This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
t/lib/croak/regcomp: Fix up for EBCDIC
[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, "")) {
929a7f8c 828 default_name = "C";
e9bc6d6b
KW
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;
bee74f4b
KW
937 while (p < e && *p != ';') {
938 if (! isGRAPH(*p)) {
939 Perl_croak(aTHX_
940 "panic: %s: %d: Unexpected character in locale name '%02X",
941 __FILE__, __LINE__, *(p-1));
942 }
e9bc6d6b
KW
943 p++;
944 }
945 name_end = p;
946
bee74f4b
KW
947 /* Space past the semi-colon */
948 if (p < e) {
949 p++;
e9bc6d6b
KW
950 }
951
952 /* Find the index of the category name in our lists */
953 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 954 char * individ_locale;
e9bc6d6b
KW
955
956 /* Keep going if this isn't the index. The strnNE() avoids a
957 * Perl_form(), but would fail if ever a category name could be
958 * a substring of another one, like if there were a
959 * "LC_TIME_DATE" */
960 if strnNE(s, category_names[i], category_end - s) {
961 continue;
962 }
963
964 /* If this index is for the single category we're changing, we
965 * have found the locale to set it to. */
966 if (category == categories[i]) {
967 locale = Perl_form(aTHX_ "%.*s",
968 (int) (name_end - name_start),
969 name_start);
970 goto ready_to_set;
971 }
972
3c76a412 973 assert(category == LC_ALL);
bee74f4b
KW
974 individ_locale = Perl_form(aTHX_ "%.*s",
975 (int) (name_end - name_start), name_start);
6435f98d
KW
976 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
977 {
978 return NULL;
979 }
e9bc6d6b
KW
980 }
981
982 s = p;
983 }
984
985 /* Here we have set all the individual categories by recursive calls.
986 * These collectively should have fixed up LC_ALL, so can just query
987 * what that now is */
988 assert(category == LC_ALL);
989
990 return do_setlocale_c(LC_ALL, NULL);
991 }
992
993 ready_to_set: ;
994
995# endif /* end of ! querylocale */
996
997 /* Ready to create a new locale by modification of the exising one */
998 new_obj = newlocale(mask, locale, old_obj);
999
1000 if (! new_obj) {
1001 dSAVE_ERRNO;
1002
1003# ifdef DEBUGGING
1004
1005 if (DEBUG_L_TEST || debug_initialization) {
1006 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1007 }
1008
1009# endif
1010
1011 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1012
1013# ifdef DEBUGGING
1014
1015 if (DEBUG_L_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1017 }
1018
1019# endif
1020
1021 }
1022 RESTORE_ERRNO;
1023 return NULL;
1024 }
1025
1026# ifdef DEBUGGING
1027
1028 if (DEBUG_Lv_TEST || debug_initialization) {
1029 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1030 }
1031
1032# endif
1033
1034 /* And switch into it */
1035 if (! uselocale(new_obj)) {
1036 dSAVE_ERRNO;
1037
1038# ifdef DEBUGGING
1039
1040 if (DEBUG_L_TEST || debug_initialization) {
1041 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1042 }
1043
1044# endif
1045
1046 if (! uselocale(old_obj)) {
1047
1048# ifdef DEBUGGING
1049
1050 if (DEBUG_L_TEST || debug_initialization) {
1051 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1052 }
1053
1054# endif
1055
1056 }
1057 freelocale(new_obj);
1058 RESTORE_ERRNO;
1059 return NULL;
1060 }
1061
1062# ifdef DEBUGGING
1063
1064 if (DEBUG_Lv_TEST || debug_initialization) {
1065 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1066 }
1067
1068# endif
1069
1070 /* We are done, except for updating our records (if the system doesn't keep
1071 * them) and in the case of locale "", we don't actually know what the
1072 * locale that got switched to is, as it came from the environment. So
1073 * have to find it */
1074
1075# ifdef HAS_QUERYLOCALE
1076
1077 if (strEQ(locale, "")) {
1078 locale = querylocale(mask, new_obj);
1079 }
1080
1081# else
1082
1083 /* Here, 'locale' is the return value */
1084
1085 /* Without querylocale(), we have to update our records */
1086
1087 if (category == LC_ALL) {
1088 unsigned int i;
1089
1090 /* For LC_ALL, we change all individual categories to correspond */
1091 /* PL_curlocales is a parallel array, so has same
1092 * length as 'categories' */
1093 for (i = 0; i <= LC_ALL_INDEX; i++) {
1094 Safefree(PL_curlocales[i]);
1095 PL_curlocales[i] = savepv(locale);
1096 }
1097 }
1098 else {
1099
1100 /* For a single category, if it's not the same as the one in LC_ALL, we
1101 * nullify LC_ALL */
1102
1103 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1104 Safefree(PL_curlocales[LC_ALL_INDEX]);
1105 PL_curlocales[LC_ALL_INDEX] = NULL;
1106 }
1107
1108 /* Then update the category's record */
1109 Safefree(PL_curlocales[index]);
1110 PL_curlocales[index] = savepv(locale);
1111 }
1112
1113# endif
1114
1115 return locale;
1116}
1117
1118#endif /* USE_POSIX_2008_LOCALE */
1119
1120#if 0 /* Code that was to emulate thread-safe locales on platforms that
1121 didn't natively support them */
1122
1123/* The way this would work is that we would keep a per-thread list of the
1124 * correct locale for that thread. Any operation that was locale-sensitive
1125 * would have to be changed so that it would look like this:
1126 *
1127 * LOCALE_LOCK;
1128 * setlocale to the correct locale for this operation
1129 * do operation
1130 * LOCALE_UNLOCK
1131 *
1132 * This leaves the global locale in the most recently used operation's, but it
1133 * was locked long enough to get the result. If that result is static, it
1134 * needs to be copied before the unlock.
1135 *
1136 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1137 * the setup, but are no-ops when not needed, and similarly,
1138 * END_LOCALE_DEPENDENT_OP for the tear-down
1139 *
1140 * But every call to a locale-sensitive function would have to be changed, and
1141 * if a module didn't cooperate by using the mutex, things would break.
1142 *
1143 * This code was abandoned before being completed or tested, and is left as-is
1144*/
1145
1146# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1147# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1148
1149STATIC char *
1150S_locking_setlocale(pTHX_
1151 const int category,
1152 const char * locale,
1153 int index,
1154 const bool is_index_valid
1155 )
1156{
1157 /* This function kind of performs a setlocale() on just the current thread;
1158 * thus it is kind of thread-safe. It does this by keeping a thread-level
1159 * array of the current locales for each category. Every time a locale is
1160 * switched to, it does the switch globally, but updates the thread's
1161 * array. A query as to what the current locale is just returns the
1162 * appropriate element from the array, and doesn't actually call the system
1163 * setlocale(). The saving into the array is done in an uninterruptible
1164 * section of code, so is unaffected by whatever any other threads might be
1165 * doing.
1166 *
1167 * All locale-sensitive operations must work by first starting a critical
1168 * section, then switching to the thread's locale as kept by this function,
1169 * and then doing the operation, then ending the critical section. Thus,
1170 * each gets done in the appropriate locale. simulating thread-safety.
1171 *
1172 * This function takes the same parameters, 'category' and 'locale', that
1173 * the regular setlocale() function does, but it also takes two additional
1174 * ones. This is because as described earlier. If we know on input the
1175 * index corresponding to the category into the array where we store the
1176 * current locales, we don't have to calculate it. If the caller knows at
1177 * compile time what the index is, it it can pass it, setting
1178 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1179 *
1180 */
1181
1182 /* If the input index might be incorrect, calculate the correct one */
1183 if (! is_index_valid) {
1184 unsigned int i;
1185
1186 if (DEBUG_Lv_TEST || debug_initialization) {
1187 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1188 }
1189
1190 for (i = 0; i <= LC_ALL_INDEX; i++) {
1191 if (category == categories[i]) {
1192 index = i;
1193 goto found_index;
1194 }
1195 }
1196
1197 /* Here, we don't know about this category, so can't handle it.
1198 * XXX best we can do is to unsafely set this
1199 * XXX warning */
1200
1201 return my_setlocale(category, locale);
1202
1203 found_index: ;
1204
1205 if (DEBUG_Lv_TEST || debug_initialization) {
1206 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1207 }
1208 }
1209
1210 /* For a query, just return what's in our records */
1211 if (new_locale == NULL) {
1212 return curlocales[index];
1213 }
1214
1215
1216 /* Otherwise, we need to do the switch, and save the result, all in a
1217 * critical section */
1218
1219 Safefree(curlocales[[index]]);
1220
1221 /* It might be that this is called from an already-locked section of code.
1222 * We would have to detect and skip the LOCK/UNLOCK if so */
1223 LOCALE_LOCK;
1224
1225 curlocales[index] = savepv(my_setlocale(category, new_locale));
1226
1227 if (strEQ(new_locale, "")) {
1228
1229#ifdef LC_ALL
1230
1231 /* The locale values come from the environment, and may not all be the
1232 * same, so for LC_ALL, we have to update all the others, while the
1233 * mutex is still locked */
1234
1235 if (category == LC_ALL) {
1236 unsigned int i;
1237 for (i = 0; i < LC_ALL_INDEX) {
1238 curlocales[i] = my_setlocale(categories[i], NULL);
1239 }
1240 }
1241 }
1242
1243#endif
1244
1245 LOCALE_UNLOCK;
1246
1247 return curlocales[index];
1248}
1249
1250#endif
837ce802 1251
a4f00dcc 1252STATIC void
86799d2d 1253S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1254{
86799d2d
KW
1255 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1256 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1257
86799d2d
KW
1258#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1259 || defined(HAS_NL_LANGINFO))
98994639 1260
87f8e8e7 1261 const char * radix = (use_locale)
4e6826bf 1262 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1263 /* FALSE => already in dest locale */
1264 : ".";
2213a3be 1265
87f8e8e7 1266 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1267
87f8e8e7
KW
1268 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1269 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1270 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1271 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1272 && _is_cur_LC_category_utf8(LC_NUMERIC))
1273 {
1274 SvUTF8_on(PL_numeric_radix_sv);
1275 }
86799d2d
KW
1276
1277# ifdef DEBUGGING
7d4bcc4a 1278
2fcc0ca9
KW
1279 if (DEBUG_L_TEST || debug_initialization) {
1280 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1281 SvPVX(PL_numeric_radix_sv),
1282 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1283 }
69014004 1284
86799d2d
KW
1285# endif
1286#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1287
98994639
HS
1288}
1289
398eeea9
KW
1290STATIC void
1291S_new_numeric(pTHX_ const char *newnum)
98994639 1292{
7d4bcc4a
KW
1293
1294#ifndef USE_LOCALE_NUMERIC
1295
1296 PERL_UNUSED_ARG(newnum);
1297
1298#else
0d071d52 1299
291a84fb 1300 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1301 * core Perl this and that 'newnum' is the name of the new locale.
1302 * It installs this locale as the current underlying default.
1303 *
1304 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1305 * set_numeric_underlying() and set_numeric_standard() functions, which
1306 * should probably not be called directly, but only via macros like
0d071d52
KW
1307 * SET_NUMERIC_STANDARD() in perl.h.
1308 *
1309 * The toggling is necessary mainly so that a non-dot radix decimal point
1310 * character can be output, while allowing internal calculations to use a
1311 * dot.
1312 *
1313 * This sets several interpreter-level variables:
bb304765 1314 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1315 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1316 * that the current locale is the program's underlying
1317 * locale
1318 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1319 * that the current locale is the C locale or
1320 * indistinguishable from the C locale. If non-zero, it
1321 * is in C; if > 1, it means it may not be toggled away
7738054c 1322 * from C.
4c68b815
KW
1323 * PL_numeric_underlying_is_standard A bool kept by this function
1324 * indicating that the underlying locale and the standard
1325 * C locale are indistinguishable for the purposes of
1326 * LC_NUMERIC. This happens when both of the above two
1327 * variables are true at the same time. (Toggling is a
1328 * no-op under these circumstances.) This variable is
1329 * used to avoid having to recalculate.
398eeea9 1330 */
0d071d52 1331
b03f34cf 1332 char *save_newnum;
98994639
HS
1333
1334 if (! newnum) {
43c5f42d
NC
1335 Safefree(PL_numeric_name);
1336 PL_numeric_name = NULL;
98994639 1337 PL_numeric_standard = TRUE;
892e6465 1338 PL_numeric_underlying = TRUE;
4c68b815 1339 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1340 return;
1341 }
1342
b03f34cf 1343 save_newnum = stdize_locale(savepv(newnum));
892e6465 1344 PL_numeric_underlying = TRUE;
4c68b815
KW
1345 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1346
69c5e0db
KW
1347#ifndef TS_W32_BROKEN_LOCALECONV
1348
4c68b815 1349 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1350 * them. But on broken Windows systems calling my_nl_langinfo() for
1351 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1352 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1353 if (! PL_numeric_standard) {
4e6826bf 1354 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1355 FALSE /* Don't toggle locale */ ))
4e6826bf 1356 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1357 }
abe1abcf 1358
69c5e0db
KW
1359#endif
1360
4c68b815 1361 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1362 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1363 Safefree(PL_numeric_name);
b03f34cf 1364 PL_numeric_name = save_newnum;
b03f34cf 1365 }
abe1abcf
KW
1366 else {
1367 Safefree(save_newnum);
1368 }
4c28b29c 1369
4c68b815
KW
1370 PL_numeric_underlying_is_standard = PL_numeric_standard;
1371
7e5377f7 1372# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1373
1374 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1375 PL_numeric_name,
1376 PL_underlying_numeric_obj);
1377
1378#endif
1379
4c68b815
KW
1380 if (DEBUG_L_TEST || debug_initialization) {
1381 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1382 }
1383
4c28b29c
KW
1384 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1385 * have to worry about the radix being a non-dot. (Core operations that
1386 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1387 if (PL_numeric_standard) {
1388 set_numeric_radix(0);
1389 }
1390 else {
1391 set_numeric_standard();
1392 }
4c28b29c 1393
98994639 1394#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1395
98994639
HS
1396}
1397
1398void
1399Perl_set_numeric_standard(pTHX)
1400{
7d4bcc4a 1401
98994639 1402#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1403
28c1bf33
KW
1404 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1405 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1406 * macro avoids calling this routine if toggling isn't necessary according
1407 * to our records (which could be wrong if some XS code has changed the
1408 * locale behind our back) */
0d071d52 1409
837ce802 1410 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1411 PL_numeric_standard = TRUE;
4c68b815 1412 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1413 set_numeric_radix(0);
7d4bcc4a
KW
1414
1415# ifdef DEBUGGING
1416
2fcc0ca9
KW
1417 if (DEBUG_L_TEST || debug_initialization) {
1418 PerlIO_printf(Perl_debug_log,
58e4a467 1419 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1420 }
98994639 1421
7d4bcc4a 1422# endif
98994639 1423#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1424
98994639
HS
1425}
1426
1427void
5792c642 1428Perl_set_numeric_underlying(pTHX)
98994639 1429{
7d4bcc4a 1430
98994639 1431#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1432
28c1bf33 1433 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1434 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1435 * instead of calling this directly. The macro avoids calling this routine
1436 * if toggling isn't necessary according to our records (which could be
1437 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1438
837ce802 1439 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1440 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1441 PL_numeric_underlying = TRUE;
3c62fd46 1442 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1443
1444# ifdef DEBUGGING
1445
2fcc0ca9
KW
1446 if (DEBUG_L_TEST || debug_initialization) {
1447 PerlIO_printf(Perl_debug_log,
58e4a467 1448 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1449 PL_numeric_name);
1450 }
98994639 1451
7d4bcc4a 1452# endif
98994639 1453#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1454
98994639
HS
1455}
1456
1457/*
1458 * Set up for a new ctype locale.
1459 */
a4f00dcc
KW
1460STATIC void
1461S_new_ctype(pTHX_ const char *newctype)
98994639 1462{
7d4bcc4a
KW
1463
1464#ifndef USE_LOCALE_CTYPE
1465
1466 PERL_ARGS_ASSERT_NEW_CTYPE;
1467 PERL_UNUSED_ARG(newctype);
1468 PERL_UNUSED_CONTEXT;
1469
1470#else
0d071d52 1471
291a84fb 1472 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1473 * core Perl this and that 'newctype' is the name of the new locale.
1474 *
1475 * This function sets up the folding arrays for all 256 bytes, assuming
1476 * that tofold() is tolc() since fold case is not a concept in POSIX,
1477 *
1478 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1479 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1480 * this function should be called directly only from this file and from
0d071d52
KW
1481 * POSIX::setlocale() */
1482
27da23d5 1483 dVAR;
013f4e03 1484 unsigned int i;
98994639 1485
8b7358b9
KW
1486 /* Don't check for problems if we are suppressing the warnings */
1487 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1488
7918f24d
NC
1489 PERL_ARGS_ASSERT_NEW_CTYPE;
1490
215c5139
KW
1491 /* We will replace any bad locale warning with 1) nothing if the new one is
1492 * ok; or 2) a new warning for the bad new locale */
1493 if (PL_warn_locale) {
1494 SvREFCNT_dec_NN(PL_warn_locale);
1495 PL_warn_locale = NULL;
1496 }
1497
c1284011 1498 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1499
1500 /* A UTF-8 locale gets standard rules. But note that code still has to
1501 * handle this specially because of the three problematic code points */
1502 if (PL_in_utf8_CTYPE_locale) {
1503 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1504 }
8b7358b9
KW
1505
1506 /* We don't populate the other lists if a UTF-8 locale, but do check that
1507 * everything works as expected, unless checking turned off */
1508 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1509 /* Assume enough space for every character being bad. 4 spaces each
1510 * for the 94 printable characters that are output like "'x' "; and 5
1511 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1512 * NUL */
8b7358b9 1513 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1514 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1515 to start */
1516 unsigned int bad_count = 0; /* Count of bad characters */
1517
baa60164 1518 for (i = 0; i < 256; i++) {
8b7358b9 1519 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1520 if (isupper(i))
1521 PL_fold_locale[i] = (U8) tolower(i);
1522 else if (islower(i))
1523 PL_fold_locale[i] = (U8) toupper(i);
1524 else
1525 PL_fold_locale[i] = (U8) i;
8b7358b9 1526 }
8c6180a9
KW
1527
1528 /* If checking for locale problems, see if the native ASCII-range
1529 * printables plus \n and \t are in their expected categories in
1530 * the new locale. If not, this could mean big trouble, upending
1531 * Perl's and most programs' assumptions, like having a
1532 * metacharacter with special meaning become a \w. Fortunately,
1533 * it's very rare to find locales that aren't supersets of ASCII
1534 * nowadays. It isn't a problem for most controls to be changed
1535 * into something else; we check only \n and \t, though perhaps \r
1536 * could be an issue as well. */
7d4bcc4a 1537 if ( check_for_problems
8c6180a9
KW
1538 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1539 {
8b7358b9
KW
1540 bool is_bad = FALSE;
1541 char name[3] = { '\0' };
1542
1543 /* Convert the name into a string */
1544 if (isPRINT_A(i)) {
1545 name[0] = i;
1546 name[1] = '\0';
1547 }
1548 else if (i == '\n') {
1549 my_strlcpy(name, "\n", sizeof(name));
1550 }
1551 else {
1552 my_strlcpy(name, "\t", sizeof(name));
1553 }
1554
1555 /* Check each possibe class */
1556 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1557 is_bad = TRUE;
1558 DEBUG_L(PerlIO_printf(Perl_debug_log,
1559 "isalnum('%s') unexpectedly is %d\n",
1560 name, cBOOL(isalnum(i))));
1561 }
1562 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1563 is_bad = TRUE;
1564 DEBUG_L(PerlIO_printf(Perl_debug_log,
1565 "isalpha('%s') unexpectedly is %d\n",
1566 name, cBOOL(isalpha(i))));
1567 }
1568 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1569 is_bad = TRUE;
1570 DEBUG_L(PerlIO_printf(Perl_debug_log,
1571 "isdigit('%s') unexpectedly is %d\n",
1572 name, cBOOL(isdigit(i))));
1573 }
1574 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1575 is_bad = TRUE;
1576 DEBUG_L(PerlIO_printf(Perl_debug_log,
1577 "isgraph('%s') unexpectedly is %d\n",
1578 name, cBOOL(isgraph(i))));
1579 }
1580 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1581 is_bad = TRUE;
1582 DEBUG_L(PerlIO_printf(Perl_debug_log,
1583 "islower('%s') unexpectedly is %d\n",
1584 name, cBOOL(islower(i))));
1585 }
1586 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1587 is_bad = TRUE;
1588 DEBUG_L(PerlIO_printf(Perl_debug_log,
1589 "isprint('%s') unexpectedly is %d\n",
1590 name, cBOOL(isprint(i))));
1591 }
1592 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1593 is_bad = TRUE;
1594 DEBUG_L(PerlIO_printf(Perl_debug_log,
1595 "ispunct('%s') unexpectedly is %d\n",
1596 name, cBOOL(ispunct(i))));
1597 }
1598 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1599 is_bad = TRUE;
1600 DEBUG_L(PerlIO_printf(Perl_debug_log,
1601 "isspace('%s') unexpectedly is %d\n",
1602 name, cBOOL(isspace(i))));
1603 }
1604 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1605 is_bad = TRUE;
1606 DEBUG_L(PerlIO_printf(Perl_debug_log,
1607 "isupper('%s') unexpectedly is %d\n",
1608 name, cBOOL(isupper(i))));
1609 }
1610 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1611 is_bad = TRUE;
1612 DEBUG_L(PerlIO_printf(Perl_debug_log,
1613 "isxdigit('%s') unexpectedly is %d\n",
1614 name, cBOOL(isxdigit(i))));
1615 }
1616 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1617 is_bad = TRUE;
1618 DEBUG_L(PerlIO_printf(Perl_debug_log,
1619 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1620 name, tolower(i), (int) toLOWER_A(i)));
1621 }
1622 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1623 is_bad = TRUE;
1624 DEBUG_L(PerlIO_printf(Perl_debug_log,
1625 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1626 name, toupper(i), (int) toUPPER_A(i)));
1627 }
1628 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1629 is_bad = TRUE;
1630 DEBUG_L(PerlIO_printf(Perl_debug_log,
1631 "'\\n' (=%02X) is not a control\n", (int) i));
1632 }
1633
1634 /* Add to the list; Separate multiple entries with a blank */
1635 if (is_bad) {
1636 if (bad_count) {
1637 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1638 }
8b7358b9
KW
1639 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1640 bad_count++;
8c6180a9
KW
1641 }
1642 }
1643 }
1644
7d4bcc4a
KW
1645# ifdef MB_CUR_MAX
1646
8c6180a9 1647 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1648 * this locale requires more than one byte, there are going to be
8c6180a9 1649 * problems. */
9c8a6dc2
KW
1650 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1651 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1652 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1653
8b7358b9
KW
1654 if ( check_for_problems && MB_CUR_MAX > 1
1655 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1656
1657 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1658 * locale. Just assume that the implementation for them (plus
1659 * for POSIX) is correct and the > 1 value is spurious. (Since
1660 * these are specially handled to never be considered UTF-8
1661 * locales, as long as this is the only problem, everything
1662 * should work fine */
1663 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1664 {
8c6180a9
KW
1665 multi_byte_locale = TRUE;
1666 }
7d4bcc4a
KW
1667
1668# endif
8c6180a9 1669
8b7358b9
KW
1670 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1671 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1672 PL_warn_locale = Perl_newSVpvf(aTHX_
1673 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1674 " which have\nunexpected meanings: %s\nThe Perl program"
1675 " will use the expected meanings",
8b7358b9 1676 newctype, bad_chars_list);
8b7358b9
KW
1677 }
1678 else {
1679 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1680 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1681 newctype,
8c6180a9
KW
1682 (multi_byte_locale)
1683 ? " Some characters in it are not recognized by"
1684 " Perl."
1685 : "",
1686 (bad_count)
1687 ? "\nThe following characters (and maybe others)"
1688 " may not have the same meaning as the Perl"
1689 " program expects:\n"
1690 : "",
1691 (bad_count)
1692 ? bad_chars_list
1693 : ""
1694 );
8b7358b9
KW
1695 }
1696
b119c2be
KW
1697# ifdef HAS_NL_LANGINFO
1698
1699 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1700 /* parameter FALSE is a don't care here */
4e6826bf 1701 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1702
1703# endif
1704
8b7358b9
KW
1705 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1706
cc9eaeb0 1707 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1708 * output the message now. If not in that scope, we save the
1709 * message to be output at the first operation using this locale,
1710 * if that actually happens. Most programs don't use locales, so
1711 * they are immune to bad ones. */
cc9eaeb0 1712 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1713
780fcc9f
KW
1714 /* The '0' below suppresses a bogus gcc compiler warning */
1715 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1716
bddebb56
KW
1717 if (IN_LC(LC_CTYPE)) {
1718 SvREFCNT_dec_NN(PL_warn_locale);
1719 PL_warn_locale = NULL;
1720 }
780fcc9f 1721 }
baa60164 1722 }
31f05a37 1723 }
98994639
HS
1724
1725#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1726
98994639
HS
1727}
1728
98994639 1729void
2726666d
KW
1730Perl__warn_problematic_locale()
1731{
2726666d
KW
1732
1733#ifdef USE_LOCALE_CTYPE
1734
5f04a188
KW
1735 dTHX;
1736
1737 /* Internal-to-core function that outputs the message in PL_warn_locale,
1738 * and then NULLS it. Should be called only through the macro
1739 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1740
2726666d 1741 if (PL_warn_locale) {
2726666d
KW
1742 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1743 SvPVX(PL_warn_locale),
1744 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1745 SvREFCNT_dec_NN(PL_warn_locale);
1746 PL_warn_locale = NULL;
1747 }
1748
1749#endif
1750
1751}
1752
a4f00dcc
KW
1753STATIC void
1754S_new_collate(pTHX_ const char *newcoll)
98994639 1755{
7d4bcc4a
KW
1756
1757#ifndef USE_LOCALE_COLLATE
1758
1759 PERL_UNUSED_ARG(newcoll);
1760 PERL_UNUSED_CONTEXT;
1761
1762#else
0d071d52 1763
291a84fb 1764 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1765 * core Perl this and that 'newcoll' is the name of the new locale.
1766 *
d35fca5f
KW
1767 * The design of locale collation is that every locale change is given an
1768 * index 'PL_collation_ix'. The first time a string particpates in an
1769 * operation that requires collation while locale collation is active, it
1770 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1771 * magic includes the collation index, and the transformation of the string
1772 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1773 * instead of the string itself. If a string changes, the magic is
1774 * cleared. The next time the locale changes, the index is incremented,
1775 * and so we know during a comparison that the transformation is not
1776 * necessarily still valid, and so is recomputed. Note that if the locale
1777 * changes enough times, the index could wrap (a U32), and it is possible
1778 * that a transformation would improperly be considered valid, leading to
1779 * an unlikely bug */
0d071d52 1780
98994639
HS
1781 if (! newcoll) {
1782 if (PL_collation_name) {
1783 ++PL_collation_ix;
1784 Safefree(PL_collation_name);
1785 PL_collation_name = NULL;
1786 }
1787 PL_collation_standard = TRUE;
00bf60ca 1788 is_standard_collation:
98994639
HS
1789 PL_collxfrm_base = 0;
1790 PL_collxfrm_mult = 2;
165a1c52 1791 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1792 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1793 PL_strxfrm_max_cp = 0;
98994639
HS
1794 return;
1795 }
1796
d35fca5f 1797 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1798 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1799 ++PL_collation_ix;
1800 Safefree(PL_collation_name);
1801 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1802 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1803 if (PL_collation_standard) {
1804 goto is_standard_collation;
1805 }
98994639 1806
165a1c52 1807 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1808 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1809 PL_strxfrm_max_cp = 0;
165a1c52 1810
59c018b9
KW
1811 /* A locale collation definition includes primary, secondary, tertiary,
1812 * etc. weights for each character. To sort, the primary weights are
1813 * used, and only if they compare equal, then the secondary weights are
1814 * used, and only if they compare equal, then the tertiary, etc.
1815 *
1816 * strxfrm() works by taking the input string, say ABC, and creating an
1817 * output transformed string consisting of first the primary weights,
1818 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1819 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1820 * may not have weights at every level. In our example, let's say B
1821 * doesn't have a tertiary weight, and A doesn't have a secondary
1822 * weight. The constructed string is then going to be
1823 * A¹B¹C¹ B²C² A³C³ ....
1824 * This has the desired effect that strcmp() will look at the secondary
1825 * or tertiary weights only if the strings compare equal at all higher
1826 * priority weights. The spaces shown here, like in
c342d20e 1827 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1828 * are not just for readability. In the general case, these must
1829 * actually be bytes, which we will call here 'separator weights'; and
1830 * they must be smaller than any other weight value, but since these
1831 * are C strings, only the terminating one can be a NUL (some
1832 * implementations may include a non-NUL separator weight just before
1833 * the NUL). Implementations tend to reserve 01 for the separator
1834 * weights. They are needed so that a shorter string's secondary
1835 * weights won't be misconstrued as primary weights of a longer string,
1836 * etc. By making them smaller than any other weight, the shorter
1837 * string will sort first. (Actually, if all secondary weights are
1838 * smaller than all primary ones, there is no need for a separator
1839 * weight between those two levels, etc.)
1840 *
1841 * The length of the transformed string is roughly a linear function of
1842 * the input string. It's not exactly linear because some characters
1843 * don't have weights at all levels. When we call strxfrm() we have to
1844 * allocate some memory to hold the transformed string. The
1845 * calculations below try to find coefficients 'm' and 'b' for this
1846 * locale so that m*x + b equals how much space we need, given the size
1847 * of the input string in 'x'. If we calculate too small, we increase
1848 * the size as needed, and call strxfrm() again, but it is better to
1849 * get it right the first time to avoid wasted expensive string
1850 * transformations. */
1851
98994639 1852 {
79f120c8
KW
1853 /* We use the string below to find how long the tranformation of it
1854 * is. Almost all locales are supersets of ASCII, or at least the
1855 * ASCII letters. We use all of them, half upper half lower,
1856 * because if we used fewer, we might hit just the ones that are
1857 * outliers in a particular locale. Most of the strings being
1858 * collated will contain a preponderance of letters, and even if
1859 * they are above-ASCII, they are likely to have the same number of
1860 * weight levels as the ASCII ones. It turns out that digits tend
1861 * to have fewer levels, and some punctuation has more, but those
1862 * are relatively sparse in text, and khw believes this gives a
1863 * reasonable result, but it could be changed if experience so
1864 * dictates. */
1865 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1866 char * x_longer; /* Transformed 'longer' */
1867 Size_t x_len_longer; /* Length of 'x_longer' */
1868
1869 char * x_shorter; /* We also transform a substring of 'longer' */
1870 Size_t x_len_shorter;
1871
a4a439fb 1872 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1873 * are interested only in its length). It is used because it has
1874 * the intelligence to handle all cases, but to work, it needs some
1875 * values of 'm' and 'b' to get it started. For the purposes of
1876 * this calculation we use a very conservative estimate of 'm' and
1877 * 'b'. This assumes a weight can be multiple bytes, enough to
1878 * hold any UV on the platform, and there are 5 levels, 4 weight
1879 * bytes, and a trailing NUL. */
1880 PL_collxfrm_base = 5;
1881 PL_collxfrm_mult = 5 * sizeof(UV);
1882
1883 /* Find out how long the transformation really is */
a4a439fb
KW
1884 x_longer = _mem_collxfrm(longer,
1885 sizeof(longer) - 1,
1886 &x_len_longer,
1887
1888 /* We avoid converting to UTF-8 in the
1889 * called function by telling it the
1890 * string is in UTF-8 if the locale is a
1891 * UTF-8 one. Since the string passed
1892 * here is invariant under UTF-8, we can
1893 * claim it's UTF-8 even though it isn't.
1894 * */
1895 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1896 Safefree(x_longer);
1897
1898 /* Find out how long the transformation of a substring of 'longer'
1899 * is. Together the lengths of these transformations are
1900 * sufficient to calculate 'm' and 'b'. The substring is all of
1901 * 'longer' except the first character. This minimizes the chances
1902 * of being swayed by outliers */
a4a439fb 1903 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1904 sizeof(longer) - 2,
a4a439fb
KW
1905 &x_len_shorter,
1906 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1907 Safefree(x_shorter);
1908
1909 /* If the results are nonsensical for this simple test, the whole
1910 * locale definition is suspect. Mark it so that locale collation
1911 * is not active at all for it. XXX Should we warn? */
1912 if ( x_len_shorter == 0
1913 || x_len_longer == 0
1914 || x_len_shorter >= x_len_longer)
1915 {
1916 PL_collxfrm_mult = 0;
1917 PL_collxfrm_base = 0;
1918 }
1919 else {
1920 SSize_t base; /* Temporary */
1921
1922 /* We have both: m * strlen(longer) + b = x_len_longer
1923 * m * strlen(shorter) + b = x_len_shorter;
1924 * subtracting yields:
1925 * m * (strlen(longer) - strlen(shorter))
1926 * = x_len_longer - x_len_shorter
1927 * But we have set things up so that 'shorter' is 1 byte smaller
1928 * than 'longer'. Hence:
1929 * m = x_len_longer - x_len_shorter
1930 *
1931 * But if something went wrong, make sure the multiplier is at
1932 * least 1.
1933 */
1934 if (x_len_longer > x_len_shorter) {
1935 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1936 }
1937 else {
1938 PL_collxfrm_mult = 1;
1939 }
1940
1941 /* mx + b = len
1942 * so: b = len - mx
1943 * but in case something has gone wrong, make sure it is
1944 * non-negative */
1945 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1946 if (base < 0) {
1947 base = 0;
1948 }
1949
1950 /* Add 1 for the trailing NUL */
1951 PL_collxfrm_base = base + 1;
1952 }
58eebef2 1953
7d4bcc4a
KW
1954# ifdef DEBUGGING
1955
58eebef2
KW
1956 if (DEBUG_L_TEST || debug_initialization) {
1957 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1958 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1959 "x_len_longer=%zu,"
1960 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1961 __FILE__, __LINE__,
1962 PL_in_utf8_COLLATE_locale,
1963 x_len_shorter, x_len_longer,
1964 PL_collxfrm_mult, PL_collxfrm_base);
1965 }
7d4bcc4a
KW
1966# endif
1967
98994639
HS
1968 }
1969 }
1970
1971#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1972
98994639
HS
1973}
1974
d2b24094 1975#ifdef WIN32
b8cc575c 1976
a4f00dcc 1977STATIC char *
b8cc575c 1978S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1979{
1980 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1981 * difference between the two unless the input locale is "", which normally
1982 * means on Windows to get the machine default, which is set via the
1983 * computer's "Regional and Language Options" (or its current equivalent).
1984 * In POSIX, it instead means to find the locale from the user's
1985 * environment. This routine changes the Windows behavior to first look in
1986 * the environment, and, if anything is found, use that instead of going to
1987 * the machine default. If there is no environment override, the machine
1988 * default is used, by calling the real setlocale() with "".
1989 *
1990 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1991 * use the particular category's variable if set; otherwise to use the LANG
1992 * variable. */
b385bb4d 1993
175c4cf9 1994 bool override_LC_ALL = FALSE;
89f7b9aa 1995 char * result;
e5f10d49 1996 unsigned int i;
89f7b9aa 1997
b385bb4d 1998 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1999
2000# ifdef LC_ALL
2001
b385bb4d
KW
2002 locale = PerlEnv_getenv("LC_ALL");
2003 if (! locale) {
e5f10d49
KW
2004 if (category == LC_ALL) {
2005 override_LC_ALL = TRUE;
2006 }
2007 else {
7d4bcc4a
KW
2008
2009# endif
7d4bcc4a 2010
e5f10d49
KW
2011 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2012 if (category == categories[i]) {
2013 locale = PerlEnv_getenv(category_names[i]);
2014 goto found_locale;
2015 }
2016 }
7d4bcc4a 2017
b385bb4d 2018 locale = PerlEnv_getenv("LANG");
481465ea 2019 if (! locale) {
b385bb4d
KW
2020 locale = "";
2021 }
e5f10d49
KW
2022
2023 found_locale: ;
7d4bcc4a
KW
2024
2025# ifdef LC_ALL
2026
e5f10d49 2027 }
b385bb4d 2028 }
7d4bcc4a
KW
2029
2030# endif
2031
b385bb4d
KW
2032 }
2033
89f7b9aa 2034 result = setlocale(category, locale);
48015184
KW
2035 DEBUG_L(STMT_START {
2036 dSAVE_ERRNO;
2037 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2038 setlocale_debug_string(category, locale, result));
2039 RESTORE_ERRNO;
2040 } STMT_END);
89f7b9aa 2041
481465ea 2042 if (! override_LC_ALL) {
89f7b9aa
KW
2043 return result;
2044 }
2045
dfd77d7a 2046 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2047 * LANG variable or the system default if there is no LANG. But these have
2048 * lower priority than the other LC_foo variables, so override it for each
2049 * one that is set. (If they are set to "", it means to use the same thing
2050 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2051
948523db 2052 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2053 result = PerlEnv_getenv(category_names[i]);
2054 if (result && strNE(result, "")) {
2055 setlocale(categories[i], result);
2056 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2057 __FILE__, __LINE__,
2058 setlocale_debug_string(categories[i], result, "not captured")));
2059 }
89f7b9aa 2060 }
7d4bcc4a 2061
bbc98134 2062 result = setlocale(LC_ALL, NULL);
48015184
KW
2063 DEBUG_L(STMT_START {
2064 dSAVE_ERRNO;
2065 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2066 __FILE__, __LINE__,
48015184
KW
2067 setlocale_debug_string(LC_ALL, NULL, result));
2068 RESTORE_ERRNO;
2069 } STMT_END);
89f7b9aa 2070
bbc98134 2071 return result;
b385bb4d
KW
2072}
2073
2074#endif
2075
9aac5db8
KW
2076/*
2077
2078=head1 Locale-related functions and macros
2079
2080=for apidoc Perl_setlocale
2081
2082This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2083taking the same parameters, and returning the same information, except that it
2084returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2085perl keeps that locale category as C<C>, changing it briefly during the
2086operations where the underlying one is required.
2087
e9bc6d6b 2088Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2089declared to return S<C<const char *>>, whereas the system setlocale omits the
2090C<const>. (If it were being written today, plain setlocale would be declared
2091const, since it is illegal to change the information it returns; doing so leads
2092to segfaults.)
2093
e9bc6d6b
KW
2094Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2095C<setlocale> can be completely ineffective on some platforms under some
2096configurations.
2097
9aac5db8 2098C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2099where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2100the system C<setlocale()> is ineffective, returning the wrong information, and
2101failing to actually change the locale. C<Perl_setlocale>, however works
2102properly in all circumstances.
9aac5db8
KW
2103
2104The return points to a per-thread static buffer, which is overwritten the next
2105time C<Perl_setlocale> is called from the same thread.
2106
2107=cut
2108
2109*/
2110
2111const char *
2112Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2113{
2114 /* This wraps POSIX::setlocale() */
2115
9aac5db8
KW
2116 const char * retval;
2117 const char * newlocale;
2118 dSAVEDERRNO;
2119 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2120 dTHX;
2121
a4f00dcc
KW
2122#ifdef USE_LOCALE_NUMERIC
2123
291a84fb
KW
2124 /* A NULL locale means only query what the current one is. We have the
2125 * LC_NUMERIC name saved, because we are normally switched into the C
2126 * locale for it. For an LC_ALL query, switch back to get the correct
2127 * results. All other categories don't require special handling */
a4f00dcc
KW
2128 if (locale == NULL) {
2129 if (category == LC_NUMERIC) {
9aac5db8
KW
2130
2131 /* We don't have to copy this return value, as it is a per-thread
2132 * variable, and won't change until a future setlocale */
2133 return PL_numeric_name;
a4f00dcc
KW
2134 }
2135
7d4bcc4a 2136# ifdef LC_ALL
a4f00dcc 2137
9aac5db8
KW
2138 else if (category == LC_ALL) {
2139 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2140 }
2141
7d4bcc4a 2142# endif
a4f00dcc
KW
2143
2144 }
2145
2146#endif
2147
33e5a354
KW
2148 retval = save_to_buffer(do_setlocale_r(category, locale),
2149 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2150 SAVE_ERRNO;
2151
2152#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2153
2154 if (locale == NULL && category == LC_ALL) {
2155 RESTORE_LC_NUMERIC();
2156 }
2157
2158#endif
a4f00dcc
KW
2159
2160 DEBUG_L(PerlIO_printf(Perl_debug_log,
2161 "%s:%d: %s\n", __FILE__, __LINE__,
2162 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2163
9aac5db8
KW
2164 RESTORE_ERRNO;
2165
2166 if (! retval) {
a4f00dcc
KW
2167 return NULL;
2168 }
2169
9aac5db8 2170 /* If locale == NULL, we are just querying the state */
a4f00dcc 2171 if (locale == NULL) {
a4f00dcc
KW
2172 return retval;
2173 }
a4f00dcc 2174
1159483a
KW
2175 /* Now that have switched locales, we have to update our records to
2176 * correspond. */
a4f00dcc 2177
1159483a 2178 switch (category) {
a4f00dcc 2179
1159483a 2180#ifdef USE_LOCALE_CTYPE
a4f00dcc 2181
1159483a
KW
2182 case LC_CTYPE:
2183 new_ctype(retval);
2184 break;
a4f00dcc 2185
1159483a 2186#endif
a4f00dcc
KW
2187#ifdef USE_LOCALE_COLLATE
2188
1159483a
KW
2189 case LC_COLLATE:
2190 new_collate(retval);
2191 break;
a4f00dcc 2192
1159483a 2193#endif
a4f00dcc
KW
2194#ifdef USE_LOCALE_NUMERIC
2195
1159483a
KW
2196 case LC_NUMERIC:
2197 new_numeric(retval);
2198 break;
a4f00dcc 2199
1159483a
KW
2200#endif
2201#ifdef LC_ALL
a4f00dcc 2202
1159483a 2203 case LC_ALL:
a4f00dcc 2204
1159483a
KW
2205 /* LC_ALL updates all the things we care about. The values may not
2206 * be the same as 'retval', as the locale "" may have set things
2207 * individually */
a4f00dcc 2208
1159483a 2209# ifdef USE_LOCALE_CTYPE
a4f00dcc 2210
1159483a
KW
2211 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2212 new_ctype(newlocale);
a4f00dcc 2213
1159483a
KW
2214# endif /* USE_LOCALE_CTYPE */
2215# ifdef USE_LOCALE_COLLATE
2216
2217 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2218 new_collate(newlocale);
a4f00dcc 2219
7d4bcc4a 2220# endif
1159483a 2221# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2222
1159483a
KW
2223 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2224 new_numeric(newlocale);
a4f00dcc 2225
1159483a
KW
2226# endif /* USE_LOCALE_NUMERIC */
2227#endif /* LC_ALL */
a4f00dcc 2228
1159483a
KW
2229 default:
2230 break;
a4f00dcc
KW
2231 }
2232
2233 return retval;
2234
f7416781
KW
2235}
2236
2237PERL_STATIC_INLINE const char *
2238S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2239{
2240 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2241 * growing it if necessary */
2242
0b6697c5 2243 Size_t string_size;
f7416781
KW
2244
2245 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2246
0b6697c5
KW
2247 if (! string) {
2248 return NULL;
2249 }
2250
2251 string_size = strlen(string) + offset + 1;
2252
f7416781
KW
2253 if (*buf_size == 0) {
2254 Newx(*buf, string_size, char);
2255 *buf_size = string_size;
2256 }
2257 else if (string_size > *buf_size) {
2258 Renew(*buf, string_size, char);
2259 *buf_size = string_size;
2260 }
2261
2262 Copy(string, *buf + offset, string_size - offset, char);
2263 return *buf;
2264}
2265
2266/*
2267
f7416781
KW
2268=for apidoc Perl_langinfo
2269
ce181bf6 2270This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2271taking the same C<item> parameter values, and returning the same information.
2272But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2273of Perl's locale handling from your code, and can be used on systems that lack
2274a native C<nl_langinfo>.
2275
2276Expanding on these:
2277
2278=over
2279
2280=item *
2281
ce181bf6
KW
2282The reason it isn't quite a drop-in replacement is actually an advantage. The
2283only difference is that it returns S<C<const char *>>, whereas plain
2284C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2285forbidden to write into the buffer. By declaring this C<const>, the compiler
2286enforces this restriction, so if it is violated, you know at compilation time,
2287rather than getting segfaults at runtime.
2288
2289=item *
2290
69e2275e 2291It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2292without you having to write extra code. The reason for the extra code would be
2293because these are from the C<LC_NUMERIC> locale category, which is normally
2294kept set to the C locale by Perl, no matter what the underlying locale is
2295supposed to be, and so to get the expected results, you have to temporarily
ce181bf6
KW
2296toggle into the underlying locale, and later toggle back. (You could use plain
2297C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2298then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2299C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2300radix (decimal point) character to be a dot.)
2301
2302=item *
2303
2304The system function it replaces can have its static return buffer trashed,
2305not only by a subesequent call to that function, but by a C<freelocale>,
2306C<setlocale>, or other locale change. The returned buffer of this function is
2307not changed until the next call to it, so the buffer is never in a trashed
2308state.
f7416781
KW
2309
2310=item *
2311
ce181bf6
KW
2312Its return buffer is per-thread, so it also is never overwritten by a call to
2313this function from another thread; unlike the function it replaces.
2314
2315=item *
2316
2317But most importantly, it works on systems that don't have C<nl_langinfo>, such
2318as Windows, hence makes your code more portable. Of the fifty-some possible
2319items specified by the POSIX 2008 standard,
f7416781 2320L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2321only one is completely unimplemented, though on non-Windows platforms, another
2322significant one is also not implemented). It uses various techniques to
2323recover the other items, including calling C<L<localeconv(3)>>, and
2324C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2325available. Later C<strftime()> versions have additional capabilities; C<""> is
2326returned for those not available on your system.
ce181bf6
KW
2327
2328It is important to note that when called with an item that is recovered by
2329using C<localeconv>, the buffer from any previous explicit call to
2330C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2331contents if you need to access them after a call to this function. (But note
2332that you might not want to be using C<localeconv()> directly anyway, because of
2333issues like the ones listed in the second item of this list (above) for
2334C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2335call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2336unpack).
9f0bc911 2337
6201c220
KW
2338The details for those items which may deviate from what this emulation returns
2339and what a native C<nl_langinfo()> would return are specified in
2340L<I18N::Langinfo>.
f7416781 2341
ce181bf6
KW
2342=back
2343
f7416781
KW
2344When using C<Perl_langinfo> on systems that don't have a native
2345C<nl_langinfo()>, you must
2346
2347 #include "perl_langinfo.h"
2348
2349before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2350C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2351C<langinfo.h> would try to import into the namespace for code that doesn't need
2352it.)
f7416781 2353
f7416781
KW
2354The original impetus for C<Perl_langinfo()> was so that code that needs to
2355find out the current currency symbol, floating point radix character, or digit
2356grouping separator can use, on all systems, the simpler and more
2357thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2358pain to make thread-friendly. For other fields returned by C<localeconv>, it
2359is better to use the methods given in L<perlcall> to call
2360L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2361
2362=cut
2363
2364*/
2365
2366const char *
2367#ifdef HAS_NL_LANGINFO
2368Perl_langinfo(const nl_item item)
2369#else
2370Perl_langinfo(const int item)
2371#endif
2372{
f61748ac
KW
2373 return my_nl_langinfo(item, TRUE);
2374}
2375
59a15af0 2376STATIC const char *
f61748ac
KW
2377#ifdef HAS_NL_LANGINFO
2378S_my_nl_langinfo(const nl_item item, bool toggle)
2379#else
2380S_my_nl_langinfo(const int item, bool toggle)
2381#endif
2382{
ae74815b 2383 dTHX;
0b6697c5 2384 const char * retval;
f7416781 2385
5a854ab3
KW
2386 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2387 * two items, and only if not already there */
4e6826bf 2388 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3
KW
2389 || PL_numeric_underlying))
2390 {
2391 toggle = FALSE;
2392 }
2393
ab340fff 2394#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2395# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2396 || ! defined(HAS_POSIX_2008_LOCALE) \
2397 || ! defined(DUPLOCALE)
f7416781 2398
ab340fff 2399 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2400 * for those items dependent on it. This must be copied to a buffer before
2401 * switching back, as some systems destroy the buffer when setlocale() is
2402 * called */
f7416781 2403
038d3702
KW
2404 {
2405 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2406
b2fee59c 2407 if (toggle) {
038d3702 2408 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2409 }
f7416781 2410
5acc4454
KW
2411 LOCALE_LOCK; /* Prevent interference from another thread executing
2412 this code section (the only call to nl_langinfo in
2413 the core) */
2414
ee90eb2d 2415
628ff75a
KW
2416 /* Copy to a per-thread buffer, which is also one that won't be
2417 * destroyed by a subsequent setlocale(), such as the
2418 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2419 retval = save_to_buffer(nl_langinfo(item),
2420 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2421
5acc4454
KW
2422 LOCALE_UNLOCK;
2423
b2fee59c 2424 if (toggle) {
038d3702 2425 RESTORE_LC_NUMERIC();
b2fee59c 2426 }
f7416781
KW
2427 }
2428
ab340fff
KW
2429# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2430
5a854ab3 2431 {
b2fee59c
KW
2432 bool do_free = FALSE;
2433 locale_t cur = uselocale((locale_t) 0);
ab340fff 2434
b2fee59c
KW
2435 if (cur == LC_GLOBAL_LOCALE) {
2436 cur = duplocale(LC_GLOBAL_LOCALE);
2437 do_free = TRUE;
2438 }
ab340fff 2439
b2fee59c 2440 if (toggle) {
e1aa2579
KW
2441 if (PL_underlying_numeric_obj) {
2442 cur = PL_underlying_numeric_obj;
2443 }
2444 else {
2445 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2446 do_free = TRUE;
2447 }
b2fee59c 2448 }
ab340fff 2449
628ff75a
KW
2450 /* We have to save it to a buffer, because the freelocale() just below
2451 * can invalidate the internal one */
0b6697c5
KW
2452 retval = save_to_buffer(nl_langinfo_l(item, cur),
2453 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2454
b2fee59c
KW
2455 if (do_free) {
2456 freelocale(cur);
2457 }
5a854ab3 2458 }
ab340fff 2459
c1566110
KW
2460# endif
2461
0b6697c5 2462 if (strEQ(retval, "")) {
4e6826bf 2463 if (item == YESSTR) {
c1566110
KW
2464 return "yes";
2465 }
4e6826bf 2466 if (item == NOSTR) {
c1566110
KW
2467 return "no";
2468 }
2469 }
2470
0b6697c5 2471 return retval;
ab340fff 2472
f7416781 2473#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2474
2475 {
2476
f7416781
KW
2477# ifdef HAS_LOCALECONV
2478
43dd6b15 2479 const struct lconv* lc;
628ff75a 2480 const char * temp;
038d3702 2481 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2482
69c5e0db
KW
2483# ifdef TS_W32_BROKEN_LOCALECONV
2484
2485 const char * save_global;
2486 const char * save_thread;
2487 int needed_size;
2488 char * ptr;
2489 char * e;
2490 char * item_start;
2491
2492# endif
f7416781
KW
2493# endif
2494# ifdef HAS_STRFTIME
2495
43dd6b15
KW
2496 struct tm tm;
2497 bool return_format = FALSE; /* Return the %format, not the value */
2498 const char * format;
f7416781
KW
2499
2500# endif
2501
43dd6b15
KW
2502 /* We copy the results to a per-thread buffer, even if not
2503 * multi-threaded. This is in part to simplify this code, and partly
2504 * because we need a buffer anyway for strftime(), and partly because a
2505 * call of localeconv() could otherwise wipe out the buffer, and the
2506 * programmer would not be expecting this, as this is a nl_langinfo()
2507 * substitute after all, so s/he might be thinking their localeconv()
2508 * is safe until another localeconv() call. */
f7416781 2509
43dd6b15
KW
2510 switch (item) {
2511 Size_t len;
f7416781 2512
8d72e74e 2513 /* This is unimplemented */
4e6826bf 2514 case ERA: /* For use with strftime() %E modifier */
f7416781 2515
43dd6b15
KW
2516 default:
2517 return "";
f7416781 2518
43dd6b15 2519 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2520 case YESEXPR: return "^[+1yY]";
2521 case YESSTR: return "yes";
2522 case NOEXPR: return "^[-0nN]";
2523 case NOSTR: return "no";
2524
8d72e74e
KW
2525 case CODESET:
2526
2527# ifndef WIN32
2528
2529 /* On non-windows, this is unimplemented, in part because of
2530 * inconsistencies between vendors. The Darwin native
2531 * nl_langinfo() implementation simply looks at everything past
2532 * any dot in the name, but that doesn't work for other
2533 * vendors. Many Linux locales that don't have UTF-8 in their
2534 * names really are UTF-8, for example; z/OS locales that do
2535 * have UTF-8 in their names, aren't really UTF-8 */
2536 return "";
2537
2538# else
2539
2540 { /* But on Windows, the name does seem to be consistent, so
2541 use that. */
2542 const char * p;
2543 const char * first;
2544 Size_t offset = 0;
2545 const char * name = my_setlocale(LC_CTYPE, NULL);
2546
2547 if (isNAME_C_OR_POSIX(name)) {
2548 return "ANSI_X3.4-1968";
2549 }
2550
2551 /* Find the dot in the locale name */
2552 first = (const char *) strchr(name, '.');
2553 if (! first) {
2554 first = name;
2555 goto has_nondigit;
2556 }
2557
2558 /* Look at everything past the dot */
2559 first++;
2560 p = first;
2561
2562 while (*p) {
2563 if (! isDIGIT(*p)) {
2564 goto has_nondigit;
2565 }
2566
2567 p++;
2568 }
2569
2570 /* Here everything past the dot is a digit. Treat it as a
2571 * code page */
2572 save_to_buffer("CP", &PL_langinfo_buf,
2573 &PL_langinfo_bufsize, 0);
2574 offset = STRLENs("CP");
f7416781 2575
8d72e74e
KW
2576 has_nondigit:
2577
2578 retval = save_to_buffer(first, &PL_langinfo_buf,
2579 &PL_langinfo_bufsize, offset);
2580 }
2581
2582 break;
2583
2584# endif
f7416781
KW
2585# ifdef HAS_LOCALECONV
2586
4e6826bf 2587 case CRNCYSTR:
f7416781 2588
43dd6b15
KW
2589 /* We don't bother with localeconv_l() because any system that
2590 * has it is likely to also have nl_langinfo() */
291a84fb 2591
69c5e0db
KW
2592 LOCALE_LOCK_V; /* Prevent interference with other threads
2593 using localeconv() */
2594
2595# ifdef TS_W32_BROKEN_LOCALECONV
2596
2597 /* This is a workaround for a Windows bug prior to VS 15.
2598 * What we do here is, while locked, switch to the global
2599 * locale so localeconv() works; then switch back just before
2600 * the unlock. This can screw things up if some thread is
2601 * already using the global locale while assuming no other is.
2602 * A different workaround would be to call GetCurrencyFormat on
2603 * a known value, and parse it; patches welcome
2604 *
2605 * We have to use LC_ALL instead of LC_MONETARY because of
2606 * another bug in Windows */
2607
2608 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2609 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2610 save_global= savepv(my_setlocale(LC_ALL, NULL));
2611 my_setlocale(LC_ALL, save_thread);
2612
2613# endif
5acc4454 2614
43dd6b15
KW
2615 lc = localeconv();
2616 if ( ! lc
2617 || ! lc->currency_symbol
2618 || strEQ("", lc->currency_symbol))
2619 {
69c5e0db 2620 LOCALE_UNLOCK_V;
43dd6b15
KW
2621 return "";
2622 }
f7416781 2623
43dd6b15 2624 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2625 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2626 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2627 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2628 { /* khw couldn't figure out how the localedef specifications
2629 would show that the $ should replace the radix; this is
2630 just a guess as to how it might work.*/
a34edee3 2631 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2632 }
2633 else if (lc->p_cs_precedes) {
a34edee3 2634 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2635 }
2636 else {
a34edee3 2637 PL_langinfo_buf[0] = '+';
43dd6b15 2638 }
f7416781 2639
69c5e0db
KW
2640# ifdef TS_W32_BROKEN_LOCALECONV
2641
2642 my_setlocale(LC_ALL, save_global);
2643 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2644 my_setlocale(LC_ALL, save_thread);
2645 Safefree(save_global);
2646 Safefree(save_thread);
2647
2648# endif
2649
2650 LOCALE_UNLOCK_V;
43dd6b15 2651 break;
f7416781 2652
69c5e0db
KW
2653# ifdef TS_W32_BROKEN_LOCALECONV
2654
4e6826bf 2655 case RADIXCHAR:
69c5e0db
KW
2656
2657 /* For this, we output a known simple floating point number to
2658 * a buffer, and parse it, looking for the radix */
2659
2660 if (toggle) {
2661 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2662 }
2663
2664 if (PL_langinfo_bufsize < 10) {
2665 PL_langinfo_bufsize = 10;
2666 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2667 }
2668
2669 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2670 "%.1f", 1.5);
2671 if (needed_size >= (int) PL_langinfo_bufsize) {
2672 PL_langinfo_bufsize = needed_size + 1;
2673 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2674 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2675 "%.1f", 1.5);
2676 assert(needed_size < (int) PL_langinfo_bufsize);
2677 }
2678
2679 ptr = PL_langinfo_buf;
2680 e = PL_langinfo_buf + PL_langinfo_bufsize;
2681
2682 /* Find the '1' */
2683 while (ptr < e && *ptr != '1') {
2684 ptr++;
2685 }
2686 ptr++;
2687
2688 /* Find the '5' */
2689 item_start = ptr;
2690 while (ptr < e && *ptr != '5') {
2691 ptr++;
2692 }
2693
2694 /* Everything in between is the radix string */
2695 if (ptr >= e) {
2696 PL_langinfo_buf[0] = '?';
2697 PL_langinfo_buf[1] = '\0';
2698 }
2699 else {
2700 *ptr = '\0';
2701 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2702 }
2703
2704 if (toggle) {
2705 RESTORE_LC_NUMERIC();
2706 }
2707
2708 retval = PL_langinfo_buf;
2709 break;
2710
2711# else
2712
2713 case RADIXCHAR: /* No special handling needed */
2714
2715# endif
2716
4e6826bf 2717 case THOUSEP:
f7416781 2718
43dd6b15 2719 if (toggle) {
038d3702 2720 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2721 }
f7416781 2722
69c5e0db
KW
2723 LOCALE_LOCK_V; /* Prevent interference with other threads
2724 using localeconv() */
2725
2726# ifdef TS_W32_BROKEN_LOCALECONV
2727
2728 /* This should only be for the thousands separator. A
2729 * different work around would be to use GetNumberFormat on a
2730 * known value and parse the result to find the separator */
2731 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2732 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2733 save_global = savepv(my_setlocale(LC_ALL, NULL));
2734 my_setlocale(LC_ALL, save_thread);
2735# if 0
2736 /* This is the start of code that for broken Windows replaces
2737 * the above and below code, and instead calls
2738 * GetNumberFormat() and then would parse that to find the
2739 * thousands separator. It needs to handle UTF-16 vs -8
2740 * issues. */
2741
2742 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2743 DEBUG_L(PerlIO_printf(Perl_debug_log,
2744 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2745 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2746
2747# endif
2748# endif
5acc4454 2749
43dd6b15
KW
2750 lc = localeconv();
2751 if (! lc) {
628ff75a 2752 temp = "";
33394adc 2753 }
43dd6b15 2754 else {
4e6826bf 2755 temp = (item == RADIXCHAR)
43dd6b15
KW
2756 ? lc->decimal_point
2757 : lc->thousands_sep;
628ff75a
KW
2758 if (! temp) {
2759 temp = "";
43dd6b15
KW
2760 }
2761 }
f7416781 2762
0b6697c5
KW
2763 retval = save_to_buffer(temp, &PL_langinfo_buf,
2764 &PL_langinfo_bufsize, 0);
f7416781 2765
69c5e0db
KW
2766# ifdef TS_W32_BROKEN_LOCALECONV
2767
2768 my_setlocale(LC_ALL, save_global);
2769 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2770 my_setlocale(LC_ALL, save_thread);
2771 Safefree(save_global);
2772 Safefree(save_thread);
2773
2774# endif
2775
2776 LOCALE_UNLOCK_V;
5acc4454 2777
43dd6b15 2778 if (toggle) {
038d3702 2779 RESTORE_LC_NUMERIC();
43dd6b15 2780 }
f7416781 2781
43dd6b15 2782 break;
f7416781
KW
2783
2784# endif
2785# ifdef HAS_STRFTIME
2786
43dd6b15
KW
2787 /* These are defined by C89, so we assume that strftime supports
2788 * them, and so are returned unconditionally; they may not be what
2789 * the locale actually says, but should give good enough results
2790 * for someone using them as formats (as opposed to trying to parse
2791 * them to figure out what the locale says). The other format
2792 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2793 case D_FMT: return "%x";
2794 case T_FMT: return "%X";
2795 case D_T_FMT: return "%c";
43dd6b15
KW
2796
2797 /* These formats are only available in later strfmtime's */
4e6826bf 2798 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2799
2800 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2801 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2802 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2803 case ALT_DIGITS:
2804 case AM_STR: case PM_STR:
2805 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2806 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2807 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2808 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2809 case DAY_5: case DAY_6: case DAY_7:
2810 case MON_1: case MON_2: case MON_3: case MON_4:
2811 case MON_5: case MON_6: case MON_7: case MON_8:
2812 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2813
2814 LOCALE_LOCK;
2815
2816 init_tm(&tm); /* Precaution against core dumps */
2817 tm.tm_sec = 30;
2818 tm.tm_min = 30;
2819 tm.tm_hour = 6;
2820 tm.tm_year = 2017 - 1900;
2821 tm.tm_wday = 0;
2822 tm.tm_mon = 0;
2823 switch (item) {
2824 default:
2825 LOCALE_UNLOCK;
2826 Perl_croak(aTHX_
2827 "panic: %s: %d: switch case: %d problem",
2828 __FILE__, __LINE__, item);
2829 NOT_REACHED; /* NOTREACHED */
2830
4e6826bf
KW
2831 case PM_STR: tm.tm_hour = 18;
2832 case AM_STR:
43dd6b15
KW
2833 format = "%p";
2834 break;
2835
4e6826bf
KW
2836 case ABDAY_7: tm.tm_wday++;
2837 case ABDAY_6: tm.tm_wday++;
2838 case ABDAY_5: tm.tm_wday++;
2839 case ABDAY_4: tm.tm_wday++;
2840 case ABDAY_3: tm.tm_wday++;
2841 case ABDAY_2: tm.tm_wday++;
2842 case ABDAY_1:
43dd6b15
KW
2843 format = "%a";
2844 break;
2845
4e6826bf
KW
2846 case DAY_7: tm.tm_wday++;
2847 case DAY_6: tm.tm_wday++;
2848 case DAY_5: tm.tm_wday++;
2849 case DAY_4: tm.tm_wday++;
2850 case DAY_3: tm.tm_wday++;
2851 case DAY_2: tm.tm_wday++;
2852 case DAY_1:
43dd6b15
KW
2853 format = "%A";
2854 break;
2855
4e6826bf
KW
2856 case ABMON_12: tm.tm_mon++;
2857 case ABMON_11: tm.tm_mon++;
2858 case ABMON_10: tm.tm_mon++;
2859 case ABMON_9: tm.tm_mon++;
2860 case ABMON_8: tm.tm_mon++;
2861 case ABMON_7: tm.tm_mon++;
2862 case ABMON_6: tm.tm_mon++;
2863 case ABMON_5: tm.tm_mon++;
2864 case ABMON_4: tm.tm_mon++;
2865 case ABMON_3: tm.tm_mon++;
2866 case ABMON_2: tm.tm_mon++;
2867 case ABMON_1:
43dd6b15
KW
2868 format = "%b";
2869 break;
2870
4e6826bf
KW
2871 case MON_12: tm.tm_mon++;
2872 case MON_11: tm.tm_mon++;
2873 case MON_10: tm.tm_mon++;
2874 case MON_9: tm.tm_mon++;
2875 case MON_8: tm.tm_mon++;
2876 case MON_7: tm.tm_mon++;
2877 case MON_6: tm.tm_mon++;
2878 case MON_5: tm.tm_mon++;
2879 case MON_4: tm.tm_mon++;
2880 case MON_3: tm.tm_mon++;
2881 case MON_2: tm.tm_mon++;
2882 case MON_1:
43dd6b15
KW
2883 format = "%B";
2884 break;
2885
4e6826bf 2886 case T_FMT_AMPM:
43dd6b15
KW
2887 format = "%r";
2888 return_format = TRUE;
2889 break;
2890
4e6826bf 2891 case ERA_D_FMT:
43dd6b15
KW
2892 format = "%Ex";
2893 return_format = TRUE;
2894 break;
2895
4e6826bf 2896 case ERA_T_FMT:
43dd6b15
KW
2897 format = "%EX";
2898 return_format = TRUE;
2899 break;
2900
4e6826bf 2901 case ERA_D_T_FMT:
43dd6b15
KW
2902 format = "%Ec";
2903 return_format = TRUE;
2904 break;
2905
4e6826bf 2906 case ALT_DIGITS:
43dd6b15
KW
2907 tm.tm_wday = 0;
2908 format = "%Ow"; /* Find the alternate digit for 0 */
2909 break;
2910 }
f7416781 2911
43dd6b15
KW
2912 /* We can't use my_strftime() because it doesn't look at
2913 * tm_wday */
2914 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2915 format, &tm))
2916 {
2917 /* A zero return means one of:
2918 * a) there wasn't enough space in PL_langinfo_buf
2919 * b) the format, like a plain %p, returns empty
2920 * c) it was an illegal format, though some
2921 * implementations of strftime will just return the
2922 * illegal format as a plain character sequence.
2923 *
2924 * To quickly test for case 'b)', try again but precede
2925 * the format with a plain character. If that result is
2926 * still empty, the problem is either 'a)' or 'c)' */
2927
2928 Size_t format_size = strlen(format) + 1;
2929 Size_t mod_size = format_size + 1;
2930 char * mod_format;
2931 char * temp_result;
2932
2933 Newx(mod_format, mod_size, char);
2934 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2935 *mod_format = ' ';
43dd6b15
KW
2936 my_strlcpy(mod_format + 1, format, mod_size);
2937 len = strftime(temp_result,
2938 PL_langinfo_bufsize,
2939 mod_format, &tm);
2940 Safefree(mod_format);
2941 Safefree(temp_result);
2942
2943 /* If 'len' is non-zero, it means that we had a case like
2944 * %p which means the current locale doesn't use a.m. or
2945 * p.m., and that is valid */
2946 if (len == 0) {
2947
2948 /* Here, still didn't work. If we get well beyond a
2949 * reasonable size, bail out to prevent an infinite
2950 * loop. */
2951
2952 if (PL_langinfo_bufsize > 100 * format_size) {
2953 *PL_langinfo_buf = '\0';
2954 }
2955 else {
2956 /* Double the buffer size to retry; Add 1 in case
2957 * original was 0, so we aren't stuck at 0. */
2958 PL_langinfo_bufsize *= 2;
2959 PL_langinfo_bufsize++;
2960 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2961 continue;
2962 }
2963 }
f7416781 2964
f7416781 2965 break;
43dd6b15 2966 }
f7416781 2967
43dd6b15
KW
2968 /* Here, we got a result.
2969 *
2970 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2971 * alternate format for wday 0. If the value is the same as
2972 * the normal 0, there isn't an alternate, so clear the buffer.
2973 * */
4e6826bf 2974 if ( item == ALT_DIGITS
43dd6b15
KW
2975 && strEQ(PL_langinfo_buf, "0"))
2976 {
2977 *PL_langinfo_buf = '\0';
2978 }
f7416781 2979
43dd6b15
KW
2980 /* ALT_DIGITS is problematic. Experiments on it showed that
2981 * strftime() did not always work properly when going from
2982 * alt-9 to alt-10. Only a few locales have this item defined,
2983 * and in all of them on Linux that khw was able to find,
2984 * nl_langinfo() merely returned the alt-0 character, possibly
2985 * doubled. Most Unicode digits are in blocks of 10
2986 * consecutive code points, so that is sufficient information
2987 * for those scripts, as we can infer alt-1, alt-2, .... But
2988 * for a Japanese locale, a CJK ideographic 0 is returned, and
2989 * the CJK digits are not in code point order, so you can't
2990 * really infer anything. The localedef for this locale did
2991 * specify the succeeding digits, so that strftime() works
2992 * properly on them, without needing to infer anything. But
2993 * the nl_langinfo() return did not give sufficient information
2994 * for the caller to understand what's going on. So until
2995 * there is evidence that it should work differently, this
2996 * returns the alt-0 string for ALT_DIGITS.
2997 *
2998 * wday was chosen because its range is all a single digit.
2999 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3000
43dd6b15 3001 LOCALE_UNLOCK;
f7416781 3002
d1b150d9
KW
3003 retval = PL_langinfo_buf;
3004
43dd6b15
KW
3005 /* If to return the format, not the value, overwrite the buffer
3006 * with it. But some strftime()s will keep the original format
3007 * if illegal, so change those to "" */
3008 if (return_format) {
3009 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3010 *PL_langinfo_buf = '\0';
3011 }
43dd6b15 3012 else {
0b6697c5
KW
3013 retval = save_to_buffer(format, &PL_langinfo_buf,
3014 &PL_langinfo_bufsize, 0);
f7416781
KW
3015 }
3016 }
3017
3018 break;
f7416781
KW
3019
3020# endif
3021
43dd6b15 3022 }
f7416781
KW
3023 }
3024
0b6697c5 3025 return retval;
f7416781
KW
3026
3027#endif
3028
a4f00dcc 3029}
b385bb4d 3030
98994639
HS
3031/*
3032 * Initialize locale awareness.
3033 */
3034int
3035Perl_init_i18nl10n(pTHX_ int printwarn)
3036{
0e92a118
KW
3037 /* printwarn is
3038 *
3039 * 0 if not to output warning when setup locale is bad
3040 * 1 if to output warning based on value of PERL_BADLANG
3041 * >1 if to output regardless of PERL_BADLANG
3042 *
3043 * returns
98994639 3044 * 1 = set ok or not applicable,
0e92a118
KW
3045 * 0 = fallback to a locale of lower priority
3046 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3047 *
3048 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3049 * set, debugging information is output.
3050 *
3051 * This looks more complicated than it is, mainly due to the #ifdefs.
3052 *
3053 * We try to set LC_ALL to the value determined by the environment. If
3054 * there is no LC_ALL on this platform, we try the individual categories we
3055 * know about. If this works, we are done.
3056 *
3057 * But if it doesn't work, we have to do something else. We search the
3058 * environment variables ourselves instead of relying on the system to do
3059 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3060 * think there is one), and the ultimate fallback "C". This is all done in
3061 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3062 * more complex. The 'trial_locales' array is initialized with just one
3063 * element; it causes the behavior described in the paragraph above this to
3064 * happen. If that fails, we add elements to 'trial_locales', and do extra
3065 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3066 *
3067 * On Ultrix, the locale MUST come from the environment, so there is
3068 * preliminary code to set it. I (khw) am not sure that it is necessary,
3069 * and that this couldn't be folded into the loop, but barring any real
3070 * platforms to test on, it's staying as-is
3071 *
3072 * A slight complication is that in embedded Perls, the locale may already
3073 * be set-up, and we don't want to get it from the normal environment
3074 * variables. This is handled by having a special environment variable
3075 * indicate we're in this situation. We simply set setlocale's 2nd
3076 * parameter to be a NULL instead of "". That indicates to setlocale that
3077 * it is not to change anything, but to return the current value,
3078 * effectively initializing perl's db to what the locale already is.
3079 *
3080 * We play the same trick with NULL if a LC_ALL succeeds. We call
3081 * setlocale() on the individual categores with NULL to get their existing
3082 * values for our db, instead of trying to change them.
3083 * */
98994639 3084
0e92a118
KW
3085 int ok = 1;
3086
7d4bcc4a
KW
3087#ifndef USE_LOCALE
3088
3089 PERL_UNUSED_ARG(printwarn);
3090
3091#else /* USE_LOCALE */
7d4bcc4a
KW
3092# ifdef __GLIBC__
3093
175c4cf9 3094 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3095
3096# endif
65ebb059 3097
ccd65d51
KW
3098 /* NULL uses the existing already set up locale */
3099 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3100 ? NULL
3101 : "";
c3fcd832
KW
3102 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3103 unsigned int trial_locales_count;
175c4cf9
KW
3104 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3105 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3106 bool setlocale_failure = FALSE;
65ebb059 3107 unsigned int i;
175c4cf9
KW
3108
3109 /* A later getenv() could zap this, so only use here */
3110 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3111
3112 const bool locwarn = (printwarn > 1
e5f10d49
KW
3113 || ( printwarn
3114 && ( ! bad_lang_use_once
22ff3130 3115 || (
e5f10d49
KW
3116 /* disallow with "" or "0" */
3117 *bad_lang_use_once
3118 && strNE("0", bad_lang_use_once)))));
ea92aad8 3119
291a84fb 3120 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3121 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3122
3123 /* current locale for given category; should have been copied so aren't
3124 * volatile */
8de4332b 3125 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3126
7d4bcc4a
KW
3127# ifdef WIN32
3128
6bce99ee
JH
3129 /* In some systems you can find out the system default locale
3130 * and use that as the fallback locale. */
7d4bcc4a
KW
3131# define SYSTEM_DEFAULT_LOCALE
3132# endif
3133# ifdef SYSTEM_DEFAULT_LOCALE
3134
65ebb059 3135 const char *system_default_locale = NULL;
98994639 3136
7d4bcc4a 3137# endif
948523db
KW
3138
3139# ifndef DEBUGGING
3140# define DEBUG_LOCALE_INIT(a,b,c)
3141# else
7d4bcc4a 3142
8298454c 3143 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3144
3145# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3146 STMT_START { \
3147 if (debug_initialization) { \
3148 PerlIO_printf(Perl_debug_log, \
3149 "%s:%d: %s\n", \
3150 __FILE__, __LINE__, \
a4f00dcc 3151 setlocale_debug_string(category, \
2fcc0ca9
KW
3152 locale, \
3153 result)); \
3154 } \
3155 } STMT_END
2fcc0ca9 3156
948523db
KW
3157/* Make sure the parallel arrays are properly set up */
3158# ifdef USE_LOCALE_NUMERIC
3159 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3160 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3161# ifdef USE_POSIX_2008_LOCALE
3162 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3163# endif
948523db
KW
3164# endif
3165# ifdef USE_LOCALE_CTYPE
3166 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3167 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3168# ifdef USE_POSIX_2008_LOCALE
3169 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3170# endif
948523db
KW
3171# endif
3172# ifdef USE_LOCALE_COLLATE
3173 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3174 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3175# ifdef USE_POSIX_2008_LOCALE
3176 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3177# endif
948523db
KW
3178# endif
3179# ifdef USE_LOCALE_TIME
3180 assert(categories[LC_TIME_INDEX] == LC_TIME);
3181 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3182# ifdef USE_POSIX_2008_LOCALE
3183 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3184# endif
948523db
KW
3185# endif
3186# ifdef USE_LOCALE_MESSAGES
3187 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3188 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3189# ifdef USE_POSIX_2008_LOCALE
3190 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3191# endif
948523db
KW
3192# endif
3193# ifdef USE_LOCALE_MONETARY
3194 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3195 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3196# ifdef USE_POSIX_2008_LOCALE
3197 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3198# endif
948523db 3199# endif
9821811f
KW
3200# ifdef USE_LOCALE_ADDRESS
3201 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3202 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3203# ifdef USE_POSIX_2008_LOCALE
3204 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3205# endif
9821811f
KW
3206# endif
3207# ifdef USE_LOCALE_IDENTIFICATION
3208 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3209 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3210# ifdef USE_POSIX_2008_LOCALE
3211 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3212# endif
9821811f
KW
3213# endif
3214# ifdef USE_LOCALE_MEASUREMENT
3215 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3216 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3217# ifdef USE_POSIX_2008_LOCALE
3218 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3219# endif
9821811f
KW
3220# endif
3221# ifdef USE_LOCALE_PAPER
3222 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3223 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3224# ifdef USE_POSIX_2008_LOCALE
3225 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3226# endif
9821811f
KW
3227# endif
3228# ifdef USE_LOCALE_TELEPHONE
3229 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3230 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3231# ifdef USE_POSIX_2008_LOCALE
3232 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3233# endif
9821811f 3234# endif
948523db
KW
3235# ifdef LC_ALL
3236 assert(categories[LC_ALL_INDEX] == LC_ALL);
3237 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3238 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3239# ifdef USE_POSIX_2008_LOCALE
3240 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3241# endif
948523db
KW
3242# endif
3243# endif /* DEBUGGING */
47280b20
KW
3244
3245 /* Initialize the cache of the program's UTF-8ness for the always known
3246 * locales C and POSIX */
3247 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3248 sizeof(PL_locale_utf8ness));
3249
e9bc6d6b
KW
3250# ifdef USE_THREAD_SAFE_LOCALE
3251# ifdef WIN32
3252
3253 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3254
3255# endif
3256# endif
39e69e77 3257# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3258
3259 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3260 if (! PL_C_locale_obj) {
3261 Perl_croak_nocontext(
3262 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3263 }
3264 if (DEBUG_Lv_TEST || debug_initialization) {
3265 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3266 }
3267
3268# endif
3269
3ca88433
KW
3270 PL_numeric_radix_sv = newSVpvs(".");
3271
e9bc6d6b
KW
3272# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3273
3274 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3275 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3276
3277# endif
b56c4436 3278# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3279
3280 /*
3281 * Ultrix setlocale(..., "") fails if there are no environment
3282 * variables from which to get a locale name.
3283 */
3284
b56c4436
KW
3285# ifndef LC_ALL
3286# error Ultrix without LC_ALL not implemented
3287# else
7d4bcc4a 3288
b56c4436
KW
3289 {
3290 bool done = FALSE;
ea92aad8
KW
3291 if (lang) {
3292 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3293 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3294 if (sl_result[LC_ALL_INDEX])
3295 done = TRUE;
3296 else
e5f10d49 3297 setlocale_failure = TRUE;
ea92aad8
KW
3298 }
3299 if (! setlocale_failure) {
3300 const char * locale_param;
3301 for (i = 0; i < LC_ALL_INDEX; i++) {
3302 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3303 ? setlocale_init
3304 : NULL;
3305 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3306 if (! sl_result[i]) {
3307 setlocale_failure = TRUE;
3308 }
3309 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3310 }
c835d6be 3311 }
7d4bcc4a
KW
3312 }
3313
3314# endif /* LC_ALL */
e5f10d49 3315# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3316
65ebb059 3317 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3318 * the list. Normally the loop is executed just once. But if setting the
3319 * locale fails, inside the loop we add fallback trials to the array and so
3320 * will execute the loop multiple times */
c3fcd832
KW
3321 trial_locales[0] = setlocale_init;
3322 trial_locales_count = 1;
7d4bcc4a 3323
65ebb059
KW
3324 for (i= 0; i < trial_locales_count; i++) {
3325 const char * trial_locale = trial_locales[i];
3326
3327 if (i > 0) {
3328
3329 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3330 * when i==0, but I (khw) don't think that behavior makes much
3331 * sense */
3332 setlocale_failure = FALSE;
3333
7d4bcc4a 3334# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3335# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3336
65ebb059
KW
3337 /* On Windows machines, an entry of "" after the 0th means to use
3338 * the system default locale, which we now proceed to get. */
3339 if (strEQ(trial_locale, "")) {
3340 unsigned int j;
3341
3342 /* Note that this may change the locale, but we are going to do
3343 * that anyway just below */
837ce802 3344 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3345 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3346
7d4bcc4a 3347 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3348 * try */
3349 if (! system_default_locale) {
3350 goto next_iteration;
3351 }
3352 for (j = 0; j < trial_locales_count; j++) {
3353 if (strEQ(system_default_locale, trial_locales[j])) {
3354 goto next_iteration;
3355 }
3356 }
3357
3358 trial_locale = system_default_locale;
3359 }
ec0202b5
KW
3360# else
3361# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3362# endif
7d4bcc4a 3363# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3364
3365 } /* For i > 0 */
65ebb059 3366
7d4bcc4a
KW
3367# ifdef LC_ALL
3368
948523db
KW
3369 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3370 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3371 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3372 setlocale_failure = TRUE;
7cd8b568
KW
3373 }
3374 else {
3375 /* Since LC_ALL succeeded, it should have changed all the other
3376 * categories it can to its value; so we massage things so that the
3377 * setlocales below just return their category's current values.
3378 * This adequately handles the case in NetBSD where LC_COLLATE may
3379 * not be defined for a locale, and setting it individually will
7d4bcc4a 3380 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3381 * the POSIX locale. */
3382 trial_locale = NULL;
3383 }
7d4bcc4a
KW
3384
3385# endif /* LC_ALL */
98994639 3386
e5f10d49
KW
3387 if (! setlocale_failure) {
3388 unsigned int j;
3389 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3390 curlocales[j]
3391 = savepv(do_setlocale_r(categories[j], trial_locale));
3392 if (! curlocales[j]) {
3393 setlocale_failure = TRUE;
3394 }
3395 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3396 }
c835d6be 3397
e5f10d49
KW
3398 if (! setlocale_failure) { /* All succeeded */
3399 break; /* Exit trial_locales loop */
49c85077 3400 }
65ebb059 3401 }
98994639 3402
49c85077
KW
3403 /* Here, something failed; will need to try a fallback. */
3404 ok = 0;
65ebb059 3405
49c85077
KW
3406 if (i == 0) {
3407 unsigned int j;
98994639 3408
65ebb059 3409 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3410
3411# ifdef LC_ALL
98994639 3412
49c85077
KW
3413 PerlIO_printf(Perl_error_log,
3414 "perl: warning: Setting locale failed.\n");
98994639 3415
7d4bcc4a 3416# else /* !LC_ALL */
98994639 3417
49c85077
KW
3418 PerlIO_printf(Perl_error_log,
3419 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3420
e5f10d49
KW
3421 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3422 if (! curlocales[j]) {
3423 PerlIO_printf(Perl_error_log, category_names[j]);
3424 }
3425 else {
3426 Safefree(curlocales[j]);
3427 }
3428 }
7d4bcc4a 3429
7d4bcc4a 3430# endif /* LC_ALL */
98994639 3431
49c85077
KW
3432 PerlIO_printf(Perl_error_log,
3433 "perl: warning: Please check that your locale settings:\n");
98994639 3434
7d4bcc4a
KW
3435# ifdef __GLIBC__
3436
49c85077
KW
3437 PerlIO_printf(Perl_error_log,
3438 "\tLANGUAGE = %c%s%c,\n",
3439 language ? '"' : '(',
3440 language ? language : "unset",
3441 language ? '"' : ')');
7d4bcc4a 3442# endif
98994639 3443
49c85077
KW
3444 PerlIO_printf(Perl_error_log,
3445 "\tLC_ALL = %c%s%c,\n",
3446 lc_all ? '"' : '(',
3447 lc_all ? lc_all : "unset",
3448 lc_all ? '"' : ')');
98994639 3449
7d4bcc4a
KW
3450# if defined(USE_ENVIRON_ARRAY)
3451
49c85077 3452 {
cd999af9 3453 char **e;
d5e32b93
KW
3454
3455 /* Look through the environment for any variables of the
3456 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3457 * already handled above. These are assumed to be locale
3458 * settings. Output them and their values. */
cd999af9 3459 for (e = environ; *e; e++) {
d5e32b93
KW
3460 const STRLEN prefix_len = sizeof("LC_") - 1;
3461 STRLEN uppers_len;
3462
cd999af9 3463 if ( strBEGINs(*e, "LC_")
c8b388b0 3464 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3465 && (uppers_len = strspn(*e + prefix_len,
3466 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3467 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3468 {
3469 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3470 (int) (prefix_len + uppers_len), *e,
3471 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3472 }
3473 }
49c85077 3474 }
7d4bcc4a
KW
3475
3476# else
3477
49c85077
KW
3478 PerlIO_printf(Perl_error_log,
3479 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3480
3481# endif
98994639 3482
49c85077
KW
3483 PerlIO_printf(Perl_error_log,
3484 "\tLANG = %c%s%c\n",
3485 lang ? '"' : '(',
3486 lang ? lang : "unset",
3487 lang ? '"' : ')');
98994639 3488
49c85077
KW
3489 PerlIO_printf(Perl_error_log,
3490 " are supported and installed on your system.\n");
3491 }
98994639 3492
65ebb059 3493 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3494 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3495 * usually change the upper bound of the loop we are in.
3496 *
3497 * Since the system's default way of setting the locale has not
3498 * found one that works, We use Perl's defined ordering: LC_ALL,
3499 * LANG, and the C locale. We don't try the same locale twice, so
3500 * don't add to the list if already there. (On POSIX systems, the
3501 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3502 * but there's no harm done by doing it explicitly.
3503 *
3504 * Note that this tries the LC_ALL environment variable even on
3505 * systems which have no LC_ALL locale setting. This may or may
3506 * not have been originally intentional, but there's no real need
3507 * to change the behavior. */
65ebb059
KW
3508 if (lc_all) {
3509 for (j = 0; j < trial_locales_count; j++) {
3510 if (strEQ(lc_all, trial_locales[j])) {
3511 goto done_lc_all;
3512 }
3513 }
3514 trial_locales[trial_locales_count++] = lc_all;
3515 }
3516 done_lc_all:
98994639 3517
65ebb059
KW
3518 if (lang) {
3519 for (j = 0; j < trial_locales_count; j++) {
3520 if (strEQ(lang, trial_locales[j])) {
3521 goto done_lang;
3522 }
3523 }
3524 trial_locales[trial_locales_count++] = lang;
3525 }
3526 done_lang:
3527
7d4bcc4a
KW
3528# if defined(WIN32) && defined(LC_ALL)
3529
65ebb059
KW
3530 /* For Windows, we also try the system default locale before "C".
3531 * (If there exists a Windows without LC_ALL we skip this because
3532 * it gets too complicated. For those, the "C" is the next
3533 * fallback possibility). The "" is the same as the 0th element of
3534 * the array, but the code at the loop above knows to treat it
3535 * differently when not the 0th */
3536 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3537
3538# endif
65ebb059
KW
3539
3540 for (j = 0; j < trial_locales_count; j++) {
3541 if (strEQ("C", trial_locales[j])) {
3542 goto done_C;
3543 }
3544 }
3545 trial_locales[trial_locales_count++] = "C";
98994639 3546
65ebb059
KW
3547 done_C: ;
3548 } /* end of first time through the loop */
98994639 3549
7d4bcc4a
KW
3550# ifdef WIN32
3551
65ebb059 3552 next_iteration: ;
7d4bcc4a
KW
3553
3554# endif
65ebb059
KW
3555
3556 } /* end of looping through the trial locales */
3557
3558 if (ok < 1) { /* If we tried to fallback */
3559 const char* msg;
3560 if (! setlocale_failure) { /* fallback succeeded */
3561 msg = "Falling back to";
3562 }
3563 else { /* fallback failed */
e5f10d49 3564 unsigned int j;
98994639 3565
65ebb059
KW
3566 /* We dropped off the end of the loop, so have to decrement i to
3567 * get back to the value the last time through */
3568 i--;
98994639 3569
65ebb059
KW
3570 ok = -1;
3571 msg = "Failed to fall back to";
3572
3573 /* To continue, we should use whatever values we've got */
7d4bcc4a 3574
e5f10d49
KW
3575 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3576 Safefree(curlocales[j]);
3577 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3578 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3579 }
65ebb059
KW
3580 }
3581
3582 if (locwarn) {
3583 const char * description;
3584 const char * name = "";
3585 if (strEQ(trial_locales[i], "C")) {
3586 description = "the standard locale";
3587 name = "C";
3588 }
7d4bcc4a
KW
3589
3590# ifdef SYSTEM_DEFAULT_LOCALE
3591
65ebb059
KW
3592 else if (strEQ(trial_locales[i], "")) {
3593 description = "the system default locale";
3594 if (system_default_locale) {
3595 name = system_default_locale;
3596 }
3597 }
7d4bcc4a
KW
3598
3599# endif /* SYSTEM_DEFAULT_LOCALE */
3600
65ebb059
KW
3601 else {
3602 description = "a fallback locale";
3603 name = trial_locales[i];
3604 }
3605 if (name && strNE(name, "")) {
3606 PerlIO_printf(Perl_error_log,
3607 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3608 }
3609 else {
3610 PerlIO_printf(Perl_error_log,
3611 "perl: warning: %s %s.\n", msg, description);
3612 }
3613 }
3614 } /* End of tried to fallback */
98994639 3615
e5f10d49
KW
3616 /* Done with finding the locales; update our records */
3617
7d4bcc4a
KW
3618# ifdef USE_LOCALE_CTYPE
3619
948523db 3620 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3621
e5f10d49 3622# endif
7d4bcc4a
KW
3623# ifdef USE_LOCALE_COLLATE
3624
948523db 3625 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3626
e5f10d49 3627# endif
7d4bcc4a
KW
3628# ifdef USE_LOCALE_NUMERIC
3629
948523db 3630 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3631
3632# endif
3633
948523db 3634 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3635
e9bc6d6b 3636# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3637
3638 /* This caches whether each category's locale is UTF-8 or not. This
3639 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3640 * initialization time before any threads have started, but not later
3641 * unless thread-safe operations are used.
47280b20
KW
3642 * Caching means that if the program heeds our dictate not to change
3643 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3644 * it may get queried without having to change locales. If the
3645 * environment is such that all categories have the same locale, this
3646 * isn't needed, as the code will not change the locale; but this
3647 * handles the uncommon case where the environment has disparate
3648 * locales for the categories */
47280b20
KW
3649 (void) _is_cur_LC_category_utf8(categories[i]);
3650
3651# endif
3652
e5f10d49
KW
3653 Safefree(curlocales[i]);
3654 }
b310b053 3655
7d4bcc4a
KW
3656# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3657
49c85077 3658 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3659 * locale is UTF-8. The call to new_ctype() just above has already
3660 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3661 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3662 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3663 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3664 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3665
a05d7ebb 3666 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3667 This is an alternative to using the -C command line switch
3668 (the -C if present will override this). */
3669 {
dd374669 3670 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3671 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3672 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3673 PL_utf8cache = -1;
b310b053
JH
3674 }
3675
7d4bcc4a 3676# endif
7d4bcc4a
KW
3677# ifdef __GLIBC__
3678
175c4cf9 3679 Safefree(language);
7d4bcc4a
KW
3680
3681# endif
175c4cf9
KW
3682
3683 Safefree(lc_all);
3684 Safefree(lang);
3685
e3305790 3686#endif /* USE_LOCALE */
2fcc0ca9 3687#ifdef DEBUGGING
7d4bcc4a 3688
2fcc0ca9 3689 /* So won't continue to output stuff */
27cdc72e 3690 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3691
2fcc0ca9
KW
3692#endif
3693
98994639
HS
3694 return ok;
3695}
3696
98994639
HS
3697#ifdef USE_LOCALE_COLLATE
3698
a4a439fb 3699char *
a4a439fb
KW
3700Perl__mem_collxfrm(pTHX_ const char *input_string,
3701 STRLEN len, /* Length of 'input_string' */
3702 STRLEN *xlen, /* Set to length of returned string
3703 (not including the collation index
3704 prefix) */
3705 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3706 )
98994639 3707{
a4a439fb
KW
3708
3709 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3710 * differences. First, it handles embedded NULs. Second, it allocates a bit
3711 * more memory than needed for the transformed data itself. The real
55e5378d 3712 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3713 * the length of that, and doesn't include the collation index size.
3714 * Please see sv_collxfrm() to see how this is used. */
3715
55e5378d
KW
3716#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3717
6696cfa7
KW
3718 char * s = (char *) input_string;
3719 STRLEN s_strlen = strlen(input_string);
79f120c8 3720 char *xbuf = NULL;
55e5378d 3721 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3722 STRLEN length_in_chars;
c664130f 3723 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3724
a4a439fb
KW
3725 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3726
3727 /* Must be NUL-terminated */
3728 assert(*(input_string + len) == '\0');
7918f24d 3729
79f120c8
KW
3730 /* If this locale has defective collation, skip */
3731 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3732 DEBUG_L(PerlIO_printf(Perl_debug_log,
3733 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3734 goto bad;
3735 }
3736
6696cfa7
KW
3737 /* Replace any embedded NULs with the control that sorts before any others.
3738 * This will give as good as possible results on strings that don't
3739 * otherwise contain that character, but otherwise there may be
3740 * less-than-perfect results with that character and NUL. This is
fdc080f3 3741 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3742 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3743 NUL */
6696cfa7
KW
3744 char * e = s + len;
3745 char * sans_nuls;
fdc080f3 3746 STRLEN sans_nuls_len;
94762aa0 3747 int try_non_controls;
afc4976f
KW
3748 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3749 making sure 2nd byte is NUL.
3750 */
3751 STRLEN this_replacement_len;
3752
1e4c9676
KW
3753 /* If we don't know what non-NUL control character sorts lowest for
3754 * this locale, find it */
f28f4d2a 3755 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3756 int j;
afc4976f 3757 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3758 includes the collation index
3759 prefixed. */
3760
91c0e2e0 3761 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3762
3763 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3764 * NUL, in which case we instead look for any character. Controls
3765 * are preferred because collation order is, in general, context
3766 * sensitive, with adjoining characters affecting the order, and
3767 * controls are less likely to have such interactions, allowing the
3768 * NUL-replacement to stand on its own. (Another way to look at it
3769 * is to imagine what would happen if the NUL were replaced by a
3770 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3771 for (try_non_controls = 0;
3772 try_non_controls < 2;
3773 try_non_controls++)
3774 {
d4ff9586
KW
3775 /* Look through all legal code points (NUL isn't) */
3776 for (j = 1; j < 256; j++) {
3777 char * x; /* j's xfrm plus collation index */
3778 STRLEN x_len; /* length of 'x' */
3779 STRLEN trial_len = 1;
736a4fed 3780 char cur_source[] = { '\0', '\0' };
d4ff9586 3781
736a4fed
KW
3782 /* Skip non-controls the first time through the loop. The
3783 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3784 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3785 ? ! isCNTRL_L1(j)
3786 : ! isCNTRL_LC(j))
3787 {
d4ff9586 3788 continue;
6696cfa7 3789 }
6696cfa7 3790
736a4fed
KW
3791 /* Create a 1-char string of the current code point */
3792 cur_source[0] = (char) j;
3793
d4ff9586
KW
3794 /* Then transform it */
3795 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3796 0 /* The string is not in UTF-8 */);
6696cfa7 3797
1e4c9676 3798 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3799 * */
3800 if (! x) {
3801 continue;
3802 }
6696cfa7 3803
d4ff9586
KW
3804 /* If this character's transformation is lower than
3805 * the current lowest, this one becomes the lowest */
3806 if ( cur_min_x == NULL
3807 || strLT(x + COLLXFRM_HDR_LEN,
3808 cur_min_x + COLLXFRM_HDR_LEN))
3809 {
f28f4d2a 3810 PL_strxfrm_NUL_replacement = j;
d4ff9586 3811 cur_min_x = x;
d4ff9586
KW
3812 }
3813 else {
3814 Safefree(x);
3815 }
1e4c9676 3816 } /* end of loop through all 255 characters */
6696cfa7 3817
1e4c9676 3818 /* Stop looking if found */
94762aa0
KW
3819 if (cur_min_x) {
3820 break;
3821 }
3822
3823 /* Unlikely, but possible, if there aren't any controls that
3824 * work in the locale, repeat the loop, looking for any
3825 * character that works */
3826 DEBUG_L(PerlIO_printf(Perl_debug_log,
3827 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3828 } /* End of loop to try first the controls, then any char */
6696cfa7 3829
94762aa0
KW
3830 if (! cur_min_x) {
3831 DEBUG_L(PerlIO_printf(Perl_debug_log,
3832 "_mem_collxfrm: Couldn't find any character to replace"
3833 " embedded NULs in locale %s with", PL_collation_name));
3834 goto bad;
58eebef2
KW
3835 }
3836
94762aa0
KW
3837 DEBUG_L(PerlIO_printf(Perl_debug_log,
3838 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3839 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3840
6696cfa7 3841 Safefree(cur_min_x);
1e4c9676 3842 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3843
3844 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3845 * UTF8-ness of the original */
f28f4d2a
KW
3846 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3847 this_replacement_char[0] =
3848 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3849 this_replacement_char[1] =
3850 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3851 this_replacement_len = 2;
3852 }
3853 else {
f28f4d2a 3854 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3855 /* this_replacement_char[1] = '\0' was done at initialization */
3856 this_replacement_len = 1;
6696cfa7
KW
3857 }
3858
3859 /* The worst case length for the replaced string would be if every
3860 * character in it is NUL. Multiply that by the length of each
3861 * replacement, and allow for a trailing NUL */
afc4976f 3862 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3863 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3864 *sans_nuls = '\0';
3865
6696cfa7
KW
3866 /* Replace each NUL with the lowest collating control. Loop until have
3867 * exhausted all the NULs */
3868 while (s + s_strlen < e) {
6069d6c5 3869 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3870
3871 /* Do the actual replacement */
6069d6c5 3872 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);