This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Change enum names for new locale function parameters
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
e9bc6d6b
KW
35 *
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
166f8a29
DM
44 */
45
98994639
HS
46#include "EXTERN.h"
47#define PERL_IN_LOCALE_C
f7416781 48#include "perl_langinfo.h"
98994639
HS
49#include "perl.h"
50
a4af207c
JH
51#include "reentr.h"
52
0dec74cd
KW
53#ifdef I_WCHAR
54# include <wchar.h>
55#endif
56
2fcc0ca9
KW
57/* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
5a4b0634
KW
60#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61# define debug_initialization 0
62# define DEBUG_INITIALIZATION_set(v)
63#else
2fcc0ca9 64static bool debug_initialization = FALSE;
5a4b0634 65# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
66#endif
67
48015184
KW
68
69/* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
71 * done generally. */
72#define GET_ERRNO saved_errno
73
291a84fb
KW
74/* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
77#define STRLENs(s) (sizeof("" s "") - 1)
78
63e5b0d7
KW
79/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
85 * cntrl: 84-97 9B-9F
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89#define isNAME_C_OR_POSIX(name) \
90 ( (name) != NULL \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
93
8ef6e574
KW
94#ifdef USE_LOCALE
95
47280b20
KW
96/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
98
99#define UTF8NESS_SEP "\v"
100#define UTF8NESS_PREFIX "\f"
101
102/* So, the string looks like:
98994639 103 *
47280b20 104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 105 *
47280b20
KW
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
108
109STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
111
112#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
114
115/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
118
98994639
HS
119STATIC char *
120S_stdize_locale(pTHX_ char *locs)
121{
47280b20
KW
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
124 *
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
130 *
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
134 * */
135
7452cf6a 136 const char * const s = strchr(locs, '=');
98994639
HS
137 bool okay = TRUE;
138
7918f24d
NC
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
140
8772537c
AL
141 if (s) {
142 const char * const t = strchr(s, '.');
98994639 143 okay = FALSE;
8772537c
AL
144 if (t) {
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
149 locs[len] = 0;
150 okay = TRUE;
98994639
HS
151 }
152 }
153 }
154
155 if (!okay)
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
157
158 return locs;
159}
160
e5f10d49
KW
161/* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
163 * order. */
164
165const int categories[] = {
166
167# ifdef USE_LOCALE_NUMERIC
168 LC_NUMERIC,
169# endif
170# ifdef USE_LOCALE_CTYPE
171 LC_CTYPE,
172# endif
173# ifdef USE_LOCALE_COLLATE
174 LC_COLLATE,
175# endif
176# ifdef USE_LOCALE_TIME
177 LC_TIME,
178# endif
179# ifdef USE_LOCALE_MESSAGES
180 LC_MESSAGES,
181# endif
182# ifdef USE_LOCALE_MONETARY
183 LC_MONETARY,
184# endif
9821811f
KW
185# ifdef USE_LOCALE_ADDRESS
186 LC_ADDRESS,
187# endif
188# ifdef USE_LOCALE_IDENTIFICATION
189 LC_IDENTIFICATION,
190# endif
191# ifdef USE_LOCALE_MEASUREMENT
192 LC_MEASUREMENT,
193# endif
194# ifdef USE_LOCALE_PAPER
195 LC_PAPER,
196# endif
197# ifdef USE_LOCALE_TELEPHONE
198 LC_TELEPHONE,
199# endif
e5f10d49
KW
200# ifdef LC_ALL
201 LC_ALL,
202# endif
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
206};
207
208/* The top-most real element is LC_ALL */
209
210const char * category_names[] = {
211
212# ifdef USE_LOCALE_NUMERIC
213 "LC_NUMERIC",
214# endif
215# ifdef USE_LOCALE_CTYPE
216 "LC_CTYPE",
217# endif
218# ifdef USE_LOCALE_COLLATE
219 "LC_COLLATE",
220# endif
221# ifdef USE_LOCALE_TIME
222 "LC_TIME",
223# endif
224# ifdef USE_LOCALE_MESSAGES
225 "LC_MESSAGES",
226# endif
227# ifdef USE_LOCALE_MONETARY
228 "LC_MONETARY",
229# endif
9821811f
KW
230# ifdef USE_LOCALE_ADDRESS
231 "LC_ADDRESS",
232# endif
233# ifdef USE_LOCALE_IDENTIFICATION
234 "LC_IDENTIFICATION",
235# endif
236# ifdef USE_LOCALE_MEASUREMENT
237 "LC_MEASUREMENT",
238# endif
239# ifdef USE_LOCALE_PAPER
240 "LC_PAPER",
241# endif
242# ifdef USE_LOCALE_TELEPHONE
243 "LC_TELEPHONE",
244# endif
e5f10d49
KW
245# ifdef LC_ALL
246 "LC_ALL",
247# endif
248 NULL /* Placeholder */
249 };
250
251# ifdef LC_ALL
252
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
256
257# else
258
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
262
263# endif
264
265/* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
948523db
KW
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
e5f10d49 272
b09aaf40
KW
273STATIC const char *
274S_category_name(const int category)
275{
276 unsigned int i;
277
278#ifdef LC_ALL
279
280 if (category == LC_ALL) {
281 return "LC_ALL";
282 }
283
284#endif
285
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
289 }
290 }
291
292 {
293 const char suffix[] = " (unknown)";
294 int temp = category;
295 Size_t length = sizeof(suffix) + 1;
296 char * unknown;
297 dTHX;
298
299 if (temp < 0) {
300 length++;
301 temp = - temp;
302 }
303
304 /* Calculate the number of digits */
305 while (temp >= 10) {
306 temp /= 10;
307 length++;
308 }
309
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
312 SAVEFREEPV(unknown);
313 return unknown;
314 }
315}
316
948523db
KW
317/* Now create LC_foo_INDEX #defines for just those categories on this system */
318# ifdef USE_LOCALE_NUMERIC
319# define LC_NUMERIC_INDEX 0
320# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
321# else
322# define _DUMMY_NUMERIC -1
323# endif
324# ifdef USE_LOCALE_CTYPE
325# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326# define _DUMMY_CTYPE LC_CTYPE_INDEX
327# else
328# define _DUMMY_CTYPE _DUMMY_NUMERIC
329# endif
330# ifdef USE_LOCALE_COLLATE
331# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332# define _DUMMY_COLLATE LC_COLLATE_INDEX
333# else
334# define _DUMMY_COLLATE _DUMMY_COLLATE
335# endif
336# ifdef USE_LOCALE_TIME
337# define LC_TIME_INDEX _DUMMY_COLLATE + 1
338# define _DUMMY_TIME LC_TIME_INDEX
339# else
340# define _DUMMY_TIME _DUMMY_COLLATE
341# endif
342# ifdef USE_LOCALE_MESSAGES
343# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
345# else
346# define _DUMMY_MESSAGES _DUMMY_TIME
347# endif
348# ifdef USE_LOCALE_MONETARY
349# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350# define _DUMMY_MONETARY LC_MONETARY_INDEX
351# else
352# define _DUMMY_MONETARY _DUMMY_MESSAGES
353# endif
9821811f
KW
354# ifdef USE_LOCALE_ADDRESS
355# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
357# else
358# define _DUMMY_ADDRESS _DUMMY_MONETARY
359# endif
360# ifdef USE_LOCALE_IDENTIFICATION
361# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
363# else
364# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
365# endif
366# ifdef USE_LOCALE_MEASUREMENT
367# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
369# else
370# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
371# endif
372# ifdef USE_LOCALE_PAPER
373# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374# define _DUMMY_PAPER LC_PAPER_INDEX
375# else
376# define _DUMMY_PAPER _DUMMY_MEASUREMENT
377# endif
378# ifdef USE_LOCALE_TELEPHONE
379# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
381# else
382# define _DUMMY_TELEPHONE _DUMMY_PAPER
383# endif
948523db 384# ifdef LC_ALL
9821811f 385# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
386# endif
387#endif /* ifdef USE_LOCALE */
8ef6e574 388
d2b24094 389/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
390#ifdef WIN32
391# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
392#else
393# define my_setlocale(cat, locale) setlocale(cat, locale)
394#endif
395
396#ifndef USE_POSIX_2008_LOCALE
397
398/* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
402
403#else /* Below uses POSIX 2008 */
404
405/* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411# define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
414
415/* A third array, parallel to the ones above to map from category to its
416 * equivalent mask */
417const int category_masks[] = {
418# ifdef USE_LOCALE_NUMERIC
419 LC_NUMERIC_MASK,
420# endif
421# ifdef USE_LOCALE_CTYPE
422 LC_CTYPE_MASK,
423# endif
424# ifdef USE_LOCALE_COLLATE
425 LC_COLLATE_MASK,
426# endif
427# ifdef USE_LOCALE_TIME
428 LC_TIME_MASK,
429# endif
430# ifdef USE_LOCALE_MESSAGES
431 LC_MESSAGES_MASK,
432# endif
433# ifdef USE_LOCALE_MONETARY
434 LC_MONETARY_MASK,
435# endif
436# ifdef USE_LOCALE_ADDRESS
437 LC_ADDRESS_MASK,
438# endif
439# ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
441# endif
442# ifdef USE_LOCALE_MEASUREMENT
443 LC_MEASUREMENT_MASK,
444# endif
445# ifdef USE_LOCALE_PAPER
446 LC_PAPER_MASK,
447# endif
448# ifdef USE_LOCALE_TELEPHONE
449 LC_TELEPHONE_MASK,
450# endif
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
455 * time */
456 LC_ALL_MASK
457 };
458
459STATIC const char *
460S_emulate_setlocale(const int category,
461 const char * locale,
462 unsigned int index,
463 const bool is_index_valid
464 )
465{
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
473 *
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
476 *
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
486 *
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
501 */
502
503 int mask;
504 locale_t old_obj;
505 locale_t new_obj;
506 dTHX;
507
508# ifdef DEBUGGING
509
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
512 }
513
514# endif
515
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
518 unsigned int i;
519
520# ifdef DEBUGGING
521
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
524 }
525
526# endif
527
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
530 index = i;
531 goto found_index;
532 }
533 }
534
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
539 category, locale);
540 return NULL;
541
542 found_index: ;
543
544# ifdef DEBUGGING
545
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
548 }
549
550# endif
551
552 }
553
554 mask = category_masks[index];
555
556# ifdef DEBUGGING
557
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
560 }
561
562# endif
563
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
567
568# ifdef DEBUGGING
569
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
572 }
573
574# endif
575
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
578 }
579
580# ifdef HAS_QUERYLOCALE
581
582 return (char *) querylocale(mask, cur_obj);
583
d2b24094 584# else
ecdda939
KW
585
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
588
e9bc6d6b
KW
589# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
590
591 {
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
595
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
598 /*
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
602 */
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
607
608# ifdef USE_C_BACKTRACE
609
610 dump_c_backtrace(Perl_debug_log, 20, 1);
611
612# endif
613
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
618 }
619
620 return temp_name;
621 }
622 }
623
624# endif
625
626 /* Without querylocale(), we have to use our record-keeping we've
627 * done. */
628
629 if (category != LC_ALL) {
630
631# ifdef DEBUGGING
632
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 }
636
637# endif
638
639 return PL_curlocales[index];
640 }
641 else { /* For LC_ALL */
642 unsigned int i;
643 Size_t names_len = 0;
644 char * all_string;
7c93a471 645 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
646
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
649
650# ifdef DEBUGGING
651
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
654 }
655
656# endif
657
658 return PL_curlocales[LC_ALL_INDEX];
659 }
660
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
e9bc6d6b
KW
666 for (i = 0; i < LC_ALL_INDEX; i++) {
667
668# ifdef DEBUGGING
669
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
672 }
673
674# endif
675
676 names_len += strlen(category_names[i])
677 + 1 /* '=' */
678 + strlen(PL_curlocales[i])
679 + 1; /* ';' */
7c93a471
KW
680
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
683 }
684 }
685
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 692 }
7c93a471 693
e9bc6d6b
KW
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
696 *all_string = '\0';
697
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
700
701# ifdef DEBUGGING
702
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
705 }
706
707# endif
708
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
713 }
714
715# ifdef DEBUGGING
716
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
719 }
720
721 #endif
722
723 return all_string;
724 }
725
726# ifdef EINVAL
727
728 SETERRNO(EINVAL, LIB_INVARG);
729
730# endif
731
732 return NULL;
733
d2b24094
KW
734# endif
735
e9bc6d6b
KW
736 }
737
738 assert(PL_C_locale_obj);
739
740 /* Otherwise, we are switching locales. This will generally entail freeing
741 * the current one's space (at the C library's discretion). We need to
742 * stop using that locale before the switch. So switch to a known locale
743 * object that we don't otherwise mess with. This returns the locale
744 * object in effect at the time of the switch. */
745 old_obj = uselocale(PL_C_locale_obj);
746
747# ifdef DEBUGGING
748
749 if (DEBUG_Lv_TEST || debug_initialization) {
750 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
751 }
752
753# endif
754
755 if (! old_obj) {
756
757# ifdef DEBUGGING
758
759 if (DEBUG_L_TEST || debug_initialization) {
760 dSAVE_ERRNO;
761 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
762 RESTORE_ERRNO;
763 }
764
765# endif
766
767 return NULL;
768 }
769
770# ifdef DEBUGGING
771
772 if (DEBUG_Lv_TEST || debug_initialization) {
773 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
774 }
775
776# endif
777
778 /* If we weren't in a thread safe locale, set so that newlocale() below
779 which uses 'old_obj', uses an empty one. Same for our reserved C object.
780 The latter is defensive coding, so that, even if there is some bug, we
781 will never end up trying to modify either of these, as if passed to
782 newlocale(), they can be. */
783 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
784 old_obj = (locale_t) 0;
785 }
786
787 /* Create the new locale (it may actually modify the current one). */
788
789# ifndef HAS_QUERYLOCALE
790
791 if (strEQ(locale, "")) {
792
793 /* For non-querylocale() systems, we do the setting of "" ourselves to
794 * be sure that we really know what's going on. We follow the Linux
795 * documented behavior (but if that differs from the actual behavior,
796 * this won't work exactly as the OS implements). We go out and
797 * examine the environment based on our understanding of how the system
798 * works, and use that to figure things out */
799
800 const char * const lc_all = PerlEnv_getenv("LC_ALL");
801
802 /* Use any "LC_ALL" environment variable, as it overrides everything
803 * else. */
804 if (lc_all && strNE(lc_all, "")) {
805 locale = lc_all;
806 }
807 else {
808
809 /* Otherwise, we need to dig deeper. Unless overridden, the
810 * default is the LANG environment variable; if it doesn't exist,
811 * then "C" */
812
813 const char * default_name;
814
815 /* To minimize other threads messing with the environment, we copy
816 * the variable, making it a temporary. But this doesn't work upon
817 * program initialization before any scopes are created, and at
818 * this time, there's nothing else going on that would interfere.
819 * So skip the copy in that case */
820 if (PL_scopestack_ix == 0) {
821 default_name = PerlEnv_getenv("LANG");
822 }
823 else {
824 default_name = savepv(PerlEnv_getenv("LANG"));
825 }
826
827 if (! default_name || strEQ(default_name, "")) {
828 default_name = "C";
829 }
830 else if (PL_scopestack_ix != 0) {
831 SAVEFREEPV(default_name);
832 }
833
834 if (category != LC_ALL) {
835 const char * const name = PerlEnv_getenv(category_names[index]);
836
837 /* Here we are setting a single category. Assume will have the
838 * default name */
839 locale = default_name;
840
841 /* But then look for an overriding environment variable */
842 if (name && strNE(name, "")) {
843 locale = name;
844 }
845 }
846 else {
847 bool did_override = FALSE;
848 unsigned int i;
849
850 /* Here, we are getting LC_ALL. Any categories that don't have
851 * a corresponding environment variable set should be set to
852 * LANG, or to "C" if there is no LANG. If no individual
853 * categories differ from this, we can just set LC_ALL. This
854 * is buggy on systems that have extra categories that we don't
855 * know about. If there is an environment variable that sets
856 * that category, we won't know to look for it, and so our use
857 * of LANG or "C" improperly overrides it. On the other hand,
858 * if we don't do what is done here, and there is no
859 * environment variable, the category's locale should be set to
860 * LANG or "C". So there is no good solution. khw thinks the
861 * best is to look at systems to see what categories they have,
862 * and include them, and then to assume that we know the
863 * complete set */
864
865 for (i = 0; i < LC_ALL_INDEX; i++) {
866 const char * const env_override
867 = savepv(PerlEnv_getenv(category_names[i]));
868 const char * this_locale = ( env_override
869 && strNE(env_override, ""))
870 ? env_override
871 : default_name;
6435f98d
KW
872 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
873 {
874 Safefree(env_override);
875 return NULL;
876 }
e9bc6d6b
KW
877
878 if (strNE(this_locale, default_name)) {
879 did_override = TRUE;
880 }
881
882 Safefree(env_override);
883 }
884
885 /* If all the categories are the same, we can set LC_ALL to
886 * that */
887 if (! did_override) {
888 locale = default_name;
889 }
890 else {
891
892 /* Here, LC_ALL is no longer valid, as some individual
893 * categories don't match it. We call ourselves
894 * recursively, as that will execute the code that
895 * generates the proper locale string for this situation.
896 * We don't do the remainder of this function, as that is
897 * to update our records, and we've just done that for the
898 * individual categories in the loop above, and doing so
899 * would cause LC_ALL to be done as well */
900 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
901 }
902 }
903 }
904 }
905 else if (strchr(locale, ';')) {
906
907 /* LC_ALL may actually incude a conglomeration of various categories.
908 * Without querylocale, this code uses the glibc (as of this writing)
909 * syntax for representing that, but that is not a stable API, and
910 * other platforms do it differently, so we have to handle all cases
911 * ourselves */
912
913 const char * s = locale;
914 const char * e = locale + strlen(locale);
915 const char * p = s;
916 const char * category_end;
917 const char * name_start;
918 const char * name_end;
919
920 while (s < e) {
921 unsigned int i;
922
923 /* Parse through the category */
924 while (isWORDCHAR(*p)) {
925 p++;
926 }
927 category_end = p;
928
929 if (*p++ != '=') {
930 Perl_croak(aTHX_
931 "panic: %s: %d: Unexpected character in locale name '%02X",
932 __FILE__, __LINE__, *(p-1));
933 }
934
935 /* Parse through the locale name */
936 name_start = p;
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
1347 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1348 * them */
1349 if (! PL_numeric_standard) {
4e6826bf 1350 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1351 FALSE /* Don't toggle locale */ ))
4e6826bf 1352 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1353 }
abe1abcf 1354
4c68b815 1355 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1356 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1357 Safefree(PL_numeric_name);
b03f34cf 1358 PL_numeric_name = save_newnum;
b03f34cf 1359 }
abe1abcf
KW
1360 else {
1361 Safefree(save_newnum);
1362 }
4c28b29c 1363
4c68b815
KW
1364 PL_numeric_underlying_is_standard = PL_numeric_standard;
1365
7e5377f7 1366# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1367
1368 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1369 PL_numeric_name,
1370 PL_underlying_numeric_obj);
1371
1372#endif
1373
4c68b815
KW
1374 if (DEBUG_L_TEST || debug_initialization) {
1375 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1376 }
1377
4c28b29c
KW
1378 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1379 * have to worry about the radix being a non-dot. (Core operations that
1380 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1381 if (PL_numeric_standard) {
1382 set_numeric_radix(0);
1383 }
1384 else {
1385 set_numeric_standard();
1386 }
4c28b29c 1387
98994639 1388#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1389
98994639
HS
1390}
1391
1392void
1393Perl_set_numeric_standard(pTHX)
1394{
7d4bcc4a 1395
98994639 1396#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1397
28c1bf33
KW
1398 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1399 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1400 * macro avoids calling this routine if toggling isn't necessary according
1401 * to our records (which could be wrong if some XS code has changed the
1402 * locale behind our back) */
0d071d52 1403
837ce802 1404 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1405 PL_numeric_standard = TRUE;
4c68b815 1406 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1407 set_numeric_radix(0);
7d4bcc4a
KW
1408
1409# ifdef DEBUGGING
1410
2fcc0ca9
KW
1411 if (DEBUG_L_TEST || debug_initialization) {
1412 PerlIO_printf(Perl_debug_log,
58e4a467 1413 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1414 }
98994639 1415
7d4bcc4a 1416# endif
98994639 1417#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1418
98994639
HS
1419}
1420
1421void
5792c642 1422Perl_set_numeric_underlying(pTHX)
98994639 1423{
7d4bcc4a 1424
98994639 1425#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1426
28c1bf33 1427 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1428 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1429 * instead of calling this directly. The macro avoids calling this routine
1430 * if toggling isn't necessary according to our records (which could be
1431 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1432
837ce802 1433 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1434 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1435 PL_numeric_underlying = TRUE;
3c62fd46 1436 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1437
1438# ifdef DEBUGGING
1439
2fcc0ca9
KW
1440 if (DEBUG_L_TEST || debug_initialization) {
1441 PerlIO_printf(Perl_debug_log,
58e4a467 1442 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1443 PL_numeric_name);
1444 }
98994639 1445
7d4bcc4a 1446# endif
98994639 1447#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1448
98994639
HS
1449}
1450
1451/*
1452 * Set up for a new ctype locale.
1453 */
a4f00dcc
KW
1454STATIC void
1455S_new_ctype(pTHX_ const char *newctype)
98994639 1456{
7d4bcc4a
KW
1457
1458#ifndef USE_LOCALE_CTYPE
1459
1460 PERL_ARGS_ASSERT_NEW_CTYPE;
1461 PERL_UNUSED_ARG(newctype);
1462 PERL_UNUSED_CONTEXT;
1463
1464#else
0d071d52 1465
291a84fb 1466 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1467 * core Perl this and that 'newctype' is the name of the new locale.
1468 *
1469 * This function sets up the folding arrays for all 256 bytes, assuming
1470 * that tofold() is tolc() since fold case is not a concept in POSIX,
1471 *
1472 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1473 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1474 * this function should be called directly only from this file and from
0d071d52
KW
1475 * POSIX::setlocale() */
1476
27da23d5 1477 dVAR;
013f4e03 1478 unsigned int i;
98994639 1479
8b7358b9
KW
1480 /* Don't check for problems if we are suppressing the warnings */
1481 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1482
7918f24d
NC
1483 PERL_ARGS_ASSERT_NEW_CTYPE;
1484
215c5139
KW
1485 /* We will replace any bad locale warning with 1) nothing if the new one is
1486 * ok; or 2) a new warning for the bad new locale */
1487 if (PL_warn_locale) {
1488 SvREFCNT_dec_NN(PL_warn_locale);
1489 PL_warn_locale = NULL;
1490 }
1491
c1284011 1492 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1493
1494 /* A UTF-8 locale gets standard rules. But note that code still has to
1495 * handle this specially because of the three problematic code points */
1496 if (PL_in_utf8_CTYPE_locale) {
1497 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1498 }
8b7358b9
KW
1499
1500 /* We don't populate the other lists if a UTF-8 locale, but do check that
1501 * everything works as expected, unless checking turned off */
1502 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1503 /* Assume enough space for every character being bad. 4 spaces each
1504 * for the 94 printable characters that are output like "'x' "; and 5
1505 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1506 * NUL */
8b7358b9 1507 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1508 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1509 to start */
1510 unsigned int bad_count = 0; /* Count of bad characters */
1511
baa60164 1512 for (i = 0; i < 256; i++) {
8b7358b9 1513 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1514 if (isupper(i))
1515 PL_fold_locale[i] = (U8) tolower(i);
1516 else if (islower(i))
1517 PL_fold_locale[i] = (U8) toupper(i);
1518 else
1519 PL_fold_locale[i] = (U8) i;
8b7358b9 1520 }
8c6180a9
KW
1521
1522 /* If checking for locale problems, see if the native ASCII-range
1523 * printables plus \n and \t are in their expected categories in
1524 * the new locale. If not, this could mean big trouble, upending
1525 * Perl's and most programs' assumptions, like having a
1526 * metacharacter with special meaning become a \w. Fortunately,
1527 * it's very rare to find locales that aren't supersets of ASCII
1528 * nowadays. It isn't a problem for most controls to be changed
1529 * into something else; we check only \n and \t, though perhaps \r
1530 * could be an issue as well. */
7d4bcc4a 1531 if ( check_for_problems
8c6180a9
KW
1532 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1533 {
8b7358b9
KW
1534 bool is_bad = FALSE;
1535 char name[3] = { '\0' };
1536
1537 /* Convert the name into a string */
1538 if (isPRINT_A(i)) {
1539 name[0] = i;
1540 name[1] = '\0';
1541 }
1542 else if (i == '\n') {
1543 my_strlcpy(name, "\n", sizeof(name));
1544 }
1545 else {
1546 my_strlcpy(name, "\t", sizeof(name));
1547 }
1548
1549 /* Check each possibe class */
1550 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1551 is_bad = TRUE;
1552 DEBUG_L(PerlIO_printf(Perl_debug_log,
1553 "isalnum('%s') unexpectedly is %d\n",
1554 name, cBOOL(isalnum(i))));
1555 }
1556 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1557 is_bad = TRUE;
1558 DEBUG_L(PerlIO_printf(Perl_debug_log,
1559 "isalpha('%s') unexpectedly is %d\n",
1560 name, cBOOL(isalpha(i))));
1561 }
1562 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1563 is_bad = TRUE;
1564 DEBUG_L(PerlIO_printf(Perl_debug_log,
1565 "isdigit('%s') unexpectedly is %d\n",
1566 name, cBOOL(isdigit(i))));
1567 }
1568 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1569 is_bad = TRUE;
1570 DEBUG_L(PerlIO_printf(Perl_debug_log,
1571 "isgraph('%s') unexpectedly is %d\n",
1572 name, cBOOL(isgraph(i))));
1573 }
1574 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1575 is_bad = TRUE;
1576 DEBUG_L(PerlIO_printf(Perl_debug_log,
1577 "islower('%s') unexpectedly is %d\n",
1578 name, cBOOL(islower(i))));
1579 }
1580 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1581 is_bad = TRUE;
1582 DEBUG_L(PerlIO_printf(Perl_debug_log,
1583 "isprint('%s') unexpectedly is %d\n",
1584 name, cBOOL(isprint(i))));
1585 }
1586 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1587 is_bad = TRUE;
1588 DEBUG_L(PerlIO_printf(Perl_debug_log,
1589 "ispunct('%s') unexpectedly is %d\n",
1590 name, cBOOL(ispunct(i))));
1591 }
1592 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1593 is_bad = TRUE;
1594 DEBUG_L(PerlIO_printf(Perl_debug_log,
1595 "isspace('%s') unexpectedly is %d\n",
1596 name, cBOOL(isspace(i))));
1597 }
1598 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1599 is_bad = TRUE;
1600 DEBUG_L(PerlIO_printf(Perl_debug_log,
1601 "isupper('%s') unexpectedly is %d\n",
1602 name, cBOOL(isupper(i))));
1603 }
1604 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1605 is_bad = TRUE;
1606 DEBUG_L(PerlIO_printf(Perl_debug_log,
1607 "isxdigit('%s') unexpectedly is %d\n",
1608 name, cBOOL(isxdigit(i))));
1609 }
1610 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1611 is_bad = TRUE;
1612 DEBUG_L(PerlIO_printf(Perl_debug_log,
1613 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1614 name, tolower(i), (int) toLOWER_A(i)));
1615 }
1616 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1617 is_bad = TRUE;
1618 DEBUG_L(PerlIO_printf(Perl_debug_log,
1619 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1620 name, toupper(i), (int) toUPPER_A(i)));
1621 }
1622 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1623 is_bad = TRUE;
1624 DEBUG_L(PerlIO_printf(Perl_debug_log,
1625 "'\\n' (=%02X) is not a control\n", (int) i));
1626 }
1627
1628 /* Add to the list; Separate multiple entries with a blank */
1629 if (is_bad) {
1630 if (bad_count) {
1631 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1632 }
8b7358b9
KW
1633 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1634 bad_count++;
8c6180a9
KW
1635 }
1636 }
1637 }
1638
7d4bcc4a
KW
1639# ifdef MB_CUR_MAX
1640
8c6180a9 1641 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1642 * this locale requires more than one byte, there are going to be
8c6180a9 1643 * problems. */
9c8a6dc2
KW
1644 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1645 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1646 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1647
8b7358b9
KW
1648 if ( check_for_problems && MB_CUR_MAX > 1
1649 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1650
1651 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1652 * locale. Just assume that the implementation for them (plus
1653 * for POSIX) is correct and the > 1 value is spurious. (Since
1654 * these are specially handled to never be considered UTF-8
1655 * locales, as long as this is the only problem, everything
1656 * should work fine */
1657 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1658 {
8c6180a9
KW
1659 multi_byte_locale = TRUE;
1660 }
7d4bcc4a
KW
1661
1662# endif
8c6180a9 1663
8b7358b9
KW
1664 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1665 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1666 PL_warn_locale = Perl_newSVpvf(aTHX_
1667 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1668 " which have\nunexpected meanings: %s\nThe Perl program"
1669 " will use the expected meanings",
8b7358b9 1670 newctype, bad_chars_list);
8b7358b9
KW
1671 }
1672 else {
1673 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1674 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1675 newctype,
8c6180a9
KW
1676 (multi_byte_locale)
1677 ? " Some characters in it are not recognized by"
1678 " Perl."
1679 : "",
1680 (bad_count)
1681 ? "\nThe following characters (and maybe others)"
1682 " may not have the same meaning as the Perl"
1683 " program expects:\n"
1684 : "",
1685 (bad_count)
1686 ? bad_chars_list
1687 : ""
1688 );
8b7358b9
KW
1689 }
1690
b119c2be
KW
1691# ifdef HAS_NL_LANGINFO
1692
1693 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1694 /* parameter FALSE is a don't care here */
4e6826bf 1695 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1696
1697# endif
1698
8b7358b9
KW
1699 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1700
cc9eaeb0 1701 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1702 * output the message now. If not in that scope, we save the
1703 * message to be output at the first operation using this locale,
1704 * if that actually happens. Most programs don't use locales, so
1705 * they are immune to bad ones. */
cc9eaeb0 1706 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1707
780fcc9f
KW
1708 /* The '0' below suppresses a bogus gcc compiler warning */
1709 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1710
bddebb56
KW
1711 if (IN_LC(LC_CTYPE)) {
1712 SvREFCNT_dec_NN(PL_warn_locale);
1713 PL_warn_locale = NULL;
1714 }
780fcc9f 1715 }
baa60164 1716 }
31f05a37 1717 }
98994639
HS
1718
1719#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1720
98994639
HS
1721}
1722
98994639 1723void
2726666d
KW
1724Perl__warn_problematic_locale()
1725{
2726666d
KW
1726
1727#ifdef USE_LOCALE_CTYPE
1728
5f04a188
KW
1729 dTHX;
1730
1731 /* Internal-to-core function that outputs the message in PL_warn_locale,
1732 * and then NULLS it. Should be called only through the macro
1733 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1734
2726666d 1735 if (PL_warn_locale) {
2726666d
KW
1736 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1737 SvPVX(PL_warn_locale),
1738 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1739 SvREFCNT_dec_NN(PL_warn_locale);
1740 PL_warn_locale = NULL;
1741 }
1742
1743#endif
1744
1745}
1746
a4f00dcc
KW
1747STATIC void
1748S_new_collate(pTHX_ const char *newcoll)
98994639 1749{
7d4bcc4a
KW
1750
1751#ifndef USE_LOCALE_COLLATE
1752
1753 PERL_UNUSED_ARG(newcoll);
1754 PERL_UNUSED_CONTEXT;
1755
1756#else
0d071d52 1757
291a84fb 1758 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1759 * core Perl this and that 'newcoll' is the name of the new locale.
1760 *
d35fca5f
KW
1761 * The design of locale collation is that every locale change is given an
1762 * index 'PL_collation_ix'. The first time a string particpates in an
1763 * operation that requires collation while locale collation is active, it
1764 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1765 * magic includes the collation index, and the transformation of the string
1766 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1767 * instead of the string itself. If a string changes, the magic is
1768 * cleared. The next time the locale changes, the index is incremented,
1769 * and so we know during a comparison that the transformation is not
1770 * necessarily still valid, and so is recomputed. Note that if the locale
1771 * changes enough times, the index could wrap (a U32), and it is possible
1772 * that a transformation would improperly be considered valid, leading to
1773 * an unlikely bug */
0d071d52 1774
98994639
HS
1775 if (! newcoll) {
1776 if (PL_collation_name) {
1777 ++PL_collation_ix;
1778 Safefree(PL_collation_name);
1779 PL_collation_name = NULL;
1780 }
1781 PL_collation_standard = TRUE;
00bf60ca 1782 is_standard_collation:
98994639
HS
1783 PL_collxfrm_base = 0;
1784 PL_collxfrm_mult = 2;
165a1c52 1785 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1786 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1787 PL_strxfrm_max_cp = 0;
98994639
HS
1788 return;
1789 }
1790
d35fca5f 1791 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1792 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1793 ++PL_collation_ix;
1794 Safefree(PL_collation_name);
1795 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1796 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1797 if (PL_collation_standard) {
1798 goto is_standard_collation;
1799 }
98994639 1800
165a1c52 1801 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1802 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1803 PL_strxfrm_max_cp = 0;
165a1c52 1804
59c018b9
KW
1805 /* A locale collation definition includes primary, secondary, tertiary,
1806 * etc. weights for each character. To sort, the primary weights are
1807 * used, and only if they compare equal, then the secondary weights are
1808 * used, and only if they compare equal, then the tertiary, etc.
1809 *
1810 * strxfrm() works by taking the input string, say ABC, and creating an
1811 * output transformed string consisting of first the primary weights,
1812 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1813 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1814 * may not have weights at every level. In our example, let's say B
1815 * doesn't have a tertiary weight, and A doesn't have a secondary
1816 * weight. The constructed string is then going to be
1817 * A¹B¹C¹ B²C² A³C³ ....
1818 * This has the desired effect that strcmp() will look at the secondary
1819 * or tertiary weights only if the strings compare equal at all higher
1820 * priority weights. The spaces shown here, like in
c342d20e 1821 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1822 * are not just for readability. In the general case, these must
1823 * actually be bytes, which we will call here 'separator weights'; and
1824 * they must be smaller than any other weight value, but since these
1825 * are C strings, only the terminating one can be a NUL (some
1826 * implementations may include a non-NUL separator weight just before
1827 * the NUL). Implementations tend to reserve 01 for the separator
1828 * weights. They are needed so that a shorter string's secondary
1829 * weights won't be misconstrued as primary weights of a longer string,
1830 * etc. By making them smaller than any other weight, the shorter
1831 * string will sort first. (Actually, if all secondary weights are
1832 * smaller than all primary ones, there is no need for a separator
1833 * weight between those two levels, etc.)
1834 *
1835 * The length of the transformed string is roughly a linear function of
1836 * the input string. It's not exactly linear because some characters
1837 * don't have weights at all levels. When we call strxfrm() we have to
1838 * allocate some memory to hold the transformed string. The
1839 * calculations below try to find coefficients 'm' and 'b' for this
1840 * locale so that m*x + b equals how much space we need, given the size
1841 * of the input string in 'x'. If we calculate too small, we increase
1842 * the size as needed, and call strxfrm() again, but it is better to
1843 * get it right the first time to avoid wasted expensive string
1844 * transformations. */
1845
98994639 1846 {
79f120c8
KW
1847 /* We use the string below to find how long the tranformation of it
1848 * is. Almost all locales are supersets of ASCII, or at least the
1849 * ASCII letters. We use all of them, half upper half lower,
1850 * because if we used fewer, we might hit just the ones that are
1851 * outliers in a particular locale. Most of the strings being
1852 * collated will contain a preponderance of letters, and even if
1853 * they are above-ASCII, they are likely to have the same number of
1854 * weight levels as the ASCII ones. It turns out that digits tend
1855 * to have fewer levels, and some punctuation has more, but those
1856 * are relatively sparse in text, and khw believes this gives a
1857 * reasonable result, but it could be changed if experience so
1858 * dictates. */
1859 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1860 char * x_longer; /* Transformed 'longer' */
1861 Size_t x_len_longer; /* Length of 'x_longer' */
1862
1863 char * x_shorter; /* We also transform a substring of 'longer' */
1864 Size_t x_len_shorter;
1865
a4a439fb 1866 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1867 * are interested only in its length). It is used because it has
1868 * the intelligence to handle all cases, but to work, it needs some
1869 * values of 'm' and 'b' to get it started. For the purposes of
1870 * this calculation we use a very conservative estimate of 'm' and
1871 * 'b'. This assumes a weight can be multiple bytes, enough to
1872 * hold any UV on the platform, and there are 5 levels, 4 weight
1873 * bytes, and a trailing NUL. */
1874 PL_collxfrm_base = 5;
1875 PL_collxfrm_mult = 5 * sizeof(UV);
1876
1877 /* Find out how long the transformation really is */
a4a439fb
KW
1878 x_longer = _mem_collxfrm(longer,
1879 sizeof(longer) - 1,
1880 &x_len_longer,
1881
1882 /* We avoid converting to UTF-8 in the
1883 * called function by telling it the
1884 * string is in UTF-8 if the locale is a
1885 * UTF-8 one. Since the string passed
1886 * here is invariant under UTF-8, we can
1887 * claim it's UTF-8 even though it isn't.
1888 * */
1889 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1890 Safefree(x_longer);
1891
1892 /* Find out how long the transformation of a substring of 'longer'
1893 * is. Together the lengths of these transformations are
1894 * sufficient to calculate 'm' and 'b'. The substring is all of
1895 * 'longer' except the first character. This minimizes the chances
1896 * of being swayed by outliers */
a4a439fb 1897 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1898 sizeof(longer) - 2,
a4a439fb
KW
1899 &x_len_shorter,
1900 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1901 Safefree(x_shorter);
1902
1903 /* If the results are nonsensical for this simple test, the whole
1904 * locale definition is suspect. Mark it so that locale collation
1905 * is not active at all for it. XXX Should we warn? */
1906 if ( x_len_shorter == 0
1907 || x_len_longer == 0
1908 || x_len_shorter >= x_len_longer)
1909 {
1910 PL_collxfrm_mult = 0;
1911 PL_collxfrm_base = 0;
1912 }
1913 else {
1914 SSize_t base; /* Temporary */
1915
1916 /* We have both: m * strlen(longer) + b = x_len_longer
1917 * m * strlen(shorter) + b = x_len_shorter;
1918 * subtracting yields:
1919 * m * (strlen(longer) - strlen(shorter))
1920 * = x_len_longer - x_len_shorter
1921 * But we have set things up so that 'shorter' is 1 byte smaller
1922 * than 'longer'. Hence:
1923 * m = x_len_longer - x_len_shorter
1924 *
1925 * But if something went wrong, make sure the multiplier is at
1926 * least 1.
1927 */
1928 if (x_len_longer > x_len_shorter) {
1929 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1930 }
1931 else {
1932 PL_collxfrm_mult = 1;
1933 }
1934
1935 /* mx + b = len
1936 * so: b = len - mx
1937 * but in case something has gone wrong, make sure it is
1938 * non-negative */
1939 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1940 if (base < 0) {
1941 base = 0;
1942 }
1943
1944 /* Add 1 for the trailing NUL */
1945 PL_collxfrm_base = base + 1;
1946 }
58eebef2 1947
7d4bcc4a
KW
1948# ifdef DEBUGGING
1949
58eebef2
KW
1950 if (DEBUG_L_TEST || debug_initialization) {
1951 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1952 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1953 "x_len_longer=%zu,"
1954 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1955 __FILE__, __LINE__,
1956 PL_in_utf8_COLLATE_locale,
1957 x_len_shorter, x_len_longer,
1958 PL_collxfrm_mult, PL_collxfrm_base);
1959 }
7d4bcc4a
KW
1960# endif
1961
98994639
HS
1962 }
1963 }
1964
1965#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1966
98994639
HS
1967}
1968
d2b24094 1969#ifdef WIN32
b8cc575c 1970
a4f00dcc 1971STATIC char *
b8cc575c 1972S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1973{
1974 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1975 * difference between the two unless the input locale is "", which normally
1976 * means on Windows to get the machine default, which is set via the
1977 * computer's "Regional and Language Options" (or its current equivalent).
1978 * In POSIX, it instead means to find the locale from the user's
1979 * environment. This routine changes the Windows behavior to first look in
1980 * the environment, and, if anything is found, use that instead of going to
1981 * the machine default. If there is no environment override, the machine
1982 * default is used, by calling the real setlocale() with "".
1983 *
1984 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1985 * use the particular category's variable if set; otherwise to use the LANG
1986 * variable. */
b385bb4d 1987
175c4cf9 1988 bool override_LC_ALL = FALSE;
89f7b9aa 1989 char * result;
e5f10d49 1990 unsigned int i;
89f7b9aa 1991
b385bb4d 1992 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1993
1994# ifdef LC_ALL
1995
b385bb4d
KW
1996 locale = PerlEnv_getenv("LC_ALL");
1997 if (! locale) {
e5f10d49
KW
1998 if (category == LC_ALL) {
1999 override_LC_ALL = TRUE;
2000 }
2001 else {
7d4bcc4a
KW
2002
2003# endif
7d4bcc4a 2004
e5f10d49
KW
2005 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2006 if (category == categories[i]) {
2007 locale = PerlEnv_getenv(category_names[i]);
2008 goto found_locale;
2009 }
2010 }
7d4bcc4a 2011
b385bb4d 2012 locale = PerlEnv_getenv("LANG");
481465ea 2013 if (! locale) {
b385bb4d
KW
2014 locale = "";
2015 }
e5f10d49
KW
2016
2017 found_locale: ;
7d4bcc4a
KW
2018
2019# ifdef LC_ALL
2020
e5f10d49 2021 }
b385bb4d 2022 }
7d4bcc4a
KW
2023
2024# endif
2025
b385bb4d
KW
2026 }
2027
89f7b9aa 2028 result = setlocale(category, locale);
48015184
KW
2029 DEBUG_L(STMT_START {
2030 dSAVE_ERRNO;
2031 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2032 setlocale_debug_string(category, locale, result));
2033 RESTORE_ERRNO;
2034 } STMT_END);
89f7b9aa 2035
481465ea 2036 if (! override_LC_ALL) {
89f7b9aa
KW
2037 return result;
2038 }
2039
dfd77d7a 2040 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2041 * LANG variable or the system default if there is no LANG. But these have
2042 * lower priority than the other LC_foo variables, so override it for each
2043 * one that is set. (If they are set to "", it means to use the same thing
2044 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2045
948523db 2046 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2047 result = PerlEnv_getenv(category_names[i]);
2048 if (result && strNE(result, "")) {
2049 setlocale(categories[i], result);
2050 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2051 __FILE__, __LINE__,
2052 setlocale_debug_string(categories[i], result, "not captured")));
2053 }
89f7b9aa 2054 }
7d4bcc4a 2055
bbc98134 2056 result = setlocale(LC_ALL, NULL);
48015184
KW
2057 DEBUG_L(STMT_START {
2058 dSAVE_ERRNO;
2059 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2060 __FILE__, __LINE__,
48015184
KW
2061 setlocale_debug_string(LC_ALL, NULL, result));
2062 RESTORE_ERRNO;
2063 } STMT_END);
89f7b9aa 2064
bbc98134 2065 return result;
b385bb4d
KW
2066}
2067
2068#endif
2069
9aac5db8
KW
2070/*
2071
2072=head1 Locale-related functions and macros
2073
2074=for apidoc Perl_setlocale
2075
2076This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2077taking the same parameters, and returning the same information, except that it
2078returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2079perl keeps that locale category as C<C>, changing it briefly during the
2080operations where the underlying one is required.
2081
e9bc6d6b 2082Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2083declared to return S<C<const char *>>, whereas the system setlocale omits the
2084C<const>. (If it were being written today, plain setlocale would be declared
2085const, since it is illegal to change the information it returns; doing so leads
2086to segfaults.)
2087
e9bc6d6b
KW
2088Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2089C<setlocale> can be completely ineffective on some platforms under some
2090configurations.
2091
9aac5db8 2092C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2093where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2094the system C<setlocale()> is ineffective, returning the wrong information, and
2095failing to actually change the locale. C<Perl_setlocale>, however works
2096properly in all circumstances.
9aac5db8
KW
2097
2098The return points to a per-thread static buffer, which is overwritten the next
2099time C<Perl_setlocale> is called from the same thread.
2100
2101=cut
2102
2103*/
2104
2105const char *
2106Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2107{
2108 /* This wraps POSIX::setlocale() */
2109
9aac5db8
KW
2110 const char * retval;
2111 const char * newlocale;
2112 dSAVEDERRNO;
2113 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2114 dTHX;
2115
a4f00dcc
KW
2116#ifdef USE_LOCALE_NUMERIC
2117
291a84fb
KW
2118 /* A NULL locale means only query what the current one is. We have the
2119 * LC_NUMERIC name saved, because we are normally switched into the C
2120 * locale for it. For an LC_ALL query, switch back to get the correct
2121 * results. All other categories don't require special handling */
a4f00dcc
KW
2122 if (locale == NULL) {
2123 if (category == LC_NUMERIC) {
9aac5db8
KW
2124
2125 /* We don't have to copy this return value, as it is a per-thread
2126 * variable, and won't change until a future setlocale */
2127 return PL_numeric_name;
a4f00dcc
KW
2128 }
2129
7d4bcc4a 2130# ifdef LC_ALL
a4f00dcc 2131
9aac5db8
KW
2132 else if (category == LC_ALL) {
2133 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2134 }
2135
7d4bcc4a 2136# endif
a4f00dcc
KW
2137
2138 }
2139
2140#endif
2141
33e5a354
KW
2142 retval = save_to_buffer(do_setlocale_r(category, locale),
2143 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2144 SAVE_ERRNO;
2145
2146#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2147
2148 if (locale == NULL && category == LC_ALL) {
2149 RESTORE_LC_NUMERIC();
2150 }
2151
2152#endif
a4f00dcc
KW
2153
2154 DEBUG_L(PerlIO_printf(Perl_debug_log,
2155 "%s:%d: %s\n", __FILE__, __LINE__,
2156 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2157
9aac5db8
KW
2158 RESTORE_ERRNO;
2159
2160 if (! retval) {
a4f00dcc
KW
2161 return NULL;
2162 }
2163
9aac5db8 2164 /* If locale == NULL, we are just querying the state */
a4f00dcc 2165 if (locale == NULL) {
a4f00dcc
KW
2166 return retval;
2167 }
a4f00dcc 2168
1159483a
KW
2169 /* Now that have switched locales, we have to update our records to
2170 * correspond. */
a4f00dcc 2171
1159483a 2172 switch (category) {
a4f00dcc 2173
1159483a 2174#ifdef USE_LOCALE_CTYPE
a4f00dcc 2175
1159483a
KW
2176 case LC_CTYPE:
2177 new_ctype(retval);
2178 break;
a4f00dcc 2179
1159483a 2180#endif
a4f00dcc
KW
2181#ifdef USE_LOCALE_COLLATE
2182
1159483a
KW
2183 case LC_COLLATE:
2184 new_collate(retval);
2185 break;
a4f00dcc 2186
1159483a 2187#endif
a4f00dcc
KW
2188#ifdef USE_LOCALE_NUMERIC
2189
1159483a
KW
2190 case LC_NUMERIC:
2191 new_numeric(retval);
2192 break;
a4f00dcc 2193
1159483a
KW
2194#endif
2195#ifdef LC_ALL
a4f00dcc 2196
1159483a 2197 case LC_ALL:
a4f00dcc 2198
1159483a
KW
2199 /* LC_ALL updates all the things we care about. The values may not
2200 * be the same as 'retval', as the locale "" may have set things
2201 * individually */
a4f00dcc 2202
1159483a 2203# ifdef USE_LOCALE_CTYPE
a4f00dcc 2204
1159483a
KW
2205 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2206 new_ctype(newlocale);
a4f00dcc 2207
1159483a
KW
2208# endif /* USE_LOCALE_CTYPE */
2209# ifdef USE_LOCALE_COLLATE
2210
2211 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2212 new_collate(newlocale);
a4f00dcc 2213
7d4bcc4a 2214# endif
1159483a 2215# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2216
1159483a
KW
2217 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2218 new_numeric(newlocale);
a4f00dcc 2219
1159483a
KW
2220# endif /* USE_LOCALE_NUMERIC */
2221#endif /* LC_ALL */
a4f00dcc 2222
1159483a
KW
2223 default:
2224 break;
a4f00dcc
KW
2225 }
2226
2227 return retval;
2228
f7416781
KW
2229}
2230
2231PERL_STATIC_INLINE const char *
2232S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2233{
2234 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2235 * growing it if necessary */
2236
0b6697c5 2237 Size_t string_size;
f7416781
KW
2238
2239 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2240
0b6697c5
KW
2241 if (! string) {
2242 return NULL;
2243 }
2244
2245 string_size = strlen(string) + offset + 1;
2246
f7416781
KW
2247 if (*buf_size == 0) {
2248 Newx(*buf, string_size, char);
2249 *buf_size = string_size;
2250 }
2251 else if (string_size > *buf_size) {
2252 Renew(*buf, string_size, char);
2253 *buf_size = string_size;
2254 }
2255
2256 Copy(string, *buf + offset, string_size - offset, char);
2257 return *buf;
2258}
2259
2260/*
2261
f7416781
KW
2262=for apidoc Perl_langinfo
2263
ce181bf6 2264This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2265taking the same C<item> parameter values, and returning the same information.
2266But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2267of Perl's locale handling from your code, and can be used on systems that lack
2268a native C<nl_langinfo>.
2269
2270Expanding on these:
2271
2272=over
2273
2274=item *
2275
ce181bf6
KW
2276The reason it isn't quite a drop-in replacement is actually an advantage. The
2277only difference is that it returns S<C<const char *>>, whereas plain
2278C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2279forbidden to write into the buffer. By declaring this C<const>, the compiler
2280enforces this restriction, so if it is violated, you know at compilation time,
2281rather than getting segfaults at runtime.
2282
2283=item *
2284
69e2275e 2285It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2286without you having to write extra code. The reason for the extra code would be
2287because these are from the C<LC_NUMERIC> locale category, which is normally
2288kept set to the C locale by Perl, no matter what the underlying locale is
2289supposed to be, and so to get the expected results, you have to temporarily
ce181bf6
KW
2290toggle into the underlying locale, and later toggle back. (You could use plain
2291C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2292then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2293C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2294radix (decimal point) character to be a dot.)
2295
2296=item *
2297
2298The system function it replaces can have its static return buffer trashed,
2299not only by a subesequent call to that function, but by a C<freelocale>,
2300C<setlocale>, or other locale change. The returned buffer of this function is
2301not changed until the next call to it, so the buffer is never in a trashed
2302state.
f7416781
KW
2303
2304=item *
2305
ce181bf6
KW
2306Its return buffer is per-thread, so it also is never overwritten by a call to
2307this function from another thread; unlike the function it replaces.
2308
2309=item *
2310
2311But most importantly, it works on systems that don't have C<nl_langinfo>, such
2312as Windows, hence makes your code more portable. Of the fifty-some possible
2313items specified by the POSIX 2008 standard,
f7416781 2314L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
ce181bf6
KW
2315only two are completely unimplemented (though the loss of one of these is
2316significant). It uses various techniques to recover the other items, including
2317calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2318in C89, so should be always be available. Later C<strftime()> versions have
2319additional capabilities; C<""> is returned for those not available on your
2320system.
2321
2322It is important to note that when called with an item that is recovered by
2323using C<localeconv>, the buffer from any previous explicit call to
2324C<localeconv> will be overwritten. This means you must save that buffer's
2325contents if you need to access them after a call to this function.
9f0bc911 2326
f7416781
KW
2327The details for those items which may differ from what this emulation returns
2328and what a native C<nl_langinfo()> would return are:
2329
2330=over
2331
2332=item C<CODESET>
2333
2334=item C<ERA>
2335
2336Unimplemented, so returns C<"">.
2337
2338=item C<YESEXPR>
2339
c1566110
KW
2340=item C<YESSTR>
2341
f7416781
KW
2342=item C<NOEXPR>
2343
c1566110
KW
2344=item C<NOSTR>
2345
2346Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
ce181bf6
KW
2347removed from POSIX 2008, and are retained here for backwards compatibility.
2348Your platform's C<nl_langinfo> may not support them.
f7416781
KW
2349
2350=item C<D_FMT>
2351
2352Always evaluates to C<%x>, the locale's appropriate date representation.
2353
2354=item C<T_FMT>
2355
2356Always evaluates to C<%X>, the locale's appropriate time representation.
2357
2358=item C<D_T_FMT>
2359
2360Always evaluates to C<%c>, the locale's appropriate date and time
2361representation.
2362
2363=item C<CRNCYSTR>
2364
2365The return may be incorrect for those rare locales where the currency symbol
2366replaces the radix character.
2367Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2368to work differently.
2369
2370=item C<ALT_DIGITS>
2371
2372Currently this gives the same results as Linux does.
2373Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2374to work differently.
2375
2376=item C<ERA_D_FMT>
2377
2378=item C<ERA_T_FMT>
2379
2380=item C<ERA_D_T_FMT>
2381
2382=item C<T_FMT_AMPM>
2383
2384These are derived by using C<strftime()>, and not all versions of that function
2385know about them. C<""> is returned for these on such systems.
2386
2387=back
2388
ce181bf6
KW
2389=back
2390
f7416781
KW
2391When using C<Perl_langinfo> on systems that don't have a native
2392C<nl_langinfo()>, you must
2393
2394 #include "perl_langinfo.h"
2395
2396before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2397C<#include> with this one. (Doing it this way keeps out the symbols that plain
2398C<langinfo.h> imports into the namespace for code that doesn't need it.)
2399
f7416781
KW
2400The original impetus for C<Perl_langinfo()> was so that code that needs to
2401find out the current currency symbol, floating point radix character, or digit
2402grouping separator can use, on all systems, the simpler and more
2403thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2404pain to make thread-friendly. For other fields returned by C<localeconv>, it
2405is better to use the methods given in L<perlcall> to call
2406L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2407
2408=cut
2409
2410*/
2411
2412const char *
2413#ifdef HAS_NL_LANGINFO
2414Perl_langinfo(const nl_item item)
2415#else
2416Perl_langinfo(const int item)
2417#endif
2418{
f61748ac
KW
2419 return my_nl_langinfo(item, TRUE);
2420}
2421
59a15af0 2422STATIC const char *
f61748ac
KW
2423#ifdef HAS_NL_LANGINFO
2424S_my_nl_langinfo(const nl_item item, bool toggle)
2425#else
2426S_my_nl_langinfo(const int item, bool toggle)
2427#endif
2428{
ae74815b 2429 dTHX;
0b6697c5 2430 const char * retval;
f7416781 2431
5a854ab3
KW
2432 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2433 * two items, and only if not already there */
4e6826bf 2434 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3
KW
2435 || PL_numeric_underlying))
2436 {
2437 toggle = FALSE;
2438 }
2439
ab340fff 2440#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d
KW
2441# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2442 || ! defined(HAS_POSIX_2008_LOCALE)
f7416781 2443
ab340fff 2444 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2445 * for those items dependent on it. This must be copied to a buffer before
2446 * switching back, as some systems destroy the buffer when setlocale() is
2447 * called */
f7416781 2448
038d3702
KW
2449 {
2450 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2451
b2fee59c 2452 if (toggle) {
038d3702 2453 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2454 }
f7416781 2455
5acc4454
KW
2456 LOCALE_LOCK; /* Prevent interference from another thread executing
2457 this code section (the only call to nl_langinfo in
2458 the core) */
2459
ee90eb2d 2460
628ff75a
KW
2461 /* Copy to a per-thread buffer, which is also one that won't be
2462 * destroyed by a subsequent setlocale(), such as the
2463 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2464 retval = save_to_buffer(nl_langinfo(item),
2465 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2466
5acc4454
KW
2467 LOCALE_UNLOCK;
2468
b2fee59c 2469 if (toggle) {
038d3702 2470 RESTORE_LC_NUMERIC();
b2fee59c 2471 }
f7416781
KW
2472 }
2473
ab340fff
KW
2474# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2475
5a854ab3 2476 {
b2fee59c
KW
2477 bool do_free = FALSE;
2478 locale_t cur = uselocale((locale_t) 0);
ab340fff 2479
b2fee59c
KW
2480 if (cur == LC_GLOBAL_LOCALE) {
2481 cur = duplocale(LC_GLOBAL_LOCALE);
2482 do_free = TRUE;
2483 }
ab340fff 2484
b2fee59c 2485 if (toggle) {
e1aa2579
KW
2486 if (PL_underlying_numeric_obj) {
2487 cur = PL_underlying_numeric_obj;
2488 }
2489 else {
2490 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2491 do_free = TRUE;
2492 }
b2fee59c 2493 }
ab340fff 2494
628ff75a
KW
2495 /* We have to save it to a buffer, because the freelocale() just below
2496 * can invalidate the internal one */
0b6697c5
KW
2497 retval = save_to_buffer(nl_langinfo_l(item, cur),
2498 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2499
b2fee59c
KW
2500 if (do_free) {
2501 freelocale(cur);
2502 }
5a854ab3 2503 }
ab340fff 2504
c1566110
KW
2505# endif
2506
0b6697c5 2507 if (strEQ(retval, "")) {
4e6826bf 2508 if (item == YESSTR) {
c1566110
KW
2509 return "yes";
2510 }
4e6826bf 2511 if (item == NOSTR) {
c1566110
KW
2512 return "no";
2513 }
2514 }
2515
0b6697c5 2516 return retval;
ab340fff 2517
f7416781 2518#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2519
2520 {
2521
f7416781
KW
2522# ifdef HAS_LOCALECONV
2523
43dd6b15 2524 const struct lconv* lc;
628ff75a 2525 const char * temp;
038d3702 2526 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
2527
2528# endif
2529# ifdef HAS_STRFTIME
2530
43dd6b15
KW
2531 struct tm tm;
2532 bool return_format = FALSE; /* Return the %format, not the value */
2533 const char * format;
f7416781
KW
2534
2535# endif
2536
43dd6b15
KW
2537 /* We copy the results to a per-thread buffer, even if not
2538 * multi-threaded. This is in part to simplify this code, and partly
2539 * because we need a buffer anyway for strftime(), and partly because a
2540 * call of localeconv() could otherwise wipe out the buffer, and the
2541 * programmer would not be expecting this, as this is a nl_langinfo()
2542 * substitute after all, so s/he might be thinking their localeconv()
2543 * is safe until another localeconv() call. */
f7416781 2544
43dd6b15
KW
2545 switch (item) {
2546 Size_t len;
f7416781 2547
43dd6b15 2548 /* These 2 are unimplemented */
4e6826bf
KW
2549 case CODESET:
2550 case ERA: /* For use with strftime() %E modifier */
f7416781 2551
43dd6b15
KW
2552 default:
2553 return "";
f7416781 2554
43dd6b15 2555 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2556 case YESEXPR: return "^[+1yY]";
2557 case YESSTR: return "yes";
2558 case NOEXPR: return "^[-0nN]";
2559 case NOSTR: return "no";
2560
f7416781
KW
2561
2562# ifdef HAS_LOCALECONV
2563
4e6826bf 2564 case CRNCYSTR:
f7416781 2565
43dd6b15
KW
2566 /* We don't bother with localeconv_l() because any system that
2567 * has it is likely to also have nl_langinfo() */
291a84fb 2568
5acc4454
KW
2569 LOCALE_LOCK; /* Prevent interference with other threads
2570 using localeconv() */
2571
43dd6b15
KW
2572 lc = localeconv();
2573 if ( ! lc
2574 || ! lc->currency_symbol
2575 || strEQ("", lc->currency_symbol))
2576 {
2577 LOCALE_UNLOCK;
2578 return "";
2579 }
f7416781 2580
43dd6b15 2581 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2582 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2583 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2584 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2585 { /* khw couldn't figure out how the localedef specifications
2586 would show that the $ should replace the radix; this is
2587 just a guess as to how it might work.*/
a34edee3 2588 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2589 }
2590 else if (lc->p_cs_precedes) {
a34edee3 2591 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2592 }
2593 else {
a34edee3 2594 PL_langinfo_buf[0] = '+';
43dd6b15 2595 }
f7416781 2596
43dd6b15
KW
2597 LOCALE_UNLOCK;
2598 break;
f7416781 2599
4e6826bf
KW
2600 case RADIXCHAR:
2601 case THOUSEP:
f7416781 2602
43dd6b15 2603 if (toggle) {
038d3702 2604 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2605 }
f7416781 2606
5acc4454
KW
2607 LOCALE_LOCK; /* Prevent interference with other threads
2608 using localeconv() */
2609
43dd6b15
KW
2610 lc = localeconv();
2611 if (! lc) {
628ff75a 2612 temp = "";
33394adc 2613 }
43dd6b15 2614 else {
4e6826bf 2615 temp = (item == RADIXCHAR)
43dd6b15
KW
2616 ? lc->decimal_point
2617 : lc->thousands_sep;
628ff75a
KW
2618 if (! temp) {
2619 temp = "";
43dd6b15
KW
2620 }
2621 }
f7416781 2622
0b6697c5
KW
2623 retval = save_to_buffer(temp, &PL_langinfo_buf,
2624 &PL_langinfo_bufsize, 0);
f7416781 2625
5acc4454
KW
2626 LOCALE_UNLOCK;
2627
43dd6b15 2628 if (toggle) {
038d3702 2629 RESTORE_LC_NUMERIC();
43dd6b15 2630 }
f7416781 2631
43dd6b15 2632 break;
f7416781
KW
2633
2634# endif
2635# ifdef HAS_STRFTIME
2636
43dd6b15
KW
2637 /* These are defined by C89, so we assume that strftime supports
2638 * them, and so are returned unconditionally; they may not be what
2639 * the locale actually says, but should give good enough results
2640 * for someone using them as formats (as opposed to trying to parse
2641 * them to figure out what the locale says). The other format
2642 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2643 case D_FMT: return "%x";
2644 case T_FMT: return "%X";
2645 case D_T_FMT: return "%c";
43dd6b15
KW
2646
2647 /* These formats are only available in later strfmtime's */
4e6826bf 2648 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2649
2650 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2651 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2652 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2653 case ALT_DIGITS:
2654 case AM_STR: case PM_STR:
2655 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2656 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2657 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2658 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2659 case DAY_5: case DAY_6: case DAY_7:
2660 case MON_1: case MON_2: case MON_3: case MON_4:
2661 case MON_5: case MON_6: case MON_7: case MON_8:
2662 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2663
2664 LOCALE_LOCK;
2665
2666 init_tm(&tm); /* Precaution against core dumps */
2667 tm.tm_sec = 30;
2668 tm.tm_min = 30;
2669 tm.tm_hour = 6;
2670 tm.tm_year = 2017 - 1900;
2671 tm.tm_wday = 0;
2672 tm.tm_mon = 0;
2673 switch (item) {
2674 default:
2675 LOCALE_UNLOCK;
2676 Perl_croak(aTHX_
2677 "panic: %s: %d: switch case: %d problem",
2678 __FILE__, __LINE__, item);
2679 NOT_REACHED; /* NOTREACHED */
2680
4e6826bf
KW
2681 case PM_STR: tm.tm_hour = 18;
2682 case AM_STR:
43dd6b15
KW
2683 format = "%p";
2684 break;
2685
4e6826bf
KW
2686 case ABDAY_7: tm.tm_wday++;
2687 case ABDAY_6: tm.tm_wday++;
2688 case ABDAY_5: tm.tm_wday++;
2689 case ABDAY_4: tm.tm_wday++;
2690 case ABDAY_3: tm.tm_wday++;
2691 case ABDAY_2: tm.tm_wday++;
2692 case ABDAY_1:
43dd6b15
KW
2693 format = "%a";
2694 break;
2695
4e6826bf
KW
2696 case DAY_7: tm.tm_wday++;
2697 case DAY_6: tm.tm_wday++;
2698 case DAY_5: tm.tm_wday++;
2699 case DAY_4: tm.tm_wday++;
2700 case DAY_3: tm.tm_wday++;
2701 case DAY_2: tm.tm_wday++;
2702 case DAY_1:
43dd6b15
KW
2703 format = "%A";
2704 break;
2705
4e6826bf
KW
2706 case ABMON_12: tm.tm_mon++;
2707 case ABMON_11: tm.tm_mon++;
2708 case ABMON_10: tm.tm_mon++;
2709 case ABMON_9: tm.tm_mon++;
2710 case ABMON_8: tm.tm_mon++;
2711 case ABMON_7: tm.tm_mon++;
2712 case ABMON_6: tm.tm_mon++;
2713 case ABMON_5: tm.tm_mon++;
2714 case ABMON_4: tm.tm_mon++;
2715 case ABMON_3: tm.tm_mon++;
2716 case ABMON_2: tm.tm_mon++;
2717 case ABMON_1:
43dd6b15
KW
2718 format = "%b";
2719 break;
2720
4e6826bf
KW
2721 case MON_12: tm.tm_mon++;
2722 case MON_11: tm.tm_mon++;
2723 case MON_10: tm.tm_mon++;
2724 case MON_9: tm.tm_mon++;
2725 case MON_8: tm.tm_mon++;
2726 case MON_7: tm.tm_mon++;
2727 case MON_6: tm.tm_mon++;
2728 case MON_5: tm.tm_mon++;
2729 case MON_4: tm.tm_mon++;
2730 case MON_3: tm.tm_mon++;
2731 case MON_2: tm.tm_mon++;
2732 case MON_1:
43dd6b15
KW
2733 format = "%B";
2734 break;
2735
4e6826bf 2736 case T_FMT_AMPM:
43dd6b15
KW
2737 format = "%r";
2738 return_format = TRUE;
2739 break;
2740
4e6826bf 2741 case ERA_D_FMT:
43dd6b15
KW
2742 format = "%Ex";
2743 return_format = TRUE;
2744 break;
2745
4e6826bf 2746 case ERA_T_FMT:
43dd6b15
KW
2747 format = "%EX";
2748 return_format = TRUE;
2749 break;
2750
4e6826bf 2751 case ERA_D_T_FMT:
43dd6b15
KW
2752 format = "%Ec";
2753 return_format = TRUE;
2754 break;
2755
4e6826bf 2756 case ALT_DIGITS:
43dd6b15
KW
2757 tm.tm_wday = 0;
2758 format = "%Ow"; /* Find the alternate digit for 0 */
2759 break;
2760 }
f7416781 2761
43dd6b15
KW
2762 /* We can't use my_strftime() because it doesn't look at
2763 * tm_wday */
2764 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2765 format, &tm))
2766 {
2767 /* A zero return means one of:
2768 * a) there wasn't enough space in PL_langinfo_buf
2769 * b) the format, like a plain %p, returns empty
2770 * c) it was an illegal format, though some
2771 * implementations of strftime will just return the
2772 * illegal format as a plain character sequence.
2773 *
2774 * To quickly test for case 'b)', try again but precede
2775 * the format with a plain character. If that result is
2776 * still empty, the problem is either 'a)' or 'c)' */
2777
2778 Size_t format_size = strlen(format) + 1;
2779 Size_t mod_size = format_size + 1;
2780 char * mod_format;
2781 char * temp_result;
2782
2783 Newx(mod_format, mod_size, char);
2784 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2785 *mod_format = ' ';
43dd6b15
KW
2786 my_strlcpy(mod_format + 1, format, mod_size);
2787 len = strftime(temp_result,
2788 PL_langinfo_bufsize,
2789 mod_format, &tm);
2790 Safefree(mod_format);
2791 Safefree(temp_result);
2792
2793 /* If 'len' is non-zero, it means that we had a case like
2794 * %p which means the current locale doesn't use a.m. or
2795 * p.m., and that is valid */
2796 if (len == 0) {
2797
2798 /* Here, still didn't work. If we get well beyond a
2799 * reasonable size, bail out to prevent an infinite
2800 * loop. */
2801
2802 if (PL_langinfo_bufsize > 100 * format_size) {
2803 *PL_langinfo_buf = '\0';
2804 }
2805 else {
2806 /* Double the buffer size to retry; Add 1 in case
2807 * original was 0, so we aren't stuck at 0. */
2808 PL_langinfo_bufsize *= 2;
2809 PL_langinfo_bufsize++;
2810 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2811 continue;
2812 }
2813 }
f7416781 2814
f7416781 2815 break;
43dd6b15 2816 }
f7416781 2817
43dd6b15
KW
2818 /* Here, we got a result.
2819 *
2820 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2821 * alternate format for wday 0. If the value is the same as
2822 * the normal 0, there isn't an alternate, so clear the buffer.
2823 * */
4e6826bf 2824 if ( item == ALT_DIGITS
43dd6b15
KW
2825 && strEQ(PL_langinfo_buf, "0"))
2826 {
2827 *PL_langinfo_buf = '\0';
2828 }
f7416781 2829
43dd6b15
KW
2830 /* ALT_DIGITS is problematic. Experiments on it showed that
2831 * strftime() did not always work properly when going from
2832 * alt-9 to alt-10. Only a few locales have this item defined,
2833 * and in all of them on Linux that khw was able to find,
2834 * nl_langinfo() merely returned the alt-0 character, possibly
2835 * doubled. Most Unicode digits are in blocks of 10
2836 * consecutive code points, so that is sufficient information
2837 * for those scripts, as we can infer alt-1, alt-2, .... But
2838 * for a Japanese locale, a CJK ideographic 0 is returned, and
2839 * the CJK digits are not in code point order, so you can't
2840 * really infer anything. The localedef for this locale did
2841 * specify the succeeding digits, so that strftime() works
2842 * properly on them, without needing to infer anything. But
2843 * the nl_langinfo() return did not give sufficient information
2844 * for the caller to understand what's going on. So until
2845 * there is evidence that it should work differently, this
2846 * returns the alt-0 string for ALT_DIGITS.
2847 *
2848 * wday was chosen because its range is all a single digit.
2849 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 2850
43dd6b15 2851 LOCALE_UNLOCK;
f7416781 2852
d1b150d9
KW
2853 retval = PL_langinfo_buf;
2854
43dd6b15
KW
2855 /* If to return the format, not the value, overwrite the buffer
2856 * with it. But some strftime()s will keep the original format
2857 * if illegal, so change those to "" */
2858 if (return_format) {
2859 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
2860 *PL_langinfo_buf = '\0';
2861 }
43dd6b15 2862 else {
0b6697c5
KW
2863 retval = save_to_buffer(format, &PL_langinfo_buf,
2864 &PL_langinfo_bufsize, 0);
f7416781
KW
2865 }
2866 }
2867
2868 break;
f7416781
KW
2869
2870# endif
2871
43dd6b15 2872 }
f7416781
KW
2873 }
2874
0b6697c5 2875 return retval;
f7416781
KW
2876
2877#endif
2878
a4f00dcc 2879}
b385bb4d 2880
98994639
HS
2881/*
2882 * Initialize locale awareness.
2883 */
2884int
2885Perl_init_i18nl10n(pTHX_ int printwarn)
2886{
0e92a118
KW
2887 /* printwarn is
2888 *
2889 * 0 if not to output warning when setup locale is bad
2890 * 1 if to output warning based on value of PERL_BADLANG
2891 * >1 if to output regardless of PERL_BADLANG
2892 *
2893 * returns
98994639 2894 * 1 = set ok or not applicable,
0e92a118
KW
2895 * 0 = fallback to a locale of lower priority
2896 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2897 *
2898 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2899 * set, debugging information is output.
2900 *
2901 * This looks more complicated than it is, mainly due to the #ifdefs.
2902 *
2903 * We try to set LC_ALL to the value determined by the environment. If
2904 * there is no LC_ALL on this platform, we try the individual categories we
2905 * know about. If this works, we are done.
2906 *
2907 * But if it doesn't work, we have to do something else. We search the
2908 * environment variables ourselves instead of relying on the system to do
2909 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2910 * think there is one), and the ultimate fallback "C". This is all done in
2911 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2912 * more complex. The 'trial_locales' array is initialized with just one
2913 * element; it causes the behavior described in the paragraph above this to
2914 * happen. If that fails, we add elements to 'trial_locales', and do extra
2915 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2916 *
2917 * On Ultrix, the locale MUST come from the environment, so there is
2918 * preliminary code to set it. I (khw) am not sure that it is necessary,
2919 * and that this couldn't be folded into the loop, but barring any real
2920 * platforms to test on, it's staying as-is
2921 *
2922 * A slight complication is that in embedded Perls, the locale may already
2923 * be set-up, and we don't want to get it from the normal environment
2924 * variables. This is handled by having a special environment variable
2925 * indicate we're in this situation. We simply set setlocale's 2nd
2926 * parameter to be a NULL instead of "". That indicates to setlocale that
2927 * it is not to change anything, but to return the current value,
2928 * effectively initializing perl's db to what the locale already is.
2929 *
2930 * We play the same trick with NULL if a LC_ALL succeeds. We call
2931 * setlocale() on the individual categores with NULL to get their existing
2932 * values for our db, instead of trying to change them.
2933 * */
98994639 2934
0e92a118
KW
2935 int ok = 1;
2936
7d4bcc4a
KW
2937#ifndef USE_LOCALE
2938
2939 PERL_UNUSED_ARG(printwarn);
2940
2941#else /* USE_LOCALE */
7d4bcc4a
KW
2942# ifdef __GLIBC__
2943
175c4cf9 2944 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2945
2946# endif
65ebb059 2947
ccd65d51
KW
2948 /* NULL uses the existing already set up locale */
2949 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2950 ? NULL
2951 : "";
c3fcd832
KW
2952 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2953 unsigned int trial_locales_count;
175c4cf9
KW
2954 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2955 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2956 bool setlocale_failure = FALSE;
65ebb059 2957 unsigned int i;
175c4cf9
KW
2958
2959 /* A later getenv() could zap this, so only use here */
2960 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2961
2962 const bool locwarn = (printwarn > 1
e5f10d49
KW
2963 || ( printwarn
2964 && ( ! bad_lang_use_once
22ff3130 2965 || (
e5f10d49
KW
2966 /* disallow with "" or "0" */
2967 *bad_lang_use_once
2968 && strNE("0", bad_lang_use_once)))));
ea92aad8 2969
291a84fb 2970 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2971 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2972
2973 /* current locale for given category; should have been copied so aren't
2974 * volatile */
8de4332b 2975 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2976
7d4bcc4a
KW
2977# ifdef WIN32
2978
6bce99ee
JH
2979 /* In some systems you can find out the system default locale
2980 * and use that as the fallback locale. */
7d4bcc4a
KW
2981# define SYSTEM_DEFAULT_LOCALE
2982# endif
2983# ifdef SYSTEM_DEFAULT_LOCALE
2984
65ebb059 2985 const char *system_default_locale = NULL;
98994639 2986
7d4bcc4a 2987# endif
948523db
KW
2988
2989# ifndef DEBUGGING
2990# define DEBUG_LOCALE_INIT(a,b,c)
2991# else
7d4bcc4a 2992
8298454c 2993 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2994
2995# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2996 STMT_START { \
2997 if (debug_initialization) { \
2998 PerlIO_printf(Perl_debug_log, \
2999 "%s:%d: %s\n", \
3000 __FILE__, __LINE__, \
a4f00dcc 3001 setlocale_debug_string(category, \
2fcc0ca9
KW
3002 locale, \
3003 result)); \
3004 } \
3005 } STMT_END
2fcc0ca9 3006
948523db
KW
3007/* Make sure the parallel arrays are properly set up */
3008# ifdef USE_LOCALE_NUMERIC
3009 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3010 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3011# ifdef USE_POSIX_2008_LOCALE
3012 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3013# endif
948523db
KW
3014# endif
3015# ifdef USE_LOCALE_CTYPE
3016 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3017 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3018# ifdef USE_POSIX_2008_LOCALE
3019 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3020# endif
948523db
KW
3021# endif
3022# ifdef USE_LOCALE_COLLATE
3023 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3024 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3025# ifdef USE_POSIX_2008_LOCALE
3026 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3027# endif
948523db
KW
3028# endif
3029# ifdef USE_LOCALE_TIME
3030 assert(categories[LC_TIME_INDEX] == LC_TIME);
3031 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3032# ifdef USE_POSIX_2008_LOCALE
3033 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3034# endif
948523db
KW
3035# endif
3036# ifdef USE_LOCALE_MESSAGES
3037 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3038 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3039# ifdef USE_POSIX_2008_LOCALE
3040 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3041# endif
948523db
KW
3042# endif
3043# ifdef USE_LOCALE_MONETARY
3044 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3045 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3046# ifdef USE_POSIX_2008_LOCALE
3047 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3048# endif
948523db 3049# endif
9821811f
KW
3050# ifdef USE_LOCALE_ADDRESS
3051 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3052 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3053# ifdef USE_POSIX_2008_LOCALE
3054 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3055# endif
9821811f
KW
3056# endif
3057# ifdef USE_LOCALE_IDENTIFICATION
3058 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3059 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3060# ifdef USE_POSIX_2008_LOCALE
3061 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3062# endif
9821811f
KW
3063# endif
3064# ifdef USE_LOCALE_MEASUREMENT
3065 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3066 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3067# ifdef USE_POSIX_2008_LOCALE
3068 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3069# endif
9821811f
KW
3070# endif
3071# ifdef USE_LOCALE_PAPER
3072 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3073 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3074# ifdef USE_POSIX_2008_LOCALE
3075 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3076# endif
9821811f
KW
3077# endif
3078# ifdef USE_LOCALE_TELEPHONE
3079 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3080 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3081# ifdef USE_POSIX_2008_LOCALE
3082 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3083# endif
9821811f 3084# endif
948523db
KW
3085# ifdef LC_ALL
3086 assert(categories[LC_ALL_INDEX] == LC_ALL);
3087 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3088 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3089# ifdef USE_POSIX_2008_LOCALE
3090 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3091# endif
948523db
KW
3092# endif
3093# endif /* DEBUGGING */
47280b20
KW
3094
3095 /* Initialize the cache of the program's UTF-8ness for the always known
3096 * locales C and POSIX */
3097 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3098 sizeof(PL_locale_utf8ness));
3099
e9bc6d6b
KW
3100# ifdef USE_THREAD_SAFE_LOCALE
3101# ifdef WIN32
3102
3103 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3104
3105# endif
3106# endif
3107# if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3108
3109 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3110 if (! PL_C_locale_obj) {
3111 Perl_croak_nocontext(
3112 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3113 }
3114 if (DEBUG_Lv_TEST || debug_initialization) {
3115 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3116 }
3117
3118# endif
3119
3ca88433
KW
3120 PL_numeric_radix_sv = newSVpvs(".");
3121
e9bc6d6b
KW
3122# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3123
3124 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3125 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3126
3127# endif
b56c4436 3128# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3129
3130 /*
3131 * Ultrix setlocale(..., "") fails if there are no environment
3132 * variables from which to get a locale name.
3133 */
3134
b56c4436
KW
3135# ifndef LC_ALL
3136# error Ultrix without LC_ALL not implemented
3137# else
7d4bcc4a 3138
b56c4436
KW
3139 {
3140 bool done = FALSE;
ea92aad8
KW
3141 if (lang) {
3142 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3143 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3144 if (sl_result[LC_ALL_INDEX])
3145 done = TRUE;
3146 else
e5f10d49 3147 setlocale_failure = TRUE;
ea92aad8
KW
3148 }
3149 if (! setlocale_failure) {
3150 const char * locale_param;
3151 for (i = 0; i < LC_ALL_INDEX; i++) {
3152 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3153 ? setlocale_init
3154 : NULL;
3155 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3156 if (! sl_result[i]) {
3157 setlocale_failure = TRUE;
3158 }
3159 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3160 }
c835d6be 3161 }
7d4bcc4a
KW
3162 }
3163
3164# endif /* LC_ALL */
e5f10d49 3165# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3166
65ebb059 3167 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3168 * the list. Normally the loop is executed just once. But if setting the
3169 * locale fails, inside the loop we add fallback trials to the array and so
3170 * will execute the loop multiple times */
c3fcd832
KW
3171 trial_locales[0] = setlocale_init;
3172 trial_locales_count = 1;
7d4bcc4a 3173
65ebb059
KW
3174 for (i= 0; i < trial_locales_count; i++) {
3175 const char * trial_locale = trial_locales[i];
3176
3177 if (i > 0) {
3178
3179 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3180 * when i==0, but I (khw) don't think that behavior makes much
3181 * sense */
3182 setlocale_failure = FALSE;
3183
7d4bcc4a 3184# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3185# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3186
65ebb059
KW
3187 /* On Windows machines, an entry of "" after the 0th means to use
3188 * the system default locale, which we now proceed to get. */
3189 if (strEQ(trial_locale, "")) {
3190 unsigned int j;
3191
3192 /* Note that this may change the locale, but we are going to do
3193 * that anyway just below */
837ce802 3194 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3195 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3196
7d4bcc4a 3197 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3198 * try */
3199 if (! system_default_locale) {
3200 goto next_iteration;
3201 }
3202 for (j = 0; j < trial_locales_count; j++) {
3203 if (strEQ(system_default_locale, trial_locales[j])) {
3204 goto next_iteration;
3205 }
3206 }
3207
3208 trial_locale = system_default_locale;
3209 }
ec0202b5
KW
3210# else
3211# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3212# endif
7d4bcc4a 3213# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3214
3215 } /* For i > 0 */
65ebb059 3216
7d4bcc4a
KW
3217# ifdef LC_ALL
3218
948523db
KW
3219 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3220 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3221 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3222 setlocale_failure = TRUE;
7cd8b568
KW
3223 }
3224 else {
3225 /* Since LC_ALL succeeded, it should have changed all the other
3226 * categories it can to its value; so we massage things so that the
3227 * setlocales below just return their category's current values.
3228 * This adequately handles the case in NetBSD where LC_COLLATE may
3229 * not be defined for a locale, and setting it individually will
7d4bcc4a 3230 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3231 * the POSIX locale. */
3232 trial_locale = NULL;
3233 }
7d4bcc4a
KW
3234
3235# endif /* LC_ALL */
98994639 3236
e5f10d49
KW
3237 if (! setlocale_failure) {
3238 unsigned int j;
3239 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3240 curlocales[j]
3241 = savepv(do_setlocale_r(categories[j], trial_locale));
3242 if (! curlocales[j]) {
3243 setlocale_failure = TRUE;
3244 }
3245 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3246 }
c835d6be 3247
e5f10d49
KW
3248 if (! setlocale_failure) { /* All succeeded */
3249 break; /* Exit trial_locales loop */
49c85077 3250 }
65ebb059 3251 }
98994639 3252
49c85077
KW
3253 /* Here, something failed; will need to try a fallback. */
3254 ok = 0;
65ebb059 3255
49c85077
KW
3256 if (i == 0) {
3257 unsigned int j;
98994639 3258
65ebb059 3259 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3260
3261# ifdef LC_ALL
98994639 3262
49c85077
KW
3263 PerlIO_printf(Perl_error_log,
3264 "perl: warning: Setting locale failed.\n");
98994639 3265
7d4bcc4a 3266# else /* !LC_ALL */
98994639 3267
49c85077
KW
3268 PerlIO_printf(Perl_error_log,
3269 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3270
e5f10d49
KW
3271 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3272 if (! curlocales[j]) {
3273 PerlIO_printf(Perl_error_log, category_names[j]);
3274 }
3275 else {
3276 Safefree(curlocales[j]);
3277 }
3278 }
7d4bcc4a 3279
7d4bcc4a 3280# endif /* LC_ALL */
98994639 3281
49c85077
KW
3282 PerlIO_printf(Perl_error_log,
3283 "perl: warning: Please check that your locale settings:\n");
98994639 3284
7d4bcc4a
KW
3285# ifdef __GLIBC__
3286
49c85077
KW
3287 PerlIO_printf(Perl_error_log,
3288 "\tLANGUAGE = %c%s%c,\n",
3289 language ? '"' : '(',
3290 language ? language : "unset",
3291 language ? '"' : ')');
7d4bcc4a 3292# endif
98994639 3293
49c85077
KW
3294 PerlIO_printf(Perl_error_log,
3295 "\tLC_ALL = %c%s%c,\n",
3296 lc_all ? '"' : '(',
3297 lc_all ? lc_all : "unset",
3298 lc_all ? '"' : ')');
98994639 3299
7d4bcc4a
KW
3300# if defined(USE_ENVIRON_ARRAY)
3301
49c85077 3302 {
cd999af9 3303 char **e;
d5e32b93
KW
3304
3305 /* Look through the environment for any variables of the
3306 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3307 * already handled above. These are assumed to be locale
3308 * settings. Output them and their values. */
cd999af9 3309 for (e = environ; *e; e++) {
d5e32b93
KW
3310 const STRLEN prefix_len = sizeof("LC_") - 1;
3311 STRLEN uppers_len;
3312
cd999af9 3313 if ( strBEGINs(*e, "LC_")
c8b388b0 3314 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3315 && (uppers_len = strspn(*e + prefix_len,
3316 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3317 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3318 {
3319 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3320 (int) (prefix_len + uppers_len), *e,
3321 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3322 }
3323 }
49c85077 3324 }
7d4bcc4a
KW
3325
3326# else
3327
49c85077
KW
3328 PerlIO_printf(Perl_error_log,
3329 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3330
3331# endif
98994639 3332
49c85077
KW
3333 PerlIO_printf(Perl_error_log,
3334 "\tLANG = %c%s%c\n",
3335 lang ? '"' : '(',
3336 lang ? lang : "unset",
3337 lang ? '"' : ')');
98994639 3338
49c85077
KW
3339 PerlIO_printf(Perl_error_log,
3340 " are supported and installed on your system.\n");
3341 }
98994639 3342
65ebb059 3343 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3344 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3345 * usually change the upper bound of the loop we are in.
3346 *
3347 * Since the system's default way of setting the locale has not
3348 * found one that works, We use Perl's defined ordering: LC_ALL,
3349 * LANG, and the C locale. We don't try the same locale twice, so
3350 * don't add to the list if already there. (On POSIX systems, the
3351 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3352 * but there's no harm done by doing it explicitly.
3353 *
3354 * Note that this tries the LC_ALL environment variable even on
3355 * systems which have no LC_ALL locale setting. This may or may
3356 * not have been originally intentional, but there's no real need
3357 * to change the behavior. */
65ebb059
KW
3358 if (lc_all) {
3359 for (j = 0; j < trial_locales_count; j++) {
3360 if (strEQ(lc_all, trial_locales[j])) {
3361 goto done_lc_all;
3362 }
3363 }
3364 trial_locales[trial_locales_count++] = lc_all;
3365 }
3366 done_lc_all:
98994639 3367
65ebb059
KW
3368 if (lang) {
3369 for (j = 0; j < trial_locales_count; j++) {
3370 if (strEQ(lang, trial_locales[j])) {
3371 goto done_lang;
3372 }
3373 }
3374 trial_locales[trial_locales_count++] = lang;
3375 }
3376 done_lang:
3377
7d4bcc4a
KW
3378# if defined(WIN32) && defined(LC_ALL)
3379
65ebb059
KW
3380 /* For Windows, we also try the system default locale before "C".
3381 * (If there exists a Windows without LC_ALL we skip this because
3382 * it gets too complicated. For those, the "C" is the next
3383 * fallback possibility). The "" is the same as the 0th element of
3384 * the array, but the code at the loop above knows to treat it
3385 * differently when not the 0th */
3386 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3387
3388# endif
65ebb059
KW
3389
3390 for (j = 0; j < trial_locales_count; j++) {
3391 if (strEQ("C", trial_locales[j])) {
3392 goto done_C;
3393 }
3394 }
3395 trial_locales[trial_locales_count++] = "C";
98994639 3396
65ebb059
KW
3397 done_C: ;
3398 } /* end of first time through the loop */
98994639 3399
7d4bcc4a
KW
3400# ifdef WIN32
3401
65ebb059 3402 next_iteration: ;
7d4bcc4a
KW
3403
3404# endif
65ebb059
KW
3405
3406 } /* end of looping through the trial locales */
3407
3408 if (ok < 1) { /* If we tried to fallback */
3409 const char* msg;
3410 if (! setlocale_failure) { /* fallback succeeded */
3411 msg = "Falling back to";
3412 }
3413 else { /* fallback failed */
e5f10d49 3414 unsigned int j;
98994639 3415
65ebb059
KW
3416 /* We dropped off the end of the loop, so have to decrement i to
3417 * get back to the value the last time through */
3418 i--;
98994639 3419
65ebb059
KW
3420 ok = -1;
3421 msg = "Failed to fall back to";
3422
3423 /* To continue, we should use whatever values we've got */
7d4bcc4a 3424
e5f10d49
KW
3425 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3426 Safefree(curlocales[j]);
3427 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3428 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3429 }
65ebb059
KW
3430 }
3431
3432 if (locwarn) {
3433 const char * description;
3434 const char * name = "";
3435 if (strEQ(trial_locales[i], "C")) {
3436 description = "the standard locale";
3437 name = "C";
3438 }
7d4bcc4a
KW
3439
3440# ifdef SYSTEM_DEFAULT_LOCALE
3441
65ebb059
KW
3442 else if (strEQ(trial_locales[i], "")) {
3443 description = "the system default locale";
3444 if (system_default_locale) {
3445 name = system_default_locale;
3446 }
3447 }
7d4bcc4a
KW
3448
3449# endif /* SYSTEM_DEFAULT_LOCALE */
3450
65ebb059
KW
3451 else {
3452 description = "a fallback locale";
3453 name = trial_locales[i];
3454 }
3455 if (name && strNE(name, "")) {
3456 PerlIO_printf(Perl_error_log,
3457 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3458 }
3459 else {
3460 PerlIO_printf(Perl_error_log,
3461 "perl: warning: %s %s.\n", msg, description);
3462 }
3463 }
3464 } /* End of tried to fallback */
98994639 3465
e5f10d49
KW
3466 /* Done with finding the locales; update our records */
3467
7d4bcc4a
KW
3468# ifdef USE_LOCALE_CTYPE
3469
948523db 3470 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3471
e5f10d49 3472# endif
7d4bcc4a
KW
3473# ifdef USE_LOCALE_COLLATE
3474
948523db 3475 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3476
e5f10d49 3477# endif
7d4bcc4a
KW
3478# ifdef USE_LOCALE_NUMERIC
3479
948523db 3480 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3481
3482# endif
3483
948523db 3484 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3485
e9bc6d6b 3486# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3487
3488 /* This caches whether each category's locale is UTF-8 or not. This
3489 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3490 * initialization time before any threads have started, but not later
3491 * unless thread-safe operations are used.
47280b20
KW
3492 * Caching means that if the program heeds our dictate not to change
3493 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3494 * it may get queried without having to change locales. If the
3495 * environment is such that all categories have the same locale, this
3496 * isn't needed, as the code will not change the locale; but this
3497 * handles the uncommon case where the environment has disparate
3498 * locales for the categories */
47280b20
KW
3499 (void) _is_cur_LC_category_utf8(categories[i]);
3500
3501# endif
3502
e5f10d49
KW
3503 Safefree(curlocales[i]);
3504 }
b310b053 3505
7d4bcc4a
KW
3506# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3507
49c85077 3508 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3509 * locale is UTF-8. The call to new_ctype() just above has already
3510 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3511 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3512 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3513 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3514 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3515
a05d7ebb 3516 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3517 This is an alternative to using the -C command line switch
3518 (the -C if present will override this). */
3519 {
dd374669 3520 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3521 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3522 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3523 PL_utf8cache = -1;
b310b053
JH
3524 }
3525
7d4bcc4a 3526# endif
7d4bcc4a
KW
3527# ifdef __GLIBC__
3528
175c4cf9 3529 Safefree(language);
7d4bcc4a
KW
3530
3531# endif
175c4cf9
KW
3532
3533 Safefree(lc_all);
3534 Safefree(lang);
3535
e3305790 3536#endif /* USE_LOCALE */
2fcc0ca9 3537#ifdef DEBUGGING
7d4bcc4a 3538
2fcc0ca9 3539 /* So won't continue to output stuff */
27cdc72e 3540 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3541
2fcc0ca9
KW
3542#endif
3543
98994639
HS
3544 return ok;
3545}
3546
98994639
HS
3547#ifdef USE_LOCALE_COLLATE
3548
a4a439fb 3549char *
a4a439fb
KW
3550Perl__mem_collxfrm(pTHX_ const char *input_string,
3551 STRLEN len, /* Length of 'input_string' */
3552 STRLEN *xlen, /* Set to length of returned string
3553 (not including the collation index
3554 prefix) */
3555 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3556 )
98994639 3557{
a4a439fb
KW
3558
3559 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3560 * differences. First, it handles embedded NULs. Second, it allocates a bit
3561 * more memory than needed for the transformed data itself. The real
55e5378d 3562 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3563 * the length of that, and doesn't include the collation index size.
3564 * Please see sv_collxfrm() to see how this is used. */
3565
55e5378d
KW
3566#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3567
6696cfa7
KW
3568 char * s = (char *) input_string;
3569 STRLEN s_strlen = strlen(input_string);
79f120c8 3570 char *xbuf = NULL;
55e5378d 3571 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3572 STRLEN length_in_chars;
c664130f 3573 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3574
a4a439fb
KW
3575 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3576
3577 /* Must be NUL-terminated */
3578 assert(*(input_string + len) == '\0');
7918f24d 3579
79f120c8
KW
3580 /* If this locale has defective collation, skip */
3581 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3582 DEBUG_L(PerlIO_printf(Perl_debug_log,
3583 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3584 goto bad;
3585 }
3586
6696cfa7
KW
3587 /* Replace any embedded NULs with the control that sorts before any others.
3588 * This will give as good as possible results on strings that don't
3589 * otherwise contain that character, but otherwise there may be
3590 * less-than-perfect results with that character and NUL. This is
fdc080f3 3591 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3592 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3593 NUL */
6696cfa7
KW
3594 char * e = s + len;
3595 char * sans_nuls;
fdc080f3 3596 STRLEN sans_nuls_len;
94762aa0 3597 int try_non_controls;
afc4976f
KW
3598 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3599 making sure 2nd byte is NUL.
3600 */
3601 STRLEN this_replacement_len;
3602
1e4c9676
KW
3603 /* If we don't know what non-NUL control character sorts lowest for
3604 * this locale, find it */
f28f4d2a 3605 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3606 int j;
afc4976f 3607 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3608 includes the collation index
3609 prefixed. */
3610
91c0e2e0 3611 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3612
3613 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3614 * NUL, in which case we instead look for any character. Controls
3615 * are preferred because collation order is, in general, context
3616 * sensitive, with adjoining characters affecting the order, and
3617 * controls are less likely to have such interactions, allowing the
3618 * NUL-replacement to stand on its own. (Another way to look at it
3619 * is to imagine what would happen if the NUL were replaced by a
3620 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3621 for (try_non_controls = 0;
3622 try_non_controls < 2;
3623 try_non_controls++)
3624 {
d4ff9586
KW
3625 /* Look through all legal code points (NUL isn't) */
3626 for (j = 1; j < 256; j++) {
3627 char * x; /* j's xfrm plus collation index */
3628 STRLEN x_len; /* length of 'x' */
3629 STRLEN trial_len = 1;
736a4fed 3630 char cur_source[] = { '\0', '\0' };
d4ff9586 3631
736a4fed
KW
3632 /* Skip non-controls the first time through the loop. The
3633 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3634 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3635 ? ! isCNTRL_L1(j)
3636 : ! isCNTRL_LC(j))
3637 {
d4ff9586 3638 continue;
6696cfa7 3639 }
6696cfa7 3640
736a4fed
KW
3641 /* Create a 1-char string of the current code point */
3642 cur_source[0] = (char) j;
3643
d4ff9586
KW
3644 /* Then transform it */
3645 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3646 0 /* The string is not in UTF-8 */);
6696cfa7 3647
1e4c9676 3648 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3649 * */
3650 if (! x) {
3651 continue;
3652 }
6696cfa7 3653
d4ff9586
KW
3654 /* If this character's transformation is lower than
3655 * the current lowest, this one becomes the lowest */
3656 if ( cur_min_x == NULL
3657 || strLT(x + COLLXFRM_HDR_LEN,
3658 cur_min_x + COLLXFRM_HDR_LEN))
3659 {
f28f4d2a 3660 PL_strxfrm_NUL_replacement = j;
d4ff9586 3661 cur_min_x = x;
d4ff9586
KW
3662 }
3663 else {
3664 Safefree(x);
3665 }
1e4c9676 3666 } /* end of loop through all 255 characters */
6696cfa7 3667
1e4c9676 3668 /* Stop looking if found */
94762aa0
KW
3669 if (cur_min_x) {
3670 break;
3671 }
3672
3673 /* Unlikely, but possible, if there aren't any controls that
3674 * work in the locale, repeat the loop, looking for any
3675 * character that works */
3676 DEBUG_L(PerlIO_printf(Perl_debug_log,
3677 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3678 } /* End of loop to try first the controls, then any char */
6696cfa7 3679
94762aa0
KW
3680 if (! cur_min_x) {
3681 DEBUG_L(PerlIO_printf(Perl_debug_log,
3682 "_mem_collxfrm: Couldn't find any character to replace"
3683 " embedded NULs in locale %s with", PL_collation_name));
3684 goto bad;
58eebef2
KW
3685 }
3686
94762aa0
KW
3687 DEBUG_L(PerlIO_printf(Perl_debug_log,
3688 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3689 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3690
6696cfa7 3691 Safefree(cur_min_x);
1e4c9676 3692 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3693
3694 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3695 * UTF8-ness of the original */
f28f4d2a
KW
3696 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3697 this_replacement_char[0] =
3698 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3699 this_replacement_char[1] =
3700 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3701 this_replacement_len = 2;
3702 }
3703 else {
f28f4d2a 3704 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3705 /* this_replacement_char[1] = '\0' was done at initialization */
3706 this_replacement_len = 1;
6696cfa7
KW
3707 }
3708
3709 /* The worst case length for the replaced string would be if every
3710 * character in it is NUL. Multiply that by the length of each
3711 * replacement, and allow for a trailing NUL */
afc4976f 3712 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3713 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3714 *sans_nuls = '\0';
3715
6696cfa7
KW
3716 /* Replace each NUL with the lowest collating control. Loop until have
3717 * exhausted all the NULs */
3718 while (s + s_strlen < e) {
6069d6c5 3719 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3720
3721 /* Do the actual replacement */
6069d6c5 3722 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
3723
3724 /* Move past the input NUL */
3725 s += s_strlen + 1;
3726 s_strlen = strlen(s);
3727 }
3728
3729 /* And add anything that trails the final NUL */
6069d6c5 3730 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3731
3732 /* Switch so below we transform this modified string */
3733 s = sans_nuls;
3734 len = strlen(s);
1e4c9676 3735 } /* End of replacing NULs */
6696cfa7 3736
a4a439fb
KW
3737 /* Make sure the UTF8ness of the string and locale match */
3738 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 3739 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
3740 const char * const t = s; /* Temporary so we can later find where the
3741 input was */
3742
3743 /* Here they don't match. Change the string's to be what the locale is
3744 * expecting */
3745
3746 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3747 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3748 utf8 = TRUE;
3749 }
3750 else { /* locale is not UTF-8; but input is; downgrade the input */
3751
3752 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3753
3754 /* If the downgrade was successful we are done, but if the input
3755 * contains things that require UTF-8 to represent, have to do
3756 * damage control ... */
3757 if (UNLIKELY(utf8)) {
3758
3759 /* What we do is construct a non-UTF-8 string with
3760 * 1) the characters representable by a single byte converted
3761 * to be so (if necessary);
3762 * 2) and the rest converted to collate the same as the
3763 * highest collating representable character. That makes
3764 * them collate at the end. This is similar to how we
3765 * handle embedded NULs, but we use the highest collating
3766 * code point instead of the smallest. Like the NUL case,
3767 * this isn't perfect, but is the best we can reasonably
3768 * do. Every above-255 code point will sort the same as
3769 * the highest-sorting 0-255 code point. If that code
3770 * point can combine in a sequence with some other code
3771 * points for weight calculations, us changing something to
3772 * be it can adversely affect the results. But in most
3773 * cases, it should work reasonably. And note that this is
3774 * really an illegal situation: using code points above 255
3775 * on a locale where only 0-255 are valid. If two strings
3776 * sort entirely equal, then the sort order for the
3777 * above-255 code points will be in code point order. */
3778
3779 utf8 = FALSE;
3780
3781 /* If we haven't calculated the code point with the maximum
3782 * collating order for this locale, do so now */
3783 if (! PL_strxfrm_max_cp) {
3784 int j;
3785
3786 /* The current transformed string that collates the
3787 * highest (except it also includes the prefixed collation
3788 * index. */
3789 char * cur_max_x = NULL;
3790
3791 /* Look through all legal code points (NUL isn't) */
3792 for (j = 1; j < 256; j++) {
3793 char * x;
3794 STRLEN x_len;
736a4fed 3795 char cur_source[] = { '\0', '\0' };
a4a439fb 3796
736a4fed
KW
3797 /* Create a 1-char string of the current code point */
3798 cur_source[0] = (char) j;
a4a439fb
KW
3799
3800 /* Then transform it */
3801 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3802
3803 /* If something went wrong (which it shouldn't), just
3804 * ignore this code point */
94762aa0 3805 if (! x) {
a4a439fb
KW
3806 continue;
3807 }
3808
3809 /* If this character's transformation is higher than
3810 * the current highest, this one becomes the highest */
3811 if ( cur_max_x == NULL
55e5378d
KW
3812 || strGT(x + COLLXFRM_HDR_LEN,
3813 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
3814 {
3815 PL_strxfrm_max_cp = j;
3816 cur_max_x = x;
3817 }
3818 else {
3819 Safefree(x);
3820 }
3821 }
3822
94762aa0
KW
3823 if (! cur_max_x) {
3824 DEBUG_L(PerlIO_printf(Perl_debug_log,
3825 "_mem_collxfrm: Couldn't find any character to"
3826 " replace above-Latin1 chars in locale %s with",
3827 PL_collation_name));
3828 goto bad;
3829 }
3830
58eebef2
KW
3831 DEBUG_L(PerlIO_printf(Perl_debug_log,
3832 "_mem_collxfrm: highest 1-byte collating character"
3833 " in locale %s is 0x%02X\n",
3834 PL_collation_name,
3835 PL_strxfrm_max_cp));
58eebef2 3836
a4a439fb
KW
3837 Safefree(cur_max_x);
3838 }
3839
3840 /* Here we know which legal code point collates the highest.
3841 * We are ready to construct the non-UTF-8 string. The length
3842 * will be at least 1 byte smaller than the input string
3843 * (because we changed at least one 2-byte character into a
3844 * single byte), but that is eaten up by the trailing NUL */
3845 Newx(s, len, char);
3846
3847 {
3848 STRLEN i;
3849 STRLEN d= 0;
042d9e50 3850 char * e = (char *) t + len;
a4a439fb
KW
3851
3852 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3853 U8 cur_char = t[i];
3854 if (UTF8_IS_INVARIANT(cur_char)) {
3855 s[d++] = cur_char;
3856 }
042d9e50 3857 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
3858 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3859 }
3860 else { /* Replace illegal cp with highest collating
3861 one */
3862 s[d++] = PL_strxfrm_max_cp;
3863 }
3864 }
3865 s[d++] = '\0';
3866 Renew(s, d, char); /* Free up unused space */
3867 }
3868 }
3869 }