This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Check return code
[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;
6435f98d
KW
872 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
873 {
874 Safefree(env_override);
875 return NULL;
876 }
e9bc6d6b
KW
877
878 if (strNE(this_locale, default_name)) {
879 did_override = TRUE;
880 }
881
882 Safefree(env_override);
883 }
884
885 /* If all the categories are the same, we can set LC_ALL to
886 * that */
887 if (! did_override) {
888 locale = default_name;
889 }
890 else {
891
892 /* Here, LC_ALL is no longer valid, as some individual
893 * categories don't match it. We call ourselves
894 * recursively, as that will execute the code that
895 * generates the proper locale string for this situation.
896 * We don't do the remainder of this function, as that is
897 * to update our records, and we've just done that for the
898 * individual categories in the loop above, and doing so
899 * would cause LC_ALL to be done as well */
900 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
901 }
902 }
903 }
904 }
905 else if (strchr(locale, ';')) {
906
907 /* LC_ALL may actually incude a conglomeration of various categories.
908 * Without querylocale, this code uses the glibc (as of this writing)
909 * syntax for representing that, but that is not a stable API, and
910 * other platforms do it differently, so we have to handle all cases
911 * ourselves */
912
913 const char * s = locale;
914 const char * e = locale + strlen(locale);
915 const char * p = s;
916 const char * category_end;
917 const char * name_start;
918 const char * name_end;
919
920 while (s < e) {
921 unsigned int i;
922
923 /* Parse through the category */
924 while (isWORDCHAR(*p)) {
925 p++;
926 }
927 category_end = p;
928
929 if (*p++ != '=') {
930 Perl_croak(aTHX_
931 "panic: %s: %d: Unexpected character in locale name '%02X",
932 __FILE__, __LINE__, *(p-1));
933 }
934
935 /* Parse through the locale name */
936 name_start = p;
937 while (isGRAPH(*p) && *p != ';') {
938 p++;
939 }
940 name_end = p;
941
942 if (*p++ != ';') {
943 Perl_croak(aTHX_
944 "panic: %s: %d: Unexpected character in locale name '%02X",
945 __FILE__, __LINE__, *(p-1));
946 }
947
948 /* Find the index of the category name in our lists */
949 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 950 char * individ_locale;
e9bc6d6b
KW
951
952 /* Keep going if this isn't the index. The strnNE() avoids a
953 * Perl_form(), but would fail if ever a category name could be
954 * a substring of another one, like if there were a
955 * "LC_TIME_DATE" */
956 if strnNE(s, category_names[i], category_end - s) {
957 continue;
958 }
959
960 /* If this index is for the single category we're changing, we
961 * have found the locale to set it to. */
962 if (category == categories[i]) {
963 locale = Perl_form(aTHX_ "%.*s",
964 (int) (name_end - name_start),
965 name_start);
966 goto ready_to_set;
967 }
968
3c76a412
KW
969 assert(category == LC_ALL);
970 individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
6435f98d
KW
971 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
972 {
973 return NULL;
974 }
e9bc6d6b
KW
975 }
976
977 s = p;
978 }
979
980 /* Here we have set all the individual categories by recursive calls.
981 * These collectively should have fixed up LC_ALL, so can just query
982 * what that now is */
983 assert(category == LC_ALL);
984
985 return do_setlocale_c(LC_ALL, NULL);
986 }
987
988 ready_to_set: ;
989
990# endif /* end of ! querylocale */
991
992 /* Ready to create a new locale by modification of the exising one */
993 new_obj = newlocale(mask, locale, old_obj);
994
995 if (! new_obj) {
996 dSAVE_ERRNO;
997
998# ifdef DEBUGGING
999
1000 if (DEBUG_L_TEST || debug_initialization) {
1001 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1002 }
1003
1004# endif
1005
1006 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1007
1008# ifdef DEBUGGING
1009
1010 if (DEBUG_L_TEST || debug_initialization) {
1011 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1012 }
1013
1014# endif
1015
1016 }
1017 RESTORE_ERRNO;
1018 return NULL;
1019 }
1020
1021# ifdef DEBUGGING
1022
1023 if (DEBUG_Lv_TEST || debug_initialization) {
1024 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1025 }
1026
1027# endif
1028
1029 /* And switch into it */
1030 if (! uselocale(new_obj)) {
1031 dSAVE_ERRNO;
1032
1033# ifdef DEBUGGING
1034
1035 if (DEBUG_L_TEST || debug_initialization) {
1036 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1037 }
1038
1039# endif
1040
1041 if (! uselocale(old_obj)) {
1042
1043# ifdef DEBUGGING
1044
1045 if (DEBUG_L_TEST || debug_initialization) {
1046 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1047 }
1048
1049# endif
1050
1051 }
1052 freelocale(new_obj);
1053 RESTORE_ERRNO;
1054 return NULL;
1055 }
1056
1057# ifdef DEBUGGING
1058
1059 if (DEBUG_Lv_TEST || debug_initialization) {
1060 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1061 }
1062
1063# endif
1064
1065 /* We are done, except for updating our records (if the system doesn't keep
1066 * them) and in the case of locale "", we don't actually know what the
1067 * locale that got switched to is, as it came from the environment. So
1068 * have to find it */
1069
1070# ifdef HAS_QUERYLOCALE
1071
1072 if (strEQ(locale, "")) {
1073 locale = querylocale(mask, new_obj);
1074 }
1075
1076# else
1077
1078 /* Here, 'locale' is the return value */
1079
1080 /* Without querylocale(), we have to update our records */
1081
1082 if (category == LC_ALL) {
1083 unsigned int i;
1084
1085 /* For LC_ALL, we change all individual categories to correspond */
1086 /* PL_curlocales is a parallel array, so has same
1087 * length as 'categories' */
1088 for (i = 0; i <= LC_ALL_INDEX; i++) {
1089 Safefree(PL_curlocales[i]);
1090 PL_curlocales[i] = savepv(locale);
1091 }
1092 }
1093 else {
1094
1095 /* For a single category, if it's not the same as the one in LC_ALL, we
1096 * nullify LC_ALL */
1097
1098 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1099 Safefree(PL_curlocales[LC_ALL_INDEX]);
1100 PL_curlocales[LC_ALL_INDEX] = NULL;
1101 }
1102
1103 /* Then update the category's record */
1104 Safefree(PL_curlocales[index]);
1105 PL_curlocales[index] = savepv(locale);
1106 }
1107
1108# endif
1109
1110 return locale;
1111}
1112
1113#endif /* USE_POSIX_2008_LOCALE */
1114
1115#if 0 /* Code that was to emulate thread-safe locales on platforms that
1116 didn't natively support them */
1117
1118/* The way this would work is that we would keep a per-thread list of the
1119 * correct locale for that thread. Any operation that was locale-sensitive
1120 * would have to be changed so that it would look like this:
1121 *
1122 * LOCALE_LOCK;
1123 * setlocale to the correct locale for this operation
1124 * do operation
1125 * LOCALE_UNLOCK
1126 *
1127 * This leaves the global locale in the most recently used operation's, but it
1128 * was locked long enough to get the result. If that result is static, it
1129 * needs to be copied before the unlock.
1130 *
1131 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1132 * the setup, but are no-ops when not needed, and similarly,
1133 * END_LOCALE_DEPENDENT_OP for the tear-down
1134 *
1135 * But every call to a locale-sensitive function would have to be changed, and
1136 * if a module didn't cooperate by using the mutex, things would break.
1137 *
1138 * This code was abandoned before being completed or tested, and is left as-is
1139*/
1140
1141# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1142# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1143
1144STATIC char *
1145S_locking_setlocale(pTHX_
1146 const int category,
1147 const char * locale,
1148 int index,
1149 const bool is_index_valid
1150 )
1151{
1152 /* This function kind of performs a setlocale() on just the current thread;
1153 * thus it is kind of thread-safe. It does this by keeping a thread-level
1154 * array of the current locales for each category. Every time a locale is
1155 * switched to, it does the switch globally, but updates the thread's
1156 * array. A query as to what the current locale is just returns the
1157 * appropriate element from the array, and doesn't actually call the system
1158 * setlocale(). The saving into the array is done in an uninterruptible
1159 * section of code, so is unaffected by whatever any other threads might be
1160 * doing.
1161 *
1162 * All locale-sensitive operations must work by first starting a critical
1163 * section, then switching to the thread's locale as kept by this function,
1164 * and then doing the operation, then ending the critical section. Thus,
1165 * each gets done in the appropriate locale. simulating thread-safety.
1166 *
1167 * This function takes the same parameters, 'category' and 'locale', that
1168 * the regular setlocale() function does, but it also takes two additional
1169 * ones. This is because as described earlier. If we know on input the
1170 * index corresponding to the category into the array where we store the
1171 * current locales, we don't have to calculate it. If the caller knows at
1172 * compile time what the index is, it it can pass it, setting
1173 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1174 *
1175 */
1176
1177 /* If the input index might be incorrect, calculate the correct one */
1178 if (! is_index_valid) {
1179 unsigned int i;
1180
1181 if (DEBUG_Lv_TEST || debug_initialization) {
1182 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1183 }
1184
1185 for (i = 0; i <= LC_ALL_INDEX; i++) {
1186 if (category == categories[i]) {
1187 index = i;
1188 goto found_index;
1189 }
1190 }
1191
1192 /* Here, we don't know about this category, so can't handle it.
1193 * XXX best we can do is to unsafely set this
1194 * XXX warning */
1195
1196 return my_setlocale(category, locale);
1197
1198 found_index: ;
1199
1200 if (DEBUG_Lv_TEST || debug_initialization) {
1201 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1202 }
1203 }
1204
1205 /* For a query, just return what's in our records */
1206 if (new_locale == NULL) {
1207 return curlocales[index];
1208 }
1209
1210
1211 /* Otherwise, we need to do the switch, and save the result, all in a
1212 * critical section */
1213
1214 Safefree(curlocales[[index]]);
1215
1216 /* It might be that this is called from an already-locked section of code.
1217 * We would have to detect and skip the LOCK/UNLOCK if so */
1218 LOCALE_LOCK;
1219
1220 curlocales[index] = savepv(my_setlocale(category, new_locale));
1221
1222 if (strEQ(new_locale, "")) {
1223
1224#ifdef LC_ALL
1225
1226 /* The locale values come from the environment, and may not all be the
1227 * same, so for LC_ALL, we have to update all the others, while the
1228 * mutex is still locked */
1229
1230 if (category == LC_ALL) {
1231 unsigned int i;
1232 for (i = 0; i < LC_ALL_INDEX) {
1233 curlocales[i] = my_setlocale(categories[i], NULL);
1234 }
1235 }
1236 }
1237
1238#endif
1239
1240 LOCALE_UNLOCK;
1241
1242 return curlocales[index];
1243}
1244
1245#endif
837ce802 1246
a4f00dcc 1247STATIC void
86799d2d 1248S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1249{
86799d2d
KW
1250 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1251 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1252
86799d2d
KW
1253#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1254 || defined(HAS_NL_LANGINFO))
98994639 1255
87f8e8e7
KW
1256 const char * radix = (use_locale)
1257 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1258 /* FALSE => already in dest locale */
1259 : ".";
2213a3be 1260
87f8e8e7 1261 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1262
87f8e8e7
KW
1263 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1264 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1265 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1266 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1267 && _is_cur_LC_category_utf8(LC_NUMERIC))
1268 {
1269 SvUTF8_on(PL_numeric_radix_sv);
1270 }
86799d2d
KW
1271
1272# ifdef DEBUGGING
7d4bcc4a 1273
2fcc0ca9
KW
1274 if (DEBUG_L_TEST || debug_initialization) {
1275 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1276 SvPVX(PL_numeric_radix_sv),
1277 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1278 }
69014004 1279
86799d2d
KW
1280# endif
1281#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1282
98994639
HS
1283}
1284
398eeea9
KW
1285STATIC void
1286S_new_numeric(pTHX_ const char *newnum)
98994639 1287{
7d4bcc4a
KW
1288
1289#ifndef USE_LOCALE_NUMERIC
1290
1291 PERL_UNUSED_ARG(newnum);
1292
1293#else
0d071d52 1294
291a84fb 1295 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1296 * core Perl this and that 'newnum' is the name of the new locale.
1297 * It installs this locale as the current underlying default.
1298 *
1299 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1300 * set_numeric_underlying() and set_numeric_standard() functions, which
1301 * should probably not be called directly, but only via macros like
0d071d52
KW
1302 * SET_NUMERIC_STANDARD() in perl.h.
1303 *
1304 * The toggling is necessary mainly so that a non-dot radix decimal point
1305 * character can be output, while allowing internal calculations to use a
1306 * dot.
1307 *
1308 * This sets several interpreter-level variables:
bb304765 1309 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1310 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1311 * that the current locale is the program's underlying
1312 * locale
1313 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1314 * that the current locale is the C locale or
1315 * indistinguishable from the C locale. If non-zero, it
1316 * is in C; if > 1, it means it may not be toggled away
7738054c 1317 * from C.
4c68b815
KW
1318 * PL_numeric_underlying_is_standard A bool kept by this function
1319 * indicating that the underlying locale and the standard
1320 * C locale are indistinguishable for the purposes of
1321 * LC_NUMERIC. This happens when both of the above two
1322 * variables are true at the same time. (Toggling is a
1323 * no-op under these circumstances.) This variable is
1324 * used to avoid having to recalculate.
398eeea9 1325 */
0d071d52 1326
b03f34cf 1327 char *save_newnum;
98994639
HS
1328
1329 if (! newnum) {
43c5f42d
NC
1330 Safefree(PL_numeric_name);
1331 PL_numeric_name = NULL;
98994639 1332 PL_numeric_standard = TRUE;
892e6465 1333 PL_numeric_underlying = TRUE;
4c68b815 1334 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1335 return;
1336 }
1337
b03f34cf 1338 save_newnum = stdize_locale(savepv(newnum));
892e6465 1339 PL_numeric_underlying = TRUE;
4c68b815
KW
1340 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1341
1342 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1343 * them */
1344 if (! PL_numeric_standard) {
1345 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1346 FALSE /* Don't toggle locale */ ))
1347 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1348 FALSE)));
1349 }
abe1abcf 1350
4c68b815 1351 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1352 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1353 Safefree(PL_numeric_name);
b03f34cf 1354 PL_numeric_name = save_newnum;
b03f34cf 1355 }
abe1abcf
KW
1356 else {
1357 Safefree(save_newnum);
1358 }
4c28b29c 1359
4c68b815
KW
1360 PL_numeric_underlying_is_standard = PL_numeric_standard;
1361
7e5377f7 1362# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1363
1364 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1365 PL_numeric_name,
1366 PL_underlying_numeric_obj);
1367
1368#endif
1369
4c68b815
KW
1370 if (DEBUG_L_TEST || debug_initialization) {
1371 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1372 }
1373
4c28b29c
KW
1374 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1375 * have to worry about the radix being a non-dot. (Core operations that
1376 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1377 if (PL_numeric_standard) {
1378 set_numeric_radix(0);
1379 }
1380 else {
1381 set_numeric_standard();
1382 }
4c28b29c 1383
98994639 1384#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1385
98994639
HS
1386}
1387
1388void
1389Perl_set_numeric_standard(pTHX)
1390{
7d4bcc4a 1391
98994639 1392#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1393
28c1bf33
KW
1394 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1395 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1396 * macro avoids calling this routine if toggling isn't necessary according
1397 * to our records (which could be wrong if some XS code has changed the
1398 * locale behind our back) */
0d071d52 1399
837ce802 1400 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1401 PL_numeric_standard = TRUE;
4c68b815 1402 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1403 set_numeric_radix(0);
7d4bcc4a
KW
1404
1405# ifdef DEBUGGING
1406
2fcc0ca9
KW
1407 if (DEBUG_L_TEST || debug_initialization) {
1408 PerlIO_printf(Perl_debug_log,
58e4a467 1409 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1410 }
98994639 1411
7d4bcc4a 1412# endif
98994639 1413#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1414
98994639
HS
1415}
1416
1417void
5792c642 1418Perl_set_numeric_underlying(pTHX)
98994639 1419{
7d4bcc4a 1420
98994639 1421#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1422
28c1bf33 1423 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1424 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1425 * instead of calling this directly. The macro avoids calling this routine
1426 * if toggling isn't necessary according to our records (which could be
1427 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1428
837ce802 1429 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1430 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1431 PL_numeric_underlying = TRUE;
3c62fd46 1432 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1433
1434# ifdef DEBUGGING
1435
2fcc0ca9
KW
1436 if (DEBUG_L_TEST || debug_initialization) {
1437 PerlIO_printf(Perl_debug_log,
58e4a467 1438 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1439 PL_numeric_name);
1440 }
98994639 1441
7d4bcc4a 1442# endif
98994639 1443#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1444
98994639
HS
1445}
1446
1447/*
1448 * Set up for a new ctype locale.
1449 */
a4f00dcc
KW
1450STATIC void
1451S_new_ctype(pTHX_ const char *newctype)
98994639 1452{
7d4bcc4a
KW
1453
1454#ifndef USE_LOCALE_CTYPE
1455
1456 PERL_ARGS_ASSERT_NEW_CTYPE;
1457 PERL_UNUSED_ARG(newctype);
1458 PERL_UNUSED_CONTEXT;
1459
1460#else
0d071d52 1461
291a84fb 1462 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1463 * core Perl this and that 'newctype' is the name of the new locale.
1464 *
1465 * This function sets up the folding arrays for all 256 bytes, assuming
1466 * that tofold() is tolc() since fold case is not a concept in POSIX,
1467 *
1468 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1469 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1470 * this function should be called directly only from this file and from
0d071d52
KW
1471 * POSIX::setlocale() */
1472
27da23d5 1473 dVAR;
013f4e03 1474 unsigned int i;
98994639 1475
8b7358b9
KW
1476 /* Don't check for problems if we are suppressing the warnings */
1477 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1478
7918f24d
NC
1479 PERL_ARGS_ASSERT_NEW_CTYPE;
1480
215c5139
KW
1481 /* We will replace any bad locale warning with 1) nothing if the new one is
1482 * ok; or 2) a new warning for the bad new locale */
1483 if (PL_warn_locale) {
1484 SvREFCNT_dec_NN(PL_warn_locale);
1485 PL_warn_locale = NULL;
1486 }
1487
c1284011 1488 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1489
1490 /* A UTF-8 locale gets standard rules. But note that code still has to
1491 * handle this specially because of the three problematic code points */
1492 if (PL_in_utf8_CTYPE_locale) {
1493 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1494 }
8b7358b9
KW
1495
1496 /* We don't populate the other lists if a UTF-8 locale, but do check that
1497 * everything works as expected, unless checking turned off */
1498 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1499 /* Assume enough space for every character being bad. 4 spaces each
1500 * for the 94 printable characters that are output like "'x' "; and 5
1501 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1502 * NUL */
8b7358b9 1503 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1504 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1505 to start */
1506 unsigned int bad_count = 0; /* Count of bad characters */
1507
baa60164 1508 for (i = 0; i < 256; i++) {
8b7358b9 1509 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1510 if (isupper(i))
1511 PL_fold_locale[i] = (U8) tolower(i);
1512 else if (islower(i))
1513 PL_fold_locale[i] = (U8) toupper(i);
1514 else
1515 PL_fold_locale[i] = (U8) i;
8b7358b9 1516 }
8c6180a9
KW
1517
1518 /* If checking for locale problems, see if the native ASCII-range
1519 * printables plus \n and \t are in their expected categories in
1520 * the new locale. If not, this could mean big trouble, upending
1521 * Perl's and most programs' assumptions, like having a
1522 * metacharacter with special meaning become a \w. Fortunately,
1523 * it's very rare to find locales that aren't supersets of ASCII
1524 * nowadays. It isn't a problem for most controls to be changed
1525 * into something else; we check only \n and \t, though perhaps \r
1526 * could be an issue as well. */
7d4bcc4a 1527 if ( check_for_problems
8c6180a9
KW
1528 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1529 {
8b7358b9
KW
1530 bool is_bad = FALSE;
1531 char name[3] = { '\0' };
1532
1533 /* Convert the name into a string */
1534 if (isPRINT_A(i)) {
1535 name[0] = i;
1536 name[1] = '\0';
1537 }
1538 else if (i == '\n') {
1539 my_strlcpy(name, "\n", sizeof(name));
1540 }
1541 else {
1542 my_strlcpy(name, "\t", sizeof(name));
1543 }
1544
1545 /* Check each possibe class */
1546 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1547 is_bad = TRUE;
1548 DEBUG_L(PerlIO_printf(Perl_debug_log,
1549 "isalnum('%s') unexpectedly is %d\n",
1550 name, cBOOL(isalnum(i))));
1551 }
1552 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1553 is_bad = TRUE;
1554 DEBUG_L(PerlIO_printf(Perl_debug_log,
1555 "isalpha('%s') unexpectedly is %d\n",
1556 name, cBOOL(isalpha(i))));
1557 }
1558 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1559 is_bad = TRUE;
1560 DEBUG_L(PerlIO_printf(Perl_debug_log,
1561 "isdigit('%s') unexpectedly is %d\n",
1562 name, cBOOL(isdigit(i))));
1563 }
1564 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1565 is_bad = TRUE;
1566 DEBUG_L(PerlIO_printf(Perl_debug_log,
1567 "isgraph('%s') unexpectedly is %d\n",
1568 name, cBOOL(isgraph(i))));
1569 }
1570 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1571 is_bad = TRUE;
1572 DEBUG_L(PerlIO_printf(Perl_debug_log,
1573 "islower('%s') unexpectedly is %d\n",
1574 name, cBOOL(islower(i))));
1575 }
1576 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1577 is_bad = TRUE;
1578 DEBUG_L(PerlIO_printf(Perl_debug_log,
1579 "isprint('%s') unexpectedly is %d\n",
1580 name, cBOOL(isprint(i))));
1581 }
1582 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1583 is_bad = TRUE;
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "ispunct('%s') unexpectedly is %d\n",
1586 name, cBOOL(ispunct(i))));
1587 }
1588 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1589 is_bad = TRUE;
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isspace('%s') unexpectedly is %d\n",
1592 name, cBOOL(isspace(i))));
1593 }
1594 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1595 is_bad = TRUE;
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isupper('%s') unexpectedly is %d\n",
1598 name, cBOOL(isupper(i))));
1599 }
1600 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1601 is_bad = TRUE;
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "isxdigit('%s') unexpectedly is %d\n",
1604 name, cBOOL(isxdigit(i))));
1605 }
1606 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1607 is_bad = TRUE;
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1610 name, tolower(i), (int) toLOWER_A(i)));
1611 }
1612 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1613 is_bad = TRUE;
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1616 name, toupper(i), (int) toUPPER_A(i)));
1617 }
1618 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1619 is_bad = TRUE;
1620 DEBUG_L(PerlIO_printf(Perl_debug_log,
1621 "'\\n' (=%02X) is not a control\n", (int) i));
1622 }
1623
1624 /* Add to the list; Separate multiple entries with a blank */
1625 if (is_bad) {
1626 if (bad_count) {
1627 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1628 }
8b7358b9
KW
1629 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1630 bad_count++;
8c6180a9
KW
1631 }
1632 }
1633 }
1634
7d4bcc4a
KW
1635# ifdef MB_CUR_MAX
1636
8c6180a9 1637 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1638 * this locale requires more than one byte, there are going to be
8c6180a9 1639 * problems. */
9c8a6dc2
KW
1640 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1641 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1642 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1643
8b7358b9
KW
1644 if ( check_for_problems && MB_CUR_MAX > 1
1645 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1646
1647 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1648 * locale. Just assume that the implementation for them (plus
1649 * for POSIX) is correct and the > 1 value is spurious. (Since
1650 * these are specially handled to never be considered UTF-8
1651 * locales, as long as this is the only problem, everything
1652 * should work fine */
1653 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1654 {
8c6180a9
KW
1655 multi_byte_locale = TRUE;
1656 }
7d4bcc4a
KW
1657
1658# endif
8c6180a9 1659
8b7358b9
KW
1660 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1661 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1662 PL_warn_locale = Perl_newSVpvf(aTHX_
1663 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1664 " which have\nunexpected meanings: %s\nThe Perl program"
1665 " will use the expected meanings",
8b7358b9 1666 newctype, bad_chars_list);
8b7358b9
KW
1667 }
1668 else {
1669 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1670 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1671 newctype,
8c6180a9
KW
1672 (multi_byte_locale)
1673 ? " Some characters in it are not recognized by"
1674 " Perl."
1675 : "",
1676 (bad_count)
1677 ? "\nThe following characters (and maybe others)"
1678 " may not have the same meaning as the Perl"
1679 " program expects:\n"
1680 : "",
1681 (bad_count)
1682 ? bad_chars_list
1683 : ""
1684 );
8b7358b9
KW
1685 }
1686
b119c2be
KW
1687# ifdef HAS_NL_LANGINFO
1688
1689 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1690 /* parameter FALSE is a don't care here */
1691 my_nl_langinfo(PERL_CODESET, FALSE));
1692
1693# endif
1694
8b7358b9
KW
1695 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1696
cc9eaeb0 1697 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1698 * output the message now. If not in that scope, we save the
1699 * message to be output at the first operation using this locale,
1700 * if that actually happens. Most programs don't use locales, so
1701 * they are immune to bad ones. */
cc9eaeb0 1702 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1703
780fcc9f
KW
1704 /* The '0' below suppresses a bogus gcc compiler warning */
1705 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1706
bddebb56
KW
1707 if (IN_LC(LC_CTYPE)) {
1708 SvREFCNT_dec_NN(PL_warn_locale);
1709 PL_warn_locale = NULL;
1710 }
780fcc9f 1711 }
baa60164 1712 }
31f05a37 1713 }
98994639
HS
1714
1715#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1716
98994639
HS
1717}
1718
98994639 1719void
2726666d
KW
1720Perl__warn_problematic_locale()
1721{
2726666d
KW
1722
1723#ifdef USE_LOCALE_CTYPE
1724
5f04a188
KW
1725 dTHX;
1726
1727 /* Internal-to-core function that outputs the message in PL_warn_locale,
1728 * and then NULLS it. Should be called only through the macro
1729 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1730
2726666d 1731 if (PL_warn_locale) {
2726666d
KW
1732 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1733 SvPVX(PL_warn_locale),
1734 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1735 SvREFCNT_dec_NN(PL_warn_locale);
1736 PL_warn_locale = NULL;
1737 }
1738
1739#endif
1740
1741}
1742
a4f00dcc
KW
1743STATIC void
1744S_new_collate(pTHX_ const char *newcoll)
98994639 1745{
7d4bcc4a
KW
1746
1747#ifndef USE_LOCALE_COLLATE
1748
1749 PERL_UNUSED_ARG(newcoll);
1750 PERL_UNUSED_CONTEXT;
1751
1752#else
0d071d52 1753
291a84fb 1754 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1755 * core Perl this and that 'newcoll' is the name of the new locale.
1756 *
d35fca5f
KW
1757 * The design of locale collation is that every locale change is given an
1758 * index 'PL_collation_ix'. The first time a string particpates in an
1759 * operation that requires collation while locale collation is active, it
1760 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1761 * magic includes the collation index, and the transformation of the string
1762 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1763 * instead of the string itself. If a string changes, the magic is
1764 * cleared. The next time the locale changes, the index is incremented,
1765 * and so we know during a comparison that the transformation is not
1766 * necessarily still valid, and so is recomputed. Note that if the locale
1767 * changes enough times, the index could wrap (a U32), and it is possible
1768 * that a transformation would improperly be considered valid, leading to
1769 * an unlikely bug */
0d071d52 1770
98994639
HS
1771 if (! newcoll) {
1772 if (PL_collation_name) {
1773 ++PL_collation_ix;
1774 Safefree(PL_collation_name);
1775 PL_collation_name = NULL;
1776 }
1777 PL_collation_standard = TRUE;
00bf60ca 1778 is_standard_collation:
98994639
HS
1779 PL_collxfrm_base = 0;
1780 PL_collxfrm_mult = 2;
165a1c52 1781 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1782 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1783 PL_strxfrm_max_cp = 0;
98994639
HS
1784 return;
1785 }
1786
d35fca5f 1787 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1788 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1789 ++PL_collation_ix;
1790 Safefree(PL_collation_name);
1791 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1792 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1793 if (PL_collation_standard) {
1794 goto is_standard_collation;
1795 }
98994639 1796
165a1c52 1797 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1798 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1799 PL_strxfrm_max_cp = 0;
165a1c52 1800
59c018b9
KW
1801 /* A locale collation definition includes primary, secondary, tertiary,
1802 * etc. weights for each character. To sort, the primary weights are
1803 * used, and only if they compare equal, then the secondary weights are
1804 * used, and only if they compare equal, then the tertiary, etc.
1805 *
1806 * strxfrm() works by taking the input string, say ABC, and creating an
1807 * output transformed string consisting of first the primary weights,
1808 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1809 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1810 * may not have weights at every level. In our example, let's say B
1811 * doesn't have a tertiary weight, and A doesn't have a secondary
1812 * weight. The constructed string is then going to be
1813 * A¹B¹C¹ B²C² A³C³ ....
1814 * This has the desired effect that strcmp() will look at the secondary
1815 * or tertiary weights only if the strings compare equal at all higher
1816 * priority weights. The spaces shown here, like in
c342d20e 1817 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1818 * are not just for readability. In the general case, these must
1819 * actually be bytes, which we will call here 'separator weights'; and
1820 * they must be smaller than any other weight value, but since these
1821 * are C strings, only the terminating one can be a NUL (some
1822 * implementations may include a non-NUL separator weight just before
1823 * the NUL). Implementations tend to reserve 01 for the separator
1824 * weights. They are needed so that a shorter string's secondary
1825 * weights won't be misconstrued as primary weights of a longer string,
1826 * etc. By making them smaller than any other weight, the shorter
1827 * string will sort first. (Actually, if all secondary weights are
1828 * smaller than all primary ones, there is no need for a separator
1829 * weight between those two levels, etc.)
1830 *
1831 * The length of the transformed string is roughly a linear function of
1832 * the input string. It's not exactly linear because some characters
1833 * don't have weights at all levels. When we call strxfrm() we have to
1834 * allocate some memory to hold the transformed string. The
1835 * calculations below try to find coefficients 'm' and 'b' for this
1836 * locale so that m*x + b equals how much space we need, given the size
1837 * of the input string in 'x'. If we calculate too small, we increase
1838 * the size as needed, and call strxfrm() again, but it is better to
1839 * get it right the first time to avoid wasted expensive string
1840 * transformations. */
1841
98994639 1842 {
79f120c8
KW
1843 /* We use the string below to find how long the tranformation of it
1844 * is. Almost all locales are supersets of ASCII, or at least the
1845 * ASCII letters. We use all of them, half upper half lower,
1846 * because if we used fewer, we might hit just the ones that are
1847 * outliers in a particular locale. Most of the strings being
1848 * collated will contain a preponderance of letters, and even if
1849 * they are above-ASCII, they are likely to have the same number of
1850 * weight levels as the ASCII ones. It turns out that digits tend
1851 * to have fewer levels, and some punctuation has more, but those
1852 * are relatively sparse in text, and khw believes this gives a
1853 * reasonable result, but it could be changed if experience so
1854 * dictates. */
1855 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1856 char * x_longer; /* Transformed 'longer' */
1857 Size_t x_len_longer; /* Length of 'x_longer' */
1858
1859 char * x_shorter; /* We also transform a substring of 'longer' */
1860 Size_t x_len_shorter;
1861
a4a439fb 1862 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1863 * are interested only in its length). It is used because it has
1864 * the intelligence to handle all cases, but to work, it needs some
1865 * values of 'm' and 'b' to get it started. For the purposes of
1866 * this calculation we use a very conservative estimate of 'm' and
1867 * 'b'. This assumes a weight can be multiple bytes, enough to
1868 * hold any UV on the platform, and there are 5 levels, 4 weight
1869 * bytes, and a trailing NUL. */
1870 PL_collxfrm_base = 5;
1871 PL_collxfrm_mult = 5 * sizeof(UV);
1872
1873 /* Find out how long the transformation really is */
a4a439fb
KW
1874 x_longer = _mem_collxfrm(longer,
1875 sizeof(longer) - 1,
1876 &x_len_longer,
1877
1878 /* We avoid converting to UTF-8 in the
1879 * called function by telling it the
1880 * string is in UTF-8 if the locale is a
1881 * UTF-8 one. Since the string passed
1882 * here is invariant under UTF-8, we can
1883 * claim it's UTF-8 even though it isn't.
1884 * */
1885 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1886 Safefree(x_longer);
1887
1888 /* Find out how long the transformation of a substring of 'longer'
1889 * is. Together the lengths of these transformations are
1890 * sufficient to calculate 'm' and 'b'. The substring is all of
1891 * 'longer' except the first character. This minimizes the chances
1892 * of being swayed by outliers */
a4a439fb 1893 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1894 sizeof(longer) - 2,
a4a439fb
KW
1895 &x_len_shorter,
1896 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1897 Safefree(x_shorter);
1898
1899 /* If the results are nonsensical for this simple test, the whole
1900 * locale definition is suspect. Mark it so that locale collation
1901 * is not active at all for it. XXX Should we warn? */
1902 if ( x_len_shorter == 0
1903 || x_len_longer == 0
1904 || x_len_shorter >= x_len_longer)
1905 {
1906 PL_collxfrm_mult = 0;
1907 PL_collxfrm_base = 0;
1908 }
1909 else {
1910 SSize_t base; /* Temporary */
1911
1912 /* We have both: m * strlen(longer) + b = x_len_longer
1913 * m * strlen(shorter) + b = x_len_shorter;
1914 * subtracting yields:
1915 * m * (strlen(longer) - strlen(shorter))
1916 * = x_len_longer - x_len_shorter
1917 * But we have set things up so that 'shorter' is 1 byte smaller
1918 * than 'longer'. Hence:
1919 * m = x_len_longer - x_len_shorter
1920 *
1921 * But if something went wrong, make sure the multiplier is at
1922 * least 1.
1923 */
1924 if (x_len_longer > x_len_shorter) {
1925 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1926 }
1927 else {
1928 PL_collxfrm_mult = 1;
1929 }
1930
1931 /* mx + b = len
1932 * so: b = len - mx
1933 * but in case something has gone wrong, make sure it is
1934 * non-negative */
1935 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1936 if (base < 0) {
1937 base = 0;
1938 }
1939
1940 /* Add 1 for the trailing NUL */
1941 PL_collxfrm_base = base + 1;
1942 }
58eebef2 1943
7d4bcc4a
KW
1944# ifdef DEBUGGING
1945
58eebef2
KW
1946 if (DEBUG_L_TEST || debug_initialization) {
1947 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1948 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1949 "x_len_longer=%zu,"
1950 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1951 __FILE__, __LINE__,
1952 PL_in_utf8_COLLATE_locale,
1953 x_len_shorter, x_len_longer,
1954 PL_collxfrm_mult, PL_collxfrm_base);
1955 }
7d4bcc4a
KW
1956# endif
1957
98994639
HS
1958 }
1959 }
1960
1961#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1962
98994639
HS
1963}
1964
d2b24094 1965#ifdef WIN32
b8cc575c 1966
a4f00dcc 1967STATIC char *
b8cc575c 1968S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1969{
1970 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1971 * difference between the two unless the input locale is "", which normally
1972 * means on Windows to get the machine default, which is set via the
1973 * computer's "Regional and Language Options" (or its current equivalent).
1974 * In POSIX, it instead means to find the locale from the user's
1975 * environment. This routine changes the Windows behavior to first look in
1976 * the environment, and, if anything is found, use that instead of going to
1977 * the machine default. If there is no environment override, the machine
1978 * default is used, by calling the real setlocale() with "".
1979 *
1980 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1981 * use the particular category's variable if set; otherwise to use the LANG
1982 * variable. */
b385bb4d 1983
175c4cf9 1984 bool override_LC_ALL = FALSE;
89f7b9aa 1985 char * result;
e5f10d49 1986 unsigned int i;
89f7b9aa 1987
b385bb4d 1988 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1989
1990# ifdef LC_ALL
1991
b385bb4d
KW
1992 locale = PerlEnv_getenv("LC_ALL");
1993 if (! locale) {
e5f10d49
KW
1994 if (category == LC_ALL) {
1995 override_LC_ALL = TRUE;
1996 }
1997 else {
7d4bcc4a
KW
1998
1999# endif
7d4bcc4a 2000
e5f10d49
KW
2001 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2002 if (category == categories[i]) {
2003 locale = PerlEnv_getenv(category_names[i]);
2004 goto found_locale;
2005 }
2006 }
7d4bcc4a 2007
b385bb4d 2008 locale = PerlEnv_getenv("LANG");
481465ea 2009 if (! locale) {
b385bb4d
KW
2010 locale = "";
2011 }
e5f10d49
KW
2012
2013 found_locale: ;
7d4bcc4a
KW
2014
2015# ifdef LC_ALL
2016
e5f10d49 2017 }
b385bb4d 2018 }
7d4bcc4a
KW
2019
2020# endif
2021
b385bb4d
KW
2022 }
2023
89f7b9aa 2024 result = setlocale(category, locale);
48015184
KW
2025 DEBUG_L(STMT_START {
2026 dSAVE_ERRNO;
2027 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2028 setlocale_debug_string(category, locale, result));
2029 RESTORE_ERRNO;
2030 } STMT_END);
89f7b9aa 2031
481465ea 2032 if (! override_LC_ALL) {
89f7b9aa
KW
2033 return result;
2034 }
2035
dfd77d7a 2036 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2037 * LANG variable or the system default if there is no LANG. But these have
2038 * lower priority than the other LC_foo variables, so override it for each
2039 * one that is set. (If they are set to "", it means to use the same thing
2040 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2041
948523db 2042 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2043 result = PerlEnv_getenv(category_names[i]);
2044 if (result && strNE(result, "")) {
2045 setlocale(categories[i], result);
2046 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2047 __FILE__, __LINE__,
2048 setlocale_debug_string(categories[i], result, "not captured")));
2049 }
89f7b9aa 2050 }
7d4bcc4a 2051
bbc98134 2052 result = setlocale(LC_ALL, NULL);
48015184
KW
2053 DEBUG_L(STMT_START {
2054 dSAVE_ERRNO;
2055 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2056 __FILE__, __LINE__,
48015184
KW
2057 setlocale_debug_string(LC_ALL, NULL, result));
2058 RESTORE_ERRNO;
2059 } STMT_END);
89f7b9aa 2060
bbc98134 2061 return result;
b385bb4d
KW
2062}
2063
2064#endif
2065
9aac5db8
KW
2066/*
2067
2068=head1 Locale-related functions and macros
2069
2070=for apidoc Perl_setlocale
2071
2072This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2073taking the same parameters, and returning the same information, except that it
2074returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2075perl keeps that locale category as C<C>, changing it briefly during the
2076operations where the underlying one is required.
2077
e9bc6d6b 2078Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2079declared to return S<C<const char *>>, whereas the system setlocale omits the
2080C<const>. (If it were being written today, plain setlocale would be declared
2081const, since it is illegal to change the information it returns; doing so leads
2082to segfaults.)
2083
e9bc6d6b
KW
2084Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2085C<setlocale> can be completely ineffective on some platforms under some
2086configurations.
2087
9aac5db8 2088C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2089where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2090the system C<setlocale()> is ineffective, returning the wrong information, and
2091failing to actually change the locale. C<Perl_setlocale>, however works
2092properly in all circumstances.
9aac5db8
KW
2093
2094The return points to a per-thread static buffer, which is overwritten the next
2095time C<Perl_setlocale> is called from the same thread.
2096
2097=cut
2098
2099*/
2100
2101const char *
2102Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2103{
2104 /* This wraps POSIX::setlocale() */
2105
9aac5db8
KW
2106 const char * retval;
2107 const char * newlocale;
2108 dSAVEDERRNO;
2109 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2110 dTHX;
2111
a4f00dcc
KW
2112#ifdef USE_LOCALE_NUMERIC
2113
291a84fb
KW
2114 /* A NULL locale means only query what the current one is. We have the
2115 * LC_NUMERIC name saved, because we are normally switched into the C
2116 * locale for it. For an LC_ALL query, switch back to get the correct
2117 * results. All other categories don't require special handling */
a4f00dcc
KW
2118 if (locale == NULL) {
2119 if (category == LC_NUMERIC) {
9aac5db8
KW
2120
2121 /* We don't have to copy this return value, as it is a per-thread
2122 * variable, and won't change until a future setlocale */
2123 return PL_numeric_name;
a4f00dcc
KW
2124 }
2125
7d4bcc4a 2126# ifdef LC_ALL
a4f00dcc 2127
9aac5db8
KW
2128 else if (category == LC_ALL) {
2129 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2130 }
2131
7d4bcc4a 2132# endif
a4f00dcc
KW
2133
2134 }
2135
2136#endif
2137
33e5a354
KW
2138 retval = save_to_buffer(do_setlocale_r(category, locale),
2139 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2140 SAVE_ERRNO;
2141
2142#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2143
2144 if (locale == NULL && category == LC_ALL) {
2145 RESTORE_LC_NUMERIC();
2146 }
2147
2148#endif
a4f00dcc
KW
2149
2150 DEBUG_L(PerlIO_printf(Perl_debug_log,
2151 "%s:%d: %s\n", __FILE__, __LINE__,
2152 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2153
9aac5db8
KW
2154 RESTORE_ERRNO;
2155
2156 if (! retval) {
a4f00dcc
KW
2157 return NULL;
2158 }
2159
9aac5db8 2160 /* If locale == NULL, we are just querying the state */
a4f00dcc 2161 if (locale == NULL) {
a4f00dcc
KW
2162 return retval;
2163 }
a4f00dcc 2164
1159483a
KW
2165 /* Now that have switched locales, we have to update our records to
2166 * correspond. */
a4f00dcc 2167
1159483a 2168 switch (category) {
a4f00dcc 2169
1159483a 2170#ifdef USE_LOCALE_CTYPE
a4f00dcc 2171
1159483a
KW
2172 case LC_CTYPE:
2173 new_ctype(retval);
2174 break;
a4f00dcc 2175
1159483a 2176#endif
a4f00dcc
KW
2177#ifdef USE_LOCALE_COLLATE
2178
1159483a
KW
2179 case LC_COLLATE:
2180 new_collate(retval);
2181 break;
a4f00dcc 2182
1159483a 2183#endif
a4f00dcc
KW
2184#ifdef USE_LOCALE_NUMERIC
2185
1159483a
KW
2186 case LC_NUMERIC:
2187 new_numeric(retval);
2188 break;
a4f00dcc 2189
1159483a
KW
2190#endif
2191#ifdef LC_ALL
a4f00dcc 2192
1159483a 2193 case LC_ALL:
a4f00dcc 2194
1159483a
KW
2195 /* LC_ALL updates all the things we care about. The values may not
2196 * be the same as 'retval', as the locale "" may have set things
2197 * individually */
a4f00dcc 2198
1159483a 2199# ifdef USE_LOCALE_CTYPE
a4f00dcc 2200
1159483a
KW
2201 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2202 new_ctype(newlocale);
a4f00dcc 2203
1159483a
KW
2204# endif /* USE_LOCALE_CTYPE */
2205# ifdef USE_LOCALE_COLLATE
2206
2207 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2208 new_collate(newlocale);
a4f00dcc 2209
7d4bcc4a 2210# endif
1159483a 2211# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2212
1159483a
KW
2213 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2214 new_numeric(newlocale);
a4f00dcc 2215
1159483a
KW
2216# endif /* USE_LOCALE_NUMERIC */
2217#endif /* LC_ALL */
a4f00dcc 2218
1159483a
KW
2219 default:
2220 break;
a4f00dcc
KW
2221 }
2222
2223 return retval;
2224
f7416781
KW
2225}
2226
2227PERL_STATIC_INLINE const char *
2228S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2229{
2230 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2231 * growing it if necessary */
2232
0b6697c5 2233 Size_t string_size;
f7416781
KW
2234
2235 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2236
0b6697c5
KW
2237 if (! string) {
2238 return NULL;
2239 }
2240
2241 string_size = strlen(string) + offset + 1;
2242
f7416781
KW
2243 if (*buf_size == 0) {
2244 Newx(*buf, string_size, char);
2245 *buf_size = string_size;
2246 }
2247 else if (string_size > *buf_size) {
2248 Renew(*buf, string_size, char);
2249 *buf_size = string_size;
2250 }
2251
2252 Copy(string, *buf + offset, string_size - offset, char);
2253 return *buf;
2254}
2255
2256/*
2257
f7416781
KW
2258=for apidoc Perl_langinfo
2259
ce181bf6 2260This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2261taking the same C<item> parameter values, and returning the same information.
2262But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2263of Perl's locale handling from your code, and can be used on systems that lack
2264a native C<nl_langinfo>.
2265
2266Expanding on these:
2267
2268=over
2269
2270=item *
2271
ce181bf6
KW
2272The reason it isn't quite a drop-in replacement is actually an advantage. The
2273only difference is that it returns S<C<const char *>>, whereas plain
2274C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2275forbidden to write into the buffer. By declaring this C<const>, the compiler
2276enforces this restriction, so if it is violated, you know at compilation time,
2277rather than getting segfaults at runtime.
2278
2279=item *
2280
69e2275e 2281It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2282without you having to write extra code. The reason for the extra code would be
2283because these are from the C<LC_NUMERIC> locale category, which is normally
2284kept set to the C locale by Perl, no matter what the underlying locale is
2285supposed to be, and so to get the expected results, you have to temporarily
ce181bf6
KW
2286toggle into the underlying locale, and later toggle back. (You could use plain
2287C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2288then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2289C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2290radix (decimal point) character to be a dot.)
2291
2292=item *
2293
2294The system function it replaces can have its static return buffer trashed,
2295not only by a subesequent call to that function, but by a C<freelocale>,
2296C<setlocale>, or other locale change. The returned buffer of this function is
2297not changed until the next call to it, so the buffer is never in a trashed
2298state.
f7416781
KW
2299
2300=item *
2301
ce181bf6
KW
2302Its return buffer is per-thread, so it also is never overwritten by a call to
2303this function from another thread; unlike the function it replaces.
2304
2305=item *
2306
2307But most importantly, it works on systems that don't have C<nl_langinfo>, such
2308as Windows, hence makes your code more portable. Of the fifty-some possible
2309items specified by the POSIX 2008 standard,
f7416781 2310L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
ce181bf6
KW
2311only two are completely unimplemented (though the loss of one of these is
2312significant). It uses various techniques to recover the other items, including
2313calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2314in C89, so should be always be available. Later C<strftime()> versions have
2315additional capabilities; C<""> is returned for those not available on your
2316system.
2317
2318It is important to note that when called with an item that is recovered by
2319using C<localeconv>, the buffer from any previous explicit call to
2320C<localeconv> will be overwritten. This means you must save that buffer's
2321contents if you need to access them after a call to this function.
9f0bc911 2322
f7416781
KW
2323The details for those items which may differ from what this emulation returns
2324and what a native C<nl_langinfo()> would return are:
2325
2326=over
2327
2328=item C<CODESET>
2329
2330=item C<ERA>
2331
2332Unimplemented, so returns C<"">.
2333
2334=item C<YESEXPR>
2335
c1566110
KW
2336=item C<YESSTR>
2337
f7416781
KW
2338=item C<NOEXPR>
2339
c1566110
KW
2340=item C<NOSTR>
2341
2342Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
ce181bf6
KW
2343removed from POSIX 2008, and are retained here for backwards compatibility.
2344Your platform's C<nl_langinfo> may not support them.
f7416781
KW
2345
2346=item C<D_FMT>
2347
2348Always evaluates to C<%x>, the locale's appropriate date representation.
2349
2350=item C<T_FMT>
2351
2352Always evaluates to C<%X>, the locale's appropriate time representation.
2353
2354=item C<D_T_FMT>
2355
2356Always evaluates to C<%c>, the locale's appropriate date and time
2357representation.
2358
2359=item C<CRNCYSTR>
2360
2361The return may be incorrect for those rare locales where the currency symbol
2362replaces the radix character.
2363Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2364to work differently.
2365
2366=item C<ALT_DIGITS>
2367
2368Currently this gives the same results as Linux does.
2369Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2370to work differently.
2371
2372=item C<ERA_D_FMT>
2373
2374=item C<ERA_T_FMT>
2375
2376=item C<ERA_D_T_FMT>
2377
2378=item C<T_FMT_AMPM>
2379
2380These are derived by using C<strftime()>, and not all versions of that function
2381know about them. C<""> is returned for these on such systems.
2382
2383=back
2384
ce181bf6
KW
2385=back
2386
f7416781
KW
2387When using C<Perl_langinfo> on systems that don't have a native
2388C<nl_langinfo()>, you must
2389
2390 #include "perl_langinfo.h"
2391
2392before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2393C<#include> with this one. (Doing it this way keeps out the symbols that plain
2394C<langinfo.h> imports into the namespace for code that doesn't need it.)
2395
2396You also should not use the bare C<langinfo.h> item names, but should preface
2397them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2398The C<PERL_I<foo>> versions will also work for this function on systems that do
2399have a native C<nl_langinfo>.
2400
f7416781
KW
2401The original impetus for C<Perl_langinfo()> was so that code that needs to
2402find out the current currency symbol, floating point radix character, or digit
2403grouping separator can use, on all systems, the simpler and more
2404thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2405pain to make thread-friendly. For other fields returned by C<localeconv>, it
2406is better to use the methods given in L<perlcall> to call
2407L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2408
2409=cut
2410
2411*/
2412
2413const char *
2414#ifdef HAS_NL_LANGINFO
2415Perl_langinfo(const nl_item item)
2416#else
2417Perl_langinfo(const int item)
2418#endif
2419{
f61748ac
KW
2420 return my_nl_langinfo(item, TRUE);
2421}
2422
2423const char *
2424#ifdef HAS_NL_LANGINFO
2425S_my_nl_langinfo(const nl_item item, bool toggle)
2426#else
2427S_my_nl_langinfo(const int item, bool toggle)
2428#endif
2429{
ae74815b 2430 dTHX;
0b6697c5 2431 const char * retval;
f7416781 2432
5a854ab3
KW
2433 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2434 * two items, and only if not already there */
2435 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2436 || PL_numeric_underlying))
2437 {
2438 toggle = FALSE;
2439 }
2440
ab340fff 2441#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d
KW
2442# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2443 || ! defined(HAS_POSIX_2008_LOCALE)
f7416781 2444
ab340fff 2445 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2446 * for those items dependent on it. This must be copied to a buffer before
2447 * switching back, as some systems destroy the buffer when setlocale() is
2448 * called */
f7416781 2449
038d3702
KW
2450 {
2451 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2452
b2fee59c 2453 if (toggle) {
038d3702 2454 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2455 }
f7416781 2456
5acc4454
KW
2457 LOCALE_LOCK; /* Prevent interference from another thread executing
2458 this code section (the only call to nl_langinfo in
2459 the core) */
2460
ee90eb2d 2461
628ff75a
KW
2462 /* Copy to a per-thread buffer, which is also one that won't be
2463 * destroyed by a subsequent setlocale(), such as the
2464 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2465 retval = save_to_buffer(nl_langinfo(item),
2466 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2467
5acc4454
KW
2468 LOCALE_UNLOCK;
2469
b2fee59c 2470 if (toggle) {
038d3702 2471 RESTORE_LC_NUMERIC();
b2fee59c 2472 }
f7416781
KW
2473 }
2474
ab340fff
KW
2475# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2476
5a854ab3 2477 {
b2fee59c
KW
2478 bool do_free = FALSE;
2479 locale_t cur = uselocale((locale_t) 0);
ab340fff 2480
b2fee59c
KW
2481 if (cur == LC_GLOBAL_LOCALE) {
2482 cur = duplocale(LC_GLOBAL_LOCALE);
2483 do_free = TRUE;
2484 }
ab340fff 2485
b2fee59c 2486 if (toggle) {
e1aa2579
KW
2487 if (PL_underlying_numeric_obj) {
2488 cur = PL_underlying_numeric_obj;
2489 }
2490 else {
2491 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2492 do_free = TRUE;
2493 }
b2fee59c 2494 }
ab340fff 2495
628ff75a
KW
2496 /* We have to save it to a buffer, because the freelocale() just below
2497 * can invalidate the internal one */
0b6697c5
KW
2498 retval = save_to_buffer(nl_langinfo_l(item, cur),
2499 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2500
b2fee59c
KW
2501 if (do_free) {
2502 freelocale(cur);
2503 }
5a854ab3 2504 }
ab340fff 2505
c1566110
KW
2506# endif
2507
0b6697c5 2508 if (strEQ(retval, "")) {
c1566110
KW
2509 if (item == PERL_YESSTR) {
2510 return "yes";
2511 }
2512 if (item == PERL_NOSTR) {
2513 return "no";
2514 }
2515 }
2516
0b6697c5 2517 return retval;
ab340fff 2518
f7416781 2519#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2520
2521 {
2522
f7416781
KW
2523# ifdef HAS_LOCALECONV
2524
43dd6b15 2525 const struct lconv* lc;
628ff75a 2526 const char * temp;
038d3702 2527 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
2528
2529# endif
2530# ifdef HAS_STRFTIME
2531
43dd6b15
KW
2532 struct tm tm;
2533 bool return_format = FALSE; /* Return the %format, not the value */
2534 const char * format;
f7416781
KW
2535
2536# endif
2537
43dd6b15
KW
2538 /* We copy the results to a per-thread buffer, even if not
2539 * multi-threaded. This is in part to simplify this code, and partly
2540 * because we need a buffer anyway for strftime(), and partly because a
2541 * call of localeconv() could otherwise wipe out the buffer, and the
2542 * programmer would not be expecting this, as this is a nl_langinfo()
2543 * substitute after all, so s/he might be thinking their localeconv()
2544 * is safe until another localeconv() call. */
f7416781 2545
43dd6b15
KW
2546 switch (item) {
2547 Size_t len;
f7416781 2548
43dd6b15
KW
2549 /* These 2 are unimplemented */
2550 case PERL_CODESET:
2551 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 2552
43dd6b15
KW
2553 default:
2554 return "";
f7416781 2555
43dd6b15
KW
2556 /* We use only an English set, since we don't know any more */
2557 case PERL_YESEXPR: return "^[+1yY]";
2558 case PERL_YESSTR: return "yes";
2559 case PERL_NOEXPR: return "^[-0nN]";
2560 case PERL_NOSTR: return "no";
f7416781
KW
2561
2562# ifdef HAS_LOCALECONV
2563
43dd6b15 2564 case PERL_CRNCYSTR:
f7416781 2565
43dd6b15
KW
2566 /* We don't bother with localeconv_l() because any system that
2567 * has it is likely to also have nl_langinfo() */
291a84fb 2568
5acc4454
KW
2569 LOCALE_LOCK; /* Prevent interference with other threads
2570 using localeconv() */
2571
43dd6b15
KW
2572 lc = localeconv();
2573 if ( ! lc
2574 || ! lc->currency_symbol
2575 || strEQ("", lc->currency_symbol))
2576 {
2577 LOCALE_UNLOCK;
2578 return "";
2579 }
f7416781 2580
43dd6b15 2581 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2582 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2583 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2584 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2585 { /* khw couldn't figure out how the localedef specifications
2586 would show that the $ should replace the radix; this is
2587 just a guess as to how it might work.*/
a34edee3 2588 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2589 }
2590 else if (lc->p_cs_precedes) {
a34edee3 2591 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2592 }
2593 else {
a34edee3 2594 PL_langinfo_buf[0] = '+';
43dd6b15 2595 }
f7416781 2596
43dd6b15
KW
2597 LOCALE_UNLOCK;
2598 break;
f7416781 2599
43dd6b15
KW
2600 case PERL_RADIXCHAR:
2601 case PERL_THOUSEP:
f7416781 2602
43dd6b15 2603 if (toggle) {
038d3702 2604 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2605 }
f7416781 2606
5acc4454
KW
2607 LOCALE_LOCK; /* Prevent interference with other threads
2608 using localeconv() */
2609
43dd6b15
KW
2610 lc = localeconv();
2611 if (! lc) {
628ff75a 2612 temp = "";
33394adc 2613 }
43dd6b15 2614 else {
628ff75a 2615 temp = (item == PERL_RADIXCHAR)
43dd6b15
KW
2616 ? lc->decimal_point
2617 : lc->thousands_sep;
628ff75a
KW
2618 if (! temp) {
2619 temp = "";
43dd6b15
KW
2620 }
2621 }
f7416781 2622
0b6697c5
KW
2623 retval = save_to_buffer(temp, &PL_langinfo_buf,
2624 &PL_langinfo_bufsize, 0);
f7416781 2625
5acc4454
KW
2626 LOCALE_UNLOCK;
2627
43dd6b15 2628 if (toggle) {
038d3702 2629 RESTORE_LC_NUMERIC();
43dd6b15 2630 }
f7416781 2631
43dd6b15 2632 break;
f7416781
KW
2633
2634# endif
2635# ifdef HAS_STRFTIME
2636
43dd6b15
KW
2637 /* These are defined by C89, so we assume that strftime supports
2638 * them, and so are returned unconditionally; they may not be what
2639 * the locale actually says, but should give good enough results
2640 * for someone using them as formats (as opposed to trying to parse
2641 * them to figure out what the locale says). The other format
2642 * items are actually tested to verify they work on the platform */
2643 case PERL_D_FMT: return "%x";
2644 case PERL_T_FMT: return "%X";
2645 case PERL_D_T_FMT: return "%c";
2646
2647 /* These formats are only available in later strfmtime's */
2648 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2649 case PERL_T_FMT_AMPM:
2650
2651 /* The rest can be gotten from most versions of strftime(). */
2652 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2653 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2654 case PERL_ABDAY_7:
2655 case PERL_ALT_DIGITS:
2656 case PERL_AM_STR: case PERL_PM_STR:
2657 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2658 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2659 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2660 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2661 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2662 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2663 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2664 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2665 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2666 case PERL_MON_12:
2667
2668 LOCALE_LOCK;
2669
2670 init_tm(&tm); /* Precaution against core dumps */
2671 tm.tm_sec = 30;
2672 tm.tm_min = 30;
2673 tm.tm_hour = 6;
2674 tm.tm_year = 2017 - 1900;
2675 tm.tm_wday = 0;
2676 tm.tm_mon = 0;
2677 switch (item) {
2678 default:
2679 LOCALE_UNLOCK;
2680 Perl_croak(aTHX_
2681 "panic: %s: %d: switch case: %d problem",
2682 __FILE__, __LINE__, item);
2683 NOT_REACHED; /* NOTREACHED */
2684
2685 case PERL_PM_STR: tm.tm_hour = 18;
2686 case PERL_AM_STR:
2687 format = "%p";
2688 break;
2689
2690 case PERL_ABDAY_7: tm.tm_wday++;
2691 case PERL_ABDAY_6: tm.tm_wday++;
2692 case PERL_ABDAY_5: tm.tm_wday++;
2693 case PERL_ABDAY_4: tm.tm_wday++;
2694 case PERL_ABDAY_3: tm.tm_wday++;
2695 case PERL_ABDAY_2: tm.tm_wday++;
2696 case PERL_ABDAY_1:
2697 format = "%a";
2698 break;
2699
2700 case PERL_DAY_7: tm.tm_wday++;
2701 case PERL_DAY_6: tm.tm_wday++;
2702 case PERL_DAY_5: tm.tm_wday++;
2703 case PERL_DAY_4: tm.tm_wday++;
2704 case PERL_DAY_3: tm.tm_wday++;
2705 case PERL_DAY_2: tm.tm_wday++;
2706 case PERL_DAY_1:
2707 format = "%A";
2708 break;
2709
2710 case PERL_ABMON_12: tm.tm_mon++;
2711 case PERL_ABMON_11: tm.tm_mon++;
2712 case PERL_ABMON_10: tm.tm_mon++;
2713 case PERL_ABMON_9: tm.tm_mon++;
2714 case PERL_ABMON_8: tm.tm_mon++;
2715 case PERL_ABMON_7: tm.tm_mon++;
2716 case PERL_ABMON_6: tm.tm_mon++;
2717 case PERL_ABMON_5: tm.tm_mon++;
2718 case PERL_ABMON_4: tm.tm_mon++;
2719 case PERL_ABMON_3: tm.tm_mon++;
2720 case PERL_ABMON_2: tm.tm_mon++;
2721 case PERL_ABMON_1:
2722 format = "%b";
2723 break;
2724
2725 case PERL_MON_12: tm.tm_mon++;
2726 case PERL_MON_11: tm.tm_mon++;
2727 case PERL_MON_10: tm.tm_mon++;
2728 case PERL_MON_9: tm.tm_mon++;
2729 case PERL_MON_8: tm.tm_mon++;
2730 case PERL_MON_7: tm.tm_mon++;
2731 case PERL_MON_6: tm.tm_mon++;
2732 case PERL_MON_5: tm.tm_mon++;
2733 case PERL_MON_4: tm.tm_mon++;
2734 case PERL_MON_3: tm.tm_mon++;
2735 case PERL_MON_2: tm.tm_mon++;
2736 case PERL_MON_1:
2737 format = "%B";
2738 break;
2739
2740 case PERL_T_FMT_AMPM:
2741 format = "%r";
2742 return_format = TRUE;
2743 break;
2744
2745 case PERL_ERA_D_FMT:
2746 format = "%Ex";
2747 return_format = TRUE;
2748 break;
2749
2750 case PERL_ERA_T_FMT:
2751 format = "%EX";
2752 return_format = TRUE;
2753 break;
2754
2755 case PERL_ERA_D_T_FMT:
2756 format = "%Ec";
2757 return_format = TRUE;
2758 break;
2759
2760 case PERL_ALT_DIGITS:
2761 tm.tm_wday = 0;
2762 format = "%Ow"; /* Find the alternate digit for 0 */
2763 break;
2764 }
f7416781 2765
43dd6b15
KW
2766 /* We can't use my_strftime() because it doesn't look at
2767 * tm_wday */
2768 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2769 format, &tm))
2770 {
2771 /* A zero return means one of:
2772 * a) there wasn't enough space in PL_langinfo_buf
2773 * b) the format, like a plain %p, returns empty
2774 * c) it was an illegal format, though some
2775 * implementations of strftime will just return the
2776 * illegal format as a plain character sequence.
2777 *
2778 * To quickly test for case 'b)', try again but precede
2779 * the format with a plain character. If that result is
2780 * still empty, the problem is either 'a)' or 'c)' */
2781
2782 Size_t format_size = strlen(format) + 1;
2783 Size_t mod_size = format_size + 1;
2784 char * mod_format;
2785 char * temp_result;
2786
2787 Newx(mod_format, mod_size, char);
2788 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2789 *mod_format = ' ';
43dd6b15
KW
2790 my_strlcpy(mod_format + 1, format, mod_size);
2791 len = strftime(temp_result,
2792 PL_langinfo_bufsize,
2793 mod_format, &tm);
2794 Safefree(mod_format);
2795 Safefree(temp_result);
2796
2797 /* If 'len' is non-zero, it means that we had a case like
2798 * %p which means the current locale doesn't use a.m. or
2799 * p.m., and that is valid */
2800 if (len == 0) {
2801
2802 /* Here, still didn't work. If we get well beyond a
2803 * reasonable size, bail out to prevent an infinite
2804 * loop. */
2805
2806 if (PL_langinfo_bufsize > 100 * format_size) {
2807 *PL_langinfo_buf = '\0';
2808 }
2809 else {
2810 /* Double the buffer size to retry; Add 1 in case
2811 * original was 0, so we aren't stuck at 0. */
2812 PL_langinfo_bufsize *= 2;
2813 PL_langinfo_bufsize++;
2814 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2815 continue;
2816 }
2817 }
f7416781 2818
f7416781 2819 break;
43dd6b15 2820 }
f7416781 2821
43dd6b15
KW
2822 /* Here, we got a result.
2823 *
2824 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2825 * alternate format for wday 0. If the value is the same as
2826 * the normal 0, there isn't an alternate, so clear the buffer.
2827 * */
2828 if ( item == PERL_ALT_DIGITS
2829 && strEQ(PL_langinfo_buf, "0"))
2830 {
2831 *PL_langinfo_buf = '\0';
2832 }
f7416781 2833
43dd6b15
KW
2834 /* ALT_DIGITS is problematic. Experiments on it showed that
2835 * strftime() did not always work properly when going from
2836 * alt-9 to alt-10. Only a few locales have this item defined,
2837 * and in all of them on Linux that khw was able to find,
2838 * nl_langinfo() merely returned the alt-0 character, possibly
2839 * doubled. Most Unicode digits are in blocks of 10
2840 * consecutive code points, so that is sufficient information
2841 * for those scripts, as we can infer alt-1, alt-2, .... But
2842 * for a Japanese locale, a CJK ideographic 0 is returned, and
2843 * the CJK digits are not in code point order, so you can't
2844 * really infer anything. The localedef for this locale did
2845 * specify the succeeding digits, so that strftime() works
2846 * properly on them, without needing to infer anything. But
2847 * the nl_langinfo() return did not give sufficient information
2848 * for the caller to understand what's going on. So until
2849 * there is evidence that it should work differently, this
2850 * returns the alt-0 string for ALT_DIGITS.
2851 *
2852 * wday was chosen because its range is all a single digit.
2853 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 2854
43dd6b15 2855 LOCALE_UNLOCK;
f7416781 2856
d1b150d9
KW
2857 retval = PL_langinfo_buf;
2858
43dd6b15
KW
2859 /* If to return the format, not the value, overwrite the buffer
2860 * with it. But some strftime()s will keep the original format
2861 * if illegal, so change those to "" */
2862 if (return_format) {
2863 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
2864 *PL_langinfo_buf = '\0';
2865 }
43dd6b15 2866 else {
0b6697c5
KW
2867 retval = save_to_buffer(format, &PL_langinfo_buf,
2868 &PL_langinfo_bufsize, 0);
f7416781
KW
2869 }
2870 }
2871
2872 break;
f7416781
KW
2873
2874# endif
2875
43dd6b15 2876 }
f7416781
KW
2877 }
2878
0b6697c5 2879 return retval;
f7416781
KW
2880
2881#endif
2882
a4f00dcc 2883}
b385bb4d 2884
98994639
HS
2885/*
2886 * Initialize locale awareness.
2887 */
2888int
2889Perl_init_i18nl10n(pTHX_ int printwarn)
2890{
0e92a118
KW
2891 /* printwarn is
2892 *
2893 * 0 if not to output warning when setup locale is bad
2894 * 1 if to output warning based on value of PERL_BADLANG
2895 * >1 if to output regardless of PERL_BADLANG
2896 *
2897 * returns
98994639 2898 * 1 = set ok or not applicable,
0e92a118
KW
2899 * 0 = fallback to a locale of lower priority
2900 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2901 *
2902 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2903 * set, debugging information is output.
2904 *
2905 * This looks more complicated than it is, mainly due to the #ifdefs.
2906 *
2907 * We try to set LC_ALL to the value determined by the environment. If
2908 * there is no LC_ALL on this platform, we try the individual categories we
2909 * know about. If this works, we are done.
2910 *
2911 * But if it doesn't work, we have to do something else. We search the
2912 * environment variables ourselves instead of relying on the system to do
2913 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2914 * think there is one), and the ultimate fallback "C". This is all done in
2915 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2916 * more complex. The 'trial_locales' array is initialized with just one
2917 * element; it causes the behavior described in the paragraph above this to
2918 * happen. If that fails, we add elements to 'trial_locales', and do extra
2919 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2920 *
2921 * On Ultrix, the locale MUST come from the environment, so there is
2922 * preliminary code to set it. I (khw) am not sure that it is necessary,
2923 * and that this couldn't be folded into the loop, but barring any real
2924 * platforms to test on, it's staying as-is
2925 *
2926 * A slight complication is that in embedded Perls, the locale may already
2927 * be set-up, and we don't want to get it from the normal environment
2928 * variables. This is handled by having a special environment variable
2929 * indicate we're in this situation. We simply set setlocale's 2nd
2930 * parameter to be a NULL instead of "". That indicates to setlocale that
2931 * it is not to change anything, but to return the current value,
2932 * effectively initializing perl's db to what the locale already is.
2933 *
2934 * We play the same trick with NULL if a LC_ALL succeeds. We call
2935 * setlocale() on the individual categores with NULL to get their existing
2936 * values for our db, instead of trying to change them.
2937 * */
98994639 2938
0e92a118
KW
2939 int ok = 1;
2940
7d4bcc4a
KW
2941#ifndef USE_LOCALE
2942
2943 PERL_UNUSED_ARG(printwarn);
2944
2945#else /* USE_LOCALE */
7d4bcc4a
KW
2946# ifdef __GLIBC__
2947
175c4cf9 2948 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2949
2950# endif
65ebb059 2951
ccd65d51
KW
2952 /* NULL uses the existing already set up locale */
2953 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2954 ? NULL
2955 : "";
c3fcd832
KW
2956 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2957 unsigned int trial_locales_count;
175c4cf9
KW
2958 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2959 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2960 bool setlocale_failure = FALSE;
65ebb059 2961 unsigned int i;
175c4cf9
KW
2962
2963 /* A later getenv() could zap this, so only use here */
2964 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2965
2966 const bool locwarn = (printwarn > 1
e5f10d49
KW
2967 || ( printwarn
2968 && ( ! bad_lang_use_once
22ff3130 2969 || (
e5f10d49
KW
2970 /* disallow with "" or "0" */
2971 *bad_lang_use_once
2972 && strNE("0", bad_lang_use_once)))));
ea92aad8 2973
291a84fb 2974 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2975 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2976
2977 /* current locale for given category; should have been copied so aren't
2978 * volatile */
8de4332b 2979 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2980
7d4bcc4a
KW
2981# ifdef WIN32
2982
6bce99ee
JH
2983 /* In some systems you can find out the system default locale
2984 * and use that as the fallback locale. */
7d4bcc4a
KW
2985# define SYSTEM_DEFAULT_LOCALE
2986# endif
2987# ifdef SYSTEM_DEFAULT_LOCALE
2988
65ebb059 2989 const char *system_default_locale = NULL;
98994639 2990
7d4bcc4a 2991# endif
948523db
KW
2992
2993# ifndef DEBUGGING
2994# define DEBUG_LOCALE_INIT(a,b,c)
2995# else
7d4bcc4a 2996
8298454c 2997 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2998
2999# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3000 STMT_START { \
3001 if (debug_initialization) { \
3002 PerlIO_printf(Perl_debug_log, \
3003 "%s:%d: %s\n", \
3004 __FILE__, __LINE__, \
a4f00dcc 3005 setlocale_debug_string(category, \
2fcc0ca9
KW
3006 locale, \
3007 result)); \
3008 } \
3009 } STMT_END
2fcc0ca9 3010
948523db
KW
3011/* Make sure the parallel arrays are properly set up */
3012# ifdef USE_LOCALE_NUMERIC
3013 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3014 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3015# ifdef USE_POSIX_2008_LOCALE
3016 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3017# endif
948523db
KW
3018# endif
3019# ifdef USE_LOCALE_CTYPE
3020 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3021 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3022# ifdef USE_POSIX_2008_LOCALE
3023 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3024# endif
948523db
KW
3025# endif
3026# ifdef USE_LOCALE_COLLATE
3027 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3028 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3029# ifdef USE_POSIX_2008_LOCALE
3030 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3031# endif
948523db
KW
3032# endif
3033# ifdef USE_LOCALE_TIME
3034 assert(categories[LC_TIME_INDEX] == LC_TIME);
3035 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3036# ifdef USE_POSIX_2008_LOCALE
3037 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3038# endif
948523db
KW
3039# endif
3040# ifdef USE_LOCALE_MESSAGES
3041 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3042 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3043# ifdef USE_POSIX_2008_LOCALE
3044 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3045# endif
948523db
KW
3046# endif
3047# ifdef USE_LOCALE_MONETARY
3048 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3049 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3050# ifdef USE_POSIX_2008_LOCALE
3051 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3052# endif
948523db 3053# endif
9821811f
KW
3054# ifdef USE_LOCALE_ADDRESS
3055 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3056 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3057# ifdef USE_POSIX_2008_LOCALE
3058 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3059# endif
9821811f
KW
3060# endif
3061# ifdef USE_LOCALE_IDENTIFICATION
3062 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3063 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3064# ifdef USE_POSIX_2008_LOCALE
3065 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3066# endif
9821811f
KW
3067# endif
3068# ifdef USE_LOCALE_MEASUREMENT
3069 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3070 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3071# ifdef USE_POSIX_2008_LOCALE
3072 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3073# endif
9821811f
KW
3074# endif
3075# ifdef USE_LOCALE_PAPER
3076 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3077 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3078# ifdef USE_POSIX_2008_LOCALE
3079 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3080# endif
9821811f
KW
3081# endif
3082# ifdef USE_LOCALE_TELEPHONE
3083 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3084 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3085# ifdef USE_POSIX_2008_LOCALE
3086 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3087# endif
9821811f 3088# endif
948523db
KW
3089# ifdef LC_ALL
3090 assert(categories[LC_ALL_INDEX] == LC_ALL);
3091 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3092 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3093# ifdef USE_POSIX_2008_LOCALE
3094 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3095# endif
948523db
KW
3096# endif
3097# endif /* DEBUGGING */
47280b20
KW
3098
3099 /* Initialize the cache of the program's UTF-8ness for the always known
3100 * locales C and POSIX */
3101 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3102 sizeof(PL_locale_utf8ness));
3103
e9bc6d6b
KW
3104# ifdef USE_THREAD_SAFE_LOCALE
3105# ifdef WIN32
3106
3107 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3108
3109# endif
3110# endif
3111# if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3112
3113 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3114 if (! PL_C_locale_obj) {
3115 Perl_croak_nocontext(
3116 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3117 }
3118 if (DEBUG_Lv_TEST || debug_initialization) {
3119 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3120 }
3121
3122# endif
3123
3ca88433
KW
3124 PL_numeric_radix_sv = newSVpvs(".");
3125
e9bc6d6b
KW
3126# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3127
3128 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3129 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3130
3131# endif
b56c4436 3132# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3133
3134 /*
3135 * Ultrix setlocale(..., "") fails if there are no environment
3136 * variables from which to get a locale name.
3137 */
3138
b56c4436
KW
3139# ifndef LC_ALL
3140# error Ultrix without LC_ALL not implemented
3141# else
7d4bcc4a 3142
b56c4436
KW
3143 {
3144 bool done = FALSE;
ea92aad8
KW
3145 if (lang) {
3146 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3147 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3148 if (sl_result[LC_ALL_INDEX])
3149 done = TRUE;
3150 else
e5f10d49 3151 setlocale_failure = TRUE;
ea92aad8
KW
3152 }
3153 if (! setlocale_failure) {
3154 const char * locale_param;
3155 for (i = 0; i < LC_ALL_INDEX; i++) {
3156 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3157 ? setlocale_init
3158 : NULL;
3159 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3160 if (! sl_result[i]) {
3161 setlocale_failure = TRUE;
3162 }
3163 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3164 }
c835d6be 3165 }
7d4bcc4a
KW
3166 }
3167
3168# endif /* LC_ALL */
e5f10d49 3169# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3170
65ebb059 3171 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3172 * the list. Normally the loop is executed just once. But if setting the
3173 * locale fails, inside the loop we add fallback trials to the array and so
3174 * will execute the loop multiple times */
c3fcd832
KW
3175 trial_locales[0] = setlocale_init;
3176 trial_locales_count = 1;
7d4bcc4a 3177
65ebb059
KW
3178 for (i= 0; i < trial_locales_count; i++) {
3179 const char * trial_locale = trial_locales[i];
3180
3181 if (i > 0) {
3182
3183 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3184 * when i==0, but I (khw) don't think that behavior makes much
3185 * sense */
3186 setlocale_failure = FALSE;
3187
7d4bcc4a 3188# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3189# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3190
65ebb059
KW
3191 /* On Windows machines, an entry of "" after the 0th means to use
3192 * the system default locale, which we now proceed to get. */
3193 if (strEQ(trial_locale, "")) {
3194 unsigned int j;
3195
3196 /* Note that this may change the locale, but we are going to do
3197 * that anyway just below */
837ce802 3198 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3199 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3200
7d4bcc4a 3201 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3202 * try */
3203 if (! system_default_locale) {
3204 goto next_iteration;
3205 }
3206 for (j = 0; j < trial_locales_count; j++) {
3207 if (strEQ(system_default_locale, trial_locales[j])) {
3208 goto next_iteration;
3209 }
3210 }
3211
3212 trial_locale = system_default_locale;
3213 }
ec0202b5
KW
3214# else
3215# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3216# endif
7d4bcc4a 3217# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3218
3219 } /* For i > 0 */
65ebb059 3220
7d4bcc4a
KW
3221# ifdef LC_ALL
3222
948523db
KW
3223 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3224 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3225 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3226 setlocale_failure = TRUE;
7cd8b568
KW
3227 }
3228 else {
3229 /* Since LC_ALL succeeded, it should have changed all the other
3230 * categories it can to its value; so we massage things so that the
3231 * setlocales below just return their category's current values.
3232 * This adequately handles the case in NetBSD where LC_COLLATE may
3233 * not be defined for a locale, and setting it individually will
7d4bcc4a 3234 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3235 * the POSIX locale. */
3236 trial_locale = NULL;
3237 }
7d4bcc4a
KW
3238
3239# endif /* LC_ALL */
98994639 3240
e5f10d49
KW
3241 if (! setlocale_failure) {
3242 unsigned int j;
3243 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3244 curlocales[j]
3245 = savepv(do_setlocale_r(categories[j], trial_locale));
3246 if (! curlocales[j]) {
3247 setlocale_failure = TRUE;
3248 }
3249 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3250 }
c835d6be 3251
e5f10d49
KW
3252 if (! setlocale_failure) { /* All succeeded */
3253 break; /* Exit trial_locales loop */
49c85077 3254 }
65ebb059 3255 }
98994639 3256
49c85077
KW
3257 /* Here, something failed; will need to try a fallback. */
3258 ok = 0;
65ebb059 3259
49c85077
KW
3260 if (i == 0) {
3261 unsigned int j;
98994639 3262
65ebb059 3263 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3264
3265# ifdef LC_ALL
98994639 3266
49c85077
KW
3267 PerlIO_printf(Perl_error_log,
3268 "perl: warning: Setting locale failed.\n");
98994639 3269
7d4bcc4a 3270# else /* !LC_ALL */
98994639 3271
49c85077
KW
3272 PerlIO_printf(Perl_error_log,
3273 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3274
e5f10d49
KW
3275 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3276 if (! curlocales[j]) {
3277 PerlIO_printf(Perl_error_log, category_names[j]);
3278 }
3279 else {
3280 Safefree(curlocales[j]);
3281 }
3282 }
7d4bcc4a 3283
7d4bcc4a 3284# endif /* LC_ALL */
98994639 3285
49c85077
KW
3286 PerlIO_printf(Perl_error_log,
3287 "perl: warning: Please check that your locale settings:\n");
98994639 3288
7d4bcc4a
KW
3289# ifdef __GLIBC__
3290
49c85077
KW
3291 PerlIO_printf(Perl_error_log,
3292 "\tLANGUAGE = %c%s%c,\n",
3293 language ? '"' : '(',
3294 language ? language : "unset",
3295 language ? '"' : ')');
7d4bcc4a 3296# endif
98994639 3297
49c85077
KW
3298 PerlIO_printf(Perl_error_log,
3299 "\tLC_ALL = %c%s%c,\n",
3300 lc_all ? '"' : '(',
3301 lc_all ? lc_all : "unset",
3302 lc_all ? '"' : ')');
98994639 3303
7d4bcc4a
KW
3304# if defined(USE_ENVIRON_ARRAY)
3305
49c85077 3306 {
cd999af9 3307 char **e;
d5e32b93
KW
3308
3309 /* Look through the environment for any variables of the
3310 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3311 * already handled above. These are assumed to be locale
3312 * settings. Output them and their values. */
cd999af9 3313 for (e = environ; *e; e++) {
d5e32b93
KW
3314 const STRLEN prefix_len = sizeof("LC_") - 1;
3315 STRLEN uppers_len;
3316
cd999af9 3317 if ( strBEGINs(*e, "LC_")
c8b388b0 3318 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3319 && (uppers_len = strspn(*e + prefix_len,
3320 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3321 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3322 {
3323 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3324 (int) (prefix_len + uppers_len), *e,
3325 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3326 }
3327 }
49c85077 3328 }
7d4bcc4a
KW
3329
3330# else
3331
49c85077
KW
3332 PerlIO_printf(Perl_error_log,
3333 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3334
3335# endif
98994639 3336
49c85077
KW
3337 PerlIO_printf(Perl_error_log,
3338 "\tLANG = %c%s%c\n",
3339 lang ? '"' : '(',
3340 lang ? lang : "unset",
3341 lang ? '"' : ')');
98994639 3342
49c85077
KW
3343 PerlIO_printf(Perl_error_log,
3344 " are supported and installed on your system.\n");
3345 }
98994639 3346
65ebb059 3347 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3348 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3349 * usually change the upper bound of the loop we are in.
3350 *
3351 * Since the system's default way of setting the locale has not
3352 * found one that works, We use Perl's defined ordering: LC_ALL,
3353 * LANG, and the C locale. We don't try the same locale twice, so
3354 * don't add to the list if already there. (On POSIX systems, the
3355 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3356 * but there's no harm done by doing it explicitly.
3357 *
3358 * Note that this tries the LC_ALL environment variable even on
3359 * systems which have no LC_ALL locale setting. This may or may
3360 * not have been originally intentional, but there's no real need
3361 * to change the behavior. */
65ebb059
KW
3362 if (lc_all) {
3363 for (j = 0; j < trial_locales_count; j++) {
3364 if (strEQ(lc_all, trial_locales[j])) {
3365 goto done_lc_all;
3366 }
3367 }
3368 trial_locales[trial_locales_count++] = lc_all;
3369 }
3370 done_lc_all:
98994639 3371
65ebb059
KW
3372 if (lang) {
3373 for (j = 0; j < trial_locales_count; j++) {
3374 if (strEQ(lang, trial_locales[j])) {
3375 goto done_lang;
3376 }
3377 }
3378 trial_locales[trial_locales_count++] = lang;
3379 }
3380 done_lang:
3381
7d4bcc4a
KW
3382# if defined(WIN32) && defined(LC_ALL)
3383
65ebb059
KW
3384 /* For Windows, we also try the system default locale before "C".
3385 * (If there exists a Windows without LC_ALL we skip this because
3386 * it gets too complicated. For those, the "C" is the next
3387 * fallback possibility). The "" is the same as the 0th element of
3388 * the array, but the code at the loop above knows to treat it
3389 * differently when not the 0th */
3390 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3391
3392# endif
65ebb059
KW
3393
3394 for (j = 0; j < trial_locales_count; j++) {
3395 if (strEQ("C", trial_locales[j])) {
3396 goto done_C;
3397 }
3398 }
3399 trial_locales[trial_locales_count++] = "C";
98994639 3400
65ebb059
KW
3401 done_C: ;
3402 } /* end of first time through the loop */
98994639 3403
7d4bcc4a
KW
3404# ifdef WIN32
3405
65ebb059 3406 next_iteration: ;
7d4bcc4a
KW
3407
3408# endif
65ebb059
KW
3409
3410 } /* end of looping through the trial locales */
3411
3412 if (ok < 1) { /* If we tried to fallback */
3413 const char* msg;
3414 if (! setlocale_failure) { /* fallback succeeded */
3415 msg = "Falling back to";
3416 }
3417 else { /* fallback failed */
e5f10d49 3418 unsigned int j;
98994639 3419
65ebb059
KW
3420 /* We dropped off the end of the loop, so have to decrement i to
3421 * get back to the value the last time through */
3422 i--;
98994639 3423
65ebb059
KW
3424 ok = -1;
3425 msg = "Failed to fall back to";
3426
3427 /* To continue, we should use whatever values we've got */
7d4bcc4a 3428
e5f10d49
KW
3429 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3430 Safefree(curlocales[j]);
3431 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3432 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3433 }
65ebb059
KW
3434 }
3435
3436 if (locwarn) {
3437 const char * description;
3438 const char * name = "";
3439 if (strEQ(trial_locales[i], "C")) {
3440 description = "the standard locale";
3441 name = "C";
3442 }
7d4bcc4a
KW
3443
3444# ifdef SYSTEM_DEFAULT_LOCALE
3445
65ebb059
KW
3446 else if (strEQ(trial_locales[i], "")) {
3447 description = "the system default locale";
3448 if (system_default_locale) {
3449 name = system_default_locale;
3450 }
3451 }
7d4bcc4a
KW
3452
3453# endif /* SYSTEM_DEFAULT_LOCALE */
3454
65ebb059
KW
3455 else {
3456 description = "a fallback locale";
3457 name = trial_locales[i];
3458 }
3459 if (name && strNE(name, "")) {
3460 PerlIO_printf(Perl_error_log,
3461 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3462 }
3463 else {
3464 PerlIO_printf(Perl_error_log,
3465 "perl: warning: %s %s.\n", msg, description);
3466 }
3467 }
3468 } /* End of tried to fallback */
98994639 3469
e5f10d49
KW
3470 /* Done with finding the locales; update our records */
3471
7d4bcc4a
KW
3472# ifdef USE_LOCALE_CTYPE
3473
948523db 3474 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3475
e5f10d49 3476# endif
7d4bcc4a
KW
3477# ifdef USE_LOCALE_COLLATE
3478
948523db 3479 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3480
e5f10d49 3481# endif
7d4bcc4a
KW
3482# ifdef USE_LOCALE_NUMERIC
3483
948523db 3484 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3485
3486# endif
3487
948523db 3488 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3489
e9bc6d6b 3490# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3491
3492 /* This caches whether each category's locale is UTF-8 or not. This
3493 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3494 * initialization time before any threads have started, but not later
3495 * unless thread-safe operations are used.
47280b20
KW
3496 * Caching means that if the program heeds our dictate not to change
3497 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3498 * it may get queried without having to change locales. If the
3499 * environment is such that all categories have the same locale, this
3500 * isn't needed, as the code will not change the locale; but this
3501 * handles the uncommon case where the environment has disparate
3502 * locales for the categories */
47280b20
KW
3503 (void) _is_cur_LC_category_utf8(categories[i]);
3504
3505# endif
3506
e5f10d49
KW
3507 Safefree(curlocales[i]);
3508 }
b310b053 3509
7d4bcc4a
KW
3510# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3511
49c85077 3512 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3513 * locale is UTF-8. The call to new_ctype() just above has already
3514 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3515 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3516 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3517 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3518 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3519
a05d7ebb 3520 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3521 This is an alternative to using the -C command line switch
3522 (the -C if present will override this). */
3523 {
dd374669 3524 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3525 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3526 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3527 PL_utf8cache = -1;
b310b053
JH
3528 }
3529
7d4bcc4a 3530# endif
7d4bcc4a
KW
3531# ifdef __GLIBC__
3532
175c4cf9 3533 Safefree(language);
7d4bcc4a
KW
3534
3535# endif
175c4cf9
KW
3536
3537 Safefree(lc_all);
3538 Safefree(lang);
3539
e3305790 3540#endif /* USE_LOCALE */
2fcc0ca9 3541#ifdef DEBUGGING
7d4bcc4a 3542
2fcc0ca9 3543 /* So won't continue to output stuff */
27cdc72e 3544 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3545
2fcc0ca9
KW
3546#endif
3547
98994639
HS
3548 return ok;
3549}
3550
98994639
HS
3551#ifdef USE_LOCALE_COLLATE
3552
a4a439fb 3553char *
a4a439fb
KW
3554Perl__mem_collxfrm(pTHX_ const char *input_string,
3555 STRLEN len, /* Length of 'input_string' */
3556 STRLEN *xlen, /* Set to length of returned string
3557 (not including the collation index
3558 prefix) */
3559 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3560 )
98994639 3561{
a4a439fb
KW
3562
3563 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3564 * differences. First, it handles embedded NULs. Second, it allocates a bit
3565 * more memory than needed for the transformed data itself. The real
55e5378d 3566 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3567 * the length of that, and doesn't include the collation index size.
3568 * Please see sv_collxfrm() to see how this is used. */
3569
55e5378d
KW
3570#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3571
6696cfa7
KW
3572 char * s = (char *) input_string;
3573 STRLEN s_strlen = strlen(input_string);
79f120c8 3574 char *xbuf = NULL;
55e5378d 3575 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3576 STRLEN length_in_chars;
c664130f 3577 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3578
a4a439fb
KW
3579 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3580
3581 /* Must be NUL-terminated */
3582 assert(*(input_string + len) == '\0');
7918f24d 3583
79f120c8
KW
3584 /* If this locale has defective collation, skip */
3585 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3586 DEBUG_L(PerlIO_printf(Perl_debug_log,
3587 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3588 goto bad;
3589 }
3590
6696cfa7
KW
3591 /* Replace any embedded NULs with the control that sorts before any others.
3592 * This will give as good as possible results on strings that don't
3593 * otherwise contain that character, but otherwise there may be
3594 * less-than-perfect results with that character and NUL. This is
fdc080f3 3595 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3596 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3597 NUL */
6696cfa7
KW
3598 char * e = s + len;
3599 char * sans_nuls;
fdc080f3 3600 STRLEN sans_nuls_len;
94762aa0 3601 int try_non_controls;
afc4976f
KW
3602 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3603 making sure 2nd byte is NUL.
3604 */
3605 STRLEN this_replacement_len;
3606
1e4c9676
KW
3607 /* If we don't know what non-NUL control character sorts lowest for
3608 * this locale, find it */
f28f4d2a 3609 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3610 int j;
afc4976f 3611 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3612 includes the collation index
3613 prefixed. */
3614
91c0e2e0 3615 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3616
3617 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3618 * NUL, in which case we instead look for any character. Controls
3619 * are preferred because collation order is, in general, context
3620 * sensitive, with adjoining characters affecting the order, and
3621 * controls are less likely to have such interactions, allowing the
3622 * NUL-replacement to stand on its own. (Another way to look at it
3623 * is to imagine what would happen if the NUL were replaced by a
3624 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3625 for (try_non_controls = 0;
3626 try_non_controls < 2;
3627 try_non_controls++)
3628 {
d4ff9586
KW
3629 /* Look through all legal code points (NUL isn't) */
3630 for (j = 1; j < 256; j++) {
3631 char * x; /* j's xfrm plus collation index */
3632 STRLEN x_len; /* length of 'x' */
3633 STRLEN trial_len = 1;
736a4fed 3634 char cur_source[] = { '\0', '\0' };
d4ff9586 3635
736a4fed
KW
3636 /* Skip non-controls the first time through the loop. The
3637 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3638 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3639 ? ! isCNTRL_L1(j)
3640 : ! isCNTRL_LC(j))
3641 {
d4ff9586 3642 continue;
6696cfa7 3643 }
6696cfa7 3644
736a4fed
KW
3645 /* Create a 1-char string of the current code point */
3646 cur_source[0] = (char) j;
3647
d4ff9586
KW
3648 /* Then transform it */
3649 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3650 0 /* The string is not in UTF-8 */);
6696cfa7 3651
1e4c9676 3652 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3653 * */
3654 if (! x) {
3655 continue;
3656 }
6696cfa7 3657
d4ff9586
KW
3658 /* If this character's transformation is lower than
3659 * the current lowest, this one becomes the lowest */
3660 if ( cur_min_x == NULL
3661 || strLT(x + COLLXFRM_HDR_LEN,
3662 cur_min_x + COLLXFRM_HDR_LEN))
3663 {
f28f4d2a 3664 PL_strxfrm_NUL_replacement = j;
d4ff9586 3665 cur_min_x = x;
d4ff9586
KW
3666 }
3667 else {
3668 Safefree(x);
3669 }
1e4c9676 3670 } /* end of loop through all 255 characters */
6696cfa7 3671
1e4c9676 3672 /* Stop looking if found */
94762aa0
KW
3673 if (cur_min_x) {
3674 break;
3675 }
3676
3677 /* Unlikely, but possible, if there aren't any controls that
3678 * work in the locale, repeat the loop, looking for any
3679 * character that works */
3680 DEBUG_L(PerlIO_printf(Perl_debug_log,
3681 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3682 } /* End of loop to try first the controls, then any char */
6696cfa7 3683
94762aa0
KW
3684 if (! cur_min_x) {
3685 DEBUG_L(PerlIO_printf(Perl_debug_log,
3686 "_mem_collxfrm: Couldn't find any character to replace"
3687 " embedded NULs in locale %s with", PL_collation_name));
3688 goto bad;
58eebef2
KW
3689 }
3690
94762aa0
KW
3691 DEBUG_L(PerlIO_printf(Perl_debug_log,
3692 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3693 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3694
6696cfa7 3695 Safefree(cur_min_x);
1e4c9676 3696 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3697
3698 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3699 * UTF8-ness of the original */
f28f4d2a
KW
3700 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3701 this_replacement_char[0] =
3702 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3703 this_replacement_char[1] =
3704 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3705 this_replacement_len = 2;
3706 }
3707 else {
f28f4d2a 3708 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3709 /* this_replacement_char[1] = '\0' was done at initialization */
3710 this_replacement_len = 1;
6696cfa7
KW
3711 }
3712
3713 /* The worst case length for the replaced string would be if every
3714 * character in it is NUL. Multiply that by the length of each
3715 * replacement, and allow for a trailing NUL */
afc4976f 3716 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3717 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3718 *sans_nuls = '\0';
3719
6696cfa7
KW
3720 /* Replace each NUL with the lowest collating control. Loop until have
3721 * exhausted all the NULs */
3722 while (s + s_strlen < e) {
6069d6c5 3723 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3724
3725 /* Do the actual replacement */
6069d6c5 3726 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
3727
3728 /* Move past the input NUL */
3729 s += s_strlen + 1;
3730 s_strlen = strlen(s);
3731 }
3732
3733 /* And add anything that trails the final NUL */
6069d6c5 3734 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3735
3736 /* Switch so below we transform this modified string */
3737 s = sans_nuls;
3738 len = strlen(s);
1e4c9676 3739 } /* End of replacing NULs */
6696cfa7 3740
a4a439fb
KW
3741 /* Make sure the UTF8ness of the string and locale match */
3742 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 3743 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
3744 const char * const t = s; /* Temporary so we can later find where the
3745 input was */
3746
3747 /* Here they don't match. Change the string's to be what the locale is
3748 * expecting */
3749
3750 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3751 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3752 utf8 = TRUE;
3753 }
3754 else { /* locale is not UTF-8; but input is; downgrade the input */
3755
3756 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3757
3758 /* If the downgrade was successful we are done, but if the input
3759 * contains things that require UTF-8 to represent, have to do
3760 * damage control ... */
3761 if (UNLIKELY(utf8)) {
3762
3763 /* What we do is construct a non-UTF-8 string with
3764 * 1) the characters representable by a single byte converted
3765 * to be so (if necessary);
3766 * 2) and the rest converted to collate the same as the
3767 * highest collating representable character. That makes
3768 * them collate at the end. This is similar to how we
3769 * handle embedded NULs, but we use the highest collating
3770 * code point instead of the smallest. Like the NUL case,
3771 * this isn't perfect, but is the best we can reasonably
3772 * do. Every above-255 code point will sort the same as
3773 * the highest-sorting 0-255 code point. If that code
3774 * point can combine in a sequence with some other code
3775 * points for weight calculations, us changing something to
3776 * be it can adversely affect the results. But in most
3777 * cases, it should work reasonably. And note that this is
3778 * really an illegal situation: using code points above 255
3779 * on a locale where only 0-255 are valid. If two strings
3780 * sort entirely equal, then the sort order for the
3781 * above-255 code points will be in code point order. */
3782
3783 utf8 = FALSE;
3784
3785 /* If we haven't calculated the code point with the maximum
3786 * collating order for this locale, do so now */
3787 if (! PL_strxfrm_max_cp) {
3788 int j;
3789
3790 /* The current transformed string that collates the
3791 * highest (except it also includes the prefixed collation
3792 * index. */
3793 char * cur_max_x = NULL;
3794
3795 /* Look through all legal code points (NUL isn't) */
3796 for (j = 1; j < 256; j++) {
3797 char * x;
3798 STRLEN x_len;
736a4fed 3799 char cur_source[] = { '\0', '\0' };
a4a439fb 3800
736a4fed
KW
3801 /* Create a 1-char string of the current code point */
3802 cur_source[0] = (char) j;
a4a439fb
KW
3803
3804 /* Then transform it */
3805 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3806
3807 /* If something went wrong (which it shouldn't), just
3808 * ignore this code point */
94762aa0 3809 if (! x) {
a4a439fb
KW
3810 continue;
3811 }
3812
3813 /* If this character's transformation is higher than
3814 * the current highest, this one becomes the highest */
3815 if ( cur_max_x == NULL
55e5378d
KW
3816 || strGT(x + COLLXFRM_HDR_LEN,
3817 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
3818 {
3819 PL_strxfrm_max_cp = j;
3820 cur_max_x = x;
3821 }
3822 else {
3823 Safefree(x);
3824 }
3825 }
3826
94762aa0
KW
3827 if (! cur_max_x) {
3828 DEBUG_L(PerlIO_printf(Perl_debug_log,
3829 "_mem_collxfrm: Couldn't find any character to"
3830 " replace above-Latin1 chars in locale %s with",
3831 PL_collation_name));
3832 goto bad;
3833 }
3834
58eebef2
KW
3835 DEBUG_L(PerlIO_printf(Perl_debug_log,
3836 "_mem_collxfrm: highest 1-byte collating character"
3837 " in locale %s is 0x%02X\n",
3838 PL_collation_name,
3839 PL_strxfrm_max_cp));
58eebef2 3840
a4a439fb
KW
3841 Safefree(cur_max_x);
3842 }
3843
3844 /* Here we know which legal code point collates the highest.
3845 * We are ready to construct the non-UTF-8 string. The length
3846 * will be at least 1 byte smaller than the input string
3847 * (because we changed at least one 2-byte character into a
3848 * single byte), but that is eaten up by the trailing NUL */
3849 Newx(s, len, char);
3850
3851 {
3852 STRLEN i;
3853 STRLEN d= 0;
042d9e50 3854 char * e = (char *) t + len;
a4a439fb
KW
3855
3856 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3857 U8 cur_char = t[i];
3858 if (UTF8_IS_INVARIANT(cur_char)) {
3859 s[d++] = cur_char;
3860 }
042d9e50 3861 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
3862 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3863 }
3864 else { /* Replace illegal cp with highest collating
3865 one */
3866 s[d++] = PL_strxfrm_max_cp;
3867 }
3868 }
3869 s[d++] = '\0';
3870 Renew(s, d, char); /* Free up unused space */
3871 }
3872 }
3873 }
3874
3875 /* Here, we have constructed a modified version of the input. It could
3876 * be that we already had a modified copy before we did this version.
3877 * If so, that copy is no longer needed */
3878 if (t != input_string) {
3879 Safefree(t);
3880 }
3881 }
3882
17f41037
KW
3883 length_in_chars = (utf8)
3884 ? utf8_length((U8 *) s, (U8 *) s + len)
3885 : len;
3886
59c018b9
KW
3887 /* The first element in the output is the collation id, used by
3888 * sv_collxfrm(); then comes the space for the transformed string. The
3889 * equation should give us a good estimate as to how much is needed */
55e5378d 3890 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 3891 + PL_collxfrm_base
17f41037 3892 + (PL_collxfrm_mult * length_in_chars);
a02a5408 3893 Newx(xbuf, xAlloc, char);
c7202dee
KW
3894 if (UNLIKELY(! xbuf)) {
3895 DEBUG_L(PerlIO_printf(Perl_debug_log,
3896 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 3897 goto bad;
c7202dee 3898 }
9899