This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Remove redundant SAVE_ERRNO
[perl5.git] / locale.c
CommitLineData
98994639
HS
1/* locale.c
2 *
1129b882
NC
3 * Copyright (C) 1993, 1994, 1995, 1996, 1997, 1998, 1999, 2000, 2001,
4 * 2002, 2003, 2005, 2006, 2007, 2008 by Larry Wall and others
98994639
HS
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550 12 * A Elbereth Gilthoniel,
cdad3b53 13 * silivren penna míriel
4ac71550 14 * o menel aglar elenath!
cdad3b53 15 * Na-chaered palan-díriel
4ac71550
TC
16 * o galadhremmin ennorath,
17 * Fanuilos, le linnathon
18 * nef aear, si nef aearon!
19 *
20 * [p.238 of _The Lord of the Rings_, II/i: "Many Meetings"]
98994639
HS
21 */
22
166f8a29
DM
23/* utility functions for handling locale-specific stuff like what
24 * character represents the decimal point.
0d071d52 25 *
7d4bcc4a
KW
26 * All C programs have an underlying locale. Perl code generally doesn't pay
27 * any attention to it except within the scope of a 'use locale'. For most
0d071d52
KW
28 * categories, it accomplishes this by just using different operations if it is
29 * in such scope than if not. However, various libc functions called by Perl
30 * are affected by the LC_NUMERIC category, so there are macros in perl.h that
31 * are used to toggle between the current locale and the C locale depending on
a9ad02a8
KW
32 * the desired behavior of those functions at the moment. And, LC_MESSAGES is
33 * switched to the C locale for outputting the message unless within the scope
34 * of 'use locale'.
e9bc6d6b
KW
35 *
36 * This code now has multi-thread-safe locale handling on systems that support
37 * that. This is completely transparent to most XS code. On earlier systems,
38 * it would be possible to emulate thread-safe locales, but this likely would
39 * involve a lot of locale switching, and would require XS code changes.
40 * Macros could be written so that the code wouldn't have to know which type of
41 * system is being used. It's unlikely that we would ever do that, since most
42 * modern systems support thread-safe locales, but there was code written to
43 * this end, and is retained, #ifdef'd out.
166f8a29
DM
44 */
45
98994639
HS
46#include "EXTERN.h"
47#define PERL_IN_LOCALE_C
f7416781 48#include "perl_langinfo.h"
98994639
HS
49#include "perl.h"
50
a4af207c
JH
51#include "reentr.h"
52
0dec74cd
KW
53#ifdef I_WCHAR
54# include <wchar.h>
55#endif
56
2fcc0ca9
KW
57/* If the environment says to, we can output debugging information during
58 * initialization. This is done before option parsing, and before any thread
59 * creation, so can be a file-level static */
5a4b0634
KW
60#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
61# define debug_initialization 0
62# define DEBUG_INITIALIZATION_set(v)
63#else
2fcc0ca9 64static bool debug_initialization = FALSE;
5a4b0634 65# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
66#endif
67
48015184
KW
68
69/* Returns the Unix errno portion; ignoring any others. This is a macro here
70 * instead of putting it into perl.h, because unclear to khw what should be
71 * done generally. */
72#define GET_ERRNO saved_errno
73
291a84fb
KW
74/* strlen() of a literal string constant. We might want this more general,
75 * but using it in just this file for now. A problem with more generality is
76 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
77#define STRLENs(s) (sizeof("" s "") - 1)
78
63e5b0d7
KW
79/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
80 * return of setlocale(), then this is extremely likely to be the C or POSIX
81 * locale. However, the output of setlocale() is documented to be opaque, but
82 * the odds are extremely small that it would return these two strings for some
83 * other locale. Note that VMS in these two locales includes many non-ASCII
84 * characters as controls and punctuation (below are hex bytes):
85 * cntrl: 84-97 9B-9F
86 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
87 * Oddly, none there are listed as alphas, though some represent alphabetics
88 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
89#define isNAME_C_OR_POSIX(name) \
90 ( (name) != NULL \
91 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
92 || strEQ((name), "POSIX")))
93
8ef6e574
KW
94#ifdef USE_LOCALE
95
47280b20
KW
96/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
97 * looked up. This is in the form of a C string: */
98
99#define UTF8NESS_SEP "\v"
100#define UTF8NESS_PREFIX "\f"
101
102/* So, the string looks like:
98994639 103 *
47280b20 104 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 105 *
47280b20
KW
106 * where the digit 0 after the \a indicates that the locale starting just
107 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
108
109STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
110STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
111
112#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
113 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
114
115/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
116 * kept there always. The remining portion of the cache is LRU, with the
117 * oldest looked-up locale at the tail end */
118
98994639
HS
119STATIC char *
120S_stdize_locale(pTHX_ char *locs)
121{
47280b20
KW
122 /* Standardize the locale name from a string returned by 'setlocale',
123 * possibly modifying that string.
124 *
125 * The typical return value of setlocale() is either
126 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
127 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
128 * (the space-separated values represent the various sublocales,
129 * in some unspecified order). This is not handled by this function.
130 *
131 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
132 * which is harmful for further use of the string in setlocale(). This
133 * function removes the trailing new line and everything up through the '='
134 * */
135
7452cf6a 136 const char * const s = strchr(locs, '=');
98994639
HS
137 bool okay = TRUE;
138
7918f24d
NC
139 PERL_ARGS_ASSERT_STDIZE_LOCALE;
140
8772537c
AL
141 if (s) {
142 const char * const t = strchr(s, '.');
98994639 143 okay = FALSE;
8772537c
AL
144 if (t) {
145 const char * const u = strchr(t, '\n');
146 if (u && (u[1] == 0)) {
147 const STRLEN len = u - s;
148 Move(s + 1, locs, len, char);
149 locs[len] = 0;
150 okay = TRUE;
98994639
HS
151 }
152 }
153 }
154
155 if (!okay)
156 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
157
158 return locs;
159}
160
e5f10d49
KW
161/* Two parallel arrays; first the locale categories Perl uses on this system;
162 * the second array is their names. These arrays are in mostly arbitrary
163 * order. */
164
165const int categories[] = {
166
167# ifdef USE_LOCALE_NUMERIC
168 LC_NUMERIC,
169# endif
170# ifdef USE_LOCALE_CTYPE
171 LC_CTYPE,
172# endif
173# ifdef USE_LOCALE_COLLATE
174 LC_COLLATE,
175# endif
176# ifdef USE_LOCALE_TIME
177 LC_TIME,
178# endif
179# ifdef USE_LOCALE_MESSAGES
180 LC_MESSAGES,
181# endif
182# ifdef USE_LOCALE_MONETARY
183 LC_MONETARY,
184# endif
9821811f
KW
185# ifdef USE_LOCALE_ADDRESS
186 LC_ADDRESS,
187# endif
188# ifdef USE_LOCALE_IDENTIFICATION
189 LC_IDENTIFICATION,
190# endif
191# ifdef USE_LOCALE_MEASUREMENT
192 LC_MEASUREMENT,
193# endif
194# ifdef USE_LOCALE_PAPER
195 LC_PAPER,
196# endif
197# ifdef USE_LOCALE_TELEPHONE
198 LC_TELEPHONE,
199# endif
e5f10d49
KW
200# ifdef LC_ALL
201 LC_ALL,
202# endif
203 -1 /* Placeholder because C doesn't allow a
204 trailing comma, and it would get complicated
205 with all the #ifdef's */
206};
207
208/* The top-most real element is LC_ALL */
209
210const char * category_names[] = {
211
212# ifdef USE_LOCALE_NUMERIC
213 "LC_NUMERIC",
214# endif
215# ifdef USE_LOCALE_CTYPE
216 "LC_CTYPE",
217# endif
218# ifdef USE_LOCALE_COLLATE
219 "LC_COLLATE",
220# endif
221# ifdef USE_LOCALE_TIME
222 "LC_TIME",
223# endif
224# ifdef USE_LOCALE_MESSAGES
225 "LC_MESSAGES",
226# endif
227# ifdef USE_LOCALE_MONETARY
228 "LC_MONETARY",
229# endif
9821811f
KW
230# ifdef USE_LOCALE_ADDRESS
231 "LC_ADDRESS",
232# endif
233# ifdef USE_LOCALE_IDENTIFICATION
234 "LC_IDENTIFICATION",
235# endif
236# ifdef USE_LOCALE_MEASUREMENT
237 "LC_MEASUREMENT",
238# endif
239# ifdef USE_LOCALE_PAPER
240 "LC_PAPER",
241# endif
242# ifdef USE_LOCALE_TELEPHONE
243 "LC_TELEPHONE",
244# endif
e5f10d49
KW
245# ifdef LC_ALL
246 "LC_ALL",
247# endif
248 NULL /* Placeholder */
249 };
250
251# ifdef LC_ALL
252
253 /* On systems with LC_ALL, it is kept in the highest index position. (-2
254 * to account for the final unused placeholder element.) */
255# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
256
257# else
258
259 /* On systems without LC_ALL, we pretend it is there, one beyond the real
260 * top element, hence in the unused placeholder element. */
261# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
262
263# endif
264
265/* Pretending there is an LC_ALL element just above allows us to avoid most
266 * special cases. Most loops through these arrays in the code below are
267 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
268 * on either type of system. But the code must be written to not access the
948523db
KW
269 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
270 * checked for at compile time by using the #define LC_ALL_INDEX which is only
271 * defined if we do have LC_ALL. */
e5f10d49 272
b09aaf40
KW
273STATIC const char *
274S_category_name(const int category)
275{
276 unsigned int i;
277
278#ifdef LC_ALL
279
280 if (category == LC_ALL) {
281 return "LC_ALL";
282 }
283
284#endif
285
286 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
287 if (category == categories[i]) {
288 return category_names[i];
289 }
290 }
291
292 {
293 const char suffix[] = " (unknown)";
294 int temp = category;
295 Size_t length = sizeof(suffix) + 1;
296 char * unknown;
297 dTHX;
298
299 if (temp < 0) {
300 length++;
301 temp = - temp;
302 }
303
304 /* Calculate the number of digits */
305 while (temp >= 10) {
306 temp /= 10;
307 length++;
308 }
309
310 Newx(unknown, length, char);
311 my_snprintf(unknown, length, "%d%s", category, suffix);
312 SAVEFREEPV(unknown);
313 return unknown;
314 }
315}
316
948523db
KW
317/* Now create LC_foo_INDEX #defines for just those categories on this system */
318# ifdef USE_LOCALE_NUMERIC
319# define LC_NUMERIC_INDEX 0
320# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
321# else
322# define _DUMMY_NUMERIC -1
323# endif
324# ifdef USE_LOCALE_CTYPE
325# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
326# define _DUMMY_CTYPE LC_CTYPE_INDEX
327# else
328# define _DUMMY_CTYPE _DUMMY_NUMERIC
329# endif
330# ifdef USE_LOCALE_COLLATE
331# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
332# define _DUMMY_COLLATE LC_COLLATE_INDEX
333# else
334# define _DUMMY_COLLATE _DUMMY_COLLATE
335# endif
336# ifdef USE_LOCALE_TIME
337# define LC_TIME_INDEX _DUMMY_COLLATE + 1
338# define _DUMMY_TIME LC_TIME_INDEX
339# else
340# define _DUMMY_TIME _DUMMY_COLLATE
341# endif
342# ifdef USE_LOCALE_MESSAGES
343# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
344# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
345# else
346# define _DUMMY_MESSAGES _DUMMY_TIME
347# endif
348# ifdef USE_LOCALE_MONETARY
349# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
350# define _DUMMY_MONETARY LC_MONETARY_INDEX
351# else
352# define _DUMMY_MONETARY _DUMMY_MESSAGES
353# endif
9821811f
KW
354# ifdef USE_LOCALE_ADDRESS
355# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
356# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
357# else
358# define _DUMMY_ADDRESS _DUMMY_MONETARY
359# endif
360# ifdef USE_LOCALE_IDENTIFICATION
361# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
362# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
363# else
364# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
365# endif
366# ifdef USE_LOCALE_MEASUREMENT
367# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
368# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
369# else
370# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
371# endif
372# ifdef USE_LOCALE_PAPER
373# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
374# define _DUMMY_PAPER LC_PAPER_INDEX
375# else
376# define _DUMMY_PAPER _DUMMY_MEASUREMENT
377# endif
378# ifdef USE_LOCALE_TELEPHONE
379# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
380# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
381# else
382# define _DUMMY_TELEPHONE _DUMMY_PAPER
383# endif
948523db 384# ifdef LC_ALL
9821811f 385# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
386# endif
387#endif /* ifdef USE_LOCALE */
8ef6e574 388
d2b24094 389/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
390#ifdef WIN32
391# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
392#else
393# define my_setlocale(cat, locale) setlocale(cat, locale)
394#endif
395
396#ifndef USE_POSIX_2008_LOCALE
397
398/* "do_setlocale_c" is intended to be called when the category is a constant
399 * known at compile time; "do_setlocale_r", not known until run time */
400# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
401# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
402
403#else /* Below uses POSIX 2008 */
404
405/* We emulate setlocale with our own function. LC_foo is not valid for the
406 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
407 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
408 * proper offset into the array 'category_masks[]'. At runtime, we have to
409 * search through the array (as the actual numbers may not be small contiguous
410 * positive integers which would lend themselves to array lookup). */
411# define do_setlocale_c(cat, locale) \
412 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
413# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
414
415/* A third array, parallel to the ones above to map from category to its
416 * equivalent mask */
417const int category_masks[] = {
418# ifdef USE_LOCALE_NUMERIC
419 LC_NUMERIC_MASK,
420# endif
421# ifdef USE_LOCALE_CTYPE
422 LC_CTYPE_MASK,
423# endif
424# ifdef USE_LOCALE_COLLATE
425 LC_COLLATE_MASK,
426# endif
427# ifdef USE_LOCALE_TIME
428 LC_TIME_MASK,
429# endif
430# ifdef USE_LOCALE_MESSAGES
431 LC_MESSAGES_MASK,
432# endif
433# ifdef USE_LOCALE_MONETARY
434 LC_MONETARY_MASK,
435# endif
436# ifdef USE_LOCALE_ADDRESS
437 LC_ADDRESS_MASK,
438# endif
439# ifdef USE_LOCALE_IDENTIFICATION
440 LC_IDENTIFICATION_MASK,
441# endif
442# ifdef USE_LOCALE_MEASUREMENT
443 LC_MEASUREMENT_MASK,
444# endif
445# ifdef USE_LOCALE_PAPER
446 LC_PAPER_MASK,
447# endif
448# ifdef USE_LOCALE_TELEPHONE
449 LC_TELEPHONE_MASK,
450# endif
451 /* LC_ALL can't be turned off by a Configure
452 * option, and in Posix 2008, should always be
453 * here, so compile it in unconditionally.
454 * This could catch some glitches at compile
455 * time */
456 LC_ALL_MASK
457 };
458
459STATIC const char *
460S_emulate_setlocale(const int category,
461 const char * locale,
462 unsigned int index,
463 const bool is_index_valid
464 )
465{
466 /* This function effectively performs a setlocale() on just the current
467 * thread; thus it is thread-safe. It does this by using the POSIX 2008
468 * locale functions to emulate the behavior of setlocale(). Similar to
469 * regular setlocale(), the return from this function points to memory that
470 * can be overwritten by other system calls, so needs to be copied
471 * immediately if you need to retain it. The difference here is that
472 * system calls besides another setlocale() can overwrite it.
473 *
474 * By doing this, most locale-sensitive functions become thread-safe. The
475 * exceptions are mostly those that return a pointer to static memory.
476 *
477 * This function takes the same parameters, 'category' and 'locale', that
478 * the regular setlocale() function does, but it also takes two additional
479 * ones. This is because the 2008 functions don't use a category; instead
480 * they use a corresponding mask. Because this function operates in both
481 * worlds, it may need one or the other or both. This function can
482 * calculate the mask from the input category, but to avoid this
483 * calculation, if the caller knows at compile time what the mask is, it
484 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
485 * parameter is ignored.
486 *
487 * POSIX 2008, for some sick reason, chose not to provide a method to find
488 * the category name of a locale. Some vendors have created a
489 * querylocale() function to do just that. This function is a lot simpler
490 * to implement on systems that have this. Otherwise, we have to keep
491 * track of what the locale has been set to, so that we can return its
492 * name to emulate setlocale(). It's also possible for C code in some
493 * library to change the locale without us knowing it, though as of
494 * September 2017, there are no occurrences in CPAN of uselocale(). Some
495 * libraries do use setlocale(), but that changes the global locale, and
496 * threads using per-thread locales will just ignore those changes.
497 * Another problem is that without querylocale(), we have to guess at what
498 * was meant by setting a locale of "". We handle this by not actually
499 * ever setting to "" (unless querylocale exists), but to emulate what we
500 * think should happen for "".
501 */
502
503 int mask;
504 locale_t old_obj;
505 locale_t new_obj;
506 dTHX;
507
508# ifdef DEBUGGING
509
510 if (DEBUG_Lv_TEST || debug_initialization) {
511 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale input=%d (%s), \"%s\", %d, %d\n", __FILE__, __LINE__, category, category_name(category), locale, index, is_index_valid);
512 }
513
514# endif
515
516 /* If the input mask might be incorrect, calculate the correct one */
517 if (! is_index_valid) {
518 unsigned int i;
519
520# ifdef DEBUGGING
521
522 if (DEBUG_Lv_TEST || debug_initialization) {
523 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
524 }
525
526# endif
527
528 for (i = 0; i <= LC_ALL_INDEX; i++) {
529 if (category == categories[i]) {
530 index = i;
531 goto found_index;
532 }
533 }
534
535 /* Here, we don't know about this category, so can't handle it.
536 * Fallback to the early POSIX usages */
537 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
538 "Unknown locale category %d; can't set it to %s\n",
539 category, locale);
540 return NULL;
541
542 found_index: ;
543
544# ifdef DEBUGGING
545
546 if (DEBUG_Lv_TEST || debug_initialization) {
547 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
548 }
549
550# endif
551
552 }
553
554 mask = category_masks[index];
555
556# ifdef DEBUGGING
557
558 if (DEBUG_Lv_TEST || debug_initialization) {
559 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
560 }
561
562# endif
563
564 /* If just querying what the existing locale is ... */
565 if (locale == NULL) {
566 locale_t cur_obj = uselocale((locale_t) 0);
567
568# ifdef DEBUGGING
569
570 if (DEBUG_Lv_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
572 }
573
574# endif
575
576 if (cur_obj == LC_GLOBAL_LOCALE) {
577 return my_setlocale(category, NULL);
578 }
579
580# ifdef HAS_QUERYLOCALE
581
582 return (char *) querylocale(mask, cur_obj);
583
d2b24094 584# else
ecdda939
KW
585
586 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
587 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
588
e9bc6d6b
KW
589# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
590
591 {
592 /* Internal glibc for querylocale(), but doesn't handle
593 * empty-string ("") locale properly; who knows what other
594 * glitches. Check it for now, under debug. */
595
596 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
597 uselocale((locale_t) 0));
598 /*
599 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
600 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
601 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
602 */
603 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
604 if ( strNE(PL_curlocales[index], temp_name)
605 && ! ( isNAME_C_OR_POSIX(temp_name)
606 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
607
608# ifdef USE_C_BACKTRACE
609
610 dump_c_backtrace(Perl_debug_log, 20, 1);
611
612# endif
613
614 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
615 " (%s) and what internal glibc thinks"
616 " (%s)\n", category_names[index],
617 PL_curlocales[index], temp_name);
618 }
619
620 return temp_name;
621 }
622 }
623
624# endif
625
626 /* Without querylocale(), we have to use our record-keeping we've
627 * done. */
628
629 if (category != LC_ALL) {
630
631# ifdef DEBUGGING
632
633 if (DEBUG_Lv_TEST || debug_initialization) {
634 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
635 }
636
637# endif
638
639 return PL_curlocales[index];
640 }
641 else { /* For LC_ALL */
642 unsigned int i;
643 Size_t names_len = 0;
644 char * all_string;
7c93a471 645 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
646
647 /* If we have a valid LC_ALL value, just return it */
648 if (PL_curlocales[LC_ALL_INDEX]) {
649
650# ifdef DEBUGGING
651
652 if (DEBUG_Lv_TEST || debug_initialization) {
653 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
654 }
655
656# endif
657
658 return PL_curlocales[LC_ALL_INDEX];
659 }
660
661 /* Otherwise, we need to construct a string of name=value pairs.
662 * We use the glibc syntax, like
663 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
664 * First calculate the needed size. Along the way, check if all
665 * the locale names are the same */
e9bc6d6b
KW
666 for (i = 0; i < LC_ALL_INDEX; i++) {
667
668# ifdef DEBUGGING
669
670 if (DEBUG_Lv_TEST || debug_initialization) {
671 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
672 }
673
674# endif
675
676 names_len += strlen(category_names[i])
677 + 1 /* '=' */
678 + strlen(PL_curlocales[i])
679 + 1; /* ';' */
7c93a471
KW
680
681 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
682 are_all_categories_the_same_locale = FALSE;
683 }
684 }
685
686 /* If they are the same, we don't actually have to construct the
687 * string; we just make the entry in LC_ALL_INDEX valid, and be
688 * that single name */
689 if (are_all_categories_the_same_locale) {
690 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
691 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 692 }
7c93a471 693
e9bc6d6b
KW
694 names_len++; /* Trailing '\0' */
695 SAVEFREEPV(Newx(all_string, names_len, char));
696 *all_string = '\0';
697
698 /* Then fill in the string */
699 for (i = 0; i < LC_ALL_INDEX; i++) {
700
701# ifdef DEBUGGING
702
703 if (DEBUG_Lv_TEST || debug_initialization) {
704 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale i=%d, name=%s, locale=%s\n", __FILE__, __LINE__, i, category_names[i], PL_curlocales[i]);
705 }
706
707# endif
708
709 my_strlcat(all_string, category_names[i], names_len);
710 my_strlcat(all_string, "=", names_len);
711 my_strlcat(all_string, PL_curlocales[i], names_len);
712 my_strlcat(all_string, ";", names_len);
713 }
714
715# ifdef DEBUGGING
716
717 if (DEBUG_L_TEST || debug_initialization) {
718 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
719 }
720
721 #endif
722
723 return all_string;
724 }
725
726# ifdef EINVAL
727
728 SETERRNO(EINVAL, LIB_INVARG);
729
730# endif
731
732 return NULL;
733
d2b24094
KW
734# endif
735
e9bc6d6b
KW
736 }
737
738 assert(PL_C_locale_obj);
739
740 /* Otherwise, we are switching locales. This will generally entail freeing
741 * the current one's space (at the C library's discretion). We need to
742 * stop using that locale before the switch. So switch to a known locale
743 * object that we don't otherwise mess with. This returns the locale
744 * object in effect at the time of the switch. */
745 old_obj = uselocale(PL_C_locale_obj);
746
747# ifdef DEBUGGING
748
749 if (DEBUG_Lv_TEST || debug_initialization) {
750 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
751 }
752
753# endif
754
755 if (! old_obj) {
756
757# ifdef DEBUGGING
758
759 if (DEBUG_L_TEST || debug_initialization) {
760 dSAVE_ERRNO;
761 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
762 RESTORE_ERRNO;
763 }
764
765# endif
766
767 return NULL;
768 }
769
770# ifdef DEBUGGING
771
772 if (DEBUG_Lv_TEST || debug_initialization) {
773 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
774 }
775
776# endif
777
778 /* If we weren't in a thread safe locale, set so that newlocale() below
779 which uses 'old_obj', uses an empty one. Same for our reserved C object.
780 The latter is defensive coding, so that, even if there is some bug, we
781 will never end up trying to modify either of these, as if passed to
782 newlocale(), they can be. */
783 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
784 old_obj = (locale_t) 0;
785 }
786
787 /* Create the new locale (it may actually modify the current one). */
788
789# ifndef HAS_QUERYLOCALE
790
791 if (strEQ(locale, "")) {
792
793 /* For non-querylocale() systems, we do the setting of "" ourselves to
794 * be sure that we really know what's going on. We follow the Linux
795 * documented behavior (but if that differs from the actual behavior,
796 * this won't work exactly as the OS implements). We go out and
797 * examine the environment based on our understanding of how the system
798 * works, and use that to figure things out */
799
800 const char * const lc_all = PerlEnv_getenv("LC_ALL");
801
802 /* Use any "LC_ALL" environment variable, as it overrides everything
803 * else. */
804 if (lc_all && strNE(lc_all, "")) {
805 locale = lc_all;
806 }
807 else {
808
809 /* Otherwise, we need to dig deeper. Unless overridden, the
810 * default is the LANG environment variable; if it doesn't exist,
811 * then "C" */
812
813 const char * default_name;
814
815 /* To minimize other threads messing with the environment, we copy
816 * the variable, making it a temporary. But this doesn't work upon
817 * program initialization before any scopes are created, and at
818 * this time, there's nothing else going on that would interfere.
819 * So skip the copy in that case */
820 if (PL_scopestack_ix == 0) {
821 default_name = PerlEnv_getenv("LANG");
822 }
823 else {
824 default_name = savepv(PerlEnv_getenv("LANG"));
825 }
826
827 if (! default_name || strEQ(default_name, "")) {
828 default_name = "C";
829 }
830 else if (PL_scopestack_ix != 0) {
831 SAVEFREEPV(default_name);
832 }
833
834 if (category != LC_ALL) {
835 const char * const name = PerlEnv_getenv(category_names[index]);
836
837 /* Here we are setting a single category. Assume will have the
838 * default name */
839 locale = default_name;
840
841 /* But then look for an overriding environment variable */
842 if (name && strNE(name, "")) {
843 locale = name;
844 }
845 }
846 else {
847 bool did_override = FALSE;
848 unsigned int i;
849
850 /* Here, we are getting LC_ALL. Any categories that don't have
851 * a corresponding environment variable set should be set to
852 * LANG, or to "C" if there is no LANG. If no individual
853 * categories differ from this, we can just set LC_ALL. This
854 * is buggy on systems that have extra categories that we don't
855 * know about. If there is an environment variable that sets
856 * that category, we won't know to look for it, and so our use
857 * of LANG or "C" improperly overrides it. On the other hand,
858 * if we don't do what is done here, and there is no
859 * environment variable, the category's locale should be set to
860 * LANG or "C". So there is no good solution. khw thinks the
861 * best is to look at systems to see what categories they have,
862 * and include them, and then to assume that we know the
863 * complete set */
864
865 for (i = 0; i < LC_ALL_INDEX; i++) {
866 const char * const env_override
867 = savepv(PerlEnv_getenv(category_names[i]));
868 const char * this_locale = ( env_override
869 && strNE(env_override, ""))
870 ? env_override
871 : default_name;
872 emulate_setlocale(categories[i], this_locale, i, TRUE);
873
874 if (strNE(this_locale, default_name)) {
875 did_override = TRUE;
876 }
877
878 Safefree(env_override);
879 }
880
881 /* If all the categories are the same, we can set LC_ALL to
882 * that */
883 if (! did_override) {
884 locale = default_name;
885 }
886 else {
887
888 /* Here, LC_ALL is no longer valid, as some individual
889 * categories don't match it. We call ourselves
890 * recursively, as that will execute the code that
891 * generates the proper locale string for this situation.
892 * We don't do the remainder of this function, as that is
893 * to update our records, and we've just done that for the
894 * individual categories in the loop above, and doing so
895 * would cause LC_ALL to be done as well */
896 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
897 }
898 }
899 }
900 }
901 else if (strchr(locale, ';')) {
902
903 /* LC_ALL may actually incude a conglomeration of various categories.
904 * Without querylocale, this code uses the glibc (as of this writing)
905 * syntax for representing that, but that is not a stable API, and
906 * other platforms do it differently, so we have to handle all cases
907 * ourselves */
908
909 const char * s = locale;
910 const char * e = locale + strlen(locale);
911 const char * p = s;
912 const char * category_end;
913 const char * name_start;
914 const char * name_end;
915
916 while (s < e) {
917 unsigned int i;
918
919 /* Parse through the category */
920 while (isWORDCHAR(*p)) {
921 p++;
922 }
923 category_end = p;
924
925 if (*p++ != '=') {
926 Perl_croak(aTHX_
927 "panic: %s: %d: Unexpected character in locale name '%02X",
928 __FILE__, __LINE__, *(p-1));
929 }
930
931 /* Parse through the locale name */
932 name_start = p;
933 while (isGRAPH(*p) && *p != ';') {
934 p++;
935 }
936 name_end = p;
937
938 if (*p++ != ';') {
939 Perl_croak(aTHX_
940 "panic: %s: %d: Unexpected character in locale name '%02X",
941 __FILE__, __LINE__, *(p-1));
942 }
943
944 /* Find the index of the category name in our lists */
945 for (i = 0; i < LC_ALL_INDEX; i++) {
946
947 /* Keep going if this isn't the index. The strnNE() avoids a
948 * Perl_form(), but would fail if ever a category name could be
949 * a substring of another one, like if there were a
950 * "LC_TIME_DATE" */
951 if strnNE(s, category_names[i], category_end - s) {
952 continue;
953 }
954
955 /* If this index is for the single category we're changing, we
956 * have found the locale to set it to. */
957 if (category == categories[i]) {
958 locale = Perl_form(aTHX_ "%.*s",
959 (int) (name_end - name_start),
960 name_start);
961 goto ready_to_set;
962 }
963
964 if (category == LC_ALL) {
965 char * individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
966 emulate_setlocale(categories[i], individ_locale, i, TRUE);
967 Safefree(individ_locale);
968 }
969 }
970
971 s = p;
972 }
973
974 /* Here we have set all the individual categories by recursive calls.
975 * These collectively should have fixed up LC_ALL, so can just query
976 * what that now is */
977 assert(category == LC_ALL);
978
979 return do_setlocale_c(LC_ALL, NULL);
980 }
981
982 ready_to_set: ;
983
984# endif /* end of ! querylocale */
985
986 /* Ready to create a new locale by modification of the exising one */
987 new_obj = newlocale(mask, locale, old_obj);
988
989 if (! new_obj) {
990 dSAVE_ERRNO;
991
992# ifdef DEBUGGING
993
994 if (DEBUG_L_TEST || debug_initialization) {
995 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
996 }
997
998# endif
999
1000 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1001
1002# ifdef DEBUGGING
1003
1004 if (DEBUG_L_TEST || debug_initialization) {
1005 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1006 }
1007
1008# endif
1009
1010 }
1011 RESTORE_ERRNO;
1012 return NULL;
1013 }
1014
1015# ifdef DEBUGGING
1016
1017 if (DEBUG_Lv_TEST || debug_initialization) {
1018 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1019 }
1020
1021# endif
1022
1023 /* And switch into it */
1024 if (! uselocale(new_obj)) {
1025 dSAVE_ERRNO;
1026
1027# ifdef DEBUGGING
1028
1029 if (DEBUG_L_TEST || debug_initialization) {
1030 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1031 }
1032
1033# endif
1034
1035 if (! uselocale(old_obj)) {
1036
1037# ifdef DEBUGGING
1038
1039 if (DEBUG_L_TEST || debug_initialization) {
1040 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1041 }
1042
1043# endif
1044
1045 }
1046 freelocale(new_obj);
1047 RESTORE_ERRNO;
1048 return NULL;
1049 }
1050
1051# ifdef DEBUGGING
1052
1053 if (DEBUG_Lv_TEST || debug_initialization) {
1054 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1055 }
1056
1057# endif
1058
1059 /* We are done, except for updating our records (if the system doesn't keep
1060 * them) and in the case of locale "", we don't actually know what the
1061 * locale that got switched to is, as it came from the environment. So
1062 * have to find it */
1063
1064# ifdef HAS_QUERYLOCALE
1065
1066 if (strEQ(locale, "")) {
1067 locale = querylocale(mask, new_obj);
1068 }
1069
1070# else
1071
1072 /* Here, 'locale' is the return value */
1073
1074 /* Without querylocale(), we have to update our records */
1075
1076 if (category == LC_ALL) {
1077 unsigned int i;
1078
1079 /* For LC_ALL, we change all individual categories to correspond */
1080 /* PL_curlocales is a parallel array, so has same
1081 * length as 'categories' */
1082 for (i = 0; i <= LC_ALL_INDEX; i++) {
1083 Safefree(PL_curlocales[i]);
1084 PL_curlocales[i] = savepv(locale);
1085 }
1086 }
1087 else {
1088
1089 /* For a single category, if it's not the same as the one in LC_ALL, we
1090 * nullify LC_ALL */
1091
1092 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1093 Safefree(PL_curlocales[LC_ALL_INDEX]);
1094 PL_curlocales[LC_ALL_INDEX] = NULL;
1095 }
1096
1097 /* Then update the category's record */
1098 Safefree(PL_curlocales[index]);
1099 PL_curlocales[index] = savepv(locale);
1100 }
1101
1102# endif
1103
1104 return locale;
1105}
1106
1107#endif /* USE_POSIX_2008_LOCALE */
1108
1109#if 0 /* Code that was to emulate thread-safe locales on platforms that
1110 didn't natively support them */
1111
1112/* The way this would work is that we would keep a per-thread list of the
1113 * correct locale for that thread. Any operation that was locale-sensitive
1114 * would have to be changed so that it would look like this:
1115 *
1116 * LOCALE_LOCK;
1117 * setlocale to the correct locale for this operation
1118 * do operation
1119 * LOCALE_UNLOCK
1120 *
1121 * This leaves the global locale in the most recently used operation's, but it
1122 * was locked long enough to get the result. If that result is static, it
1123 * needs to be copied before the unlock.
1124 *
1125 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1126 * the setup, but are no-ops when not needed, and similarly,
1127 * END_LOCALE_DEPENDENT_OP for the tear-down
1128 *
1129 * But every call to a locale-sensitive function would have to be changed, and
1130 * if a module didn't cooperate by using the mutex, things would break.
1131 *
1132 * This code was abandoned before being completed or tested, and is left as-is
1133*/
1134
1135# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1136# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1137
1138STATIC char *
1139S_locking_setlocale(pTHX_
1140 const int category,
1141 const char * locale,
1142 int index,
1143 const bool is_index_valid
1144 )
1145{
1146 /* This function kind of performs a setlocale() on just the current thread;
1147 * thus it is kind of thread-safe. It does this by keeping a thread-level
1148 * array of the current locales for each category. Every time a locale is
1149 * switched to, it does the switch globally, but updates the thread's
1150 * array. A query as to what the current locale is just returns the
1151 * appropriate element from the array, and doesn't actually call the system
1152 * setlocale(). The saving into the array is done in an uninterruptible
1153 * section of code, so is unaffected by whatever any other threads might be
1154 * doing.
1155 *
1156 * All locale-sensitive operations must work by first starting a critical
1157 * section, then switching to the thread's locale as kept by this function,
1158 * and then doing the operation, then ending the critical section. Thus,
1159 * each gets done in the appropriate locale. simulating thread-safety.
1160 *
1161 * This function takes the same parameters, 'category' and 'locale', that
1162 * the regular setlocale() function does, but it also takes two additional
1163 * ones. This is because as described earlier. If we know on input the
1164 * index corresponding to the category into the array where we store the
1165 * current locales, we don't have to calculate it. If the caller knows at
1166 * compile time what the index is, it it can pass it, setting
1167 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1168 *
1169 */
1170
1171 /* If the input index might be incorrect, calculate the correct one */
1172 if (! is_index_valid) {
1173 unsigned int i;
1174
1175 if (DEBUG_Lv_TEST || debug_initialization) {
1176 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1177 }
1178
1179 for (i = 0; i <= LC_ALL_INDEX; i++) {
1180 if (category == categories[i]) {
1181 index = i;
1182 goto found_index;
1183 }
1184 }
1185
1186 /* Here, we don't know about this category, so can't handle it.
1187 * XXX best we can do is to unsafely set this
1188 * XXX warning */
1189
1190 return my_setlocale(category, locale);
1191
1192 found_index: ;
1193
1194 if (DEBUG_Lv_TEST || debug_initialization) {
1195 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1196 }
1197 }
1198
1199 /* For a query, just return what's in our records */
1200 if (new_locale == NULL) {
1201 return curlocales[index];
1202 }
1203
1204
1205 /* Otherwise, we need to do the switch, and save the result, all in a
1206 * critical section */
1207
1208 Safefree(curlocales[[index]]);
1209
1210 /* It might be that this is called from an already-locked section of code.
1211 * We would have to detect and skip the LOCK/UNLOCK if so */
1212 LOCALE_LOCK;
1213
1214 curlocales[index] = savepv(my_setlocale(category, new_locale));
1215
1216 if (strEQ(new_locale, "")) {
1217
1218#ifdef LC_ALL
1219
1220 /* The locale values come from the environment, and may not all be the
1221 * same, so for LC_ALL, we have to update all the others, while the
1222 * mutex is still locked */
1223
1224 if (category == LC_ALL) {
1225 unsigned int i;
1226 for (i = 0; i < LC_ALL_INDEX) {
1227 curlocales[i] = my_setlocale(categories[i], NULL);
1228 }
1229 }
1230 }
1231
1232#endif
1233
1234 LOCALE_UNLOCK;
1235
1236 return curlocales[index];
1237}
1238
1239#endif
837ce802 1240
a4f00dcc 1241STATIC void
86799d2d 1242S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1243{
86799d2d
KW
1244 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1245 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1246
86799d2d
KW
1247#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1248 || defined(HAS_NL_LANGINFO))
98994639 1249
87f8e8e7
KW
1250 const char * radix = (use_locale)
1251 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1252 /* FALSE => already in dest locale */
1253 : ".";
2213a3be 1254
87f8e8e7 1255 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1256
87f8e8e7
KW
1257 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1258 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1259 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1260 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1261 && _is_cur_LC_category_utf8(LC_NUMERIC))
1262 {
1263 SvUTF8_on(PL_numeric_radix_sv);
1264 }
86799d2d
KW
1265
1266# ifdef DEBUGGING
7d4bcc4a 1267
2fcc0ca9
KW
1268 if (DEBUG_L_TEST || debug_initialization) {
1269 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1270 SvPVX(PL_numeric_radix_sv),
1271 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1272 }
69014004 1273
86799d2d
KW
1274# endif
1275#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1276
98994639
HS
1277}
1278
398eeea9
KW
1279STATIC void
1280S_new_numeric(pTHX_ const char *newnum)
98994639 1281{
7d4bcc4a
KW
1282
1283#ifndef USE_LOCALE_NUMERIC
1284
1285 PERL_UNUSED_ARG(newnum);
1286
1287#else
0d071d52 1288
291a84fb 1289 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1290 * core Perl this and that 'newnum' is the name of the new locale.
1291 * It installs this locale as the current underlying default.
1292 *
1293 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1294 * set_numeric_underlying() and set_numeric_standard() functions, which
1295 * should probably not be called directly, but only via macros like
0d071d52
KW
1296 * SET_NUMERIC_STANDARD() in perl.h.
1297 *
1298 * The toggling is necessary mainly so that a non-dot radix decimal point
1299 * character can be output, while allowing internal calculations to use a
1300 * dot.
1301 *
1302 * This sets several interpreter-level variables:
bb304765 1303 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1304 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1305 * that the current locale is the program's underlying
1306 * locale
1307 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1308 * that the current locale is the C locale or
1309 * indistinguishable from the C locale. If non-zero, it
1310 * is in C; if > 1, it means it may not be toggled away
7738054c 1311 * from C.
4c68b815
KW
1312 * PL_numeric_underlying_is_standard A bool kept by this function
1313 * indicating that the underlying locale and the standard
1314 * C locale are indistinguishable for the purposes of
1315 * LC_NUMERIC. This happens when both of the above two
1316 * variables are true at the same time. (Toggling is a
1317 * no-op under these circumstances.) This variable is
1318 * used to avoid having to recalculate.
398eeea9 1319 */
0d071d52 1320
b03f34cf 1321 char *save_newnum;
98994639
HS
1322
1323 if (! newnum) {
43c5f42d
NC
1324 Safefree(PL_numeric_name);
1325 PL_numeric_name = NULL;
98994639 1326 PL_numeric_standard = TRUE;
892e6465 1327 PL_numeric_underlying = TRUE;
4c68b815 1328 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1329 return;
1330 }
1331
b03f34cf 1332 save_newnum = stdize_locale(savepv(newnum));
892e6465 1333 PL_numeric_underlying = TRUE;
4c68b815
KW
1334 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1335
1336 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1337 * them */
1338 if (! PL_numeric_standard) {
1339 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1340 FALSE /* Don't toggle locale */ ))
1341 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1342 FALSE)));
1343 }
abe1abcf 1344
4c68b815 1345 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1346 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1347 Safefree(PL_numeric_name);
b03f34cf 1348 PL_numeric_name = save_newnum;
b03f34cf 1349 }
abe1abcf
KW
1350 else {
1351 Safefree(save_newnum);
1352 }
4c28b29c 1353
4c68b815
KW
1354 PL_numeric_underlying_is_standard = PL_numeric_standard;
1355
7e5377f7 1356# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1357
1358 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1359 PL_numeric_name,
1360 PL_underlying_numeric_obj);
1361
1362#endif
1363
4c68b815
KW
1364 if (DEBUG_L_TEST || debug_initialization) {
1365 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1366 }
1367
4c28b29c
KW
1368 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1369 * have to worry about the radix being a non-dot. (Core operations that
1370 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1371 if (PL_numeric_standard) {
1372 set_numeric_radix(0);
1373 }
1374 else {
1375 set_numeric_standard();
1376 }
4c28b29c 1377
98994639 1378#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1379
98994639
HS
1380}
1381
1382void
1383Perl_set_numeric_standard(pTHX)
1384{
7d4bcc4a 1385
98994639 1386#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1387
28c1bf33
KW
1388 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1389 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1390 * macro avoids calling this routine if toggling isn't necessary according
1391 * to our records (which could be wrong if some XS code has changed the
1392 * locale behind our back) */
0d071d52 1393
837ce802 1394 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1395 PL_numeric_standard = TRUE;
4c68b815 1396 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1397 set_numeric_radix(0);
7d4bcc4a
KW
1398
1399# ifdef DEBUGGING
1400
2fcc0ca9
KW
1401 if (DEBUG_L_TEST || debug_initialization) {
1402 PerlIO_printf(Perl_debug_log,
58e4a467 1403 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1404 }
98994639 1405
7d4bcc4a 1406# endif
98994639 1407#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1408
98994639
HS
1409}
1410
1411void
5792c642 1412Perl_set_numeric_underlying(pTHX)
98994639 1413{
7d4bcc4a 1414
98994639 1415#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1416
28c1bf33 1417 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1418 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1419 * instead of calling this directly. The macro avoids calling this routine
1420 * if toggling isn't necessary according to our records (which could be
1421 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1422
837ce802 1423 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1424 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1425 PL_numeric_underlying = TRUE;
3c62fd46 1426 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1427
1428# ifdef DEBUGGING
1429
2fcc0ca9
KW
1430 if (DEBUG_L_TEST || debug_initialization) {
1431 PerlIO_printf(Perl_debug_log,
58e4a467 1432 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1433 PL_numeric_name);
1434 }
98994639 1435
7d4bcc4a 1436# endif
98994639 1437#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1438
98994639
HS
1439}
1440
1441/*
1442 * Set up for a new ctype locale.
1443 */
a4f00dcc
KW
1444STATIC void
1445S_new_ctype(pTHX_ const char *newctype)
98994639 1446{
7d4bcc4a
KW
1447
1448#ifndef USE_LOCALE_CTYPE
1449
1450 PERL_ARGS_ASSERT_NEW_CTYPE;
1451 PERL_UNUSED_ARG(newctype);
1452 PERL_UNUSED_CONTEXT;
1453
1454#else
0d071d52 1455
291a84fb 1456 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1457 * core Perl this and that 'newctype' is the name of the new locale.
1458 *
1459 * This function sets up the folding arrays for all 256 bytes, assuming
1460 * that tofold() is tolc() since fold case is not a concept in POSIX,
1461 *
1462 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1463 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1464 * this function should be called directly only from this file and from
0d071d52
KW
1465 * POSIX::setlocale() */
1466
27da23d5 1467 dVAR;
013f4e03 1468 unsigned int i;
98994639 1469
8b7358b9
KW
1470 /* Don't check for problems if we are suppressing the warnings */
1471 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1472
7918f24d
NC
1473 PERL_ARGS_ASSERT_NEW_CTYPE;
1474
215c5139
KW
1475 /* We will replace any bad locale warning with 1) nothing if the new one is
1476 * ok; or 2) a new warning for the bad new locale */
1477 if (PL_warn_locale) {
1478 SvREFCNT_dec_NN(PL_warn_locale);
1479 PL_warn_locale = NULL;
1480 }
1481
c1284011 1482 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1483
1484 /* A UTF-8 locale gets standard rules. But note that code still has to
1485 * handle this specially because of the three problematic code points */
1486 if (PL_in_utf8_CTYPE_locale) {
1487 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1488 }
8b7358b9
KW
1489
1490 /* We don't populate the other lists if a UTF-8 locale, but do check that
1491 * everything works as expected, unless checking turned off */
1492 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1493 /* Assume enough space for every character being bad. 4 spaces each
1494 * for the 94 printable characters that are output like "'x' "; and 5
1495 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1496 * NUL */
8b7358b9 1497 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1498 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1499 to start */
1500 unsigned int bad_count = 0; /* Count of bad characters */
1501
baa60164 1502 for (i = 0; i < 256; i++) {
8b7358b9 1503 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1504 if (isupper(i))
1505 PL_fold_locale[i] = (U8) tolower(i);
1506 else if (islower(i))
1507 PL_fold_locale[i] = (U8) toupper(i);
1508 else
1509 PL_fold_locale[i] = (U8) i;
8b7358b9 1510 }
8c6180a9
KW
1511
1512 /* If checking for locale problems, see if the native ASCII-range
1513 * printables plus \n and \t are in their expected categories in
1514 * the new locale. If not, this could mean big trouble, upending
1515 * Perl's and most programs' assumptions, like having a
1516 * metacharacter with special meaning become a \w. Fortunately,
1517 * it's very rare to find locales that aren't supersets of ASCII
1518 * nowadays. It isn't a problem for most controls to be changed
1519 * into something else; we check only \n and \t, though perhaps \r
1520 * could be an issue as well. */
7d4bcc4a 1521 if ( check_for_problems
8c6180a9
KW
1522 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1523 {
8b7358b9
KW
1524 bool is_bad = FALSE;
1525 char name[3] = { '\0' };
1526
1527 /* Convert the name into a string */
1528 if (isPRINT_A(i)) {
1529 name[0] = i;
1530 name[1] = '\0';
1531 }
1532 else if (i == '\n') {
1533 my_strlcpy(name, "\n", sizeof(name));
1534 }
1535 else {
1536 my_strlcpy(name, "\t", sizeof(name));
1537 }
1538
1539 /* Check each possibe class */
1540 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1541 is_bad = TRUE;
1542 DEBUG_L(PerlIO_printf(Perl_debug_log,
1543 "isalnum('%s') unexpectedly is %d\n",
1544 name, cBOOL(isalnum(i))));
1545 }
1546 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1547 is_bad = TRUE;
1548 DEBUG_L(PerlIO_printf(Perl_debug_log,
1549 "isalpha('%s') unexpectedly is %d\n",
1550 name, cBOOL(isalpha(i))));
1551 }
1552 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1553 is_bad = TRUE;
1554 DEBUG_L(PerlIO_printf(Perl_debug_log,
1555 "isdigit('%s') unexpectedly is %d\n",
1556 name, cBOOL(isdigit(i))));
1557 }
1558 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1559 is_bad = TRUE;
1560 DEBUG_L(PerlIO_printf(Perl_debug_log,
1561 "isgraph('%s') unexpectedly is %d\n",
1562 name, cBOOL(isgraph(i))));
1563 }
1564 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1565 is_bad = TRUE;
1566 DEBUG_L(PerlIO_printf(Perl_debug_log,
1567 "islower('%s') unexpectedly is %d\n",
1568 name, cBOOL(islower(i))));
1569 }
1570 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1571 is_bad = TRUE;
1572 DEBUG_L(PerlIO_printf(Perl_debug_log,
1573 "isprint('%s') unexpectedly is %d\n",
1574 name, cBOOL(isprint(i))));
1575 }
1576 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1577 is_bad = TRUE;
1578 DEBUG_L(PerlIO_printf(Perl_debug_log,
1579 "ispunct('%s') unexpectedly is %d\n",
1580 name, cBOOL(ispunct(i))));
1581 }
1582 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1583 is_bad = TRUE;
1584 DEBUG_L(PerlIO_printf(Perl_debug_log,
1585 "isspace('%s') unexpectedly is %d\n",
1586 name, cBOOL(isspace(i))));
1587 }
1588 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1589 is_bad = TRUE;
1590 DEBUG_L(PerlIO_printf(Perl_debug_log,
1591 "isupper('%s') unexpectedly is %d\n",
1592 name, cBOOL(isupper(i))));
1593 }
1594 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1595 is_bad = TRUE;
1596 DEBUG_L(PerlIO_printf(Perl_debug_log,
1597 "isxdigit('%s') unexpectedly is %d\n",
1598 name, cBOOL(isxdigit(i))));
1599 }
1600 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1601 is_bad = TRUE;
1602 DEBUG_L(PerlIO_printf(Perl_debug_log,
1603 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1604 name, tolower(i), (int) toLOWER_A(i)));
1605 }
1606 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1607 is_bad = TRUE;
1608 DEBUG_L(PerlIO_printf(Perl_debug_log,
1609 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1610 name, toupper(i), (int) toUPPER_A(i)));
1611 }
1612 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1613 is_bad = TRUE;
1614 DEBUG_L(PerlIO_printf(Perl_debug_log,
1615 "'\\n' (=%02X) is not a control\n", (int) i));
1616 }
1617
1618 /* Add to the list; Separate multiple entries with a blank */
1619 if (is_bad) {
1620 if (bad_count) {
1621 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1622 }
8b7358b9
KW
1623 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1624 bad_count++;
8c6180a9
KW
1625 }
1626 }
1627 }
1628
7d4bcc4a
KW
1629# ifdef MB_CUR_MAX
1630
8c6180a9 1631 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1632 * this locale requires more than one byte, there are going to be
8c6180a9 1633 * problems. */
9c8a6dc2
KW
1634 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1635 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1636 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1637
8b7358b9
KW
1638 if ( check_for_problems && MB_CUR_MAX > 1
1639 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1640
1641 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1642 * locale. Just assume that the implementation for them (plus
1643 * for POSIX) is correct and the > 1 value is spurious. (Since
1644 * these are specially handled to never be considered UTF-8
1645 * locales, as long as this is the only problem, everything
1646 * should work fine */
1647 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1648 {
8c6180a9
KW
1649 multi_byte_locale = TRUE;
1650 }
7d4bcc4a
KW
1651
1652# endif
8c6180a9 1653
8b7358b9
KW
1654 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1655 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1656 PL_warn_locale = Perl_newSVpvf(aTHX_
1657 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1658 " which have\nunexpected meanings: %s\nThe Perl program"
1659 " will use the expected meanings",
8b7358b9 1660 newctype, bad_chars_list);
8b7358b9
KW
1661 }
1662 else {
1663 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1664 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1665 newctype,
8c6180a9
KW
1666 (multi_byte_locale)
1667 ? " Some characters in it are not recognized by"
1668 " Perl."
1669 : "",
1670 (bad_count)
1671 ? "\nThe following characters (and maybe others)"
1672 " may not have the same meaning as the Perl"
1673 " program expects:\n"
1674 : "",
1675 (bad_count)
1676 ? bad_chars_list
1677 : ""
1678 );
8b7358b9
KW
1679 }
1680
b119c2be
KW
1681# ifdef HAS_NL_LANGINFO
1682
1683 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1684 /* parameter FALSE is a don't care here */
1685 my_nl_langinfo(PERL_CODESET, FALSE));
1686
1687# endif
1688
8b7358b9
KW
1689 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1690
cc9eaeb0 1691 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1692 * output the message now. If not in that scope, we save the
1693 * message to be output at the first operation using this locale,
1694 * if that actually happens. Most programs don't use locales, so
1695 * they are immune to bad ones. */
cc9eaeb0 1696 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1697
780fcc9f
KW
1698 /* The '0' below suppresses a bogus gcc compiler warning */
1699 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1700
bddebb56
KW
1701 if (IN_LC(LC_CTYPE)) {
1702 SvREFCNT_dec_NN(PL_warn_locale);
1703 PL_warn_locale = NULL;
1704 }
780fcc9f 1705 }
baa60164 1706 }
31f05a37 1707 }
98994639
HS
1708
1709#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1710
98994639
HS
1711}
1712
98994639 1713void
2726666d
KW
1714Perl__warn_problematic_locale()
1715{
2726666d
KW
1716
1717#ifdef USE_LOCALE_CTYPE
1718
5f04a188
KW
1719 dTHX;
1720
1721 /* Internal-to-core function that outputs the message in PL_warn_locale,
1722 * and then NULLS it. Should be called only through the macro
1723 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1724
2726666d 1725 if (PL_warn_locale) {
2726666d
KW
1726 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1727 SvPVX(PL_warn_locale),
1728 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1729 SvREFCNT_dec_NN(PL_warn_locale);
1730 PL_warn_locale = NULL;
1731 }
1732
1733#endif
1734
1735}
1736
a4f00dcc
KW
1737STATIC void
1738S_new_collate(pTHX_ const char *newcoll)
98994639 1739{
7d4bcc4a
KW
1740
1741#ifndef USE_LOCALE_COLLATE
1742
1743 PERL_UNUSED_ARG(newcoll);
1744 PERL_UNUSED_CONTEXT;
1745
1746#else
0d071d52 1747
291a84fb 1748 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1749 * core Perl this and that 'newcoll' is the name of the new locale.
1750 *
d35fca5f
KW
1751 * The design of locale collation is that every locale change is given an
1752 * index 'PL_collation_ix'. The first time a string particpates in an
1753 * operation that requires collation while locale collation is active, it
1754 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1755 * magic includes the collation index, and the transformation of the string
1756 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1757 * instead of the string itself. If a string changes, the magic is
1758 * cleared. The next time the locale changes, the index is incremented,
1759 * and so we know during a comparison that the transformation is not
1760 * necessarily still valid, and so is recomputed. Note that if the locale
1761 * changes enough times, the index could wrap (a U32), and it is possible
1762 * that a transformation would improperly be considered valid, leading to
1763 * an unlikely bug */
0d071d52 1764
98994639
HS
1765 if (! newcoll) {
1766 if (PL_collation_name) {
1767 ++PL_collation_ix;
1768 Safefree(PL_collation_name);
1769 PL_collation_name = NULL;
1770 }
1771 PL_collation_standard = TRUE;
00bf60ca 1772 is_standard_collation:
98994639
HS
1773 PL_collxfrm_base = 0;
1774 PL_collxfrm_mult = 2;
165a1c52 1775 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1776 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1777 PL_strxfrm_max_cp = 0;
98994639
HS
1778 return;
1779 }
1780
d35fca5f 1781 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1782 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1783 ++PL_collation_ix;
1784 Safefree(PL_collation_name);
1785 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1786 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1787 if (PL_collation_standard) {
1788 goto is_standard_collation;
1789 }
98994639 1790
165a1c52 1791 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1792 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1793 PL_strxfrm_max_cp = 0;
165a1c52 1794
59c018b9
KW
1795 /* A locale collation definition includes primary, secondary, tertiary,
1796 * etc. weights for each character. To sort, the primary weights are
1797 * used, and only if they compare equal, then the secondary weights are
1798 * used, and only if they compare equal, then the tertiary, etc.
1799 *
1800 * strxfrm() works by taking the input string, say ABC, and creating an
1801 * output transformed string consisting of first the primary weights,
1802 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1803 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1804 * may not have weights at every level. In our example, let's say B
1805 * doesn't have a tertiary weight, and A doesn't have a secondary
1806 * weight. The constructed string is then going to be
1807 * A¹B¹C¹ B²C² A³C³ ....
1808 * This has the desired effect that strcmp() will look at the secondary
1809 * or tertiary weights only if the strings compare equal at all higher
1810 * priority weights. The spaces shown here, like in
c342d20e 1811 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1812 * are not just for readability. In the general case, these must
1813 * actually be bytes, which we will call here 'separator weights'; and
1814 * they must be smaller than any other weight value, but since these
1815 * are C strings, only the terminating one can be a NUL (some
1816 * implementations may include a non-NUL separator weight just before
1817 * the NUL). Implementations tend to reserve 01 for the separator
1818 * weights. They are needed so that a shorter string's secondary
1819 * weights won't be misconstrued as primary weights of a longer string,
1820 * etc. By making them smaller than any other weight, the shorter
1821 * string will sort first. (Actually, if all secondary weights are
1822 * smaller than all primary ones, there is no need for a separator
1823 * weight between those two levels, etc.)
1824 *
1825 * The length of the transformed string is roughly a linear function of
1826 * the input string. It's not exactly linear because some characters
1827 * don't have weights at all levels. When we call strxfrm() we have to
1828 * allocate some memory to hold the transformed string. The
1829 * calculations below try to find coefficients 'm' and 'b' for this
1830 * locale so that m*x + b equals how much space we need, given the size
1831 * of the input string in 'x'. If we calculate too small, we increase
1832 * the size as needed, and call strxfrm() again, but it is better to
1833 * get it right the first time to avoid wasted expensive string
1834 * transformations. */
1835
98994639 1836 {
79f120c8
KW
1837 /* We use the string below to find how long the tranformation of it
1838 * is. Almost all locales are supersets of ASCII, or at least the
1839 * ASCII letters. We use all of them, half upper half lower,
1840 * because if we used fewer, we might hit just the ones that are
1841 * outliers in a particular locale. Most of the strings being
1842 * collated will contain a preponderance of letters, and even if
1843 * they are above-ASCII, they are likely to have the same number of
1844 * weight levels as the ASCII ones. It turns out that digits tend
1845 * to have fewer levels, and some punctuation has more, but those
1846 * are relatively sparse in text, and khw believes this gives a
1847 * reasonable result, but it could be changed if experience so
1848 * dictates. */
1849 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1850 char * x_longer; /* Transformed 'longer' */
1851 Size_t x_len_longer; /* Length of 'x_longer' */
1852
1853 char * x_shorter; /* We also transform a substring of 'longer' */
1854 Size_t x_len_shorter;
1855
a4a439fb 1856 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1857 * are interested only in its length). It is used because it has
1858 * the intelligence to handle all cases, but to work, it needs some
1859 * values of 'm' and 'b' to get it started. For the purposes of
1860 * this calculation we use a very conservative estimate of 'm' and
1861 * 'b'. This assumes a weight can be multiple bytes, enough to
1862 * hold any UV on the platform, and there are 5 levels, 4 weight
1863 * bytes, and a trailing NUL. */
1864 PL_collxfrm_base = 5;
1865 PL_collxfrm_mult = 5 * sizeof(UV);
1866
1867 /* Find out how long the transformation really is */
a4a439fb
KW
1868 x_longer = _mem_collxfrm(longer,
1869 sizeof(longer) - 1,
1870 &x_len_longer,
1871
1872 /* We avoid converting to UTF-8 in the
1873 * called function by telling it the
1874 * string is in UTF-8 if the locale is a
1875 * UTF-8 one. Since the string passed
1876 * here is invariant under UTF-8, we can
1877 * claim it's UTF-8 even though it isn't.
1878 * */
1879 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1880 Safefree(x_longer);
1881
1882 /* Find out how long the transformation of a substring of 'longer'
1883 * is. Together the lengths of these transformations are
1884 * sufficient to calculate 'm' and 'b'. The substring is all of
1885 * 'longer' except the first character. This minimizes the chances
1886 * of being swayed by outliers */
a4a439fb 1887 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1888 sizeof(longer) - 2,
a4a439fb
KW
1889 &x_len_shorter,
1890 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1891 Safefree(x_shorter);
1892
1893 /* If the results are nonsensical for this simple test, the whole
1894 * locale definition is suspect. Mark it so that locale collation
1895 * is not active at all for it. XXX Should we warn? */
1896 if ( x_len_shorter == 0
1897 || x_len_longer == 0
1898 || x_len_shorter >= x_len_longer)
1899 {
1900 PL_collxfrm_mult = 0;
1901 PL_collxfrm_base = 0;
1902 }
1903 else {
1904 SSize_t base; /* Temporary */
1905
1906 /* We have both: m * strlen(longer) + b = x_len_longer
1907 * m * strlen(shorter) + b = x_len_shorter;
1908 * subtracting yields:
1909 * m * (strlen(longer) - strlen(shorter))
1910 * = x_len_longer - x_len_shorter
1911 * But we have set things up so that 'shorter' is 1 byte smaller
1912 * than 'longer'. Hence:
1913 * m = x_len_longer - x_len_shorter
1914 *
1915 * But if something went wrong, make sure the multiplier is at
1916 * least 1.
1917 */
1918 if (x_len_longer > x_len_shorter) {
1919 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1920 }
1921 else {
1922 PL_collxfrm_mult = 1;
1923 }
1924
1925 /* mx + b = len
1926 * so: b = len - mx
1927 * but in case something has gone wrong, make sure it is
1928 * non-negative */
1929 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1930 if (base < 0) {
1931 base = 0;
1932 }
1933
1934 /* Add 1 for the trailing NUL */
1935 PL_collxfrm_base = base + 1;
1936 }
58eebef2 1937
7d4bcc4a
KW
1938# ifdef DEBUGGING
1939
58eebef2
KW
1940 if (DEBUG_L_TEST || debug_initialization) {
1941 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1942 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1943 "x_len_longer=%zu,"
1944 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1945 __FILE__, __LINE__,
1946 PL_in_utf8_COLLATE_locale,
1947 x_len_shorter, x_len_longer,
1948 PL_collxfrm_mult, PL_collxfrm_base);
1949 }
7d4bcc4a
KW
1950# endif
1951
98994639
HS
1952 }
1953 }
1954
1955#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1956
98994639
HS
1957}
1958
d2b24094 1959#ifdef WIN32
b8cc575c 1960
a4f00dcc 1961STATIC char *
b8cc575c 1962S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1963{
1964 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1965 * difference between the two unless the input locale is "", which normally
1966 * means on Windows to get the machine default, which is set via the
1967 * computer's "Regional and Language Options" (or its current equivalent).
1968 * In POSIX, it instead means to find the locale from the user's
1969 * environment. This routine changes the Windows behavior to first look in
1970 * the environment, and, if anything is found, use that instead of going to
1971 * the machine default. If there is no environment override, the machine
1972 * default is used, by calling the real setlocale() with "".
1973 *
1974 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1975 * use the particular category's variable if set; otherwise to use the LANG
1976 * variable. */
b385bb4d 1977
175c4cf9 1978 bool override_LC_ALL = FALSE;
89f7b9aa 1979 char * result;
e5f10d49 1980 unsigned int i;
89f7b9aa 1981
b385bb4d 1982 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1983
1984# ifdef LC_ALL
1985
b385bb4d
KW
1986 locale = PerlEnv_getenv("LC_ALL");
1987 if (! locale) {
e5f10d49
KW
1988 if (category == LC_ALL) {
1989 override_LC_ALL = TRUE;
1990 }
1991 else {
7d4bcc4a
KW
1992
1993# endif
7d4bcc4a 1994
e5f10d49
KW
1995 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1996 if (category == categories[i]) {
1997 locale = PerlEnv_getenv(category_names[i]);
1998 goto found_locale;
1999 }
2000 }
7d4bcc4a 2001
b385bb4d 2002 locale = PerlEnv_getenv("LANG");
481465ea 2003 if (! locale) {
b385bb4d
KW
2004 locale = "";
2005 }
e5f10d49
KW
2006
2007 found_locale: ;
7d4bcc4a
KW
2008
2009# ifdef LC_ALL
2010
e5f10d49 2011 }
b385bb4d 2012 }
7d4bcc4a
KW
2013
2014# endif
2015
b385bb4d
KW
2016 }
2017
89f7b9aa 2018 result = setlocale(category, locale);
48015184
KW
2019 DEBUG_L(STMT_START {
2020 dSAVE_ERRNO;
2021 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2022 setlocale_debug_string(category, locale, result));
2023 RESTORE_ERRNO;
2024 } STMT_END);
89f7b9aa 2025
481465ea 2026 if (! override_LC_ALL) {
89f7b9aa
KW
2027 return result;
2028 }
2029
dfd77d7a 2030 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2031 * LANG variable or the system default if there is no LANG. But these have
2032 * lower priority than the other LC_foo variables, so override it for each
2033 * one that is set. (If they are set to "", it means to use the same thing
2034 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2035
948523db 2036 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2037 result = PerlEnv_getenv(category_names[i]);
2038 if (result && strNE(result, "")) {
2039 setlocale(categories[i], result);
2040 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2041 __FILE__, __LINE__,
2042 setlocale_debug_string(categories[i], result, "not captured")));
2043 }
89f7b9aa 2044 }
7d4bcc4a 2045
bbc98134 2046 result = setlocale(LC_ALL, NULL);
48015184
KW
2047 DEBUG_L(STMT_START {
2048 dSAVE_ERRNO;
2049 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2050 __FILE__, __LINE__,
48015184
KW
2051 setlocale_debug_string(LC_ALL, NULL, result));
2052 RESTORE_ERRNO;
2053 } STMT_END);
89f7b9aa 2054
bbc98134 2055 return result;
b385bb4d
KW
2056}
2057
2058#endif
2059
9aac5db8
KW
2060/*
2061
2062=head1 Locale-related functions and macros
2063
2064=for apidoc Perl_setlocale
2065
2066This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2067taking the same parameters, and returning the same information, except that it
2068returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2069perl keeps that locale category as C<C>, changing it briefly during the
2070operations where the underlying one is required.
2071
e9bc6d6b 2072Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2073declared to return S<C<const char *>>, whereas the system setlocale omits the
2074C<const>. (If it were being written today, plain setlocale would be declared
2075const, since it is illegal to change the information it returns; doing so leads
2076to segfaults.)
2077
e9bc6d6b
KW
2078Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2079C<setlocale> can be completely ineffective on some platforms under some
2080configurations.
2081
9aac5db8 2082C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2083where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2084the system C<setlocale()> is ineffective, returning the wrong information, and
2085failing to actually change the locale. C<Perl_setlocale>, however works
2086properly in all circumstances.
9aac5db8
KW
2087
2088The return points to a per-thread static buffer, which is overwritten the next
2089time C<Perl_setlocale> is called from the same thread.
2090
2091=cut
2092
2093*/
2094
2095const char *
2096Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2097{
2098 /* This wraps POSIX::setlocale() */
2099
9aac5db8
KW
2100 const char * retval;
2101 const char * newlocale;
2102 dSAVEDERRNO;
2103 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2104 dTHX;
2105
a4f00dcc
KW
2106#ifdef USE_LOCALE_NUMERIC
2107
291a84fb
KW
2108 /* A NULL locale means only query what the current one is. We have the
2109 * LC_NUMERIC name saved, because we are normally switched into the C
2110 * locale for it. For an LC_ALL query, switch back to get the correct
2111 * results. All other categories don't require special handling */
a4f00dcc
KW
2112 if (locale == NULL) {
2113 if (category == LC_NUMERIC) {
9aac5db8
KW
2114
2115 /* We don't have to copy this return value, as it is a per-thread
2116 * variable, and won't change until a future setlocale */
2117 return PL_numeric_name;
a4f00dcc
KW
2118 }
2119
7d4bcc4a 2120# ifdef LC_ALL
a4f00dcc 2121
9aac5db8
KW
2122 else if (category == LC_ALL) {
2123 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2124 }
2125
7d4bcc4a 2126# endif
a4f00dcc
KW
2127
2128 }
2129
2130#endif
2131
33e5a354
KW
2132 retval = save_to_buffer(do_setlocale_r(category, locale),
2133 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2134 SAVE_ERRNO;
2135
2136#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2137
2138 if (locale == NULL && category == LC_ALL) {
2139 RESTORE_LC_NUMERIC();
2140 }
2141
2142#endif
a4f00dcc
KW
2143
2144 DEBUG_L(PerlIO_printf(Perl_debug_log,
2145 "%s:%d: %s\n", __FILE__, __LINE__,
2146 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2147
9aac5db8
KW
2148 RESTORE_ERRNO;
2149
2150 if (! retval) {
a4f00dcc
KW
2151 return NULL;
2152 }
2153
9aac5db8 2154 /* If locale == NULL, we are just querying the state */
a4f00dcc 2155 if (locale == NULL) {
a4f00dcc
KW
2156 return retval;
2157 }
a4f00dcc 2158
1159483a
KW
2159 /* Now that have switched locales, we have to update our records to
2160 * correspond. */
a4f00dcc 2161
1159483a 2162 switch (category) {
a4f00dcc 2163
1159483a 2164#ifdef USE_LOCALE_CTYPE
a4f00dcc 2165
1159483a
KW
2166 case LC_CTYPE:
2167 new_ctype(retval);
2168 break;
a4f00dcc 2169
1159483a 2170#endif
a4f00dcc
KW
2171#ifdef USE_LOCALE_COLLATE
2172
1159483a
KW
2173 case LC_COLLATE:
2174 new_collate(retval);
2175 break;
a4f00dcc 2176
1159483a 2177#endif
a4f00dcc
KW
2178#ifdef USE_LOCALE_NUMERIC
2179
1159483a
KW
2180 case LC_NUMERIC:
2181 new_numeric(retval);
2182 break;
a4f00dcc 2183
1159483a
KW
2184#endif
2185#ifdef LC_ALL
a4f00dcc 2186
1159483a 2187 case LC_ALL:
a4f00dcc 2188
1159483a
KW
2189 /* LC_ALL updates all the things we care about. The values may not
2190 * be the same as 'retval', as the locale "" may have set things
2191 * individually */
a4f00dcc 2192
1159483a 2193# ifdef USE_LOCALE_CTYPE
a4f00dcc 2194
1159483a
KW
2195 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2196 new_ctype(newlocale);
a4f00dcc 2197
1159483a
KW
2198# endif /* USE_LOCALE_CTYPE */
2199# ifdef USE_LOCALE_COLLATE
2200
2201 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2202 new_collate(newlocale);
a4f00dcc 2203
7d4bcc4a 2204# endif
1159483a 2205# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2206
1159483a
KW
2207 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2208 new_numeric(newlocale);
a4f00dcc 2209
1159483a
KW
2210# endif /* USE_LOCALE_NUMERIC */
2211#endif /* LC_ALL */
a4f00dcc 2212
1159483a
KW
2213 default:
2214 break;
a4f00dcc
KW
2215 }
2216
2217 return retval;
2218
f7416781
KW
2219}
2220
2221PERL_STATIC_INLINE const char *
2222S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2223{
2224 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2225 * growing it if necessary */
2226
0b6697c5 2227 Size_t string_size;
f7416781
KW
2228
2229 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2230
0b6697c5
KW
2231 if (! string) {
2232 return NULL;
2233 }
2234
2235 string_size = strlen(string) + offset + 1;
2236
f7416781
KW
2237 if (*buf_size == 0) {
2238 Newx(*buf, string_size, char);
2239 *buf_size = string_size;
2240 }
2241 else if (string_size > *buf_size) {
2242 Renew(*buf, string_size, char);
2243 *buf_size = string_size;
2244 }
2245
2246 Copy(string, *buf + offset, string_size - offset, char);
2247 return *buf;
2248}
2249
2250/*
2251
f7416781
KW
2252=for apidoc Perl_langinfo
2253
ce181bf6 2254This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2255taking the same C<item> parameter values, and returning the same information.
2256But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2257of Perl's locale handling from your code, and can be used on systems that lack
2258a native C<nl_langinfo>.
2259
2260Expanding on these:
2261
2262=over
2263
2264=item *
2265
ce181bf6
KW
2266The reason it isn't quite a drop-in replacement is actually an advantage. The
2267only difference is that it returns S<C<const char *>>, whereas plain
2268C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2269forbidden to write into the buffer. By declaring this C<const>, the compiler
2270enforces this restriction, so if it is violated, you know at compilation time,
2271rather than getting segfaults at runtime.
2272
2273=item *
2274
69e2275e 2275It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2276without you having to write extra code. The reason for the extra code would be
2277because these are from the C<LC_NUMERIC> locale category, which is normally
2278kept set to the C locale by Perl, no matter what the underlying locale is
2279supposed to be, and so to get the expected results, you have to temporarily
ce181bf6
KW
2280toggle into the underlying locale, and later toggle back. (You could use plain
2281C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2282then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2283C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2284radix (decimal point) character to be a dot.)
2285
2286=item *
2287
2288The system function it replaces can have its static return buffer trashed,
2289not only by a subesequent call to that function, but by a C<freelocale>,
2290C<setlocale>, or other locale change. The returned buffer of this function is
2291not changed until the next call to it, so the buffer is never in a trashed
2292state.
f7416781
KW
2293
2294=item *
2295
ce181bf6
KW
2296Its return buffer is per-thread, so it also is never overwritten by a call to
2297this function from another thread; unlike the function it replaces.
2298
2299=item *
2300
2301But most importantly, it works on systems that don't have C<nl_langinfo>, such
2302as Windows, hence makes your code more portable. Of the fifty-some possible
2303items specified by the POSIX 2008 standard,
f7416781 2304L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
ce181bf6
KW
2305only two are completely unimplemented (though the loss of one of these is
2306significant). It uses various techniques to recover the other items, including
2307calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2308in C89, so should be always be available. Later C<strftime()> versions have
2309additional capabilities; C<""> is returned for those not available on your
2310system.
2311
2312It is important to note that when called with an item that is recovered by
2313using C<localeconv>, the buffer from any previous explicit call to
2314C<localeconv> will be overwritten. This means you must save that buffer's
2315contents if you need to access them after a call to this function.
9f0bc911 2316
f7416781
KW
2317The details for those items which may differ from what this emulation returns
2318and what a native C<nl_langinfo()> would return are:
2319
2320=over
2321
2322=item C<CODESET>
2323
2324=item C<ERA>
2325
2326Unimplemented, so returns C<"">.
2327
2328=item C<YESEXPR>
2329
c1566110
KW
2330=item C<YESSTR>
2331
f7416781
KW
2332=item C<NOEXPR>
2333
c1566110
KW
2334=item C<NOSTR>
2335
2336Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
ce181bf6
KW
2337removed from POSIX 2008, and are retained here for backwards compatibility.
2338Your platform's C<nl_langinfo> may not support them.
f7416781
KW
2339
2340=item C<D_FMT>
2341
2342Always evaluates to C<%x>, the locale's appropriate date representation.
2343
2344=item C<T_FMT>
2345
2346Always evaluates to C<%X>, the locale's appropriate time representation.
2347
2348=item C<D_T_FMT>
2349
2350Always evaluates to C<%c>, the locale's appropriate date and time
2351representation.
2352
2353=item C<CRNCYSTR>
2354
2355The return may be incorrect for those rare locales where the currency symbol
2356replaces the radix character.
2357Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2358to work differently.
2359
2360=item C<ALT_DIGITS>
2361
2362Currently this gives the same results as Linux does.
2363Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2364to work differently.
2365
2366=item C<ERA_D_FMT>
2367
2368=item C<ERA_T_FMT>
2369
2370=item C<ERA_D_T_FMT>
2371
2372=item C<T_FMT_AMPM>
2373
2374These are derived by using C<strftime()>, and not all versions of that function
2375know about them. C<""> is returned for these on such systems.
2376
2377=back
2378
ce181bf6
KW
2379=back
2380
f7416781
KW
2381When using C<Perl_langinfo> on systems that don't have a native
2382C<nl_langinfo()>, you must
2383
2384 #include "perl_langinfo.h"
2385
2386before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2387C<#include> with this one. (Doing it this way keeps out the symbols that plain
2388C<langinfo.h> imports into the namespace for code that doesn't need it.)
2389
2390You also should not use the bare C<langinfo.h> item names, but should preface
2391them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2392The C<PERL_I<foo>> versions will also work for this function on systems that do
2393have a native C<nl_langinfo>.
2394
f7416781
KW
2395The original impetus for C<Perl_langinfo()> was so that code that needs to
2396find out the current currency symbol, floating point radix character, or digit
2397grouping separator can use, on all systems, the simpler and more
2398thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2399pain to make thread-friendly. For other fields returned by C<localeconv>, it
2400is better to use the methods given in L<perlcall> to call
2401L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2402
2403=cut
2404
2405*/
2406
2407const char *
2408#ifdef HAS_NL_LANGINFO
2409Perl_langinfo(const nl_item item)
2410#else
2411Perl_langinfo(const int item)
2412#endif
2413{
f61748ac
KW
2414 return my_nl_langinfo(item, TRUE);
2415}
2416
2417const char *
2418#ifdef HAS_NL_LANGINFO
2419S_my_nl_langinfo(const nl_item item, bool toggle)
2420#else
2421S_my_nl_langinfo(const int item, bool toggle)
2422#endif
2423{
ae74815b 2424 dTHX;
0b6697c5 2425 const char * retval;
f7416781 2426
5a854ab3
KW
2427 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2428 * two items, and only if not already there */
2429 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2430 || PL_numeric_underlying))
2431 {
2432 toggle = FALSE;
2433 }
2434
ab340fff 2435#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d
KW
2436# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2437 || ! defined(HAS_POSIX_2008_LOCALE)
f7416781 2438
ab340fff 2439 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2440 * for those items dependent on it. This must be copied to a buffer before
2441 * switching back, as some systems destroy the buffer when setlocale() is
2442 * called */
f7416781 2443
038d3702
KW
2444 {
2445 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2446
b2fee59c 2447 if (toggle) {
038d3702 2448 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2449 }
f7416781 2450
5acc4454
KW
2451 LOCALE_LOCK; /* Prevent interference from another thread executing
2452 this code section (the only call to nl_langinfo in
2453 the core) */
2454
ee90eb2d 2455
628ff75a
KW
2456 /* Copy to a per-thread buffer, which is also one that won't be
2457 * destroyed by a subsequent setlocale(), such as the
2458 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2459 retval = save_to_buffer(nl_langinfo(item),
2460 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2461
5acc4454
KW
2462 LOCALE_UNLOCK;
2463
b2fee59c 2464 if (toggle) {
038d3702 2465 RESTORE_LC_NUMERIC();
b2fee59c 2466 }
f7416781
KW
2467 }
2468
ab340fff
KW
2469# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2470
5a854ab3 2471 {
b2fee59c
KW
2472 bool do_free = FALSE;
2473 locale_t cur = uselocale((locale_t) 0);
ab340fff 2474
b2fee59c
KW
2475 if (cur == LC_GLOBAL_LOCALE) {
2476 cur = duplocale(LC_GLOBAL_LOCALE);
2477 do_free = TRUE;
2478 }
ab340fff 2479
b2fee59c 2480 if (toggle) {
e1aa2579
KW
2481 if (PL_underlying_numeric_obj) {
2482 cur = PL_underlying_numeric_obj;
2483 }
2484 else {
2485 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2486 do_free = TRUE;
2487 }
b2fee59c 2488 }
ab340fff 2489
628ff75a
KW
2490 /* We have to save it to a buffer, because the freelocale() just below
2491 * can invalidate the internal one */
0b6697c5
KW
2492 retval = save_to_buffer(nl_langinfo_l(item, cur),
2493 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2494
b2fee59c
KW
2495 if (do_free) {
2496 freelocale(cur);
2497 }
5a854ab3 2498 }
ab340fff 2499
c1566110
KW
2500# endif
2501
0b6697c5 2502 if (strEQ(retval, "")) {
c1566110
KW
2503 if (item == PERL_YESSTR) {
2504 return "yes";
2505 }
2506 if (item == PERL_NOSTR) {
2507 return "no";
2508 }
2509 }
2510
0b6697c5 2511 return retval;
ab340fff 2512
f7416781 2513#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2514
2515 {
2516
f7416781
KW
2517# ifdef HAS_LOCALECONV
2518
43dd6b15 2519 const struct lconv* lc;
628ff75a 2520 const char * temp;
038d3702 2521 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
2522
2523# endif
2524# ifdef HAS_STRFTIME
2525
43dd6b15
KW
2526 struct tm tm;
2527 bool return_format = FALSE; /* Return the %format, not the value */
2528 const char * format;
f7416781
KW
2529
2530# endif
2531
43dd6b15
KW
2532 /* We copy the results to a per-thread buffer, even if not
2533 * multi-threaded. This is in part to simplify this code, and partly
2534 * because we need a buffer anyway for strftime(), and partly because a
2535 * call of localeconv() could otherwise wipe out the buffer, and the
2536 * programmer would not be expecting this, as this is a nl_langinfo()
2537 * substitute after all, so s/he might be thinking their localeconv()
2538 * is safe until another localeconv() call. */
f7416781 2539
43dd6b15
KW
2540 switch (item) {
2541 Size_t len;
f7416781 2542
43dd6b15
KW
2543 /* These 2 are unimplemented */
2544 case PERL_CODESET:
2545 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 2546
43dd6b15
KW
2547 default:
2548 return "";
f7416781 2549
43dd6b15
KW
2550 /* We use only an English set, since we don't know any more */
2551 case PERL_YESEXPR: return "^[+1yY]";
2552 case PERL_YESSTR: return "yes";
2553 case PERL_NOEXPR: return "^[-0nN]";
2554 case PERL_NOSTR: return "no";
f7416781
KW
2555
2556# ifdef HAS_LOCALECONV
2557
43dd6b15 2558 case PERL_CRNCYSTR:
f7416781 2559
43dd6b15
KW
2560 /* We don't bother with localeconv_l() because any system that
2561 * has it is likely to also have nl_langinfo() */
291a84fb 2562
5acc4454
KW
2563 LOCALE_LOCK; /* Prevent interference with other threads
2564 using localeconv() */
2565
43dd6b15
KW
2566 lc = localeconv();
2567 if ( ! lc
2568 || ! lc->currency_symbol
2569 || strEQ("", lc->currency_symbol))
2570 {
2571 LOCALE_UNLOCK;
2572 return "";
2573 }
f7416781 2574
43dd6b15 2575 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2576 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2577 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2578 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2579 { /* khw couldn't figure out how the localedef specifications
2580 would show that the $ should replace the radix; this is
2581 just a guess as to how it might work.*/
a34edee3 2582 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2583 }
2584 else if (lc->p_cs_precedes) {
a34edee3 2585 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2586 }
2587 else {
a34edee3 2588 PL_langinfo_buf[0] = '+';
43dd6b15 2589 }
f7416781 2590
43dd6b15
KW
2591 LOCALE_UNLOCK;
2592 break;
f7416781 2593
43dd6b15
KW
2594 case PERL_RADIXCHAR:
2595 case PERL_THOUSEP:
f7416781 2596
43dd6b15 2597 if (toggle) {
038d3702 2598 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2599 }
f7416781 2600
5acc4454
KW
2601 LOCALE_LOCK; /* Prevent interference with other threads
2602 using localeconv() */
2603
43dd6b15
KW
2604 lc = localeconv();
2605 if (! lc) {
628ff75a 2606 temp = "";
33394adc 2607 }
43dd6b15 2608 else {
628ff75a 2609 temp = (item == PERL_RADIXCHAR)
43dd6b15
KW
2610 ? lc->decimal_point
2611 : lc->thousands_sep;
628ff75a
KW
2612 if (! temp) {
2613 temp = "";
43dd6b15
KW
2614 }
2615 }
f7416781 2616
0b6697c5
KW
2617 retval = save_to_buffer(temp, &PL_langinfo_buf,
2618 &PL_langinfo_bufsize, 0);
f7416781 2619
5acc4454
KW
2620 LOCALE_UNLOCK;
2621
43dd6b15 2622 if (toggle) {
038d3702 2623 RESTORE_LC_NUMERIC();
43dd6b15 2624 }
f7416781 2625
43dd6b15 2626 break;
f7416781
KW
2627
2628# endif
2629# ifdef HAS_STRFTIME
2630
43dd6b15
KW
2631 /* These are defined by C89, so we assume that strftime supports
2632 * them, and so are returned unconditionally; they may not be what
2633 * the locale actually says, but should give good enough results
2634 * for someone using them as formats (as opposed to trying to parse
2635 * them to figure out what the locale says). The other format
2636 * items are actually tested to verify they work on the platform */
2637 case PERL_D_FMT: return "%x";
2638 case PERL_T_FMT: return "%X";
2639 case PERL_D_T_FMT: return "%c";
2640
2641 /* These formats are only available in later strfmtime's */
2642 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2643 case PERL_T_FMT_AMPM:
2644
2645 /* The rest can be gotten from most versions of strftime(). */
2646 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2647 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2648 case PERL_ABDAY_7:
2649 case PERL_ALT_DIGITS:
2650 case PERL_AM_STR: case PERL_PM_STR:
2651 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2652 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2653 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2654 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2655 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2656 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2657 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2658 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2659 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2660 case PERL_MON_12:
2661
2662 LOCALE_LOCK;
2663
2664 init_tm(&tm); /* Precaution against core dumps */
2665 tm.tm_sec = 30;
2666 tm.tm_min = 30;
2667 tm.tm_hour = 6;
2668 tm.tm_year = 2017 - 1900;
2669 tm.tm_wday = 0;
2670 tm.tm_mon = 0;
2671 switch (item) {
2672 default:
2673 LOCALE_UNLOCK;
2674 Perl_croak(aTHX_
2675 "panic: %s: %d: switch case: %d problem",
2676 __FILE__, __LINE__, item);
2677 NOT_REACHED; /* NOTREACHED */
2678
2679 case PERL_PM_STR: tm.tm_hour = 18;
2680 case PERL_AM_STR:
2681 format = "%p";
2682 break;
2683
2684 case PERL_ABDAY_7: tm.tm_wday++;
2685 case PERL_ABDAY_6: tm.tm_wday++;
2686 case PERL_ABDAY_5: tm.tm_wday++;
2687 case PERL_ABDAY_4: tm.tm_wday++;
2688 case PERL_ABDAY_3: tm.tm_wday++;
2689 case PERL_ABDAY_2: tm.tm_wday++;
2690 case PERL_ABDAY_1:
2691 format = "%a";
2692 break;
2693
2694 case PERL_DAY_7: tm.tm_wday++;
2695 case PERL_DAY_6: tm.tm_wday++;
2696 case PERL_DAY_5: tm.tm_wday++;
2697 case PERL_DAY_4: tm.tm_wday++;
2698 case PERL_DAY_3: tm.tm_wday++;
2699 case PERL_DAY_2: tm.tm_wday++;
2700 case PERL_DAY_1:
2701 format = "%A";
2702 break;
2703
2704 case PERL_ABMON_12: tm.tm_mon++;
2705 case PERL_ABMON_11: tm.tm_mon++;
2706 case PERL_ABMON_10: tm.tm_mon++;
2707 case PERL_ABMON_9: tm.tm_mon++;
2708 case PERL_ABMON_8: tm.tm_mon++;
2709 case PERL_ABMON_7: tm.tm_mon++;
2710 case PERL_ABMON_6: tm.tm_mon++;
2711 case PERL_ABMON_5: tm.tm_mon++;
2712 case PERL_ABMON_4: tm.tm_mon++;
2713 case PERL_ABMON_3: tm.tm_mon++;
2714 case PERL_ABMON_2: tm.tm_mon++;
2715 case PERL_ABMON_1:
2716 format = "%b";
2717 break;
2718
2719 case PERL_MON_12: tm.tm_mon++;
2720 case PERL_MON_11: tm.tm_mon++;
2721 case PERL_MON_10: tm.tm_mon++;
2722 case PERL_MON_9: tm.tm_mon++;
2723 case PERL_MON_8: tm.tm_mon++;
2724 case PERL_MON_7: tm.tm_mon++;
2725 case PERL_MON_6: tm.tm_mon++;
2726 case PERL_MON_5: tm.tm_mon++;
2727 case PERL_MON_4: tm.tm_mon++;
2728 case PERL_MON_3: tm.tm_mon++;
2729 case PERL_MON_2: tm.tm_mon++;
2730 case PERL_MON_1:
2731 format = "%B";
2732 break;
2733
2734 case PERL_T_FMT_AMPM:
2735 format = "%r";
2736 return_format = TRUE;
2737 break;
2738
2739 case PERL_ERA_D_FMT:
2740 format = "%Ex";
2741 return_format = TRUE;
2742 break;
2743
2744 case PERL_ERA_T_FMT:
2745 format = "%EX";
2746 return_format = TRUE;
2747 break;
2748
2749 case PERL_ERA_D_T_FMT:
2750 format = "%Ec";
2751 return_format = TRUE;
2752 break;
2753
2754 case PERL_ALT_DIGITS:
2755 tm.tm_wday = 0;
2756 format = "%Ow"; /* Find the alternate digit for 0 */
2757 break;
2758 }
f7416781 2759
43dd6b15
KW
2760 /* We can't use my_strftime() because it doesn't look at
2761 * tm_wday */
2762 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2763 format, &tm))
2764 {
2765 /* A zero return means one of:
2766 * a) there wasn't enough space in PL_langinfo_buf
2767 * b) the format, like a plain %p, returns empty
2768 * c) it was an illegal format, though some
2769 * implementations of strftime will just return the
2770 * illegal format as a plain character sequence.
2771 *
2772 * To quickly test for case 'b)', try again but precede
2773 * the format with a plain character. If that result is
2774 * still empty, the problem is either 'a)' or 'c)' */
2775
2776 Size_t format_size = strlen(format) + 1;
2777 Size_t mod_size = format_size + 1;
2778 char * mod_format;
2779 char * temp_result;
2780
2781 Newx(mod_format, mod_size, char);
2782 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2783 *mod_format = ' ';
43dd6b15
KW
2784 my_strlcpy(mod_format + 1, format, mod_size);
2785 len = strftime(temp_result,
2786 PL_langinfo_bufsize,
2787 mod_format, &tm);
2788 Safefree(mod_format);
2789 Safefree(temp_result);
2790
2791 /* If 'len' is non-zero, it means that we had a case like
2792 * %p which means the current locale doesn't use a.m. or
2793 * p.m., and that is valid */
2794 if (len == 0) {
2795
2796 /* Here, still didn't work. If we get well beyond a
2797 * reasonable size, bail out to prevent an infinite
2798 * loop. */
2799
2800 if (PL_langinfo_bufsize > 100 * format_size) {
2801 *PL_langinfo_buf = '\0';
2802 }
2803 else {
2804 /* Double the buffer size to retry; Add 1 in case
2805 * original was 0, so we aren't stuck at 0. */
2806 PL_langinfo_bufsize *= 2;
2807 PL_langinfo_bufsize++;
2808 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2809 continue;
2810 }
2811 }
f7416781 2812
f7416781 2813 break;
43dd6b15 2814 }
f7416781 2815
43dd6b15
KW
2816 /* Here, we got a result.
2817 *
2818 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2819 * alternate format for wday 0. If the value is the same as
2820 * the normal 0, there isn't an alternate, so clear the buffer.
2821 * */
2822 if ( item == PERL_ALT_DIGITS
2823 && strEQ(PL_langinfo_buf, "0"))
2824 {
2825 *PL_langinfo_buf = '\0';
2826 }
f7416781 2827
43dd6b15
KW
2828 /* ALT_DIGITS is problematic. Experiments on it showed that
2829 * strftime() did not always work properly when going from
2830 * alt-9 to alt-10. Only a few locales have this item defined,
2831 * and in all of them on Linux that khw was able to find,
2832 * nl_langinfo() merely returned the alt-0 character, possibly
2833 * doubled. Most Unicode digits are in blocks of 10
2834 * consecutive code points, so that is sufficient information
2835 * for those scripts, as we can infer alt-1, alt-2, .... But
2836 * for a Japanese locale, a CJK ideographic 0 is returned, and
2837 * the CJK digits are not in code point order, so you can't
2838 * really infer anything. The localedef for this locale did
2839 * specify the succeeding digits, so that strftime() works
2840 * properly on them, without needing to infer anything. But
2841 * the nl_langinfo() return did not give sufficient information
2842 * for the caller to understand what's going on. So until
2843 * there is evidence that it should work differently, this
2844 * returns the alt-0 string for ALT_DIGITS.
2845 *
2846 * wday was chosen because its range is all a single digit.
2847 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 2848
43dd6b15 2849 LOCALE_UNLOCK;
f7416781 2850
d1b150d9
KW
2851 retval = PL_langinfo_buf;
2852
43dd6b15
KW
2853 /* If to return the format, not the value, overwrite the buffer
2854 * with it. But some strftime()s will keep the original format
2855 * if illegal, so change those to "" */
2856 if (return_format) {
2857 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
2858 *PL_langinfo_buf = '\0';
2859 }
43dd6b15 2860 else {
0b6697c5
KW
2861 retval = save_to_buffer(format, &PL_langinfo_buf,
2862 &PL_langinfo_bufsize, 0);
f7416781
KW
2863 }
2864 }
2865
2866 break;
f7416781
KW
2867
2868# endif
2869
43dd6b15 2870 }
f7416781
KW
2871 }
2872
0b6697c5 2873 return retval;
f7416781
KW
2874
2875#endif
2876
a4f00dcc 2877}
b385bb4d 2878
98994639
HS
2879/*
2880 * Initialize locale awareness.
2881 */
2882int
2883Perl_init_i18nl10n(pTHX_ int printwarn)
2884{
0e92a118
KW
2885 /* printwarn is
2886 *
2887 * 0 if not to output warning when setup locale is bad
2888 * 1 if to output warning based on value of PERL_BADLANG
2889 * >1 if to output regardless of PERL_BADLANG
2890 *
2891 * returns
98994639 2892 * 1 = set ok or not applicable,
0e92a118
KW
2893 * 0 = fallback to a locale of lower priority
2894 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2895 *
2896 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2897 * set, debugging information is output.
2898 *
2899 * This looks more complicated than it is, mainly due to the #ifdefs.
2900 *
2901 * We try to set LC_ALL to the value determined by the environment. If
2902 * there is no LC_ALL on this platform, we try the individual categories we
2903 * know about. If this works, we are done.
2904 *
2905 * But if it doesn't work, we have to do something else. We search the
2906 * environment variables ourselves instead of relying on the system to do
2907 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2908 * think there is one), and the ultimate fallback "C". This is all done in
2909 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2910 * more complex. The 'trial_locales' array is initialized with just one
2911 * element; it causes the behavior described in the paragraph above this to
2912 * happen. If that fails, we add elements to 'trial_locales', and do extra
2913 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2914 *
2915 * On Ultrix, the locale MUST come from the environment, so there is
2916 * preliminary code to set it. I (khw) am not sure that it is necessary,
2917 * and that this couldn't be folded into the loop, but barring any real
2918 * platforms to test on, it's staying as-is
2919 *
2920 * A slight complication is that in embedded Perls, the locale may already
2921 * be set-up, and we don't want to get it from the normal environment
2922 * variables. This is handled by having a special environment variable
2923 * indicate we're in this situation. We simply set setlocale's 2nd
2924 * parameter to be a NULL instead of "". That indicates to setlocale that
2925 * it is not to change anything, but to return the current value,
2926 * effectively initializing perl's db to what the locale already is.
2927 *
2928 * We play the same trick with NULL if a LC_ALL succeeds. We call
2929 * setlocale() on the individual categores with NULL to get their existing
2930 * values for our db, instead of trying to change them.
2931 * */
98994639 2932
0e92a118
KW
2933 int ok = 1;
2934
7d4bcc4a
KW
2935#ifndef USE_LOCALE
2936
2937 PERL_UNUSED_ARG(printwarn);
2938
2939#else /* USE_LOCALE */
7d4bcc4a
KW
2940# ifdef __GLIBC__
2941
175c4cf9 2942 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2943
2944# endif
65ebb059 2945
ccd65d51
KW
2946 /* NULL uses the existing already set up locale */
2947 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2948 ? NULL
2949 : "";
c3fcd832
KW
2950 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2951 unsigned int trial_locales_count;
175c4cf9
KW
2952 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2953 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2954 bool setlocale_failure = FALSE;
65ebb059 2955 unsigned int i;
175c4cf9
KW
2956
2957 /* A later getenv() could zap this, so only use here */
2958 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2959
2960 const bool locwarn = (printwarn > 1
e5f10d49
KW
2961 || ( printwarn
2962 && ( ! bad_lang_use_once
22ff3130 2963 || (
e5f10d49
KW
2964 /* disallow with "" or "0" */
2965 *bad_lang_use_once
2966 && strNE("0", bad_lang_use_once)))));
ea92aad8 2967
291a84fb 2968 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2969 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2970
2971 /* current locale for given category; should have been copied so aren't
2972 * volatile */
8de4332b 2973 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2974
7d4bcc4a
KW
2975# ifdef WIN32
2976
6bce99ee
JH
2977 /* In some systems you can find out the system default locale
2978 * and use that as the fallback locale. */
7d4bcc4a
KW
2979# define SYSTEM_DEFAULT_LOCALE
2980# endif
2981# ifdef SYSTEM_DEFAULT_LOCALE
2982
65ebb059 2983 const char *system_default_locale = NULL;
98994639 2984
7d4bcc4a 2985# endif
948523db
KW
2986
2987# ifndef DEBUGGING
2988# define DEBUG_LOCALE_INIT(a,b,c)
2989# else
7d4bcc4a 2990
8298454c 2991 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2992
2993# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2994 STMT_START { \
2995 if (debug_initialization) { \
2996 PerlIO_printf(Perl_debug_log, \
2997 "%s:%d: %s\n", \
2998 __FILE__, __LINE__, \
a4f00dcc 2999 setlocale_debug_string(category, \
2fcc0ca9
KW
3000 locale, \
3001 result)); \
3002 } \
3003 } STMT_END
2fcc0ca9 3004
948523db
KW
3005/* Make sure the parallel arrays are properly set up */
3006# ifdef USE_LOCALE_NUMERIC
3007 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3008 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3009# ifdef USE_POSIX_2008_LOCALE
3010 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3011# endif
948523db
KW
3012# endif
3013# ifdef USE_LOCALE_CTYPE
3014 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3015 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3016# ifdef USE_POSIX_2008_LOCALE
3017 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3018# endif
948523db
KW
3019# endif
3020# ifdef USE_LOCALE_COLLATE
3021 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3022 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3023# ifdef USE_POSIX_2008_LOCALE
3024 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3025# endif
948523db
KW
3026# endif
3027# ifdef USE_LOCALE_TIME
3028 assert(categories[LC_TIME_INDEX] == LC_TIME);
3029 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3030# ifdef USE_POSIX_2008_LOCALE
3031 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3032# endif
948523db
KW
3033# endif
3034# ifdef USE_LOCALE_MESSAGES
3035 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3036 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3037# ifdef USE_POSIX_2008_LOCALE
3038 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3039# endif
948523db
KW
3040# endif
3041# ifdef USE_LOCALE_MONETARY
3042 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3043 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3044# ifdef USE_POSIX_2008_LOCALE
3045 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3046# endif
948523db 3047# endif
9821811f
KW
3048# ifdef USE_LOCALE_ADDRESS
3049 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3050 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3051# ifdef USE_POSIX_2008_LOCALE
3052 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3053# endif
9821811f
KW
3054# endif
3055# ifdef USE_LOCALE_IDENTIFICATION
3056 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3057 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3058# ifdef USE_POSIX_2008_LOCALE
3059 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3060# endif
9821811f
KW
3061# endif
3062# ifdef USE_LOCALE_MEASUREMENT
3063 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3064 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3065# ifdef USE_POSIX_2008_LOCALE
3066 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3067# endif
9821811f
KW
3068# endif
3069# ifdef USE_LOCALE_PAPER
3070 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3071 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3072# ifdef USE_POSIX_2008_LOCALE
3073 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3074# endif
9821811f
KW
3075# endif
3076# ifdef USE_LOCALE_TELEPHONE
3077 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3078 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3079# ifdef USE_POSIX_2008_LOCALE
3080 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3081# endif
9821811f 3082# endif
948523db
KW
3083# ifdef LC_ALL
3084 assert(categories[LC_ALL_INDEX] == LC_ALL);
3085 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3086 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3087# ifdef USE_POSIX_2008_LOCALE
3088 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3089# endif
948523db
KW
3090# endif
3091# endif /* DEBUGGING */
47280b20
KW
3092
3093 /* Initialize the cache of the program's UTF-8ness for the always known
3094 * locales C and POSIX */
3095 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3096 sizeof(PL_locale_utf8ness));
3097
e9bc6d6b
KW
3098# ifdef USE_THREAD_SAFE_LOCALE
3099# ifdef WIN32
3100
3101 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3102
3103# endif
3104# endif
3105# if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3106
3107 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3108 if (! PL_C_locale_obj) {
3109 Perl_croak_nocontext(
3110 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3111 }
3112 if (DEBUG_Lv_TEST || debug_initialization) {
3113 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3114 }
3115
3116# endif
3117
3ca88433
KW
3118 PL_numeric_radix_sv = newSVpvs(".");
3119
e9bc6d6b
KW
3120# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3121
3122 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3123 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3124
3125# endif
b56c4436 3126# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3127
3128 /*
3129 * Ultrix setlocale(..., "") fails if there are no environment
3130 * variables from which to get a locale name.
3131 */
3132
b56c4436
KW
3133# ifndef LC_ALL
3134# error Ultrix without LC_ALL not implemented
3135# else
7d4bcc4a 3136
b56c4436
KW
3137 {
3138 bool done = FALSE;
ea92aad8
KW
3139 if (lang) {
3140 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3141 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3142 if (sl_result[LC_ALL_INDEX])
3143 done = TRUE;
3144 else
e5f10d49 3145 setlocale_failure = TRUE;
ea92aad8
KW
3146 }
3147 if (! setlocale_failure) {
3148 const char * locale_param;
3149 for (i = 0; i < LC_ALL_INDEX; i++) {
3150 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3151 ? setlocale_init
3152 : NULL;
3153 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3154 if (! sl_result[i]) {
3155 setlocale_failure = TRUE;
3156 }
3157 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3158 }
c835d6be 3159 }
7d4bcc4a
KW
3160 }
3161
3162# endif /* LC_ALL */
e5f10d49 3163# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3164
65ebb059 3165 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3166 * the list. Normally the loop is executed just once. But if setting the
3167 * locale fails, inside the loop we add fallback trials to the array and so
3168 * will execute the loop multiple times */
c3fcd832
KW
3169 trial_locales[0] = setlocale_init;
3170 trial_locales_count = 1;
7d4bcc4a 3171
65ebb059
KW
3172 for (i= 0; i < trial_locales_count; i++) {
3173 const char * trial_locale = trial_locales[i];
3174
3175 if (i > 0) {
3176
3177 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3178 * when i==0, but I (khw) don't think that behavior makes much
3179 * sense */
3180 setlocale_failure = FALSE;
3181
7d4bcc4a 3182# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3183# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3184
65ebb059
KW
3185 /* On Windows machines, an entry of "" after the 0th means to use
3186 * the system default locale, which we now proceed to get. */
3187 if (strEQ(trial_locale, "")) {
3188 unsigned int j;
3189
3190 /* Note that this may change the locale, but we are going to do
3191 * that anyway just below */
837ce802 3192 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3193 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3194
7d4bcc4a 3195 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3196 * try */
3197 if (! system_default_locale) {
3198 goto next_iteration;
3199 }
3200 for (j = 0; j < trial_locales_count; j++) {
3201 if (strEQ(system_default_locale, trial_locales[j])) {
3202 goto next_iteration;
3203 }
3204 }
3205
3206 trial_locale = system_default_locale;
3207 }
ec0202b5
KW
3208# else
3209# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3210# endif
7d4bcc4a 3211# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3212
3213 } /* For i > 0 */
65ebb059 3214
7d4bcc4a
KW
3215# ifdef LC_ALL
3216
948523db
KW
3217 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3218 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3219 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3220 setlocale_failure = TRUE;
7cd8b568
KW
3221 }
3222 else {
3223 /* Since LC_ALL succeeded, it should have changed all the other
3224 * categories it can to its value; so we massage things so that the
3225 * setlocales below just return their category's current values.
3226 * This adequately handles the case in NetBSD where LC_COLLATE may
3227 * not be defined for a locale, and setting it individually will
7d4bcc4a 3228 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3229 * the POSIX locale. */
3230 trial_locale = NULL;
3231 }
7d4bcc4a
KW
3232
3233# endif /* LC_ALL */
98994639 3234
e5f10d49
KW
3235 if (! setlocale_failure) {
3236 unsigned int j;
3237 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3238 curlocales[j]
3239 = savepv(do_setlocale_r(categories[j], trial_locale));
3240 if (! curlocales[j]) {
3241 setlocale_failure = TRUE;
3242 }
3243 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3244 }
c835d6be 3245
e5f10d49
KW
3246 if (! setlocale_failure) { /* All succeeded */
3247 break; /* Exit trial_locales loop */
49c85077 3248 }
65ebb059 3249 }
98994639 3250
49c85077
KW
3251 /* Here, something failed; will need to try a fallback. */
3252 ok = 0;
65ebb059 3253
49c85077
KW
3254 if (i == 0) {
3255 unsigned int j;
98994639 3256
65ebb059 3257 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3258
3259# ifdef LC_ALL
98994639 3260
49c85077
KW
3261 PerlIO_printf(Perl_error_log,
3262 "perl: warning: Setting locale failed.\n");
98994639 3263
7d4bcc4a 3264# else /* !LC_ALL */
98994639 3265
49c85077
KW
3266 PerlIO_printf(Perl_error_log,
3267 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3268
e5f10d49
KW
3269 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3270 if (! curlocales[j]) {
3271 PerlIO_printf(Perl_error_log, category_names[j]);
3272 }
3273 else {
3274 Safefree(curlocales[j]);
3275 }
3276 }
7d4bcc4a 3277
7d4bcc4a 3278# endif /* LC_ALL */
98994639 3279
49c85077
KW
3280 PerlIO_printf(Perl_error_log,
3281 "perl: warning: Please check that your locale settings:\n");
98994639 3282
7d4bcc4a
KW
3283# ifdef __GLIBC__
3284
49c85077
KW
3285 PerlIO_printf(Perl_error_log,
3286 "\tLANGUAGE = %c%s%c,\n",
3287 language ? '"' : '(',
3288 language ? language : "unset",
3289 language ? '"' : ')');
7d4bcc4a 3290# endif
98994639 3291
49c85077
KW
3292 PerlIO_printf(Perl_error_log,
3293 "\tLC_ALL = %c%s%c,\n",
3294 lc_all ? '"' : '(',
3295 lc_all ? lc_all : "unset",
3296 lc_all ? '"' : ')');
98994639 3297
7d4bcc4a
KW
3298# if defined(USE_ENVIRON_ARRAY)
3299
49c85077 3300 {
cd999af9 3301 char **e;
d5e32b93
KW
3302
3303 /* Look through the environment for any variables of the
3304 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3305 * already handled above. These are assumed to be locale
3306 * settings. Output them and their values. */
cd999af9 3307 for (e = environ; *e; e++) {
d5e32b93
KW
3308 const STRLEN prefix_len = sizeof("LC_") - 1;
3309 STRLEN uppers_len;
3310
cd999af9 3311 if ( strBEGINs(*e, "LC_")
c8b388b0 3312 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3313 && (uppers_len = strspn(*e + prefix_len,
3314 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3315 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3316 {
3317 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3318 (int) (prefix_len + uppers_len), *e,
3319 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3320 }
3321 }
49c85077 3322 }
7d4bcc4a
KW
3323
3324# else
3325
49c85077
KW
3326 PerlIO_printf(Perl_error_log,
3327 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3328
3329# endif
98994639 3330
49c85077
KW
3331 PerlIO_printf(Perl_error_log,
3332 "\tLANG = %c%s%c\n",
3333 lang ? '"' : '(',
3334 lang ? lang : "unset",
3335 lang ? '"' : ')');
98994639 3336
49c85077
KW
3337 PerlIO_printf(Perl_error_log,
3338 " are supported and installed on your system.\n");
3339 }
98994639 3340
65ebb059 3341 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3342 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3343 * usually change the upper bound of the loop we are in.
3344 *
3345 * Since the system's default way of setting the locale has not
3346 * found one that works, We use Perl's defined ordering: LC_ALL,
3347 * LANG, and the C locale. We don't try the same locale twice, so
3348 * don't add to the list if already there. (On POSIX systems, the
3349 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3350 * but there's no harm done by doing it explicitly.
3351 *
3352 * Note that this tries the LC_ALL environment variable even on
3353 * systems which have no LC_ALL locale setting. This may or may
3354 * not have been originally intentional, but there's no real need
3355 * to change the behavior. */
65ebb059
KW
3356 if (lc_all) {
3357 for (j = 0; j < trial_locales_count; j++) {
3358 if (strEQ(lc_all, trial_locales[j])) {
3359 goto done_lc_all;
3360 }
3361 }
3362 trial_locales[trial_locales_count++] = lc_all;
3363 }
3364 done_lc_all:
98994639 3365
65ebb059
KW
3366 if (lang) {
3367 for (j = 0; j < trial_locales_count; j++) {
3368 if (strEQ(lang, trial_locales[j])) {
3369 goto done_lang;
3370 }
3371 }
3372 trial_locales[trial_locales_count++] = lang;
3373 }
3374 done_lang:
3375
7d4bcc4a
KW
3376# if defined(WIN32) && defined(LC_ALL)
3377
65ebb059
KW
3378 /* For Windows, we also try the system default locale before "C".
3379 * (If there exists a Windows without LC_ALL we skip this because
3380 * it gets too complicated. For those, the "C" is the next
3381 * fallback possibility). The "" is the same as the 0th element of
3382 * the array, but the code at the loop above knows to treat it
3383 * differently when not the 0th */
3384 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3385
3386# endif
65ebb059
KW
3387
3388 for (j = 0; j < trial_locales_count; j++) {
3389 if (strEQ("C", trial_locales[j])) {
3390 goto done_C;
3391 }
3392 }
3393 trial_locales[trial_locales_count++] = "C";
98994639 3394
65ebb059
KW
3395 done_C: ;
3396 } /* end of first time through the loop */
98994639 3397
7d4bcc4a
KW
3398# ifdef WIN32
3399
65ebb059 3400 next_iteration: ;
7d4bcc4a
KW
3401
3402# endif
65ebb059
KW
3403
3404 } /* end of looping through the trial locales */
3405
3406 if (ok < 1) { /* If we tried to fallback */
3407 const char* msg;
3408 if (! setlocale_failure) { /* fallback succeeded */
3409 msg = "Falling back to";
3410 }
3411 else { /* fallback failed */
e5f10d49 3412 unsigned int j;
98994639 3413
65ebb059
KW
3414 /* We dropped off the end of the loop, so have to decrement i to
3415 * get back to the value the last time through */
3416 i--;
98994639 3417
65ebb059
KW
3418 ok = -1;
3419 msg = "Failed to fall back to";
3420
3421 /* To continue, we should use whatever values we've got */
7d4bcc4a 3422
e5f10d49
KW
3423 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3424 Safefree(curlocales[j]);
3425 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3426 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3427 }
65ebb059
KW
3428 }
3429
3430 if (locwarn) {
3431 const char * description;
3432 const char * name = "";
3433 if (strEQ(trial_locales[i], "C")) {
3434 description = "the standard locale";
3435 name = "C";
3436 }
7d4bcc4a
KW
3437
3438# ifdef SYSTEM_DEFAULT_LOCALE
3439
65ebb059
KW
3440 else if (strEQ(trial_locales[i], "")) {
3441 description = "the system default locale";
3442 if (system_default_locale) {
3443 name = system_default_locale;
3444 }
3445 }
7d4bcc4a
KW
3446
3447# endif /* SYSTEM_DEFAULT_LOCALE */
3448
65ebb059
KW
3449 else {
3450 description = "a fallback locale";
3451 name = trial_locales[i];
3452 }
3453 if (name && strNE(name, "")) {
3454 PerlIO_printf(Perl_error_log,
3455 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3456 }
3457 else {
3458 PerlIO_printf(Perl_error_log,
3459 "perl: warning: %s %s.\n", msg, description);
3460 }
3461 }
3462 } /* End of tried to fallback */
98994639 3463
e5f10d49
KW
3464 /* Done with finding the locales; update our records */
3465
7d4bcc4a
KW
3466# ifdef USE_LOCALE_CTYPE
3467
948523db 3468 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3469
e5f10d49 3470# endif
7d4bcc4a
KW
3471# ifdef USE_LOCALE_COLLATE
3472
948523db 3473 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3474
e5f10d49 3475# endif
7d4bcc4a
KW
3476# ifdef USE_LOCALE_NUMERIC
3477
948523db 3478 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3479
3480# endif
3481
948523db 3482 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3483
e9bc6d6b 3484# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3485
3486 /* This caches whether each category's locale is UTF-8 or not. This
3487 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3488 * initialization time before any threads have started, but not later
3489 * unless thread-safe operations are used.
47280b20
KW
3490 * Caching means that if the program heeds our dictate not to change
3491 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3492 * it may get queried without having to change locales. If the
3493 * environment is such that all categories have the same locale, this
3494 * isn't needed, as the code will not change the locale; but this
3495 * handles the uncommon case where the environment has disparate
3496 * locales for the categories */
47280b20
KW
3497 (void) _is_cur_LC_category_utf8(categories[i]);
3498
3499# endif
3500
e5f10d49
KW
3501 Safefree(curlocales[i]);
3502 }
b310b053 3503
7d4bcc4a
KW
3504# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3505
49c85077 3506 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3507 * locale is UTF-8. The call to new_ctype() just above has already
3508 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3509 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3510 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3511 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3512 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3513
a05d7ebb 3514 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3515 This is an alternative to using the -C command line switch
3516 (the -C if present will override this). */
3517 {
dd374669 3518 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3519 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3520 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3521 PL_utf8cache = -1;
b310b053
JH
3522 }
3523
7d4bcc4a 3524# endif
7d4bcc4a
KW
3525# ifdef __GLIBC__
3526
175c4cf9 3527 Safefree(language);
7d4bcc4a
KW
3528
3529# endif
175c4cf9
KW
3530
3531 Safefree(lc_all);
3532 Safefree(lang);
3533
e3305790 3534#endif /* USE_LOCALE */
2fcc0ca9 3535#ifdef DEBUGGING
7d4bcc4a 3536
2fcc0ca9 3537 /* So won't continue to output stuff */
27cdc72e 3538 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3539
2fcc0ca9
KW
3540#endif
3541
98994639
HS
3542 return ok;
3543}
3544
98994639
HS
3545#ifdef USE_LOCALE_COLLATE
3546
a4a439fb 3547char *
a4a439fb
KW
3548Perl__mem_collxfrm(pTHX_ const char *input_string,
3549 STRLEN len, /* Length of 'input_string' */
3550 STRLEN *xlen, /* Set to length of returned string
3551 (not including the collation index
3552 prefix) */
3553 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3554 )
98994639 3555{
a4a439fb
KW
3556
3557 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3558 * differences. First, it handles embedded NULs. Second, it allocates a bit
3559 * more memory than needed for the transformed data itself. The real
55e5378d 3560 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3561 * the length of that, and doesn't include the collation index size.
3562 * Please see sv_collxfrm() to see how this is used. */
3563
55e5378d
KW
3564#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3565
6696cfa7
KW
3566 char * s = (char *) input_string;
3567 STRLEN s_strlen = strlen(input_string);
79f120c8 3568 char *xbuf = NULL;
55e5378d 3569 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3570 STRLEN length_in_chars;
c664130f 3571 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3572
a4a439fb
KW
3573 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3574
3575 /* Must be NUL-terminated */
3576 assert(*(input_string + len) == '\0');
7918f24d 3577
79f120c8
KW
3578 /* If this locale has defective collation, skip */
3579 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3580 DEBUG_L(PerlIO_printf(Perl_debug_log,
3581 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3582 goto bad;
3583 }
3584
6696cfa7
KW
3585 /* Replace any embedded NULs with the control that sorts before any others.
3586 * This will give as good as possible results on strings that don't
3587 * otherwise contain that character, but otherwise there may be
3588 * less-than-perfect results with that character and NUL. This is
fdc080f3 3589 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3590 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3591 NUL */
6696cfa7
KW
3592 char * e = s + len;
3593 char * sans_nuls;
fdc080f3 3594 STRLEN sans_nuls_len;
94762aa0 3595 int try_non_controls;
afc4976f
KW
3596 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3597 making sure 2nd byte is NUL.
3598 */
3599 STRLEN this_replacement_len;
3600
1e4c9676
KW
3601 /* If we don't know what non-NUL control character sorts lowest for
3602 * this locale, find it */
f28f4d2a 3603 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3604 int j;
afc4976f 3605 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3606 includes the collation index
3607 prefixed. */
3608
91c0e2e0 3609 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3610
3611 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3612 * NUL, in which case we instead look for any character. Controls
3613 * are preferred because collation order is, in general, context
3614 * sensitive, with adjoining characters affecting the order, and
3615 * controls are less likely to have such interactions, allowing the
3616 * NUL-replacement to stand on its own. (Another way to look at it
3617 * is to imagine what would happen if the NUL were replaced by a
3618 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3619 for (try_non_controls = 0;
3620 try_non_controls < 2;
3621 try_non_controls++)
3622 {
d4ff9586
KW
3623 /* Look through all legal code points (NUL isn't) */
3624 for (j = 1; j < 256; j++) {
3625 char * x; /* j's xfrm plus collation index */
3626 STRLEN x_len; /* length of 'x' */
3627 STRLEN trial_len = 1;
736a4fed 3628 char cur_source[] = { '\0', '\0' };
d4ff9586 3629
736a4fed
KW
3630 /* Skip non-controls the first time through the loop. The
3631 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3632 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3633 ? ! isCNTRL_L1(j)
3634 : ! isCNTRL_LC(j))
3635 {
d4ff9586 3636 continue;
6696cfa7 3637 }
6696cfa7 3638
736a4fed
KW
3639 /* Create a 1-char string of the current code point */
3640 cur_source[0] = (char) j;
3641
d4ff9586
KW
3642 /* Then transform it */
3643 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3644 0 /* The string is not in UTF-8 */);
6696cfa7 3645
1e4c9676 3646 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3647 * */
3648 if (! x) {
3649 continue;
3650 }
6696cfa7 3651
d4ff9586
KW
3652 /* If this character's transformation is lower than
3653 * the current lowest, this one becomes the lowest */
3654 if ( cur_min_x == NULL
3655 || strLT(x + COLLXFRM_HDR_LEN,
3656 cur_min_x + COLLXFRM_HDR_LEN))
3657 {
f28f4d2a 3658 PL_strxfrm_NUL_replacement = j;
d4ff9586 3659 cur_min_x = x;
d4ff9586
KW
3660 }
3661 else {
3662 Safefree(x);
3663 }
1e4c9676 3664 } /* end of loop through all 255 characters */
6696cfa7 3665
1e4c9676 3666 /* Stop looking if found */
94762aa0
KW
3667 if (cur_min_x) {
3668 break;
3669 }
3670
3671 /* Unlikely, but possible, if there aren't any controls that
3672 * work in the locale, repeat the loop, looking for any
3673 * character that works */
3674 DEBUG_L(PerlIO_printf(Perl_debug_log,
3675 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3676 } /* End of loop to try first the controls, then any char */
6696cfa7 3677
94762aa0
KW
3678 if (! cur_min_x) {
3679 DEBUG_L(PerlIO_printf(Perl_debug_log,
3680 "_mem_collxfrm: Couldn't find any character to replace"
3681 " embedded NULs in locale %s with", PL_collation_name));
3682 goto bad;
58eebef2
KW
3683 }
3684
94762aa0
KW
3685 DEBUG_L(PerlIO_printf(Perl_debug_log,
3686 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3687 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3688
6696cfa7 3689 Safefree(cur_min_x);
1e4c9676 3690 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3691
3692 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3693 * UTF8-ness of the original */
f28f4d2a
KW
3694 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3695 this_replacement_char[0] =
3696 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3697 this_replacement_char[1] =
3698 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3699 this_replacement_len = 2;
3700 }
3701 else {
f28f4d2a 3702 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3703 /* this_replacement_char[1] = '\0' was done at initialization */
3704 this_replacement_len = 1;
6696cfa7
KW
3705 }
3706
3707 /* The worst case length for the replaced string would be if every
3708 * character in it is NUL. Multiply that by the length of each
3709 * replacement, and allow for a trailing NUL */
afc4976f 3710 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3711 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3712 *sans_nuls = '\0';
3713
6696cfa7
KW
3714 /* Replace each NUL with the lowest collating control. Loop until have
3715 * exhausted all the NULs */
3716 while (s + s_strlen < e) {
6069d6c5 3717 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3718
3719 /* Do the actual replacement */
6069d6c5 3720 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
3721
3722 /* Move past the input NUL */
3723 s += s_strlen + 1;
3724 s_strlen = strlen(s);
3725 }
3726
3727 /* And add anything that trails the final NUL */
6069d6c5 3728 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3729
3730 /* Switch so below we transform this modified string */
3731 s = sans_nuls;
3732 len = strlen(s);
1e4c9676 3733 } /* End of replacing NULs */
6696cfa7 3734
a4a439fb
KW
3735 /* Make sure the UTF8ness of the string and locale match */
3736 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 3737 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
3738 const char * const t = s; /* Temporary so we can later find where the
3739 input was */
3740
3741 /* Here they don't match. Change the string's to be what the locale is
3742 * expecting */
3743
3744 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3745 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3746 utf8 = TRUE;
3747 }
3748 else { /* locale is not UTF-8; but input is; downgrade the input */
3749
3750 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3751
3752 /* If the downgrade was successful we are done, but if the input
3753 * contains things that require UTF-8 to represent, have to do
3754 * damage control ... */
3755 if (UNLIKELY(utf8)) {
3756
3757 /* What we do is construct a non-UTF-8 string with
3758 * 1) the characters representable by a single byte converted
3759 * to be so (if necessary);
3760 * 2) and the rest converted to collate the same as the
3761 * highest collating representable character. That makes
3762 * them collate at the end. This is similar to how we
3763 * handle embedded NULs, but we use the highest collating
3764 * code point instead of the smallest. Like the NUL case,
3765 * this isn't perfect, but is the best we can reasonably
3766 * do. Every above-255 code point will sort the same as
3767 * the highest-sorting 0-255 code point. If that code
3768 * point can combine in a sequence with some other code
3769 * points for weight calculations, us changing something to
3770 * be it can adversely affect the results. But in most
3771 * cases, it should work reasonably. And note that this is
3772 * really an illegal situation: using code points above 255
3773 * on a locale where only 0-255 are valid. If two strings
3774 * sort entirely equal, then the sort order for the
3775 * above-255 code points will be in code point order. */
3776
3777 utf8 = FALSE;
3778
3779 /* If we haven't calculated the code point with the maximum
3780 * collating order for this locale, do so now */
3781 if (! PL_strxfrm_max_cp) {
3782 int j;
3783
3784 /* The current transformed string that collates the
3785 * highest (except it also includes the prefixed collation
3786 * index. */
3787 char * cur_max_x = NULL;
3788
3789 /* Look through all legal code points (NUL isn't) */
3790 for (j = 1; j < 256; j++) {
3791 char * x;
3792 STRLEN x_len;
736a4fed 3793 char cur_source[] = { '\0', '\0' };
a4a439fb 3794
736a4fed
KW
3795 /* Create a 1-char string of the current code point */
3796 cur_source[0] = (char) j;
a4a439fb
KW
3797
3798 /* Then transform it */
3799 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3800
3801 /* If something went wrong (which it shouldn't), just
3802 * ignore this code point */
94762aa0 3803 if (! x) {
a4a439fb
KW
3804 continue;
3805 }
3806
3807 /* If this character's transformation is higher than
3808 * the current highest, this one becomes the highest */
3809 if ( cur_max_x == NULL
55e5378d
KW
3810 || strGT(x + COLLXFRM_HDR_LEN,
3811 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
3812 {
3813 PL_strxfrm_max_cp = j;
3814 cur_max_x = x;
3815 }
3816 else {
3817 Safefree(x);
3818 }
3819 }
3820
94762aa0
KW
3821 if (! cur_max_x) {
3822 DEBUG_L(PerlIO_printf(Perl_debug_log,
3823 "_mem_collxfrm: Couldn't find any character to"
3824 " replace above-Latin1 chars in locale %s with",
3825 PL_collation_name));
3826 goto bad;
3827 }
3828
58eebef2
KW
3829 DEBUG_L(PerlIO_printf(Perl_debug_log,
3830 "_mem_collxfrm: highest 1-byte collating character"
3831 " in locale %s is 0x%02X\n",
3832 PL_collation_name,
3833 PL_strxfrm_max_cp));
58eebef2 3834
a4a439fb
KW
3835 Safefree(cur_max_x);
3836 }
3837
3838 /* Here we know which legal code point collates the highest.
3839 * We are ready to construct the non-UTF-8 string. The length
3840 * will be at least 1 byte smaller than the input string
3841 * (because we changed at least one 2-byte character into a
3842 * single byte), but that is eaten up by the trailing NUL */
3843 Newx(s, len, char);
3844
3845 {
3846 STRLEN i;
3847 STRLEN d= 0;
042d9e50 3848 char * e = (char *) t + len;
a4a439fb
KW
3849
3850 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3851 U8 cur_char = t[i];
3852 if (UTF8_IS_INVARIANT(cur_char)) {
3853 s[d++] = cur_char;
3854 }
042d9e50 3855 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
3856 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3857 }
3858 else { /* Replace illegal cp with highest collating
3859 one */
3860 s[d++] = PL_strxfrm_max_cp;
3861 }
3862 }
3863 s[d++] = '\0';
3864 Renew(s, d, char); /* Free up unused space */
3865 }
3866 }
3867 }
3868
3869 /* Here, we have constructed a modified version of the input. It could
3870 * be that we already had a modified copy before we did this version.
3871 * If so, that copy is no longer needed */
3872 if (t != input_string) {
3873 Safefree(t);
3874 }
3875 }
3876
17f41037
KW
3877 length_in_chars = (utf8)
3878 ? utf8_length((U8 *) s, (U8 *) s + len)
3879 : len;
3880
59c018b9
KW
3881 /* The first element in the output is the collation id, used by
3882 * sv_collxfrm(); then comes the space for the transformed string. The
3883 * equation should give us a good estimate as to how much is needed */
55e5378d 3884 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 3885 + PL_collxfrm_base
17f41037 3886 + (PL_collxfrm_mult * length_in_chars);
a02a5408 3887 Newx(xbuf, xAlloc, char);
c7202dee
KW
3888 if (UNLIKELY(! xbuf)) {
3889 DEBUG_L(PerlIO_printf(Perl_debug_log,
3890 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 3891 goto bad;
c7202dee 3892 }
98994639 3893
d35fca5f 3894 /* Store the collation id */
98994639 3895 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
3896
3897 /* Then the transformation of the input. We loop until successful, or we
3898 * give up */
4ebeff16 3899 for (;;) {
1adab0a7 3900
55e5378d