
1558 }
1559 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1560 is_bad = TRUE;
1561 DEBUG_L(PerlIO_printf(Perl_debug_log,
1562 "isgraph('%s') unexpectedly is %d\n",
1563 name, cBOOL(isgraph(i))));
1564 }
1565 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1566 is_bad = TRUE;
1567 DEBUG_L(PerlIO_printf(Perl_debug_log,
1568 "islower('%s') unexpectedly is %d\n",
1569 name, cBOOL(islower(i))));
1570 }
1571 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1572 is_bad = TRUE;
1573 DEBUG_L(PerlIO_printf(Perl_debug_log,
1574 "isprint('%s') unexpectedly is %d\n",
1575 name, cBOOL(isprint(i))));
1576 }
1577 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1578 is_bad = TRUE;
1579 DEBUG_L(PerlIO_printf(Perl_debug_log,
1580 "ispunct('%s') unexpectedly is %d\n",
1581 name, cBOOL(ispunct(i))));
1582 }
1583 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1584 is_bad = TRUE;
1585 DEBUG_L(PerlIO_printf(Perl_debug_log,
1586 "isspace('%s') unexpectedly is %d\n",
1587 name, cBOOL(isspace(i))));
1588 }
1589 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1590 is_bad = TRUE;
1591 DEBUG_L(PerlIO_printf(Perl_debug_log,
1592 "isupper('%s') unexpectedly is %d\n",
1593 name, cBOOL(isupper(i))));
1594 }
1595 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1596 is_bad = TRUE;
1597 DEBUG_L(PerlIO_printf(Perl_debug_log,
1598 "isxdigit('%s') unexpectedly is %d\n",
1599 name, cBOOL(isxdigit(i))));
1600 }
1601 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1602 is_bad = TRUE;
1603 DEBUG_L(PerlIO_printf(Perl_debug_log,
1604 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1605 name, tolower(i), (int) toLOWER_A(i)));
1606 }
1607 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1608 is_bad = TRUE;
1609 DEBUG_L(PerlIO_printf(Perl_debug_log,
1610 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1611 name, toupper(i), (int) toUPPER_A(i)));
1612 }
1613 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1614 is_bad = TRUE;
1615 DEBUG_L(PerlIO_printf(Perl_debug_log,
1616 "'\\n' (=%02X) is not a control\n", (int) i));
1617 }
1618
1619 /* Add to the list; Separate multiple entries with a blank */
1620 if (is_bad) {
1621 if (bad_count) {
1622 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1623 }
8b7358b9
KW
1624 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1625 bad_count++;
8c6180a9
KW
1626 }
1627 }
1628 }
1629
7d4bcc4a
KW
1630# ifdef MB_CUR_MAX
1631
8c6180a9 1632 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1633 * this locale requires more than one byte, there are going to be
8c6180a9 1634 * problems. */
9c8a6dc2
KW
1635 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1636 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1637 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1638
8b7358b9
KW
1639 if ( check_for_problems && MB_CUR_MAX > 1
1640 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1641
1642 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1643 * locale. Just assume that the implementation for them (plus
1644 * for POSIX) is correct and the > 1 value is spurious. (Since
1645 * these are specially handled to never be considered UTF-8
1646 * locales, as long as this is the only problem, everything
1647 * should work fine */
1648 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1649 {
8c6180a9
KW
1650 multi_byte_locale = TRUE;
1651 }
7d4bcc4a
KW
1652
1653# endif
8c6180a9 1654
8b7358b9
KW
1655 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1656 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1657 PL_warn_locale = Perl_newSVpvf(aTHX_
1658 "Locale '%s' contains (at least) the following characters"
1659 " which have\nnon-standard meanings: %s\nThe Perl program"
1660 " will use the standard meanings",
1661 newctype, bad_chars_list);
1662
1663 }
1664 else {
1665 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1666 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1667 newctype,
8c6180a9
KW
1668 (multi_byte_locale)
1669 ? " Some characters in it are not recognized by"
1670 " Perl."
1671 : "",
1672 (bad_count)
1673 ? "\nThe following characters (and maybe others)"
1674 " may not have the same meaning as the Perl"
1675 " program expects:\n"
1676 : "",
1677 (bad_count)
1678 ? bad_chars_list
1679 : ""
1680 );
8b7358b9
KW
1681 }
1682
b119c2be
KW
1683# ifdef HAS_NL_LANGINFO
1684
1685 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1686 /* parameter FALSE is a don't care here */
1687 my_nl_langinfo(PERL_CODESET, FALSE));
1688
1689# endif
1690
8b7358b9
KW
1691 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1692
cc9eaeb0 1693 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1694 * output the message now. If not in that scope, we save the
1695 * message to be output at the first operation using this locale,
1696 * if that actually happens. Most programs don't use locales, so
1697 * they are immune to bad ones. */
cc9eaeb0 1698 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1699
780fcc9f
KW
1700 /* The '0' below suppresses a bogus gcc compiler warning */
1701 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1702
bddebb56
KW
1703 if (IN_LC(LC_CTYPE)) {
1704 SvREFCNT_dec_NN(PL_warn_locale);
1705 PL_warn_locale = NULL;
1706 }
780fcc9f 1707 }
baa60164 1708 }
31f05a37 1709 }
98994639
HS
1710
1711#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1712
98994639
HS
1713}
1714
98994639 1715void
2726666d
KW
1716Perl__warn_problematic_locale()
1717{
2726666d
KW
1718
1719#ifdef USE_LOCALE_CTYPE
1720
5f04a188
KW
1721 dTHX;
1722
1723 /* Internal-to-core function that outputs the message in PL_warn_locale,
1724 * and then NULLS it. Should be called only through the macro
1725 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1726
2726666d 1727 if (PL_warn_locale) {
2726666d
KW
1728 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1729 SvPVX(PL_warn_locale),
1730 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1731 SvREFCNT_dec_NN(PL_warn_locale);
1732 PL_warn_locale = NULL;
1733 }
1734
1735#endif
1736
1737}
1738
a4f00dcc
KW
1739STATIC void
1740S_new_collate(pTHX_ const char *newcoll)
98994639 1741{
7d4bcc4a
KW
1742
1743#ifndef USE_LOCALE_COLLATE
1744
1745 PERL_UNUSED_ARG(newcoll);
1746 PERL_UNUSED_CONTEXT;
1747
1748#else
0d071d52 1749
291a84fb 1750 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1751 * core Perl this and that 'newcoll' is the name of the new locale.
1752 *
d35fca5f
KW
1753 * The design of locale collation is that every locale change is given an
1754 * index 'PL_collation_ix'. The first time a string particpates in an
1755 * operation that requires collation while locale collation is active, it
1756 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1757 * magic includes the collation index, and the transformation of the string
1758 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1759 * instead of the string itself. If a string changes, the magic is
1760 * cleared. The next time the locale changes, the index is incremented,
1761 * and so we know during a comparison that the transformation is not
1762 * necessarily still valid, and so is recomputed. Note that if the locale
1763 * changes enough times, the index could wrap (a U32), and it is possible
1764 * that a transformation would improperly be considered valid, leading to
1765 * an unlikely bug */
0d071d52 1766
98994639
HS
1767 if (! newcoll) {
1768 if (PL_collation_name) {
1769 ++PL_collation_ix;
1770 Safefree(PL_collation_name);
1771 PL_collation_name = NULL;
1772 }
1773 PL_collation_standard = TRUE;
00bf60ca 1774 is_standard_collation:
98994639
HS
1775 PL_collxfrm_base = 0;
1776 PL_collxfrm_mult = 2;
165a1c52 1777 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1778 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1779 PL_strxfrm_max_cp = 0;
98994639
HS
1780 return;
1781 }
1782
d35fca5f 1783 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1784 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1785 ++PL_collation_ix;
1786 Safefree(PL_collation_name);
1787 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1788 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1789 if (PL_collation_standard) {
1790 goto is_standard_collation;
1791 }
98994639 1792
165a1c52 1793 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1794 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1795 PL_strxfrm_max_cp = 0;
165a1c52 1796
59c018b9
KW
1797 /* A locale collation definition includes primary, secondary, tertiary,
1798 * etc. weights for each character. To sort, the primary weights are
1799 * used, and only if they compare equal, then the secondary weights are
1800 * used, and only if they compare equal, then the tertiary, etc.
1801 *
1802 * strxfrm() works by taking the input string, say ABC, and creating an
1803 * output transformed string consisting of first the primary weights,
1804 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1805 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1806 * may not have weights at every level. In our example, let's say B
1807 * doesn't have a tertiary weight, and A doesn't have a secondary
1808 * weight. The constructed string is then going to be
1809 * A¹B¹C¹ B²C² A³C³ ....
1810 * This has the desired effect that strcmp() will look at the secondary
1811 * or tertiary weights only if the strings compare equal at all higher
1812 * priority weights. The spaces shown here, like in
c342d20e 1813 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1814 * are not just for readability. In the general case, these must
1815 * actually be bytes, which we will call here 'separator weights'; and
1816 * they must be smaller than any other weight value, but since these
1817 * are C strings, only the terminating one can be a NUL (some
1818 * implementations may include a non-NUL separator weight just before
1819 * the NUL). Implementations tend to reserve 01 for the separator
1820 * weights. They are needed so that a shorter string's secondary
1821 * weights won't be misconstrued as primary weights of a longer string,
1822 * etc. By making them smaller than any other weight, the shorter
1823 * string will sort first. (Actually, if all secondary weights are
1824 * smaller than all primary ones, there is no need for a separator
1825 * weight between those two levels, etc.)
1826 *
1827 * The length of the transformed string is roughly a linear function of
1828 * the input string. It's not exactly linear because some characters
1829 * don't have weights at all levels. When we call strxfrm() we have to
1830 * allocate some memory to hold the transformed string. The
1831 * calculations below try to find coefficients 'm' and 'b' for this
1832 * locale so that m*x + b equals how much space we need, given the size
1833 * of the input string in 'x'. If we calculate too small, we increase
1834 * the size as needed, and call strxfrm() again, but it is better to
1835 * get it right the first time to avoid wasted expensive string
1836 * transformations. */
1837
98994639 1838 {
79f120c8
KW
1839 /* We use the string below to find how long the tranformation of it
1840 * is. Almost all locales are supersets of ASCII, or at least the
1841 * ASCII letters. We use all of them, half upper half lower,
1842 * because if we used fewer, we might hit just the ones that are
1843 * outliers in a particular locale. Most of the strings being
1844 * collated will contain a preponderance of letters, and even if
1845 * they are above-ASCII, they are likely to have the same number of
1846 * weight levels as the ASCII ones. It turns out that digits tend
1847 * to have fewer levels, and some punctuation has more, but those
1848 * are relatively sparse in text, and khw believes this gives a
1849 * reasonable result, but it could be changed if experience so
1850 * dictates. */
1851 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1852 char * x_longer; /* Transformed 'longer' */
1853 Size_t x_len_longer; /* Length of 'x_longer' */
1854
1855 char * x_shorter; /* We also transform a substring of 'longer' */
1856 Size_t x_len_shorter;
1857
a4a439fb 1858 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1859 * are interested only in its length). It is used because it has
1860 * the intelligence to handle all cases, but to work, it needs some
1861 * values of 'm' and 'b' to get it started. For the purposes of
1862 * this calculation we use a very conservative estimate of 'm' and
1863 * 'b'. This assumes a weight can be multiple bytes, enough to
1864 * hold any UV on the platform, and there are 5 levels, 4 weight
1865 * bytes, and a trailing NUL. */
1866 PL_collxfrm_base = 5;
1867 PL_collxfrm_mult = 5 * sizeof(UV);
1868
1869 /* Find out how long the transformation really is */
a4a439fb
KW
1870 x_longer = _mem_collxfrm(longer,
1871 sizeof(longer) - 1,
1872 &x_len_longer,
1873
1874 /* We avoid converting to UTF-8 in the
1875 * called function by telling it the
1876 * string is in UTF-8 if the locale is a
1877 * UTF-8 one. Since the string passed
1878 * here is invariant under UTF-8, we can
1879 * claim it's UTF-8 even though it isn't.
1880 * */
1881 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1882 Safefree(x_longer);
1883
1884 /* Find out how long the transformation of a substring of 'longer'
1885 * is. Together the lengths of these transformations are
1886 * sufficient to calculate 'm' and 'b'. The substring is all of
1887 * 'longer' except the first character. This minimizes the chances
1888 * of being swayed by outliers */
a4a439fb 1889 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1890 sizeof(longer) - 2,
a4a439fb
KW
1891 &x_len_shorter,
1892 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1893 Safefree(x_shorter);
1894
1895 /* If the results are nonsensical for this simple test, the whole
1896 * locale definition is suspect. Mark it so that locale collation
1897 * is not active at all for it. XXX Should we warn? */
1898 if ( x_len_shorter == 0
1899 || x_len_longer == 0
1900 || x_len_shorter >= x_len_longer)
1901 {
1902 PL_collxfrm_mult = 0;
1903 PL_collxfrm_base = 0;
1904 }
1905 else {
1906 SSize_t base; /* Temporary */
1907
1908 /* We have both: m * strlen(longer) + b = x_len_longer
1909 * m * strlen(shorter) + b = x_len_shorter;
1910 * subtracting yields:
1911 * m * (strlen(longer) - strlen(shorter))
1912 * = x_len_longer - x_len_shorter
1913 * But we have set things up so that 'shorter' is 1 byte smaller
1914 * than 'longer'. Hence:
1915 * m = x_len_longer - x_len_shorter
1916 *
1917 * But if something went wrong, make sure the multiplier is at
1918 * least 1.
1919 */
1920 if (x_len_longer > x_len_shorter) {
1921 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1922 }
1923 else {
1924 PL_collxfrm_mult = 1;
1925 }
1926
1927 /* mx + b = len
1928 * so: b = len - mx
1929 * but in case something has gone wrong, make sure it is
1930 * non-negative */
1931 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1932 if (base < 0) {
1933 base = 0;
1934 }
1935
1936 /* Add 1 for the trailing NUL */
1937 PL_collxfrm_base = base + 1;
1938 }
58eebef2 1939
7d4bcc4a
KW
1940# ifdef DEBUGGING
1941
58eebef2
KW
1942 if (DEBUG_L_TEST || debug_initialization) {
1943 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1944 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1945 "x_len_longer=%zu,"
1946 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1947 __FILE__, __LINE__,
1948 PL_in_utf8_COLLATE_locale,
1949 x_len_shorter, x_len_longer,
1950 PL_collxfrm_mult, PL_collxfrm_base);
1951 }
7d4bcc4a
KW
1952# endif
1953
98994639
HS
1954 }
1955 }
1956
1957#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1958
98994639
HS
1959}
1960
d2b24094 1961#ifdef WIN32
b8cc575c 1962
a4f00dcc 1963STATIC char *
b8cc575c 1964S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1965{
1966 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1967 * difference between the two unless the input locale is "", which normally
1968 * means on Windows to get the machine default, which is set via the
1969 * computer's "Regional and Language Options" (or its current equivalent).
1970 * In POSIX, it instead means to find the locale from the user's
1971 * environment. This routine changes the Windows behavior to first look in
1972 * the environment, and, if anything is found, use that instead of going to
1973 * the machine default. If there is no environment override, the machine
1974 * default is used, by calling the real setlocale() with "".
1975 *
1976 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1977 * use the particular category's variable if set; otherwise to use the LANG
1978 * variable. */
b385bb4d 1979
175c4cf9 1980 bool override_LC_ALL = FALSE;
89f7b9aa 1981 char * result;
e5f10d49 1982 unsigned int i;
89f7b9aa 1983
b385bb4d 1984 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1985
1986# ifdef LC_ALL
1987
b385bb4d
KW
1988 locale = PerlEnv_getenv("LC_ALL");
1989 if (! locale) {
e5f10d49
KW
1990 if (category == LC_ALL) {
1991 override_LC_ALL = TRUE;
1992 }
1993 else {
7d4bcc4a
KW
1994
1995# endif
7d4bcc4a 1996
e5f10d49
KW
1997 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1998 if (category == categories[i]) {
1999 locale = PerlEnv_getenv(category_names[i]);
2000 goto found_locale;
2001 }
2002 }
7d4bcc4a 2003
b385bb4d 2004 locale = PerlEnv_getenv("LANG");
481465ea 2005 if (! locale) {
b385bb4d
KW
2006 locale = "";
2007 }
e5f10d49
KW
2008
2009 found_locale: ;
7d4bcc4a
KW
2010
2011# ifdef LC_ALL
2012
e5f10d49 2013 }
b385bb4d 2014 }
7d4bcc4a
KW
2015
2016# endif
2017
b385bb4d
KW
2018 }
2019
89f7b9aa 2020 result = setlocale(category, locale);
48015184
KW
2021 DEBUG_L(STMT_START {
2022 dSAVE_ERRNO;
2023 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2024 setlocale_debug_string(category, locale, result));
2025 RESTORE_ERRNO;
2026 } STMT_END);
89f7b9aa 2027
481465ea 2028 if (! override_LC_ALL) {
89f7b9aa
KW
2029 return result;
2030 }
2031
dfd77d7a 2032 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2033 * LANG variable or the system default if there is no LANG. But these have
2034 * lower priority than the other LC_foo variables, so override it for each
2035 * one that is set. (If they are set to "", it means to use the same thing
2036 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2037
948523db 2038 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2039 result = PerlEnv_getenv(category_names[i]);
2040 if (result && strNE(result, "")) {
2041 setlocale(categories[i], result);
2042 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2043 __FILE__, __LINE__,
2044 setlocale_debug_string(categories[i], result, "not captured")));
2045 }
89f7b9aa 2046 }
7d4bcc4a 2047
bbc98134 2048 result = setlocale(LC_ALL, NULL);
48015184
KW
2049 DEBUG_L(STMT_START {
2050 dSAVE_ERRNO;
2051 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2052 __FILE__, __LINE__,
48015184
KW
2053 setlocale_debug_string(LC_ALL, NULL, result));
2054 RESTORE_ERRNO;
2055 } STMT_END);
89f7b9aa 2056
bbc98134 2057 return result;
b385bb4d
KW
2058}
2059
2060#endif
2061
9aac5db8
KW
2062/*
2063
2064=head1 Locale-related functions and macros
2065
2066=for apidoc Perl_setlocale
2067
2068This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2069taking the same parameters, and returning the same information, except that it
2070returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2071perl keeps that locale category as C<C>, changing it briefly during the
2072operations where the underlying one is required.
2073
e9bc6d6b 2074Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2075declared to return S<C<const char *>>, whereas the system setlocale omits the
2076C<const>. (If it were being written today, plain setlocale would be declared
2077const, since it is illegal to change the information it returns; doing so leads
2078to segfaults.)
2079
e9bc6d6b
KW
2080Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2081C<setlocale> can be completely ineffective on some platforms under some
2082configurations.
2083
9aac5db8 2084C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2085where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2086the system C<setlocale()> is ineffective, returning the wrong information, and
2087failing to actually change the locale. C<Perl_setlocale>, however works
2088properly in all circumstances.
9aac5db8
KW
2089
2090The return points to a per-thread static buffer, which is overwritten the next
2091time C<Perl_setlocale> is called from the same thread.
2092
2093=cut
2094
2095*/
2096
2097const char *
2098Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2099{
2100 /* This wraps POSIX::setlocale() */
2101
9aac5db8
KW
2102 const char * retval;
2103 const char * newlocale;
2104 dSAVEDERRNO;
2105 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2106 dTHX;
2107
a4f00dcc
KW
2108#ifdef USE_LOCALE_NUMERIC
2109
291a84fb
KW
2110 /* A NULL locale means only query what the current one is. We have the
2111 * LC_NUMERIC name saved, because we are normally switched into the C
2112 * locale for it. For an LC_ALL query, switch back to get the correct
2113 * results. All other categories don't require special handling */
a4f00dcc
KW
2114 if (locale == NULL) {
2115 if (category == LC_NUMERIC) {
9aac5db8
KW
2116
2117 /* We don't have to copy this return value, as it is a per-thread
2118 * variable, and won't change until a future setlocale */
2119 return PL_numeric_name;
a4f00dcc
KW
2120 }
2121
7d4bcc4a 2122# ifdef LC_ALL
a4f00dcc 2123
9aac5db8
KW
2124 else if (category == LC_ALL) {
2125 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2126 }
2127
7d4bcc4a 2128# endif
a4f00dcc
KW
2129
2130 }
2131
2132#endif
2133
33e5a354
KW
2134 retval = save_to_buffer(do_setlocale_r(category, locale),
2135 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2136 SAVE_ERRNO;
2137
2138#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2139
2140 if (locale == NULL && category == LC_ALL) {
2141 RESTORE_LC_NUMERIC();
2142 }
2143
2144#endif
a4f00dcc
KW
2145
2146 DEBUG_L(PerlIO_printf(Perl_debug_log,
2147 "%s:%d: %s\n", __FILE__, __LINE__,
2148 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2149
9aac5db8
KW
2150 RESTORE_ERRNO;
2151
2152 if (! retval) {
a4f00dcc
KW
2153 return NULL;
2154 }
2155
9aac5db8 2156 /* If locale == NULL, we are just querying the state */
a4f00dcc 2157 if (locale == NULL) {
a4f00dcc
KW
2158 return retval;
2159 }
a4f00dcc 2160
1159483a
KW
2161 /* Now that have switched locales, we have to update our records to
2162 * correspond. */
a4f00dcc 2163
1159483a 2164 switch (category) {
a4f00dcc 2165
1159483a 2166#ifdef USE_LOCALE_CTYPE
a4f00dcc 2167
1159483a
KW
2168 case LC_CTYPE:
2169 new_ctype(retval);
2170 break;
a4f00dcc 2171
1159483a 2172#endif
a4f00dcc
KW
2173#ifdef USE_LOCALE_COLLATE
2174
1159483a
KW
2175 case LC_COLLATE:
2176 new_collate(retval);
2177 break;
a4f00dcc 2178
1159483a 2179#endif
a4f00dcc
KW
2180#ifdef USE_LOCALE_NUMERIC
2181
1159483a
KW
2182 case LC_NUMERIC:
2183 new_numeric(retval);
2184 break;
a4f00dcc 2185
1159483a
KW
2186#endif
2187#ifdef LC_ALL
a4f00dcc 2188
1159483a 2189 case LC_ALL:
a4f00dcc 2190
1159483a
KW
2191 /* LC_ALL updates all the things we care about. The values may not
2192 * be the same as 'retval', as the locale "" may have set things
2193 * individually */
a4f00dcc 2194
1159483a 2195# ifdef USE_LOCALE_CTYPE
a4f00dcc 2196
1159483a
KW
2197 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2198 new_ctype(newlocale);
a4f00dcc 2199
1159483a
KW
2200# endif /* USE_LOCALE_CTYPE */
2201# ifdef USE_LOCALE_COLLATE
2202
2203 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2204 new_collate(newlocale);
a4f00dcc 2205
7d4bcc4a 2206# endif
1159483a 2207# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2208
1159483a
KW
2209 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2210 new_numeric(newlocale);
a4f00dcc 2211
1159483a
KW
2212# endif /* USE_LOCALE_NUMERIC */
2213#endif /* LC_ALL */
a4f00dcc 2214
1159483a
KW
2215 default:
2216 break;
a4f00dcc
KW
2217 }
2218
2219 return retval;
2220
f7416781
KW
2221}
2222
2223PERL_STATIC_INLINE const char *
2224S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2225{
2226 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2227 * growing it if necessary */
2228
0b6697c5 2229 Size_t string_size;
f7416781
KW
2230
2231 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2232
0b6697c5
KW
2233 if (! string) {
2234 return NULL;
2235 }
2236
2237 string_size = strlen(string) + offset + 1;
2238
f7416781
KW
2239 if (*buf_size == 0) {
2240 Newx(*buf, string_size, char);
2241 *buf_size = string_size;
2242 }
2243 else if (string_size > *buf_size) {
2244 Renew(*buf, string_size, char);
2245 *buf_size = string_size;
2246 }
2247
2248 Copy(string, *buf + offset, string_size - offset, char);
2249 return *buf;
2250}
2251
2252/*
2253
f7416781
KW
2254=for apidoc Perl_langinfo
2255
ce181bf6 2256This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2257taking the same C<item> parameter values, and returning the same information.
2258But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2259of Perl's locale handling from your code, and can be used on systems that lack
2260a native C<nl_langinfo>.
2261
2262Expanding on these:
2263
2264=over
2265
2266=item *
2267
ce181bf6
KW
2268The reason it isn't quite a drop-in replacement is actually an advantage. The
2269only difference is that it returns S<C<const char *>>, whereas plain
2270C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2271forbidden to write into the buffer. By declaring this C<const>, the compiler
2272enforces this restriction, so if it is violated, you know at compilation time,
2273rather than getting segfaults at runtime.
2274
2275=item *
2276
f7416781
KW
2277It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
2278without you having to write extra code. The reason for the extra code would be
2279because these are from the C<LC_NUMERIC> locale category, which is normally
2280kept set to the C locale by Perl, no matter what the underlying locale is
2281supposed to be, and so to get the expected results, you have to temporarily
ce181bf6
KW
2282toggle into the underlying locale, and later toggle back. (You could use plain
2283C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2284then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2285C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2286radix (decimal point) character to be a dot.)
2287
2288=item *
2289
2290The system function it replaces can have its static return buffer trashed,
2291not only by a subesequent call to that function, but by a C<freelocale>,
2292C<setlocale>, or other locale change. The returned buffer of this function is
2293not changed until the next call to it, so the buffer is never in a trashed
2294state.
f7416781
KW
2295
2296=item *
2297
ce181bf6
KW
2298Its return buffer is per-thread, so it also is never overwritten by a call to
2299this function from another thread; unlike the function it replaces.
2300
2301=item *
2302
2303But most importantly, it works on systems that don't have C<nl_langinfo>, such
2304as Windows, hence makes your code more portable. Of the fifty-some possible
2305items specified by the POSIX 2008 standard,
f7416781 2306L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
ce181bf6
KW
2307only two are completely unimplemented (though the loss of one of these is
2308significant). It uses various techniques to recover the other items, including
2309calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2310in C89, so should be always be available. Later C<strftime()> versions have
2311additional capabilities; C<""> is returned for those not available on your
2312system.
2313
2314It is important to note that when called with an item that is recovered by
2315using C<localeconv>, the buffer from any previous explicit call to
2316C<localeconv> will be overwritten. This means you must save that buffer's
2317contents if you need to access them after a call to this function.
9f0bc911 2318
f7416781
KW
2319The details for those items which may differ from what this emulation returns
2320and what a native C<nl_langinfo()> would return are:
2321
2322=over
2323
2324=item C<CODESET>
2325
2326=item C<ERA>
2327
2328Unimplemented, so returns C<"">.
2329
2330=item C<YESEXPR>
2331
c1566110
KW
2332=item C<YESSTR>
2333
f7416781
KW
2334=item C<NOEXPR>
2335
c1566110
KW
2336=item C<NOSTR>
2337
2338Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
ce181bf6
KW
2339removed from POSIX 2008, and are retained here for backwards compatibility.
2340Your platform's C<nl_langinfo> may not support them.
f7416781
KW
2341
2342=item C<D_FMT>
2343
2344Always evaluates to C<%x>, the locale's appropriate date representation.
2345
2346=item C<T_FMT>
2347
2348Always evaluates to C<%X>, the locale's appropriate time representation.
2349
2350=item C<D_T_FMT>
2351
2352Always evaluates to C<%c>, the locale's appropriate date and time
2353representation.
2354
2355=item C<CRNCYSTR>
2356
2357The return may be incorrect for those rare locales where the currency symbol
2358replaces the radix character.
2359Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2360to work differently.
2361
2362=item C<ALT_DIGITS>
2363
2364Currently this gives the same results as Linux does.
2365Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2366to work differently.
2367
2368=item C<ERA_D_FMT>
2369
2370=item C<ERA_T_FMT>
2371
2372=item C<ERA_D_T_FMT>
2373
2374=item C<T_FMT_AMPM>
2375
2376These are derived by using C<strftime()>, and not all versions of that function
2377know about them. C<""> is returned for these on such systems.
2378
2379=back
2380
ce181bf6
KW
2381=back
2382
f7416781
KW
2383When using C<Perl_langinfo> on systems that don't have a native
2384C<nl_langinfo()>, you must
2385
2386 #include "perl_langinfo.h"
2387
2388before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2389C<#include> with this one. (Doing it this way keeps out the symbols that plain
2390C<langinfo.h> imports into the namespace for code that doesn't need it.)
2391
2392You also should not use the bare C<langinfo.h> item names, but should preface
2393them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2394The C<PERL_I<foo>> versions will also work for this function on systems that do
2395have a native C<nl_langinfo>.
2396
f7416781
KW
2397The original impetus for C<Perl_langinfo()> was so that code that needs to
2398find out the current currency symbol, floating point radix character, or digit
2399grouping separator can use, on all systems, the simpler and more
2400thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2401pain to make thread-friendly. For other fields returned by C<localeconv>, it
2402is better to use the methods given in L<perlcall> to call
2403L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2404
2405=cut
2406
2407*/
2408
2409const char *
2410#ifdef HAS_NL_LANGINFO
2411Perl_langinfo(const nl_item item)
2412#else
2413Perl_langinfo(const int item)
2414#endif
2415{
f61748ac
KW
2416 return my_nl_langinfo(item, TRUE);
2417}
2418
2419const char *
2420#ifdef HAS_NL_LANGINFO
2421S_my_nl_langinfo(const nl_item item, bool toggle)
2422#else
2423S_my_nl_langinfo(const int item, bool toggle)
2424#endif
2425{
ae74815b 2426 dTHX;
0b6697c5 2427 const char * retval;
f7416781 2428
5a854ab3
KW
2429 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2430 * two items, and only if not already there */
2431 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2432 || PL_numeric_underlying))
2433 {
2434 toggle = FALSE;
2435 }
2436
ab340fff 2437#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d
KW
2438# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2439 || ! defined(HAS_POSIX_2008_LOCALE)
f7416781 2440
ab340fff 2441 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2442 * for those items dependent on it. This must be copied to a buffer before
2443 * switching back, as some systems destroy the buffer when setlocale() is
2444 * called */
f7416781 2445
038d3702
KW
2446 {
2447 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2448
b2fee59c 2449 if (toggle) {
038d3702 2450 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2451 }
f7416781 2452
5acc4454
KW
2453 LOCALE_LOCK; /* Prevent interference from another thread executing
2454 this code section (the only call to nl_langinfo in
2455 the core) */
2456
ee90eb2d 2457
628ff75a
KW
2458 /* Copy to a per-thread buffer, which is also one that won't be
2459 * destroyed by a subsequent setlocale(), such as the
2460 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2461 retval = save_to_buffer(nl_langinfo(item),
2462 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2463
5acc4454
KW
2464 LOCALE_UNLOCK;
2465
b2fee59c 2466 if (toggle) {
038d3702 2467 RESTORE_LC_NUMERIC();
b2fee59c 2468 }
f7416781
KW
2469 }
2470
ab340fff
KW
2471# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2472
5a854ab3 2473 {
b2fee59c
KW
2474 bool do_free = FALSE;
2475 locale_t cur = uselocale((locale_t) 0);
ab340fff 2476
b2fee59c
KW
2477 if (cur == LC_GLOBAL_LOCALE) {
2478 cur = duplocale(LC_GLOBAL_LOCALE);
2479 do_free = TRUE;
2480 }
ab340fff 2481
b2fee59c 2482 if (toggle) {
e1aa2579
KW
2483 if (PL_underlying_numeric_obj) {
2484 cur = PL_underlying_numeric_obj;
2485 }
2486 else {
2487 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2488 do_free = TRUE;
2489 }
b2fee59c 2490 }
ab340fff 2491
628ff75a
KW
2492 /* We have to save it to a buffer, because the freelocale() just below
2493 * can invalidate the internal one */
0b6697c5
KW
2494 retval = save_to_buffer(nl_langinfo_l(item, cur),
2495 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2496
b2fee59c
KW
2497 if (do_free) {
2498 freelocale(cur);
2499 }
5a854ab3 2500 }
ab340fff 2501
c1566110
KW
2502# endif
2503
0b6697c5 2504 if (strEQ(retval, "")) {
c1566110
KW
2505 if (item == PERL_YESSTR) {
2506 return "yes";
2507 }
2508 if (item == PERL_NOSTR) {
2509 return "no";
2510 }
2511 }
2512
0b6697c5 2513 return retval;
ab340fff 2514
f7416781 2515#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2516
2517 {
2518
f7416781
KW
2519# ifdef HAS_LOCALECONV
2520
43dd6b15 2521 const struct lconv* lc;
628ff75a 2522 const char * temp;
038d3702 2523 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
2524
2525# endif
2526# ifdef HAS_STRFTIME
2527
43dd6b15
KW
2528 struct tm tm;
2529 bool return_format = FALSE; /* Return the %format, not the value */
2530 const char * format;
f7416781
KW
2531
2532# endif
2533
43dd6b15
KW
2534 /* We copy the results to a per-thread buffer, even if not
2535 * multi-threaded. This is in part to simplify this code, and partly
2536 * because we need a buffer anyway for strftime(), and partly because a
2537 * call of localeconv() could otherwise wipe out the buffer, and the
2538 * programmer would not be expecting this, as this is a nl_langinfo()
2539 * substitute after all, so s/he might be thinking their localeconv()
2540 * is safe until another localeconv() call. */
f7416781 2541
43dd6b15
KW
2542 switch (item) {
2543 Size_t len;
f7416781 2544
43dd6b15
KW
2545 /* These 2 are unimplemented */
2546 case PERL_CODESET:
2547 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 2548
43dd6b15
KW
2549 default:
2550 return "";
f7416781 2551
43dd6b15
KW
2552 /* We use only an English set, since we don't know any more */
2553 case PERL_YESEXPR: return "^[+1yY]";
2554 case PERL_YESSTR: return "yes";
2555 case PERL_NOEXPR: return "^[-0nN]";
2556 case PERL_NOSTR: return "no";
f7416781
KW
2557
2558# ifdef HAS_LOCALECONV
2559
43dd6b15 2560 case PERL_CRNCYSTR:
f7416781 2561
43dd6b15
KW
2562 /* We don't bother with localeconv_l() because any system that
2563 * has it is likely to also have nl_langinfo() */
291a84fb 2564
5acc4454
KW
2565 LOCALE_LOCK; /* Prevent interference with other threads
2566 using localeconv() */
2567
43dd6b15
KW
2568 lc = localeconv();
2569 if ( ! lc
2570 || ! lc->currency_symbol
2571 || strEQ("", lc->currency_symbol))
2572 {
2573 LOCALE_UNLOCK;
2574 return "";
2575 }
f7416781 2576
43dd6b15 2577 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2578 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2579 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2580 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2581 { /* khw couldn't figure out how the localedef specifications
2582 would show that the $ should replace the radix; this is
2583 just a guess as to how it might work.*/
0b6697c5 2584 retval = ".";
43dd6b15
KW
2585 }
2586 else if (lc->p_cs_precedes) {
0b6697c5 2587 retval = "-";
43dd6b15
KW
2588 }
2589 else {
0b6697c5 2590 retval = "+";
43dd6b15 2591 }
f7416781 2592
43dd6b15
KW
2593 LOCALE_UNLOCK;
2594 break;
f7416781 2595
43dd6b15
KW
2596 case PERL_RADIXCHAR:
2597 case PERL_THOUSEP:
f7416781 2598
43dd6b15 2599 if (toggle) {
038d3702 2600 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2601 }
f7416781 2602
5acc4454
KW
2603 LOCALE_LOCK; /* Prevent interference with other threads
2604 using localeconv() */
2605
43dd6b15
KW
2606 lc = localeconv();
2607 if (! lc) {
628ff75a 2608 temp = "";
33394adc 2609 }
43dd6b15 2610 else {
628ff75a 2611 temp = (item == PERL_RADIXCHAR)
43dd6b15
KW
2612 ? lc->decimal_point
2613 : lc->thousands_sep;
628ff75a
KW
2614 if (! temp) {
2615 temp = "";
43dd6b15
KW
2616 }
2617 }
f7416781 2618
0b6697c5
KW
2619 retval = save_to_buffer(temp, &PL_langinfo_buf,
2620 &PL_langinfo_bufsize, 0);
f7416781 2621
5acc4454
KW
2622 LOCALE_UNLOCK;
2623
43dd6b15 2624 if (toggle) {
038d3702 2625 RESTORE_LC_NUMERIC();
43dd6b15 2626 }
f7416781 2627
43dd6b15 2628 break;
f7416781
KW
2629
2630# endif
2631# ifdef HAS_STRFTIME
2632
43dd6b15
KW
2633 /* These are defined by C89, so we assume that strftime supports
2634 * them, and so are returned unconditionally; they may not be what
2635 * the locale actually says, but should give good enough results
2636 * for someone using them as formats (as opposed to trying to parse
2637 * them to figure out what the locale says). The other format
2638 * items are actually tested to verify they work on the platform */
2639 case PERL_D_FMT: return "%x";
2640 case PERL_T_FMT: return "%X";
2641 case PERL_D_T_FMT: return "%c";
2642
2643 /* These formats are only available in later strfmtime's */
2644 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2645 case PERL_T_FMT_AMPM:
2646
2647 /* The rest can be gotten from most versions of strftime(). */
2648 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2649 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2650 case PERL_ABDAY_7:
2651 case PERL_ALT_DIGITS:
2652 case PERL_AM_STR: case PERL_PM_STR:
2653 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2654 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2655 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2656 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2657 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2658 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2659 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2660 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2661 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2662 case PERL_MON_12:
2663
2664 LOCALE_LOCK;
2665
2666 init_tm(&tm); /* Precaution against core dumps */
2667 tm.tm_sec = 30;
2668 tm.tm_min = 30;
2669 tm.tm_hour = 6;
2670 tm.tm_year = 2017 - 1900;
2671 tm.tm_wday = 0;
2672 tm.tm_mon = 0;
2673 switch (item) {
2674 default:
2675 LOCALE_UNLOCK;
2676 Perl_croak(aTHX_
2677 "panic: %s: %d: switch case: %d problem",
2678 __FILE__, __LINE__, item);
2679 NOT_REACHED; /* NOTREACHED */
2680
2681 case PERL_PM_STR: tm.tm_hour = 18;
2682 case PERL_AM_STR:
2683 format = "%p";
2684 break;
2685
2686 case PERL_ABDAY_7: tm.tm_wday++;
2687 case PERL_ABDAY_6: tm.tm_wday++;
2688 case PERL_ABDAY_5: tm.tm_wday++;
2689 case PERL_ABDAY_4: tm.tm_wday++;
2690 case PERL_ABDAY_3: tm.tm_wday++;
2691 case PERL_ABDAY_2: tm.tm_wday++;
2692 case PERL_ABDAY_1:
2693 format = "%a";
2694 break;
2695
2696 case PERL_DAY_7: tm.tm_wday++;
2697 case PERL_DAY_6: tm.tm_wday++;
2698 case PERL_DAY_5: tm.tm_wday++;
2699 case PERL_DAY_4: tm.tm_wday++;
2700 case PERL_DAY_3: tm.tm_wday++;
2701 case PERL_DAY_2: tm.tm_wday++;
2702 case PERL_DAY_1:
2703 format = "%A";
2704 break;
2705
2706 case PERL_ABMON_12: tm.tm_mon++;
2707 case PERL_ABMON_11: tm.tm_mon++;
2708 case PERL_ABMON_10: tm.tm_mon++;
2709 case PERL_ABMON_9: tm.tm_mon++;
2710 case PERL_ABMON_8: tm.tm_mon++;
2711 case PERL_ABMON_7: tm.tm_mon++;
2712 case PERL_ABMON_6: tm.tm_mon++;
2713 case PERL_ABMON_5: tm.tm_mon++;
2714 case PERL_ABMON_4: tm.tm_mon++;
2715 case PERL_ABMON_3: tm.tm_mon++;
2716 case PERL_ABMON_2: tm.tm_mon++;
2717 case PERL_ABMON_1:
2718 format = "%b";
2719 break;
2720
2721 case PERL_MON_12: tm.tm_mon++;
2722 case PERL_MON_11: tm.tm_mon++;
2723 case PERL_MON_10: tm.tm_mon++;
2724 case PERL_MON_9: tm.tm_mon++;
2725 case PERL_MON_8: tm.tm_mon++;
2726 case PERL_MON_7: tm.tm_mon++;
2727 case PERL_MON_6: tm.tm_mon++;
2728 case PERL_MON_5: tm.tm_mon++;
2729 case PERL_MON_4: tm.tm_mon++;
2730 case PERL_MON_3: tm.tm_mon++;
2731 case PERL_MON_2: tm.tm_mon++;
2732 case PERL_MON_1:
2733 format = "%B";
2734 break;
2735
2736 case PERL_T_FMT_AMPM:
2737 format = "%r";
2738 return_format = TRUE;
2739 break;
2740
2741 case PERL_ERA_D_FMT:
2742 format = "%Ex";
2743 return_format = TRUE;
2744 break;
2745
2746 case PERL_ERA_T_FMT:
2747 format = "%EX";
2748 return_format = TRUE;
2749 break;
2750
2751 case PERL_ERA_D_T_FMT:
2752 format = "%Ec";
2753 return_format = TRUE;
2754 break;
2755
2756 case PERL_ALT_DIGITS:
2757 tm.tm_wday = 0;
2758 format = "%Ow"; /* Find the alternate digit for 0 */
2759 break;
2760 }
f7416781 2761
43dd6b15
KW
2762 /* We can't use my_strftime() because it doesn't look at
2763 * tm_wday */
2764 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2765 format, &tm))
2766 {
2767 /* A zero return means one of:
2768 * a) there wasn't enough space in PL_langinfo_buf
2769 * b) the format, like a plain %p, returns empty
2770 * c) it was an illegal format, though some
2771 * implementations of strftime will just return the
2772 * illegal format as a plain character sequence.
2773 *
2774 * To quickly test for case 'b)', try again but precede
2775 * the format with a plain character. If that result is
2776 * still empty, the problem is either 'a)' or 'c)' */
2777
2778 Size_t format_size = strlen(format) + 1;
2779 Size_t mod_size = format_size + 1;
2780 char * mod_format;
2781 char * temp_result;
2782
2783 Newx(mod_format, mod_size, char);
2784 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2785 *mod_format = ' ';
43dd6b15
KW
2786 my_strlcpy(mod_format + 1, format, mod_size);
2787 len = strftime(temp_result,
2788 PL_langinfo_bufsize,
2789 mod_format, &tm);
2790 Safefree(mod_format);
2791 Safefree(temp_result);
2792
2793 /* If 'len' is non-zero, it means that we had a case like
2794 * %p which means the current locale doesn't use a.m. or
2795 * p.m., and that is valid */
2796 if (len == 0) {
2797
2798 /* Here, still didn't work. If we get well beyond a
2799 * reasonable size, bail out to prevent an infinite
2800 * loop. */
2801
2802 if (PL_langinfo_bufsize > 100 * format_size) {
2803 *PL_langinfo_buf = '\0';
2804 }
2805 else {
2806 /* Double the buffer size to retry; Add 1 in case
2807 * original was 0, so we aren't stuck at 0. */
2808 PL_langinfo_bufsize *= 2;
2809 PL_langinfo_bufsize++;
2810 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2811 continue;
2812 }
2813 }
f7416781 2814
f7416781 2815 break;
43dd6b15 2816 }
f7416781 2817
43dd6b15
KW
2818 /* Here, we got a result.
2819 *
2820 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2821 * alternate format for wday 0. If the value is the same as
2822 * the normal 0, there isn't an alternate, so clear the buffer.
2823 * */
2824 if ( item == PERL_ALT_DIGITS
2825 && strEQ(PL_langinfo_buf, "0"))
2826 {
2827 *PL_langinfo_buf = '\0';
2828 }
f7416781 2829
43dd6b15
KW
2830 /* ALT_DIGITS is problematic. Experiments on it showed that
2831 * strftime() did not always work properly when going from
2832 * alt-9 to alt-10. Only a few locales have this item defined,
2833 * and in all of them on Linux that khw was able to find,
2834 * nl_langinfo() merely returned the alt-0 character, possibly
2835 * doubled. Most Unicode digits are in blocks of 10
2836 * consecutive code points, so that is sufficient information
2837 * for those scripts, as we can infer alt-1, alt-2, .... But
2838 * for a Japanese locale, a CJK ideographic 0 is returned, and
2839 * the CJK digits are not in code point order, so you can't
2840 * really infer anything. The localedef for this locale did
2841 * specify the succeeding digits, so that strftime() works
2842 * properly on them, without needing to infer anything. But
2843 * the nl_langinfo() return did not give sufficient information
2844 * for the caller to understand what's going on. So until
2845 * there is evidence that it should work differently, this
2846 * returns the alt-0 string for ALT_DIGITS.
2847 *
2848 * wday was chosen because its range is all a single digit.
2849 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 2850
43dd6b15 2851 LOCALE_UNLOCK;
f7416781 2852
43dd6b15
KW
2853 /* If to return the format, not the value, overwrite the buffer
2854 * with it. But some strftime()s will keep the original format
2855 * if illegal, so change those to "" */
2856 if (return_format) {
2857 if (strEQ(PL_langinfo_buf, format)) {
f7416781 2858 *PL_langinfo_buf = '\0';
0b6697c5 2859 retval = PL_langinfo_buf;
f7416781 2860 }
43dd6b15 2861 else {
0b6697c5
KW
2862 retval = save_to_buffer(format, &PL_langinfo_buf,
2863 &PL_langinfo_bufsize, 0);
f7416781
KW
2864 }
2865 }
2866
2867 break;
f7416781
KW
2868
2869# endif
2870
43dd6b15 2871 }
f7416781
KW
2872 }
2873
0b6697c5 2874 return retval;
f7416781
KW
2875
2876#endif
2877
a4f00dcc 2878}
b385bb4d 2879
98994639
HS
2880/*
2881 * Initialize locale awareness.
2882 */
2883int
2884Perl_init_i18nl10n(pTHX_ int printwarn)
2885{
0e92a118
KW
2886 /* printwarn is
2887 *
2888 * 0 if not to output warning when setup locale is bad
2889 * 1 if to output warning based on value of PERL_BADLANG
2890 * >1 if to output regardless of PERL_BADLANG
2891 *
2892 * returns
98994639 2893 * 1 = set ok or not applicable,
0e92a118
KW
2894 * 0 = fallback to a locale of lower priority
2895 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2896 *
2897 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2898 * set, debugging information is output.
2899 *
2900 * This looks more complicated than it is, mainly due to the #ifdefs.
2901 *
2902 * We try to set LC_ALL to the value determined by the environment. If
2903 * there is no LC_ALL on this platform, we try the individual categories we
2904 * know about. If this works, we are done.
2905 *
2906 * But if it doesn't work, we have to do something else. We search the
2907 * environment variables ourselves instead of relying on the system to do
2908 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2909 * think there is one), and the ultimate fallback "C". This is all done in
2910 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2911 * more complex. The 'trial_locales' array is initialized with just one
2912 * element; it causes the behavior described in the paragraph above this to
2913 * happen. If that fails, we add elements to 'trial_locales', and do extra
2914 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2915 *
2916 * On Ultrix, the locale MUST come from the environment, so there is
2917 * preliminary code to set it. I (khw) am not sure that it is necessary,
2918 * and that this couldn't be folded into the loop, but barring any real
2919 * platforms to test on, it's staying as-is
2920 *
2921 * A slight complication is that in embedded Perls, the locale may already
2922 * be set-up, and we don't want to get it from the normal environment
2923 * variables. This is handled by having a special environment variable
2924 * indicate we're in this situation. We simply set setlocale's 2nd
2925 * parameter to be a NULL instead of "". That indicates to setlocale that
2926 * it is not to change anything, but to return the current value,
2927 * effectively initializing perl's db to what the locale already is.
2928 *
2929 * We play the same trick with NULL if a LC_ALL succeeds. We call
2930 * setlocale() on the individual categores with NULL to get their existing
2931 * values for our db, instead of trying to change them.
2932 * */
98994639 2933
0e92a118
KW
2934 int ok = 1;
2935
7d4bcc4a
KW
2936#ifndef USE_LOCALE
2937
2938 PERL_UNUSED_ARG(printwarn);
2939
2940#else /* USE_LOCALE */
7d4bcc4a
KW
2941# ifdef __GLIBC__
2942
175c4cf9 2943 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2944
2945# endif
65ebb059 2946
ccd65d51
KW
2947 /* NULL uses the existing already set up locale */
2948 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2949 ? NULL
2950 : "";
c3fcd832
KW
2951 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2952 unsigned int trial_locales_count;
175c4cf9
KW
2953 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2954 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2955 bool setlocale_failure = FALSE;
65ebb059 2956 unsigned int i;
175c4cf9
KW
2957
2958 /* A later getenv() could zap this, so only use here */
2959 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2960
2961 const bool locwarn = (printwarn > 1
e5f10d49
KW
2962 || ( printwarn
2963 && ( ! bad_lang_use_once
22ff3130 2964 || (
e5f10d49
KW
2965 /* disallow with "" or "0" */
2966 *bad_lang_use_once
2967 && strNE("0", bad_lang_use_once)))));
ea92aad8 2968
291a84fb 2969 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2970 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2971
2972 /* current locale for given category; should have been copied so aren't
2973 * volatile */
8de4332b 2974 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2975
7d4bcc4a
KW
2976# ifdef WIN32
2977
6bce99ee
JH
2978 /* In some systems you can find out the system default locale
2979 * and use that as the fallback locale. */
7d4bcc4a
KW
2980# define SYSTEM_DEFAULT_LOCALE
2981# endif
2982# ifdef SYSTEM_DEFAULT_LOCALE
2983
65ebb059 2984 const char *system_default_locale = NULL;
98994639 2985
7d4bcc4a 2986# endif
948523db
KW
2987
2988# ifndef DEBUGGING
2989# define DEBUG_LOCALE_INIT(a,b,c)
2990# else
7d4bcc4a 2991
8298454c 2992 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2993
2994# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2995 STMT_START { \
2996 if (debug_initialization) { \
2997 PerlIO_printf(Perl_debug_log, \
2998 "%s:%d: %s\n", \
2999 __FILE__, __LINE__, \
a4f00dcc 3000 setlocale_debug_string(category, \
2fcc0ca9
KW
3001 locale, \
3002 result)); \
3003 } \
3004 } STMT_END
2fcc0ca9 3005
948523db
KW
3006/* Make sure the parallel arrays are properly set up */
3007# ifdef USE_LOCALE_NUMERIC
3008 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3009 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3010# ifdef USE_POSIX_2008_LOCALE
3011 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3012# endif
948523db
KW
3013# endif
3014# ifdef USE_LOCALE_CTYPE
3015 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3016 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3017# ifdef USE_POSIX_2008_LOCALE
3018 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3019# endif
948523db
KW
3020# endif
3021# ifdef USE_LOCALE_COLLATE
3022 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3023 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3024# ifdef USE_POSIX_2008_LOCALE
3025 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3026# endif
948523db
KW
3027# endif
3028# ifdef USE_LOCALE_TIME
3029 assert(categories[LC_TIME_INDEX] == LC_TIME);
3030 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3031# ifdef USE_POSIX_2008_LOCALE
3032 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3033# endif
948523db
KW
3034# endif
3035# ifdef USE_LOCALE_MESSAGES
3036 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3037 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3038# ifdef USE_POSIX_2008_LOCALE
3039 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3040# endif
948523db
KW
3041# endif
3042# ifdef USE_LOCALE_MONETARY
3043 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3044 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3045# ifdef USE_POSIX_2008_LOCALE
3046 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3047# endif
948523db 3048# endif
9821811f
KW
3049# ifdef USE_LOCALE_ADDRESS
3050 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3051 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3052# ifdef USE_POSIX_2008_LOCALE
3053 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3054# endif
9821811f
KW
3055# endif
3056# ifdef USE_LOCALE_IDENTIFICATION
3057 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3058 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3059# ifdef USE_POSIX_2008_LOCALE
3060 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3061# endif
9821811f
KW
3062# endif
3063# ifdef USE_LOCALE_MEASUREMENT
3064 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3065 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3066# ifdef USE_POSIX_2008_LOCALE
3067 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3068# endif
9821811f
KW
3069# endif
3070# ifdef USE_LOCALE_PAPER
3071 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3072 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3073# ifdef USE_POSIX_2008_LOCALE
3074 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3075# endif
9821811f
KW
3076# endif
3077# ifdef USE_LOCALE_TELEPHONE
3078 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3079 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3080# ifdef USE_POSIX_2008_LOCALE
3081 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3082# endif
9821811f 3083# endif
948523db
KW
3084# ifdef LC_ALL
3085 assert(categories[LC_ALL_INDEX] == LC_ALL);
3086 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3087 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3088# ifdef USE_POSIX_2008_LOCALE
3089 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3090# endif
948523db
KW
3091# endif
3092# endif /* DEBUGGING */
47280b20
KW
3093
3094 /* Initialize the cache of the program's UTF-8ness for the always known
3095 * locales C and POSIX */
3096 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3097 sizeof(PL_locale_utf8ness));
3098
e9bc6d6b
KW
3099# ifdef USE_THREAD_SAFE_LOCALE
3100# ifdef WIN32
3101
3102 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3103
3104# endif
3105# endif
3106# if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3107
3108 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3109 if (! PL_C_locale_obj) {
3110 Perl_croak_nocontext(
3111 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3112 }
3113 if (DEBUG_Lv_TEST || debug_initialization) {
3114 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3115 }
3116
3117# endif
3118
3ca88433
KW
3119 PL_numeric_radix_sv = newSVpvs(".");
3120
e9bc6d6b
KW
3121# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3122
3123 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3124 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3125
3126# endif
b56c4436 3127# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3128
3129 /*
3130 * Ultrix setlocale(..., "") fails if there are no environment
3131 * variables from which to get a locale name.
3132 */
3133
b56c4436
KW
3134# ifndef LC_ALL
3135# error Ultrix without LC_ALL not implemented
3136# else
7d4bcc4a 3137
b56c4436
KW
3138 {
3139 bool done = FALSE;
ea92aad8
KW
3140 if (lang) {
3141 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3142 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3143 if (sl_result[LC_ALL_INDEX])
3144 done = TRUE;
3145 else
e5f10d49 3146 setlocale_failure = TRUE;
ea92aad8
KW
3147 }
3148 if (! setlocale_failure) {
3149 const char * locale_param;
3150 for (i = 0; i < LC_ALL_INDEX; i++) {
3151 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3152 ? setlocale_init
3153 : NULL;
3154 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3155 if (! sl_result[i]) {
3156 setlocale_failure = TRUE;
3157 }
3158 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3159 }
c835d6be 3160 }
7d4bcc4a
KW
3161 }
3162
3163# endif /* LC_ALL */
e5f10d49 3164# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3165
65ebb059 3166 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3167 * the list. Normally the loop is executed just once. But if setting the
3168 * locale fails, inside the loop we add fallback trials to the array and so
3169 * will execute the loop multiple times */
c3fcd832
KW
3170 trial_locales[0] = setlocale_init;
3171 trial_locales_count = 1;
7d4bcc4a 3172
65ebb059
KW
3173 for (i= 0; i < trial_locales_count; i++) {
3174 const char * trial_locale = trial_locales[i];
3175
3176 if (i > 0) {
3177
3178 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3179 * when i==0, but I (khw) don't think that behavior makes much
3180 * sense */
3181 setlocale_failure = FALSE;
3182
7d4bcc4a 3183# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3184# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3185
65ebb059
KW
3186 /* On Windows machines, an entry of "" after the 0th means to use
3187 * the system default locale, which we now proceed to get. */
3188 if (strEQ(trial_locale, "")) {
3189 unsigned int j;
3190
3191 /* Note that this may change the locale, but we are going to do
3192 * that anyway just below */
837ce802 3193 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3194 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3195
7d4bcc4a 3196 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3197 * try */
3198 if (! system_default_locale) {
3199 goto next_iteration;
3200 }
3201 for (j = 0; j < trial_locales_count; j++) {
3202 if (strEQ(system_default_locale, trial_locales[j])) {
3203 goto next_iteration;
3204 }
3205 }
3206
3207 trial_locale = system_default_locale;
3208 }
ec0202b5
KW
3209# else
3210# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3211# endif
7d4bcc4a 3212# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3213
3214 } /* For i > 0 */
65ebb059 3215
7d4bcc4a
KW
3216# ifdef LC_ALL
3217
948523db
KW
3218 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3219 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3220 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3221 setlocale_failure = TRUE;
7cd8b568
KW
3222 }
3223 else {
3224 /* Since LC_ALL succeeded, it should have changed all the other
3225 * categories it can to its value; so we massage things so that the
3226 * setlocales below just return their category's current values.
3227 * This adequately handles the case in NetBSD where LC_COLLATE may
3228 * not be defined for a locale, and setting it individually will
7d4bcc4a 3229 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3230 * the POSIX locale. */
3231 trial_locale = NULL;
3232 }
7d4bcc4a
KW
3233
3234# endif /* LC_ALL */
98994639 3235
e5f10d49
KW
3236 if (! setlocale_failure) {
3237 unsigned int j;
3238 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3239 curlocales[j]
3240 = savepv(do_setlocale_r(categories[j], trial_locale));
3241 if (! curlocales[j]) {
3242 setlocale_failure = TRUE;
3243 }
3244 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3245 }
c835d6be 3246
e5f10d49
KW
3247 if (! setlocale_failure) { /* All succeeded */
3248 break; /* Exit trial_locales loop */
49c85077 3249 }
65ebb059 3250 }
98994639 3251
49c85077
KW
3252 /* Here, something failed; will need to try a fallback. */
3253 ok = 0;
65ebb059 3254
49c85077
KW
3255 if (i == 0) {
3256 unsigned int j;
98994639 3257
65ebb059 3258 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3259
3260# ifdef LC_ALL
98994639 3261
49c85077
KW
3262 PerlIO_printf(Perl_error_log,
3263 "perl: warning: Setting locale failed.\n");
98994639 3264
7d4bcc4a 3265# else /* !LC_ALL */
98994639 3266
49c85077
KW
3267 PerlIO_printf(Perl_error_log,
3268 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3269
e5f10d49
KW
3270 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3271 if (! curlocales[j]) {
3272 PerlIO_printf(Perl_error_log, category_names[j]);
3273 }
3274 else {
3275 Safefree(curlocales[j]);
3276 }
3277 }
7d4bcc4a 3278
7d4bcc4a 3279# endif /* LC_ALL */
98994639 3280
49c85077
KW
3281 PerlIO_printf(Perl_error_log,
3282 "perl: warning: Please check that your locale settings:\n");
98994639 3283
7d4bcc4a
KW
3284# ifdef __GLIBC__
3285
49c85077
KW
3286 PerlIO_printf(Perl_error_log,
3287 "\tLANGUAGE = %c%s%c,\n",
3288 language ? '"' : '(',
3289 language ? language : "unset",
3290 language ? '"' : ')');
7d4bcc4a 3291# endif
98994639 3292
49c85077
KW
3293 PerlIO_printf(Perl_error_log,
3294 "\tLC_ALL = %c%s%c,\n",
3295 lc_all ? '"' : '(',
3296 lc_all ? lc_all : "unset",
3297 lc_all ? '"' : ')');
98994639 3298
7d4bcc4a
KW
3299# if defined(USE_ENVIRON_ARRAY)
3300
49c85077 3301 {
cd999af9 3302 char **e;
d5e32b93
KW
3303
3304 /* Look through the environment for any variables of the
3305 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3306 * already handled above. These are assumed to be locale
3307 * settings. Output them and their values. */
cd999af9 3308 for (e = environ; *e; e++) {
d5e32b93
KW
3309 const STRLEN prefix_len = sizeof("LC_") - 1;
3310 STRLEN uppers_len;
3311
cd999af9 3312 if ( strBEGINs(*e, "LC_")
c8b388b0 3313 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3314 && (uppers_len = strspn(*e + prefix_len,
3315 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3316 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3317 {
3318 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3319 (int) (prefix_len + uppers_len), *e,
3320 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3321 }
3322 }
49c85077 3323 }
7d4bcc4a
KW
3324
3325# else
3326
49c85077
KW
3327 PerlIO_printf(Perl_error_log,
3328 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3329
3330# endif
98994639 3331
49c85077
KW
3332 PerlIO_printf(Perl_error_log,
3333 "\tLANG = %c%s%c\n",
3334 lang ? '"' : '(',
3335 lang ? lang : "unset",
3336 lang ? '"' : ')');
98994639 3337
49c85077
KW
3338 PerlIO_printf(Perl_error_log,
3339 " are supported and installed on your system.\n");
3340 }
98994639 3341
65ebb059 3342 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3343 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3344 * usually change the upper bound of the loop we are in.
3345 *
3346 * Since the system's default way of setting the locale has not
3347 * found one that works, We use Perl's defined ordering: LC_ALL,
3348 * LANG, and the C locale. We don't try the same locale twice, so
3349 * don't add to the list if already there. (On POSIX systems, the
3350 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3351 * but there's no harm done by doing it explicitly.
3352 *
3353 * Note that this tries the LC_ALL environment variable even on
3354 * systems which have no LC_ALL locale setting. This may or may
3355 * not have been originally intentional, but there's no real need
3356 * to change the behavior. */
65ebb059
KW
3357 if (lc_all) {
3358 for (j = 0; j < trial_locales_count; j++) {
3359 if (strEQ(lc_all, trial_locales[j])) {
3360 goto done_lc_all;
3361 }
3362 }
3363 trial_locales[trial_locales_count++] = lc_all;
3364 }
3365 done_lc_all:
98994639 3366
65ebb059
KW
3367 if (lang) {
3368 for (j = 0; j < trial_locales_count; j++) {
3369 if (strEQ(lang, trial_locales[j])) {
3370 goto done_lang;
3371 }
3372 }
3373 trial_locales[trial_locales_count++] = lang;
3374 }
3375 done_lang:
3376
7d4bcc4a
KW
3377# if defined(WIN32) && defined(LC_ALL)
3378
65ebb059
KW
3379 /* For Windows, we also try the system default locale before "C".
3380 * (If there exists a Windows without LC_ALL we skip this because
3381 * it gets too complicated. For those, the "C" is the next
3382 * fallback possibility). The "" is the same as the 0th element of
3383 * the array, but the code at the loop above knows to treat it
3384 * differently when not the 0th */
3385 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3386
3387# endif
65ebb059
KW
3388
3389 for (j = 0; j < trial_locales_count; j++) {
3390 if (strEQ("C", trial_locales[j])) {
3391 goto done_C;
3392 }
3393 }
3394 trial_locales[trial_locales_count++] = "C";
98994639 3395
65ebb059
KW
3396 done_C: ;
3397 } /* end of first time through the loop */
98994639 3398
7d4bcc4a
KW
3399# ifdef WIN32
3400
65ebb059 3401 next_iteration: ;
7d4bcc4a
KW
3402
3403# endif
65ebb059
KW
3404
3405 } /* end of looping through the trial locales */
3406
3407 if (ok < 1) { /* If we tried to fallback */
3408 const char* msg;
3409 if (! setlocale_failure) { /* fallback succeeded */
3410 msg = "Falling back to";
3411 }
3412 else { /* fallback failed */
e5f10d49 3413 unsigned int j;
98994639 3414
65ebb059
KW
3415 /* We dropped off the end of the loop, so have to decrement i to
3416 * get back to the value the last time through */
3417 i--;
98994639 3418
65ebb059
KW
3419 ok = -1;
3420 msg = "Failed to fall back to";
3421
3422 /* To continue, we should use whatever values we've got */
7d4bcc4a 3423
e5f10d49
KW
3424 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3425 Safefree(curlocales[j]);
3426 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3427 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3428 }
65ebb059
KW
3429 }
3430
3431 if (locwarn) {
3432 const char * description;
3433 const char * name = "";
3434 if (strEQ(trial_locales[i], "C")) {
3435 description = "the standard locale";
3436 name = "C";
3437 }
7d4bcc4a
KW
3438
3439# ifdef SYSTEM_DEFAULT_LOCALE
3440
65ebb059
KW
3441 else if (strEQ(trial_locales[i], "")) {
3442 description = "the system default locale";
3443 if (system_default_locale) {
3444 name = system_default_locale;
3445 }
3446 }
7d4bcc4a
KW
3447
3448# endif /* SYSTEM_DEFAULT_LOCALE */
3449
65ebb059
KW
3450 else {
3451 description = "a fallback locale";
3452 name = trial_locales[i];
3453 }
3454 if (name && strNE(name, "")) {
3455 PerlIO_printf(Perl_error_log,
3456 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3457 }
3458 else {
3459 PerlIO_printf(Perl_error_log,
3460 "perl: warning: %s %s.\n", msg, description);
3461 }
3462 }
3463 } /* End of tried to fallback */
98994639 3464
e5f10d49
KW
3465 /* Done with finding the locales; update our records */
3466
7d4bcc4a
KW
3467# ifdef USE_LOCALE_CTYPE
3468
948523db 3469 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3470
e5f10d49 3471# endif
7d4bcc4a
KW
3472# ifdef USE_LOCALE_COLLATE
3473
948523db 3474 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3475
e5f10d49 3476# endif
7d4bcc4a
KW
3477# ifdef USE_LOCALE_NUMERIC
3478
948523db 3479 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3480
3481# endif
3482
948523db 3483 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3484
e9bc6d6b 3485# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3486
3487 /* This caches whether each category's locale is UTF-8 or not. This
3488 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3489 * initialization time before any threads have started, but not later
3490 * unless thread-safe operations are used.
47280b20
KW
3491 * Caching means that if the program heeds our dictate not to change
3492 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3493 * it may get queried without having to change locales. If the
3494 * environment is such that all categories have the same locale, this
3495 * isn't needed, as the code will not change the locale; but this
3496 * handles the uncommon case where the environment has disparate
3497 * locales for the categories */
47280b20
KW
3498 (void) _is_cur_LC_category_utf8(categories[i]);
3499
3500# endif
3501
e5f10d49
KW
3502 Safefree(curlocales[i]);
3503 }
b310b053 3504
7d4bcc4a
KW
3505# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3506
49c85077 3507 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3508 * locale is UTF-8. The call to new_ctype() just above has already
3509 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3510 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3511 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3512 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3513 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3514
a05d7ebb 3515 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3516 This is an alternative to using the -C command line switch
3517 (the -C if present will override this). */
3518 {
dd374669 3519 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3520 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3521 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3522 PL_utf8cache = -1;
b310b053
JH
3523 }
3524
7d4bcc4a 3525# endif
7d4bcc4a
KW
3526# ifdef __GLIBC__
3527
175c4cf9 3528 Safefree(language);
7d4bcc4a
KW
3529
3530# endif
175c4cf9
KW
3531
3532 Safefree(lc_all);
3533 Safefree(lang);
3534
e3305790 3535#endif /* USE_LOCALE */
2fcc0ca9 3536#ifdef DEBUGGING
7d4bcc4a 3537
2fcc0ca9 3538 /* So won't continue to output stuff */
27cdc72e 3539 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3540
2fcc0ca9
KW
3541#endif
3542
98994639
HS
3543 return ok;
3544}
3545
98994639
HS
3546#ifdef USE_LOCALE_COLLATE
3547
a4a439fb 3548char *
a4a439fb
KW
3549Perl__mem_collxfrm(pTHX_ const char *input_string,
3550 STRLEN len, /* Length of 'input_string' */
3551 STRLEN *xlen, /* Set to length of returned string
3552 (not including the collation index
3553 prefix) */
3554 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3555 )
98994639 3556{
a4a439fb
KW
3557
3558 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3559 * differences. First, it handles embedded NULs. Second, it allocates a bit
3560 * more memory than needed for the transformed data itself. The real
55e5378d 3561 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3562 * the length of that, and doesn't include the collation index size.
3563 * Please see sv_collxfrm() to see how this is used. */
3564
55e5378d
KW
3565#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3566
6696cfa7
KW
3567 char * s = (char *) input_string;
3568 STRLEN s_strlen = strlen(input_string);
79f120c8 3569 char *xbuf = NULL;
55e5378d 3570 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3571 STRLEN length_in_chars;
c664130f 3572 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3573
a4a439fb
KW
3574 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3575
3576 /* Must be NUL-terminated */
3577 assert(*(input_string + len) == '\0');
7918f24d 3578
79f120c8
KW
3579 /* If this locale has defective collation, skip */
3580 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3581 DEBUG_L(PerlIO_printf(Perl_debug_log,
3582 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3583 goto bad;
3584 }
3585
6696cfa7
KW
3586 /* Replace any embedded NULs with the control that sorts before any others.
3587 * This will give as good as possible results on strings that don't
3588 * otherwise contain that character, but otherwise there may be
3589 * less-than-perfect results with that character and NUL. This is
fdc080f3 3590 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3591 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3592 NUL */
6696cfa7
KW
3593 char * e = s + len;
3594 char * sans_nuls;
fdc080f3 3595 STRLEN sans_nuls_len;
94762aa0 3596 int try_non_controls;
afc4976f
KW
3597 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3598 making sure 2nd byte is NUL.
3599 */
3600 STRLEN this_replacement_len;
3601
1e4c9676
KW
3602 /* If we don't know what non-NUL control character sorts lowest for
3603 * this locale, find it */
f28f4d2a 3604 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3605 int j;
afc4976f 3606 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3607 includes the collation index
3608 prefixed. */
3609
91c0e2e0 3610 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3611
3612 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3613 * NUL, in which case we instead look for any character. Controls
3614 * are preferred because collation order is, in general, context
3615 * sensitive, with adjoining characters affecting the order, and
3616 * controls are less likely to have such interactions, allowing the
3617 * NUL-replacement to stand on its own. (Another way to look at it
3618 * is to imagine what would happen if the NUL were replaced by a
3619 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3620 for (try_non_controls = 0;
3621 try_non_controls < 2;
3622 try_non_controls++)
3623 {
d4ff9586
KW
3624 /* Look through all legal code points (NUL isn't) */
3625 for (j = 1; j < 256; j++) {
3626 char * x; /* j's xfrm plus collation index */
3627 STRLEN x_len; /* length of 'x' */
3628 STRLEN trial_len = 1;
736a4fed 3629 char cur_source[] = { '\0', '\0' };
d4ff9586 3630
736a4fed
KW
3631 /* Skip non-controls the first time through the loop. The
3632 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3633 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3634 ? ! isCNTRL_L1(j)
3635 : ! isCNTRL_LC(j))
3636 {
d4ff9586 3637 continue;
6696cfa7 3638 }
6696cfa7 3639
736a4fed
KW
3640 /* Create a 1-char string of the current code point */
3641 cur_source[0] = (char) j;
3642
d4ff9586
KW
3643 /* Then transform it */
3644 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3645 0 /* The string is not in UTF-8 */);
6696cfa7 3646
1e4c9676 3647 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3648 * */
3649 if (! x) {
3650 continue;
3651 }
6696cfa7 3652
d4ff9586
KW
3653 /* If this character's transformation is lower than
3654 * the current lowest, this one becomes the lowest */
3655 if ( cur_min_x == NULL
3656 || strLT(x + COLLXFRM_HDR_LEN,
3657 cur_min_x + COLLXFRM_HDR_LEN))
3658 {
f28f4d2a 3659 PL_strxfrm_NUL_replacement = j;
d4ff9586 3660 cur_min_x = x;
d4ff9586
KW
3661 }
3662 else {
3663 Safefree(x);
3664 }
1e4c9676 3665 } /* end of loop through all 255 characters */
6696cfa7 3666
1e4c9676 3667 /* Stop looking if found */
94762aa0
KW
3668 if (cur_min_x) {
3669 break;
3670 }
3671
3672 /* Unlikely, but possible, if there aren't any controls that
3673 * work in the locale, repeat the loop, looking for any
3674 * character that works */
3675 DEBUG_L(PerlIO_printf(Perl_debug_log,
3676 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3677 } /* End of loop to try first the controls, then any char */
6696cfa7 3678
94762aa0
KW
3679 if (! cur_min_x) {
3680 DEBUG_L(PerlIO_printf(Perl_debug_log,
3681 "_mem_collxfrm: Couldn't find any character to replace"
3682 " embedded NULs in locale %s with", PL_collation_name));
3683 goto bad;
58eebef2
KW
3684 }
3685
94762aa0
KW
3686 DEBUG_L(PerlIO_printf(Perl_debug_log,
3687 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3688 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3689
6696cfa7 3690 Safefree(cur_min_x);
1e4c9676 3691 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3692
3693 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3694 * UTF8-ness of the original */
f28f4d2a
KW
3695 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3696 this_replacement_char[0] =
3697 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3698 this_replacement_char[1] =
3699 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3700 this_replacement_len = 2;
3701 }
3702 else {
f28f4d2a 3703 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3704 /* this_replacement_char[1] = '\0' was done at initialization */
3705 this_replacement_len = 1;
6696cfa7
KW
3706 }
3707
3708 /* The worst case length for the replaced string would be if every
3709 * character in it is NUL. Multiply that by the length of each
3710 * replacement, and allow for a trailing NUL */
afc4976f 3711 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3712 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3713 *sans_nuls = '\0';
3714
6696cfa7
KW
3715 /* Replace each NUL with the lowest collating control. Loop until have
3716 * exhausted all the NULs */
3717 while (s + s_strlen < e) {
6069d6c5 3718 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3719
3720 /* Do the actual replacement */
6069d6c5 3721 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
3722
3723 /* Move past the input NUL */
3724 s += s_strlen + 1;
3725 s_strlen = strlen(s);
3726 }
3727
3728 /* And add anything that trails the final NUL */
6069d6c5 3729 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3730
3731 /* Switch so below we transform this modified string */
3732 s = sans_nuls;
3733 len = strlen(s);
1e4c9676 3734 } /* End of replacing NULs */
6696cfa7 3735
a4a439fb
KW
3736 /* Make sure the UTF8ness of the string and locale match */
3737 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 3738 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
3739 const char * const t = s; /* Temporary so we can later find where the
3740 input was */
3741
3742 /* Here they don't match. Change the string's to be what the locale is
3743 * expecting */
3744
3745 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3746 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3747 utf8 = TRUE;
3748 }
3749 else { /* locale is not UTF-8; but input is; downgrade the input */
3750
3751 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3752
3753 /* If the downgrade was successful we are done, but if the input
3754 * contains things that require UTF-8 to represent, have to do
3755 * damage control ... */
3756 if (UNLIKELY(utf8)) {
3757
3758 /* What we do is construct a non-UTF-8 string with
3759 * 1) the characters representable by a single byte converted
3760 * to be so (if necessary);
3761 * 2) and the rest converted to collate the same as the
3762 * highest collating representable character. That makes
3763 * them collate at the end. This is similar to how we
3764 * handle embedded NULs, but we use the highest collating
3765 * code point instead of the smallest. Like the NUL case,
3766 * this isn't perfect, but is the best we can reasonably
3767 * do. Every above-255 code point will sort the same as
3768 * the highest-sorting 0-255 code point. If that code
3769 * point can combine in a sequence with some other code
3770 * points for weight calculations, us changing something to
3771 * be it can adversely affect the results. But in most
3772 * cases, it should work reasonably. And note that this is
3773 * really an illegal situation: using code points above 255
3774 * on a locale where only 0-255 are valid. If two strings
3775 * sort entirely equal, then the sort order for the
3776 * above-255 code points will be in code point order. */
3777
3778 utf8 = FALSE;
3779
3780 /* If we haven't calculated the code point with the maximum
3781 * collating order for this locale, do so now */
3782 if (! PL_strxfrm_max_cp) {
3783 int j;
3784
3785 /* The current transformed string that collates the
3786 * highest (except it also includes the prefixed collation
3787 * index. */
3788 char * cur_max_x = NULL;
3789
3790 /* Look through all legal code points (NUL isn't) */
3791 for (j = 1; j < 256; j++) {
3792 char * x;
3793 STRLEN x_len;
736a4fed 3794 char cur_source[] = { '\0', '\0' };
a4a439fb 3795
736a4fed
KW
3796 /* Create a 1-char string of the current code point */
3797 cur_source[0] = (char) j;
a4a439fb
KW
3798
3799 /* Then transform it */
3800 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3801
3802 /* If something went wrong (which it shouldn't), just
3803 * ignore this code point */
94762aa0 3804 if (! x) {
a4a439fb
KW
3805 continue;
3806 }
3807
3808 /* If this character's transformation is higher than
3809 * the current highest, this one becomes the highest */
3810 if ( cur_max_x == NULL
55e5378d
KW
3811 || strGT(x + COLLXFRM_HDR_LEN,
3812 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
3813 {
3814 PL_strxfrm_max_cp = j;
3815 cur_max_x = x;
3816 }
3817 else {
3818 Safefree(x);
3819 }
3820 }
3821
94762aa0
KW
3822 if (! cur_max_x) {
3823 DEBUG_L(PerlIO_printf(Perl_debug_log,
3824 "_mem_collxfrm: Couldn't find any character to"
3825 " replace above-Latin1 chars in locale %s with",
3826 PL_collation_name));
3827 goto bad;
3828 }
3829
58eebef2
KW
3830 DEBUG_L(PerlIO_printf(Perl_debug_log,
3831 "_mem_collxfrm: highest 1-byte collating character"
3832 " in locale %s is 0x%02X\n",
3833 PL_collation_name,
3834 PL_strxfrm_max_cp));
58eebef2 3835
a4a439fb
KW
3836 Safefree(cur_max_x);
3837 }
3838
3839 /* Here we know which legal code point collates the highest.
3840 * We are ready to construct the non-UTF-8 string. The length
3841 * will be at least 1 byte smaller than the input string
3842 * (because we changed at least one 2-byte character into a
3843 * single byte), but that is eaten up by the trailing NUL */
3844 Newx(s, len, char);
3845
3846 {
3847 STRLEN i;
3848 STRLEN d= 0;
042d9e50 3849 char * e = (char *) t + len;
a4a439fb
KW
3850
3851 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3852 U8 cur_char = t[i];
3853 if (UTF8_IS_INVARIANT(cur_char)) {
3854 s[d++] = cur_char;
3855 }
042d9e50 3856 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
3857 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3858 }
3859 else { /* Replace illegal cp with highest collating
3860 one */
3861 s[d++] = PL_strxfrm_max_cp;
3862 }
3863 }
3864 s[d++] = '\0';
3865 Renew(s, d, char); /* Free up unused space */
3866 }
3867 }
3868 }
3869
3870 /* Here, we have constructed a modified version of the input. It could
3871 * be that we already had a modified copy before we did this version.
3872 * If so, that copy is no longer needed */
3873 if (t != input_string) {
3874 Safefree(t);
3875 }
3876 }
3877
17f41037
KW
3878 length_in_chars = (utf8)
3879 ? utf8_length((U8 *) s, (U8 *) s + len)
3880 : len;
3881
59c018b9
KW
3882 /* The first element in the output is the collation id, used by
3883 * sv_collxfrm(); then comes the space for the transformed string. The
3884 * equation should give us a good estimate as to how much is needed */
55e5378d 3885 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 3886 + PL_collxfrm_base
17f41037 3887 + (PL_collxfrm_mult * length_in_chars);
a02a5408 3888 Newx(xbuf, xAlloc, char);
c7202dee
KW
3889 if (UNLIKELY(! xbuf)) {
3890 DEBUG_L(PerlIO_printf(Perl_debug_log,
3891 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 3892 goto bad;
c7202dee 3893 }
98994639 3894
d35fca5f 3895 /* Store the collation id */
98994639 3896 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
3897
3898 /* Then the transformation of the input. We loop until successful, or we
3899 * give up */
4ebeff16 3900 for (;;) {
1adab0a7 3901
55e5378d 3902 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);