This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add switch_to_globale_locale()
[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
e9bc6d6b
KW
585# if defined(_NL_LOCALE_NAME) && defined(DEBUGGING)
586
587 {
588 /* Internal glibc for querylocale(), but doesn't handle
589 * empty-string ("") locale properly; who knows what other
590 * glitches. Check it for now, under debug. */
591
592 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
593 uselocale((locale_t) 0));
594 /*
595 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
596 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
597 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
598 */
599 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
600 if ( strNE(PL_curlocales[index], temp_name)
601 && ! ( isNAME_C_OR_POSIX(temp_name)
602 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
603
604# ifdef USE_C_BACKTRACE
605
606 dump_c_backtrace(Perl_debug_log, 20, 1);
607
608# endif
609
610 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
611 " (%s) and what internal glibc thinks"
612 " (%s)\n", category_names[index],
613 PL_curlocales[index], temp_name);
614 }
615
616 return temp_name;
617 }
618 }
619
620# endif
621
622 /* Without querylocale(), we have to use our record-keeping we've
623 * done. */
624
625 if (category != LC_ALL) {
626
627# ifdef DEBUGGING
628
629 if (DEBUG_Lv_TEST || debug_initialization) {
630 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
631 }
632
633# endif
634
635 return PL_curlocales[index];
636 }
637 else { /* For LC_ALL */
638 unsigned int i;
639 Size_t names_len = 0;
640 char * all_string;
641
642 /* If we have a valid LC_ALL value, just return it */
643 if (PL_curlocales[LC_ALL_INDEX]) {
644
645# ifdef DEBUGGING
646
647 if (DEBUG_Lv_TEST || debug_initialization) {
648 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
649 }
650
651# endif
652
653 return PL_curlocales[LC_ALL_INDEX];
654 }
655
656 /* Otherwise, we need to construct a string of name=value pairs.
657 * We use the glibc syntax, like
658 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
659 * First calculate the needed size. */
660 for (i = 0; i < LC_ALL_INDEX; i++) {
661
662# ifdef DEBUGGING
663
664 if (DEBUG_Lv_TEST || debug_initialization) {
665 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]);
666 }
667
668# endif
669
670 names_len += strlen(category_names[i])
671 + 1 /* '=' */
672 + strlen(PL_curlocales[i])
673 + 1; /* ';' */
674 }
675 names_len++; /* Trailing '\0' */
676 SAVEFREEPV(Newx(all_string, names_len, char));
677 *all_string = '\0';
678
679 /* Then fill in the string */
680 for (i = 0; i < LC_ALL_INDEX; i++) {
681
682# ifdef DEBUGGING
683
684 if (DEBUG_Lv_TEST || debug_initialization) {
685 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]);
686 }
687
688# endif
689
690 my_strlcat(all_string, category_names[i], names_len);
691 my_strlcat(all_string, "=", names_len);
692 my_strlcat(all_string, PL_curlocales[i], names_len);
693 my_strlcat(all_string, ";", names_len);
694 }
695
696# ifdef DEBUGGING
697
698 if (DEBUG_L_TEST || debug_initialization) {
699 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
700 }
701
702 #endif
703
704 return all_string;
705 }
706
707# ifdef EINVAL
708
709 SETERRNO(EINVAL, LIB_INVARG);
710
711# endif
712
713 return NULL;
714
d2b24094
KW
715# endif
716
e9bc6d6b
KW
717 }
718
719 assert(PL_C_locale_obj);
720
721 /* Otherwise, we are switching locales. This will generally entail freeing
722 * the current one's space (at the C library's discretion). We need to
723 * stop using that locale before the switch. So switch to a known locale
724 * object that we don't otherwise mess with. This returns the locale
725 * object in effect at the time of the switch. */
726 old_obj = uselocale(PL_C_locale_obj);
727
728# ifdef DEBUGGING
729
730 if (DEBUG_Lv_TEST || debug_initialization) {
731 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
732 }
733
734# endif
735
736 if (! old_obj) {
737
738# ifdef DEBUGGING
739
740 if (DEBUG_L_TEST || debug_initialization) {
741 dSAVE_ERRNO;
742 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
743 RESTORE_ERRNO;
744 }
745
746# endif
747
748 return NULL;
749 }
750
751# ifdef DEBUGGING
752
753 if (DEBUG_Lv_TEST || debug_initialization) {
754 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
755 }
756
757# endif
758
759 /* If we weren't in a thread safe locale, set so that newlocale() below
760 which uses 'old_obj', uses an empty one. Same for our reserved C object.
761 The latter is defensive coding, so that, even if there is some bug, we
762 will never end up trying to modify either of these, as if passed to
763 newlocale(), they can be. */
764 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
765 old_obj = (locale_t) 0;
766 }
767
768 /* Create the new locale (it may actually modify the current one). */
769
770# ifndef HAS_QUERYLOCALE
771
772 if (strEQ(locale, "")) {
773
774 /* For non-querylocale() systems, we do the setting of "" ourselves to
775 * be sure that we really know what's going on. We follow the Linux
776 * documented behavior (but if that differs from the actual behavior,
777 * this won't work exactly as the OS implements). We go out and
778 * examine the environment based on our understanding of how the system
779 * works, and use that to figure things out */
780
781 const char * const lc_all = PerlEnv_getenv("LC_ALL");
782
783 /* Use any "LC_ALL" environment variable, as it overrides everything
784 * else. */
785 if (lc_all && strNE(lc_all, "")) {
786 locale = lc_all;
787 }
788 else {
789
790 /* Otherwise, we need to dig deeper. Unless overridden, the
791 * default is the LANG environment variable; if it doesn't exist,
792 * then "C" */
793
794 const char * default_name;
795
796 /* To minimize other threads messing with the environment, we copy
797 * the variable, making it a temporary. But this doesn't work upon
798 * program initialization before any scopes are created, and at
799 * this time, there's nothing else going on that would interfere.
800 * So skip the copy in that case */
801 if (PL_scopestack_ix == 0) {
802 default_name = PerlEnv_getenv("LANG");
803 }
804 else {
805 default_name = savepv(PerlEnv_getenv("LANG"));
806 }
807
808 if (! default_name || strEQ(default_name, "")) {
809 default_name = "C";
810 }
811 else if (PL_scopestack_ix != 0) {
812 SAVEFREEPV(default_name);
813 }
814
815 if (category != LC_ALL) {
816 const char * const name = PerlEnv_getenv(category_names[index]);
817
818 /* Here we are setting a single category. Assume will have the
819 * default name */
820 locale = default_name;
821
822 /* But then look for an overriding environment variable */
823 if (name && strNE(name, "")) {
824 locale = name;
825 }
826 }
827 else {
828 bool did_override = FALSE;
829 unsigned int i;
830
831 /* Here, we are getting LC_ALL. Any categories that don't have
832 * a corresponding environment variable set should be set to
833 * LANG, or to "C" if there is no LANG. If no individual
834 * categories differ from this, we can just set LC_ALL. This
835 * is buggy on systems that have extra categories that we don't
836 * know about. If there is an environment variable that sets
837 * that category, we won't know to look for it, and so our use
838 * of LANG or "C" improperly overrides it. On the other hand,
839 * if we don't do what is done here, and there is no
840 * environment variable, the category's locale should be set to
841 * LANG or "C". So there is no good solution. khw thinks the
842 * best is to look at systems to see what categories they have,
843 * and include them, and then to assume that we know the
844 * complete set */
845
846 for (i = 0; i < LC_ALL_INDEX; i++) {
847 const char * const env_override
848 = savepv(PerlEnv_getenv(category_names[i]));
849 const char * this_locale = ( env_override
850 && strNE(env_override, ""))
851 ? env_override
852 : default_name;
853 emulate_setlocale(categories[i], this_locale, i, TRUE);
854
855 if (strNE(this_locale, default_name)) {
856 did_override = TRUE;
857 }
858
859 Safefree(env_override);
860 }
861
862 /* If all the categories are the same, we can set LC_ALL to
863 * that */
864 if (! did_override) {
865 locale = default_name;
866 }
867 else {
868
869 /* Here, LC_ALL is no longer valid, as some individual
870 * categories don't match it. We call ourselves
871 * recursively, as that will execute the code that
872 * generates the proper locale string for this situation.
873 * We don't do the remainder of this function, as that is
874 * to update our records, and we've just done that for the
875 * individual categories in the loop above, and doing so
876 * would cause LC_ALL to be done as well */
877 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
878 }
879 }
880 }
881 }
882 else if (strchr(locale, ';')) {
883
884 /* LC_ALL may actually incude a conglomeration of various categories.
885 * Without querylocale, this code uses the glibc (as of this writing)
886 * syntax for representing that, but that is not a stable API, and
887 * other platforms do it differently, so we have to handle all cases
888 * ourselves */
889
890 const char * s = locale;
891 const char * e = locale + strlen(locale);
892 const char * p = s;
893 const char * category_end;
894 const char * name_start;
895 const char * name_end;
896
897 while (s < e) {
898 unsigned int i;
899
900 /* Parse through the category */
901 while (isWORDCHAR(*p)) {
902 p++;
903 }
904 category_end = p;
905
906 if (*p++ != '=') {
907 Perl_croak(aTHX_
908 "panic: %s: %d: Unexpected character in locale name '%02X",
909 __FILE__, __LINE__, *(p-1));
910 }
911
912 /* Parse through the locale name */
913 name_start = p;
914 while (isGRAPH(*p) && *p != ';') {
915 p++;
916 }
917 name_end = p;
918
919 if (*p++ != ';') {
920 Perl_croak(aTHX_
921 "panic: %s: %d: Unexpected character in locale name '%02X",
922 __FILE__, __LINE__, *(p-1));
923 }
924
925 /* Find the index of the category name in our lists */
926 for (i = 0; i < LC_ALL_INDEX; i++) {
927
928 /* Keep going if this isn't the index. The strnNE() avoids a
929 * Perl_form(), but would fail if ever a category name could be
930 * a substring of another one, like if there were a
931 * "LC_TIME_DATE" */
932 if strnNE(s, category_names[i], category_end - s) {
933 continue;
934 }
935
936 /* If this index is for the single category we're changing, we
937 * have found the locale to set it to. */
938 if (category == categories[i]) {
939 locale = Perl_form(aTHX_ "%.*s",
940 (int) (name_end - name_start),
941 name_start);
942 goto ready_to_set;
943 }
944
945 if (category == LC_ALL) {
946 char * individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
947 emulate_setlocale(categories[i], individ_locale, i, TRUE);
948 Safefree(individ_locale);
949 }
950 }
951
952 s = p;
953 }
954
955 /* Here we have set all the individual categories by recursive calls.
956 * These collectively should have fixed up LC_ALL, so can just query
957 * what that now is */
958 assert(category == LC_ALL);
959
960 return do_setlocale_c(LC_ALL, NULL);
961 }
962
963 ready_to_set: ;
964
965# endif /* end of ! querylocale */
966
967 /* Ready to create a new locale by modification of the exising one */
968 new_obj = newlocale(mask, locale, old_obj);
969
970 if (! new_obj) {
971 dSAVE_ERRNO;
972
973# ifdef DEBUGGING
974
975 if (DEBUG_L_TEST || debug_initialization) {
976 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
977 }
978
979# endif
980
981 if (! uselocale(old_obj)) {
982 SAVE_ERRNO;
983
984# ifdef DEBUGGING
985
986 if (DEBUG_L_TEST || debug_initialization) {
987 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
988 }
989
990# endif
991
992 }
993 RESTORE_ERRNO;
994 return NULL;
995 }
996
997# ifdef DEBUGGING
998
999 if (DEBUG_Lv_TEST || debug_initialization) {
1000 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1001 }
1002
1003# endif
1004
1005 /* And switch into it */
1006 if (! uselocale(new_obj)) {
1007 dSAVE_ERRNO;
1008
1009# ifdef DEBUGGING
1010
1011 if (DEBUG_L_TEST || debug_initialization) {
1012 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1013 }
1014
1015# endif
1016
1017 if (! uselocale(old_obj)) {
1018
1019# ifdef DEBUGGING
1020
1021 if (DEBUG_L_TEST || debug_initialization) {
1022 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1023 }
1024
1025# endif
1026
1027 }
1028 freelocale(new_obj);
1029 RESTORE_ERRNO;
1030 return NULL;
1031 }
1032
1033# ifdef DEBUGGING
1034
1035 if (DEBUG_Lv_TEST || debug_initialization) {
1036 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1037 }
1038
1039# endif
1040
1041 /* We are done, except for updating our records (if the system doesn't keep
1042 * them) and in the case of locale "", we don't actually know what the
1043 * locale that got switched to is, as it came from the environment. So
1044 * have to find it */
1045
1046# ifdef HAS_QUERYLOCALE
1047
1048 if (strEQ(locale, "")) {
1049 locale = querylocale(mask, new_obj);
1050 }
1051
1052# else
1053
1054 /* Here, 'locale' is the return value */
1055
1056 /* Without querylocale(), we have to update our records */
1057
1058 if (category == LC_ALL) {
1059 unsigned int i;
1060
1061 /* For LC_ALL, we change all individual categories to correspond */
1062 /* PL_curlocales is a parallel array, so has same
1063 * length as 'categories' */
1064 for (i = 0; i <= LC_ALL_INDEX; i++) {
1065 Safefree(PL_curlocales[i]);
1066 PL_curlocales[i] = savepv(locale);
1067 }
1068 }
1069 else {
1070
1071 /* For a single category, if it's not the same as the one in LC_ALL, we
1072 * nullify LC_ALL */
1073
1074 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1075 Safefree(PL_curlocales[LC_ALL_INDEX]);
1076 PL_curlocales[LC_ALL_INDEX] = NULL;
1077 }
1078
1079 /* Then update the category's record */
1080 Safefree(PL_curlocales[index]);
1081 PL_curlocales[index] = savepv(locale);
1082 }
1083
1084# endif
1085
1086 return locale;
1087}
1088
1089#endif /* USE_POSIX_2008_LOCALE */
1090
1091#if 0 /* Code that was to emulate thread-safe locales on platforms that
1092 didn't natively support them */
1093
1094/* The way this would work is that we would keep a per-thread list of the
1095 * correct locale for that thread. Any operation that was locale-sensitive
1096 * would have to be changed so that it would look like this:
1097 *
1098 * LOCALE_LOCK;
1099 * setlocale to the correct locale for this operation
1100 * do operation
1101 * LOCALE_UNLOCK
1102 *
1103 * This leaves the global locale in the most recently used operation's, but it
1104 * was locked long enough to get the result. If that result is static, it
1105 * needs to be copied before the unlock.
1106 *
1107 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1108 * the setup, but are no-ops when not needed, and similarly,
1109 * END_LOCALE_DEPENDENT_OP for the tear-down
1110 *
1111 * But every call to a locale-sensitive function would have to be changed, and
1112 * if a module didn't cooperate by using the mutex, things would break.
1113 *
1114 * This code was abandoned before being completed or tested, and is left as-is
1115*/
1116
1117# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1118# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1119
1120STATIC char *
1121S_locking_setlocale(pTHX_
1122 const int category,
1123 const char * locale,
1124 int index,
1125 const bool is_index_valid
1126 )
1127{
1128 /* This function kind of performs a setlocale() on just the current thread;
1129 * thus it is kind of thread-safe. It does this by keeping a thread-level
1130 * array of the current locales for each category. Every time a locale is
1131 * switched to, it does the switch globally, but updates the thread's
1132 * array. A query as to what the current locale is just returns the
1133 * appropriate element from the array, and doesn't actually call the system
1134 * setlocale(). The saving into the array is done in an uninterruptible
1135 * section of code, so is unaffected by whatever any other threads might be
1136 * doing.
1137 *
1138 * All locale-sensitive operations must work by first starting a critical
1139 * section, then switching to the thread's locale as kept by this function,
1140 * and then doing the operation, then ending the critical section. Thus,
1141 * each gets done in the appropriate locale. simulating thread-safety.
1142 *
1143 * This function takes the same parameters, 'category' and 'locale', that
1144 * the regular setlocale() function does, but it also takes two additional
1145 * ones. This is because as described earlier. If we know on input the
1146 * index corresponding to the category into the array where we store the
1147 * current locales, we don't have to calculate it. If the caller knows at
1148 * compile time what the index is, it it can pass it, setting
1149 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1150 *
1151 */
1152
1153 /* If the input index might be incorrect, calculate the correct one */
1154 if (! is_index_valid) {
1155 unsigned int i;
1156
1157 if (DEBUG_Lv_TEST || debug_initialization) {
1158 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1159 }
1160
1161 for (i = 0; i <= LC_ALL_INDEX; i++) {
1162 if (category == categories[i]) {
1163 index = i;
1164 goto found_index;
1165 }
1166 }
1167
1168 /* Here, we don't know about this category, so can't handle it.
1169 * XXX best we can do is to unsafely set this
1170 * XXX warning */
1171
1172 return my_setlocale(category, locale);
1173
1174 found_index: ;
1175
1176 if (DEBUG_Lv_TEST || debug_initialization) {
1177 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1178 }
1179 }
1180
1181 /* For a query, just return what's in our records */
1182 if (new_locale == NULL) {
1183 return curlocales[index];
1184 }
1185
1186
1187 /* Otherwise, we need to do the switch, and save the result, all in a
1188 * critical section */
1189
1190 Safefree(curlocales[[index]]);
1191
1192 /* It might be that this is called from an already-locked section of code.
1193 * We would have to detect and skip the LOCK/UNLOCK if so */
1194 LOCALE_LOCK;
1195
1196 curlocales[index] = savepv(my_setlocale(category, new_locale));
1197
1198 if (strEQ(new_locale, "")) {
1199
1200#ifdef LC_ALL
1201
1202 /* The locale values come from the environment, and may not all be the
1203 * same, so for LC_ALL, we have to update all the others, while the
1204 * mutex is still locked */
1205
1206 if (category == LC_ALL) {
1207 unsigned int i;
1208 for (i = 0; i < LC_ALL_INDEX) {
1209 curlocales[i] = my_setlocale(categories[i], NULL);
1210 }
1211 }
1212 }
1213
1214#endif
1215
1216 LOCALE_UNLOCK;
1217
1218 return curlocales[index];
1219}
1220
1221#endif
837ce802 1222
a4f00dcc 1223STATIC void
86799d2d 1224S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1225{
86799d2d
KW
1226 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1227 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1228
86799d2d
KW
1229#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1230 || defined(HAS_NL_LANGINFO))
98994639 1231
87f8e8e7
KW
1232 const char * radix = (use_locale)
1233 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1234 /* FALSE => already in dest locale */
1235 : ".";
2213a3be 1236
87f8e8e7 1237 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1238
87f8e8e7
KW
1239 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1240 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1241 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1242 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1243 && _is_cur_LC_category_utf8(LC_NUMERIC))
1244 {
1245 SvUTF8_on(PL_numeric_radix_sv);
1246 }
86799d2d
KW
1247
1248# ifdef DEBUGGING
7d4bcc4a 1249
2fcc0ca9
KW
1250 if (DEBUG_L_TEST || debug_initialization) {
1251 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1252 SvPVX(PL_numeric_radix_sv),
1253 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1254 }
69014004 1255
86799d2d
KW
1256# endif
1257#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1258
98994639
HS
1259}
1260
398eeea9
KW
1261STATIC void
1262S_new_numeric(pTHX_ const char *newnum)
98994639 1263{
7d4bcc4a
KW
1264
1265#ifndef USE_LOCALE_NUMERIC
1266
1267 PERL_UNUSED_ARG(newnum);
1268
1269#else
0d071d52 1270
291a84fb 1271 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1272 * core Perl this and that 'newnum' is the name of the new locale.
1273 * It installs this locale as the current underlying default.
1274 *
1275 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1276 * set_numeric_underlying() and set_numeric_standard() functions, which
1277 * should probably not be called directly, but only via macros like
0d071d52
KW
1278 * SET_NUMERIC_STANDARD() in perl.h.
1279 *
1280 * The toggling is necessary mainly so that a non-dot radix decimal point
1281 * character can be output, while allowing internal calculations to use a
1282 * dot.
1283 *
1284 * This sets several interpreter-level variables:
bb304765 1285 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1286 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1287 * that the current locale is the program's underlying
1288 * locale
1289 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1290 * that the current locale is the C locale or
1291 * indistinguishable from the C locale. If non-zero, it
1292 * is in C; if > 1, it means it may not be toggled away
7738054c 1293 * from C.
4c68b815
KW
1294 * PL_numeric_underlying_is_standard A bool kept by this function
1295 * indicating that the underlying locale and the standard
1296 * C locale are indistinguishable for the purposes of
1297 * LC_NUMERIC. This happens when both of the above two
1298 * variables are true at the same time. (Toggling is a
1299 * no-op under these circumstances.) This variable is
1300 * used to avoid having to recalculate.
398eeea9 1301 */
0d071d52 1302
b03f34cf 1303 char *save_newnum;
98994639
HS
1304
1305 if (! newnum) {
43c5f42d
NC
1306 Safefree(PL_numeric_name);
1307 PL_numeric_name = NULL;
98994639 1308 PL_numeric_standard = TRUE;
892e6465 1309 PL_numeric_underlying = TRUE;
4c68b815 1310 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1311 return;
1312 }
1313
b03f34cf 1314 save_newnum = stdize_locale(savepv(newnum));
892e6465 1315 PL_numeric_underlying = TRUE;
4c68b815
KW
1316 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1317
1318 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1319 * them */
1320 if (! PL_numeric_standard) {
1321 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1322 FALSE /* Don't toggle locale */ ))
1323 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1324 FALSE)));
1325 }
abe1abcf 1326
4c68b815 1327 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1328 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1329 Safefree(PL_numeric_name);
b03f34cf 1330 PL_numeric_name = save_newnum;
b03f34cf 1331 }
abe1abcf
KW
1332 else {
1333 Safefree(save_newnum);
1334 }
4c28b29c 1335
4c68b815
KW
1336 PL_numeric_underlying_is_standard = PL_numeric_standard;
1337
7e5377f7 1338# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1339
1340 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1341 PL_numeric_name,
1342 PL_underlying_numeric_obj);
1343
1344#endif
1345
4c68b815
KW
1346 if (DEBUG_L_TEST || debug_initialization) {
1347 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1348 }
1349
4c28b29c
KW
1350 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1351 * have to worry about the radix being a non-dot. (Core operations that
1352 * need the underlying locale change to it temporarily). */
1353 set_numeric_standard();
1354
98994639 1355#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1356
98994639
HS
1357}
1358
1359void
1360Perl_set_numeric_standard(pTHX)
1361{
7d4bcc4a 1362
98994639 1363#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1364
28c1bf33
KW
1365 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1366 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1367 * macro avoids calling this routine if toggling isn't necessary according
1368 * to our records (which could be wrong if some XS code has changed the
1369 * locale behind our back) */
0d071d52 1370
837ce802 1371 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1372 PL_numeric_standard = TRUE;
4c68b815 1373 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1374 set_numeric_radix(0);
7d4bcc4a
KW
1375
1376# ifdef DEBUGGING
1377
2fcc0ca9
KW
1378 if (DEBUG_L_TEST || debug_initialization) {
1379 PerlIO_printf(Perl_debug_log,
58e4a467 1380 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1381 }
98994639 1382
7d4bcc4a 1383# endif
98994639 1384#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1385
98994639
HS
1386}
1387
1388void
5792c642 1389Perl_set_numeric_underlying(pTHX)
98994639 1390{
7d4bcc4a 1391
98994639 1392#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1393
28c1bf33 1394 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1395 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1396 * instead of calling this directly. The macro avoids calling this routine
1397 * if toggling isn't necessary according to our records (which could be
1398 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1399
837ce802 1400 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1401 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1402 PL_numeric_underlying = TRUE;
3c62fd46 1403 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1404
1405# ifdef DEBUGGING
1406
2fcc0ca9
KW
1407 if (DEBUG_L_TEST || debug_initialization) {
1408 PerlIO_printf(Perl_debug_log,
58e4a467 1409 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1410 PL_numeric_name);
1411 }
98994639 1412
7d4bcc4a 1413# endif
98994639 1414#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1415
98994639
HS
1416}
1417
1418/*
1419 * Set up for a new ctype locale.
1420 */
a4f00dcc
KW
1421STATIC void
1422S_new_ctype(pTHX_ const char *newctype)
98994639 1423{
7d4bcc4a
KW
1424
1425#ifndef USE_LOCALE_CTYPE
1426
1427 PERL_ARGS_ASSERT_NEW_CTYPE;
1428 PERL_UNUSED_ARG(newctype);
1429 PERL_UNUSED_CONTEXT;
1430
1431#else
0d071d52 1432
291a84fb 1433 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1434 * core Perl this and that 'newctype' is the name of the new locale.
1435 *
1436 * This function sets up the folding arrays for all 256 bytes, assuming
1437 * that tofold() is tolc() since fold case is not a concept in POSIX,
1438 *
1439 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1440 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1441 * this function should be called directly only from this file and from
0d071d52
KW
1442 * POSIX::setlocale() */
1443
27da23d5 1444 dVAR;
68067e4e 1445 UV i;
98994639 1446
8b7358b9
KW
1447 /* Don't check for problems if we are suppressing the warnings */
1448 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1449
7918f24d
NC
1450 PERL_ARGS_ASSERT_NEW_CTYPE;
1451
215c5139
KW
1452 /* We will replace any bad locale warning with 1) nothing if the new one is
1453 * ok; or 2) a new warning for the bad new locale */
1454 if (PL_warn_locale) {
1455 SvREFCNT_dec_NN(PL_warn_locale);
1456 PL_warn_locale = NULL;
1457 }
1458
c1284011 1459 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1460
1461 /* A UTF-8 locale gets standard rules. But note that code still has to
1462 * handle this specially because of the three problematic code points */
1463 if (PL_in_utf8_CTYPE_locale) {
1464 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1465 }
8b7358b9
KW
1466
1467 /* We don't populate the other lists if a UTF-8 locale, but do check that
1468 * everything works as expected, unless checking turned off */
1469 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1470 /* Assume enough space for every character being bad. 4 spaces each
1471 * for the 94 printable characters that are output like "'x' "; and 5
1472 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1473 * NUL */
8b7358b9 1474 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1475 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1476 to start */
1477 unsigned int bad_count = 0; /* Count of bad characters */
1478
baa60164 1479 for (i = 0; i < 256; i++) {
8b7358b9 1480 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1481 if (isupper(i))
1482 PL_fold_locale[i] = (U8) tolower(i);
1483 else if (islower(i))
1484 PL_fold_locale[i] = (U8) toupper(i);
1485 else
1486 PL_fold_locale[i] = (U8) i;
8b7358b9 1487 }
8c6180a9
KW
1488
1489 /* If checking for locale problems, see if the native ASCII-range
1490 * printables plus \n and \t are in their expected categories in
1491 * the new locale. If not, this could mean big trouble, upending
1492 * Perl's and most programs' assumptions, like having a
1493 * metacharacter with special meaning become a \w. Fortunately,
1494 * it's very rare to find locales that aren't supersets of ASCII
1495 * nowadays. It isn't a problem for most controls to be changed
1496 * into something else; we check only \n and \t, though perhaps \r
1497 * could be an issue as well. */
7d4bcc4a 1498 if ( check_for_problems
8c6180a9
KW
1499 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1500 {
8b7358b9
KW
1501 bool is_bad = FALSE;
1502 char name[3] = { '\0' };
1503
1504 /* Convert the name into a string */
1505 if (isPRINT_A(i)) {
1506 name[0] = i;
1507 name[1] = '\0';
1508 }
1509 else if (i == '\n') {
1510 my_strlcpy(name, "\n", sizeof(name));
1511 }
1512 else {
1513 my_strlcpy(name, "\t", sizeof(name));
1514 }
1515
1516 /* Check each possibe class */
1517 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1518 is_bad = TRUE;
1519 DEBUG_L(PerlIO_printf(Perl_debug_log,
1520 "isalnum('%s') unexpectedly is %d\n",
1521 name, cBOOL(isalnum(i))));
1522 }
1523 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1524 is_bad = TRUE;
1525 DEBUG_L(PerlIO_printf(Perl_debug_log,
1526 "isalpha('%s') unexpectedly is %d\n",
1527 name, cBOOL(isalpha(i))));
1528 }
1529 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1530 is_bad = TRUE;
1531 DEBUG_L(PerlIO_printf(Perl_debug_log,
1532 "isdigit('%s') unexpectedly is %d\n",
1533 name, cBOOL(isdigit(i))));
1534 }
1535 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1536 is_bad = TRUE;
1537 DEBUG_L(PerlIO_printf(Perl_debug_log,
1538 "isgraph('%s') unexpectedly is %d\n",
1539 name, cBOOL(isgraph(i))));
1540 }
1541 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1542 is_bad = TRUE;
1543 DEBUG_L(PerlIO_printf(Perl_debug_log,
1544 "islower('%s') unexpectedly is %d\n",
1545 name, cBOOL(islower(i))));
1546 }
1547 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1548 is_bad = TRUE;
1549 DEBUG_L(PerlIO_printf(Perl_debug_log,
1550 "isprint('%s') unexpectedly is %d\n",
1551 name, cBOOL(isprint(i))));
1552 }
1553 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1554 is_bad = TRUE;
1555 DEBUG_L(PerlIO_printf(Perl_debug_log,
1556 "ispunct('%s') unexpectedly is %d\n",
1557 name, cBOOL(ispunct(i))));
1558 }
1559 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1560 is_bad = TRUE;
1561 DEBUG_L(PerlIO_printf(Perl_debug_log,
1562 "isspace('%s') unexpectedly is %d\n",
1563 name, cBOOL(isspace(i))));
1564 }
1565 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1566 is_bad = TRUE;
1567 DEBUG_L(PerlIO_printf(Perl_debug_log,
1568 "isupper('%s') unexpectedly is %d\n",
1569 name, cBOOL(isupper(i))));
1570 }
1571 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1572 is_bad = TRUE;
1573 DEBUG_L(PerlIO_printf(Perl_debug_log,
1574 "isxdigit('%s') unexpectedly is %d\n",
1575 name, cBOOL(isxdigit(i))));
1576 }
1577 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1578 is_bad = TRUE;
1579 DEBUG_L(PerlIO_printf(Perl_debug_log,
1580 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1581 name, tolower(i), (int) toLOWER_A(i)));
1582 }
1583 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1584 is_bad = TRUE;
1585 DEBUG_L(PerlIO_printf(Perl_debug_log,
1586 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1587 name, toupper(i), (int) toUPPER_A(i)));
1588 }
1589 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1590 is_bad = TRUE;
1591 DEBUG_L(PerlIO_printf(Perl_debug_log,
1592 "'\\n' (=%02X) is not a control\n", (int) i));
1593 }
1594
1595 /* Add to the list; Separate multiple entries with a blank */
1596 if (is_bad) {
1597 if (bad_count) {
1598 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1599 }
8b7358b9
KW
1600 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1601 bad_count++;
8c6180a9
KW
1602 }
1603 }
1604 }
1605
7d4bcc4a
KW
1606# ifdef MB_CUR_MAX
1607
8c6180a9 1608 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1609 * this locale requires more than one byte, there are going to be
8c6180a9 1610 * problems. */
9c8a6dc2
KW
1611 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1612 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1613 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1614
8b7358b9
KW
1615 if ( check_for_problems && MB_CUR_MAX > 1
1616 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1617
1618 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1619 * locale. Just assume that the implementation for them (plus
1620 * for POSIX) is correct and the > 1 value is spurious. (Since
1621 * these are specially handled to never be considered UTF-8
1622 * locales, as long as this is the only problem, everything
1623 * should work fine */
1624 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1625 {
8c6180a9
KW
1626 multi_byte_locale = TRUE;
1627 }
7d4bcc4a
KW
1628
1629# endif
8c6180a9 1630
8b7358b9
KW
1631 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1632 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1633 PL_warn_locale = Perl_newSVpvf(aTHX_
1634 "Locale '%s' contains (at least) the following characters"
1635 " which have\nnon-standard meanings: %s\nThe Perl program"
1636 " will use the standard meanings",
1637 newctype, bad_chars_list);
1638
1639 }
1640 else {
1641 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1642 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1643 newctype,
8c6180a9
KW
1644 (multi_byte_locale)
1645 ? " Some characters in it are not recognized by"
1646 " Perl."
1647 : "",
1648 (bad_count)
1649 ? "\nThe following characters (and maybe others)"
1650 " may not have the same meaning as the Perl"
1651 " program expects:\n"
1652 : "",
1653 (bad_count)
1654 ? bad_chars_list
1655 : ""
1656 );
8b7358b9
KW
1657 }
1658
b119c2be
KW
1659# ifdef HAS_NL_LANGINFO
1660
1661 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1662 /* parameter FALSE is a don't care here */
1663 my_nl_langinfo(PERL_CODESET, FALSE));
1664
1665# endif
1666
8b7358b9
KW
1667 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1668
cc9eaeb0 1669 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1670 * output the message now. If not in that scope, we save the
1671 * message to be output at the first operation using this locale,
1672 * if that actually happens. Most programs don't use locales, so
1673 * they are immune to bad ones. */
cc9eaeb0 1674 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1675
780fcc9f
KW
1676 /* The '0' below suppresses a bogus gcc compiler warning */
1677 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1678
bddebb56
KW
1679 if (IN_LC(LC_CTYPE)) {
1680 SvREFCNT_dec_NN(PL_warn_locale);
1681 PL_warn_locale = NULL;
1682 }
780fcc9f 1683 }
baa60164 1684 }
31f05a37 1685 }
98994639
HS
1686
1687#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1688
98994639
HS
1689}
1690
98994639 1691void
2726666d
KW
1692Perl__warn_problematic_locale()
1693{
2726666d
KW
1694
1695#ifdef USE_LOCALE_CTYPE
1696
5f04a188
KW
1697 dTHX;
1698
1699 /* Internal-to-core function that outputs the message in PL_warn_locale,
1700 * and then NULLS it. Should be called only through the macro
1701 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1702
2726666d 1703 if (PL_warn_locale) {
2726666d
KW
1704 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1705 SvPVX(PL_warn_locale),
1706 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1707 SvREFCNT_dec_NN(PL_warn_locale);
1708 PL_warn_locale = NULL;
1709 }
1710
1711#endif
1712
1713}
1714
a4f00dcc
KW
1715STATIC void
1716S_new_collate(pTHX_ const char *newcoll)
98994639 1717{
7d4bcc4a
KW
1718
1719#ifndef USE_LOCALE_COLLATE
1720
1721 PERL_UNUSED_ARG(newcoll);
1722 PERL_UNUSED_CONTEXT;
1723
1724#else
0d071d52 1725
291a84fb 1726 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1727 * core Perl this and that 'newcoll' is the name of the new locale.
1728 *
d35fca5f
KW
1729 * The design of locale collation is that every locale change is given an
1730 * index 'PL_collation_ix'. The first time a string particpates in an
1731 * operation that requires collation while locale collation is active, it
1732 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1733 * magic includes the collation index, and the transformation of the string
1734 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1735 * instead of the string itself. If a string changes, the magic is
1736 * cleared. The next time the locale changes, the index is incremented,
1737 * and so we know during a comparison that the transformation is not
1738 * necessarily still valid, and so is recomputed. Note that if the locale
1739 * changes enough times, the index could wrap (a U32), and it is possible
1740 * that a transformation would improperly be considered valid, leading to
1741 * an unlikely bug */
0d071d52 1742
98994639
HS
1743 if (! newcoll) {
1744 if (PL_collation_name) {
1745 ++PL_collation_ix;
1746 Safefree(PL_collation_name);
1747 PL_collation_name = NULL;
1748 }
1749 PL_collation_standard = TRUE;
00bf60ca 1750 is_standard_collation:
98994639
HS
1751 PL_collxfrm_base = 0;
1752 PL_collxfrm_mult = 2;
165a1c52 1753 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1754 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1755 PL_strxfrm_max_cp = 0;
98994639
HS
1756 return;
1757 }
1758
d35fca5f 1759 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1760 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1761 ++PL_collation_ix;
1762 Safefree(PL_collation_name);
1763 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1764 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1765 if (PL_collation_standard) {
1766 goto is_standard_collation;
1767 }
98994639 1768
165a1c52 1769 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1770 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1771 PL_strxfrm_max_cp = 0;
165a1c52 1772
59c018b9
KW
1773 /* A locale collation definition includes primary, secondary, tertiary,
1774 * etc. weights for each character. To sort, the primary weights are
1775 * used, and only if they compare equal, then the secondary weights are
1776 * used, and only if they compare equal, then the tertiary, etc.
1777 *
1778 * strxfrm() works by taking the input string, say ABC, and creating an
1779 * output transformed string consisting of first the primary weights,
1780 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1781 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1782 * may not have weights at every level. In our example, let's say B
1783 * doesn't have a tertiary weight, and A doesn't have a secondary
1784 * weight. The constructed string is then going to be
1785 * A¹B¹C¹ B²C² A³C³ ....
1786 * This has the desired effect that strcmp() will look at the secondary
1787 * or tertiary weights only if the strings compare equal at all higher
1788 * priority weights. The spaces shown here, like in
c342d20e 1789 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1790 * are not just for readability. In the general case, these must
1791 * actually be bytes, which we will call here 'separator weights'; and
1792 * they must be smaller than any other weight value, but since these
1793 * are C strings, only the terminating one can be a NUL (some
1794 * implementations may include a non-NUL separator weight just before
1795 * the NUL). Implementations tend to reserve 01 for the separator
1796 * weights. They are needed so that a shorter string's secondary
1797 * weights won't be misconstrued as primary weights of a longer string,
1798 * etc. By making them smaller than any other weight, the shorter
1799 * string will sort first. (Actually, if all secondary weights are
1800 * smaller than all primary ones, there is no need for a separator
1801 * weight between those two levels, etc.)
1802 *
1803 * The length of the transformed string is roughly a linear function of
1804 * the input string. It's not exactly linear because some characters
1805 * don't have weights at all levels. When we call strxfrm() we have to
1806 * allocate some memory to hold the transformed string. The
1807 * calculations below try to find coefficients 'm' and 'b' for this
1808 * locale so that m*x + b equals how much space we need, given the size
1809 * of the input string in 'x'. If we calculate too small, we increase
1810 * the size as needed, and call strxfrm() again, but it is better to
1811 * get it right the first time to avoid wasted expensive string
1812 * transformations. */
1813
98994639 1814 {
79f120c8
KW
1815 /* We use the string below to find how long the tranformation of it
1816 * is. Almost all locales are supersets of ASCII, or at least the
1817 * ASCII letters. We use all of them, half upper half lower,
1818 * because if we used fewer, we might hit just the ones that are
1819 * outliers in a particular locale. Most of the strings being
1820 * collated will contain a preponderance of letters, and even if
1821 * they are above-ASCII, they are likely to have the same number of
1822 * weight levels as the ASCII ones. It turns out that digits tend
1823 * to have fewer levels, and some punctuation has more, but those
1824 * are relatively sparse in text, and khw believes this gives a
1825 * reasonable result, but it could be changed if experience so
1826 * dictates. */
1827 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1828 char * x_longer; /* Transformed 'longer' */
1829 Size_t x_len_longer; /* Length of 'x_longer' */
1830
1831 char * x_shorter; /* We also transform a substring of 'longer' */
1832 Size_t x_len_shorter;
1833
a4a439fb 1834 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1835 * are interested only in its length). It is used because it has
1836 * the intelligence to handle all cases, but to work, it needs some
1837 * values of 'm' and 'b' to get it started. For the purposes of
1838 * this calculation we use a very conservative estimate of 'm' and
1839 * 'b'. This assumes a weight can be multiple bytes, enough to
1840 * hold any UV on the platform, and there are 5 levels, 4 weight
1841 * bytes, and a trailing NUL. */
1842 PL_collxfrm_base = 5;
1843 PL_collxfrm_mult = 5 * sizeof(UV);
1844
1845 /* Find out how long the transformation really is */
a4a439fb
KW
1846 x_longer = _mem_collxfrm(longer,
1847 sizeof(longer) - 1,
1848 &x_len_longer,
1849
1850 /* We avoid converting to UTF-8 in the
1851 * called function by telling it the
1852 * string is in UTF-8 if the locale is a
1853 * UTF-8 one. Since the string passed
1854 * here is invariant under UTF-8, we can
1855 * claim it's UTF-8 even though it isn't.
1856 * */
1857 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1858 Safefree(x_longer);
1859
1860 /* Find out how long the transformation of a substring of 'longer'
1861 * is. Together the lengths of these transformations are
1862 * sufficient to calculate 'm' and 'b'. The substring is all of
1863 * 'longer' except the first character. This minimizes the chances
1864 * of being swayed by outliers */
a4a439fb 1865 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1866 sizeof(longer) - 2,
a4a439fb
KW
1867 &x_len_shorter,
1868 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1869 Safefree(x_shorter);
1870
1871 /* If the results are nonsensical for this simple test, the whole
1872 * locale definition is suspect. Mark it so that locale collation
1873 * is not active at all for it. XXX Should we warn? */
1874 if ( x_len_shorter == 0
1875 || x_len_longer == 0
1876 || x_len_shorter >= x_len_longer)
1877 {
1878 PL_collxfrm_mult = 0;
1879 PL_collxfrm_base = 0;
1880 }
1881 else {
1882 SSize_t base; /* Temporary */
1883
1884 /* We have both: m * strlen(longer) + b = x_len_longer
1885 * m * strlen(shorter) + b = x_len_shorter;
1886 * subtracting yields:
1887 * m * (strlen(longer) - strlen(shorter))
1888 * = x_len_longer - x_len_shorter
1889 * But we have set things up so that 'shorter' is 1 byte smaller
1890 * than 'longer'. Hence:
1891 * m = x_len_longer - x_len_shorter
1892 *
1893 * But if something went wrong, make sure the multiplier is at
1894 * least 1.
1895 */
1896 if (x_len_longer > x_len_shorter) {
1897 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1898 }
1899 else {
1900 PL_collxfrm_mult = 1;
1901 }
1902
1903 /* mx + b = len
1904 * so: b = len - mx
1905 * but in case something has gone wrong, make sure it is
1906 * non-negative */
1907 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1908 if (base < 0) {
1909 base = 0;
1910 }
1911
1912 /* Add 1 for the trailing NUL */
1913 PL_collxfrm_base = base + 1;
1914 }
58eebef2 1915
7d4bcc4a
KW
1916# ifdef DEBUGGING
1917
58eebef2
KW
1918 if (DEBUG_L_TEST || debug_initialization) {
1919 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1920 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1921 "x_len_longer=%zu,"
1922 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1923 __FILE__, __LINE__,
1924 PL_in_utf8_COLLATE_locale,
1925 x_len_shorter, x_len_longer,
1926 PL_collxfrm_mult, PL_collxfrm_base);
1927 }
7d4bcc4a
KW
1928# endif
1929
98994639
HS
1930 }
1931 }
1932
1933#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1934
98994639
HS
1935}
1936
d2b24094 1937#ifdef WIN32
b8cc575c 1938
a4f00dcc 1939STATIC char *
b8cc575c 1940S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1941{
1942 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1943 * difference between the two unless the input locale is "", which normally
1944 * means on Windows to get the machine default, which is set via the
1945 * computer's "Regional and Language Options" (or its current equivalent).
1946 * In POSIX, it instead means to find the locale from the user's
1947 * environment. This routine changes the Windows behavior to first look in
1948 * the environment, and, if anything is found, use that instead of going to
1949 * the machine default. If there is no environment override, the machine
1950 * default is used, by calling the real setlocale() with "".
1951 *
1952 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1953 * use the particular category's variable if set; otherwise to use the LANG
1954 * variable. */
b385bb4d 1955
175c4cf9 1956 bool override_LC_ALL = FALSE;
89f7b9aa 1957 char * result;
e5f10d49 1958 unsigned int i;
89f7b9aa 1959
b385bb4d 1960 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1961
1962# ifdef LC_ALL
1963
b385bb4d
KW
1964 locale = PerlEnv_getenv("LC_ALL");
1965 if (! locale) {
e5f10d49
KW
1966 if (category == LC_ALL) {
1967 override_LC_ALL = TRUE;
1968 }
1969 else {
7d4bcc4a
KW
1970
1971# endif
7d4bcc4a 1972
e5f10d49
KW
1973 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1974 if (category == categories[i]) {
1975 locale = PerlEnv_getenv(category_names[i]);
1976 goto found_locale;
1977 }
1978 }
7d4bcc4a 1979
b385bb4d 1980 locale = PerlEnv_getenv("LANG");
481465ea 1981 if (! locale) {
b385bb4d
KW
1982 locale = "";
1983 }
e5f10d49
KW
1984
1985 found_locale: ;
7d4bcc4a
KW
1986
1987# ifdef LC_ALL
1988
e5f10d49 1989 }
b385bb4d 1990 }
7d4bcc4a
KW
1991
1992# endif
1993
b385bb4d
KW
1994 }
1995
89f7b9aa 1996 result = setlocale(category, locale);
48015184
KW
1997 DEBUG_L(STMT_START {
1998 dSAVE_ERRNO;
1999 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2000 setlocale_debug_string(category, locale, result));
2001 RESTORE_ERRNO;
2002 } STMT_END);
89f7b9aa 2003
481465ea 2004 if (! override_LC_ALL) {
89f7b9aa
KW
2005 return result;
2006 }
2007
dfd77d7a 2008 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2009 * LANG variable or the system default if there is no LANG. But these have
2010 * lower priority than the other LC_foo variables, so override it for each
2011 * one that is set. (If they are set to "", it means to use the same thing
2012 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2013
948523db 2014 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2015 result = PerlEnv_getenv(category_names[i]);
2016 if (result && strNE(result, "")) {
2017 setlocale(categories[i], result);
2018 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2019 __FILE__, __LINE__,
2020 setlocale_debug_string(categories[i], result, "not captured")));
2021 }
89f7b9aa 2022 }
7d4bcc4a 2023
bbc98134 2024 result = setlocale(LC_ALL, NULL);
48015184
KW
2025 DEBUG_L(STMT_START {
2026 dSAVE_ERRNO;
2027 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2028 __FILE__, __LINE__,
48015184
KW
2029 setlocale_debug_string(LC_ALL, NULL, result));
2030 RESTORE_ERRNO;
2031 } STMT_END);
89f7b9aa 2032
bbc98134 2033 return result;
b385bb4d
KW
2034}
2035
2036#endif
2037
9aac5db8
KW
2038/*
2039
2040=head1 Locale-related functions and macros
2041
2042=for apidoc Perl_setlocale
2043
2044This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2045taking the same parameters, and returning the same information, except that it
2046returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2047perl keeps that locale category as C<C>, changing it briefly during the
2048operations where the underlying one is required.
2049
e9bc6d6b 2050Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2051declared to return S<C<const char *>>, whereas the system setlocale omits the
2052C<const>. (If it were being written today, plain setlocale would be declared
2053const, since it is illegal to change the information it returns; doing so leads
2054to segfaults.)
2055
e9bc6d6b
KW
2056Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2057C<setlocale> can be completely ineffective on some platforms under some
2058configurations.
2059
9aac5db8 2060C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2061where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2062the system C<setlocale()> is ineffective, returning the wrong information, and
2063failing to actually change the locale. C<Perl_setlocale>, however works
2064properly in all circumstances.
9aac5db8
KW
2065
2066The return points to a per-thread static buffer, which is overwritten the next
2067time C<Perl_setlocale> is called from the same thread.
2068
2069=cut
2070
2071*/
2072
2073const char *
2074Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2075{
2076 /* This wraps POSIX::setlocale() */
2077
9aac5db8
KW
2078 const char * retval;
2079 const char * newlocale;
2080 dSAVEDERRNO;
2081 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2082 dTHX;
2083
a4f00dcc
KW
2084#ifdef USE_LOCALE_NUMERIC
2085
291a84fb
KW
2086 /* A NULL locale means only query what the current one is. We have the
2087 * LC_NUMERIC name saved, because we are normally switched into the C
2088 * locale for it. For an LC_ALL query, switch back to get the correct
2089 * results. All other categories don't require special handling */
a4f00dcc
KW
2090 if (locale == NULL) {
2091 if (category == LC_NUMERIC) {
9aac5db8
KW
2092
2093 /* We don't have to copy this return value, as it is a per-thread
2094 * variable, and won't change until a future setlocale */
2095 return PL_numeric_name;
a4f00dcc
KW
2096 }
2097
7d4bcc4a 2098# ifdef LC_ALL
a4f00dcc 2099
9aac5db8
KW
2100 else if (category == LC_ALL) {
2101 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2102 }
2103
7d4bcc4a 2104# endif
a4f00dcc
KW
2105
2106 }
2107
2108#endif
2109
9aac5db8
KW
2110 retval = do_setlocale_r(category, locale);
2111 SAVE_ERRNO;
2112
2113#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2114
2115 if (locale == NULL && category == LC_ALL) {
2116 RESTORE_LC_NUMERIC();
2117 }
2118
2119#endif
a4f00dcc
KW
2120
2121 DEBUG_L(PerlIO_printf(Perl_debug_log,
2122 "%s:%d: %s\n", __FILE__, __LINE__,
2123 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2124
9aac5db8
KW
2125 RESTORE_ERRNO;
2126
2127 if (! retval) {
a4f00dcc
KW
2128 return NULL;
2129 }
2130
9aac5db8
KW
2131 save_to_buffer(retval, &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2132 retval = PL_setlocale_buf;
2133
2134 /* If locale == NULL, we are just querying the state */
a4f00dcc 2135 if (locale == NULL) {
a4f00dcc
KW
2136 return retval;
2137 }
a4f00dcc 2138
1159483a
KW
2139 /* Now that have switched locales, we have to update our records to
2140 * correspond. */
a4f00dcc 2141
1159483a 2142 switch (category) {
a4f00dcc 2143
1159483a 2144#ifdef USE_LOCALE_CTYPE
a4f00dcc 2145
1159483a
KW
2146 case LC_CTYPE:
2147 new_ctype(retval);
2148 break;
a4f00dcc 2149
1159483a 2150#endif
a4f00dcc
KW
2151#ifdef USE_LOCALE_COLLATE
2152
1159483a
KW
2153 case LC_COLLATE:
2154 new_collate(retval);
2155 break;
a4f00dcc 2156
1159483a 2157#endif
a4f00dcc
KW
2158#ifdef USE_LOCALE_NUMERIC
2159
1159483a
KW
2160 case LC_NUMERIC:
2161 new_numeric(retval);
2162 break;
a4f00dcc 2163
1159483a
KW
2164#endif
2165#ifdef LC_ALL
a4f00dcc 2166
1159483a 2167 case LC_ALL:
a4f00dcc 2168
1159483a
KW
2169 /* LC_ALL updates all the things we care about. The values may not
2170 * be the same as 'retval', as the locale "" may have set things
2171 * individually */
a4f00dcc 2172
1159483a 2173# ifdef USE_LOCALE_CTYPE
a4f00dcc 2174
1159483a
KW
2175 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2176 new_ctype(newlocale);
a4f00dcc 2177
1159483a
KW
2178# endif /* USE_LOCALE_CTYPE */
2179# ifdef USE_LOCALE_COLLATE
2180
2181 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2182 new_collate(newlocale);
a4f00dcc 2183
7d4bcc4a 2184# endif
1159483a 2185# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2186
1159483a
KW
2187 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2188 new_numeric(newlocale);
a4f00dcc 2189
1159483a
KW
2190# endif /* USE_LOCALE_NUMERIC */
2191#endif /* LC_ALL */
a4f00dcc 2192
1159483a
KW
2193 default:
2194 break;
a4f00dcc
KW
2195 }
2196
2197 return retval;
2198
f7416781
KW
2199}
2200
2201PERL_STATIC_INLINE const char *
2202S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2203{
2204 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2205 * growing it if necessary */
2206
2207 const Size_t string_size = strlen(string) + offset + 1;
2208
2209 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2210
2211 if (*buf_size == 0) {
2212 Newx(*buf, string_size, char);
2213 *buf_size = string_size;
2214 }
2215 else if (string_size > *buf_size) {
2216 Renew(*buf, string_size, char);
2217 *buf_size = string_size;
2218 }
2219
2220 Copy(string, *buf + offset, string_size - offset, char);
2221 return *buf;
2222}
2223
2224/*
2225
f7416781
KW
2226=for apidoc Perl_langinfo
2227
7d4bcc4a 2228This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2229taking the same C<item> parameter values, and returning the same information.
2230But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2231of Perl's locale handling from your code, and can be used on systems that lack
2232a native C<nl_langinfo>.
2233
2234Expanding on these:
2235
2236=over
2237
2238=item *
2239
2240It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
2241without you having to write extra code. The reason for the extra code would be
2242because these are from the C<LC_NUMERIC> locale category, which is normally
2243kept set to the C locale by Perl, no matter what the underlying locale is
2244supposed to be, and so to get the expected results, you have to temporarily
2245toggle into the underlying locale, and later toggle back. (You could use
2246plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
2247but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
2248keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
2249expecting the radix (decimal point) character to be a dot.)
2250
2251=item *
2252
2253Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
2254makes your code more portable. Of the fifty-some possible items specified by
2255the POSIX 2008 standard,
2256L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
2257only two are completely unimplemented. It uses various techniques to recover
2258the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
2259both of which are specified in C89, so should be always be available. Later
2260C<strftime()> versions have additional capabilities; C<""> is returned for
2261those not available on your system.
2262
9f0bc911
KW
2263It is important to note that on such systems, this calls C<localeconv>, and so
2264overwrites the static buffer returned from previous explicit calls to that
2265function. Thus, if the program doesn't use or save the information from an
2266explicit C<localeconv> call (which good practice suggests should be done
2267anyway), use of this function can break it.
2268
f7416781
KW
2269The details for those items which may differ from what this emulation returns
2270and what a native C<nl_langinfo()> would return are:
2271
2272=over
2273
2274=item C<CODESET>
2275
2276=item C<ERA>
2277
2278Unimplemented, so returns C<"">.
2279
2280=item C<YESEXPR>
2281
c1566110
KW
2282=item C<YESSTR>
2283
f7416781
KW
2284=item C<NOEXPR>
2285
c1566110
KW
2286=item C<NOSTR>
2287
2288Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
2289removed from POSIX 2008, and are retained for backwards compatibility. Your
2290platform's C<nl_langinfo> may not support them.
f7416781
KW
2291
2292=item C<D_FMT>
2293
2294Always evaluates to C<%x>, the locale's appropriate date representation.
2295
2296=item C<T_FMT>
2297
2298Always evaluates to C<%X>, the locale's appropriate time representation.
2299
2300=item C<D_T_FMT>
2301
2302Always evaluates to C<%c>, the locale's appropriate date and time
2303representation.
2304
2305=item C<CRNCYSTR>
2306
2307The return may be incorrect for those rare locales where the currency symbol
2308replaces the radix character.
2309Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2310to work differently.
2311
2312=item C<ALT_DIGITS>
2313
2314Currently this gives the same results as Linux does.
2315Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2316to work differently.
2317
2318=item C<ERA_D_FMT>
2319
2320=item C<ERA_T_FMT>
2321
2322=item C<ERA_D_T_FMT>
2323
2324=item C<T_FMT_AMPM>
2325
2326These are derived by using C<strftime()>, and not all versions of that function
2327know about them. C<""> is returned for these on such systems.
2328
2329=back
2330
2331When using C<Perl_langinfo> on systems that don't have a native
2332C<nl_langinfo()>, you must
2333
2334 #include "perl_langinfo.h"
2335
2336before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2337C<#include> with this one. (Doing it this way keeps out the symbols that plain
2338C<langinfo.h> imports into the namespace for code that doesn't need it.)
2339
2340You also should not use the bare C<langinfo.h> item names, but should preface
2341them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2342The C<PERL_I<foo>> versions will also work for this function on systems that do
2343have a native C<nl_langinfo>.
2344
2345=item *
2346
2347It is thread-friendly, returning its result in a buffer that won't be
2348overwritten by another thread, so you don't have to code for that possibility.
2349The buffer can be overwritten by the next call to C<nl_langinfo> or
2350C<Perl_langinfo> in the same thread.
2351
2352=item *
2353
7d4bcc4a 2354ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
2355*>>, but you are (only by documentation) forbidden to write into the buffer.
2356By declaring this C<const>, the compiler enforces this restriction. The extra
2357C<const> is why this isn't an unequivocal drop-in replacement for
2358C<nl_langinfo>.
2359
2360=back
2361
2362The original impetus for C<Perl_langinfo()> was so that code that needs to
2363find out the current currency symbol, floating point radix character, or digit
2364grouping separator can use, on all systems, the simpler and more
2365thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2366pain to make thread-friendly. For other fields returned by C<localeconv>, it
2367is better to use the methods given in L<perlcall> to call
2368L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2369
2370=cut
2371
2372*/
2373
2374const char *
2375#ifdef HAS_NL_LANGINFO
2376Perl_langinfo(const nl_item item)
2377#else
2378Perl_langinfo(const int item)
2379#endif
2380{
f61748ac
KW
2381 return my_nl_langinfo(item, TRUE);
2382}
2383
2384const char *
2385#ifdef HAS_NL_LANGINFO
2386S_my_nl_langinfo(const nl_item item, bool toggle)
2387#else
2388S_my_nl_langinfo(const int item, bool toggle)
2389#endif
2390{
ae74815b 2391 dTHX;
f7416781 2392
5a854ab3
KW
2393 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2394 * two items, and only if not already there */
2395 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2396 || PL_numeric_underlying))
2397 {
2398 toggle = FALSE;
2399 }
2400
ab340fff
KW
2401#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
2402#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 2403
ab340fff 2404 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2405 * for those items dependent on it. This must be copied to a buffer before
2406 * switching back, as some systems destroy the buffer when setlocale() is
2407 * called */
f7416781 2408
038d3702
KW
2409 {
2410 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2411
b2fee59c 2412 if (toggle) {
038d3702 2413 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2414 }
f7416781 2415
5acc4454
KW
2416 LOCALE_LOCK; /* Prevent interference from another thread executing
2417 this code section (the only call to nl_langinfo in
2418 the core) */
2419
b2fee59c
KW
2420 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf,
2421 &PL_langinfo_bufsize, 0);
f7416781 2422
5acc4454
KW
2423 LOCALE_UNLOCK;
2424
b2fee59c 2425 if (toggle) {
038d3702 2426 RESTORE_LC_NUMERIC();
b2fee59c 2427 }
f7416781
KW
2428 }
2429
ab340fff
KW
2430# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2431
5a854ab3 2432 {
b2fee59c
KW
2433 bool do_free = FALSE;
2434 locale_t cur = uselocale((locale_t) 0);
ab340fff 2435
b2fee59c
KW
2436 if (cur == LC_GLOBAL_LOCALE) {
2437 cur = duplocale(LC_GLOBAL_LOCALE);
2438 do_free = TRUE;
2439 }
ab340fff 2440
b2fee59c 2441 if (toggle) {
e1aa2579
KW
2442 if (PL_underlying_numeric_obj) {
2443 cur = PL_underlying_numeric_obj;
2444 }
2445 else {
2446 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2447 do_free = TRUE;
2448 }
b2fee59c 2449 }
ab340fff 2450
b2fee59c
KW
2451 save_to_buffer(nl_langinfo_l(item, cur),
2452 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
2453 if (do_free) {
2454 freelocale(cur);
2455 }
5a854ab3 2456 }
ab340fff 2457
c1566110
KW
2458# endif
2459
2460 if (strEQ(PL_langinfo_buf, "")) {
2461 if (item == PERL_YESSTR) {
2462 return "yes";
2463 }
2464 if (item == PERL_NOSTR) {
2465 return "no";
2466 }
2467 }
2468
ab340fff
KW
2469 return PL_langinfo_buf;
2470
f7416781 2471#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2472
2473 {
2474
f7416781
KW
2475# ifdef HAS_LOCALECONV
2476
43dd6b15 2477 const struct lconv* lc;
038d3702 2478 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
2479
2480# endif
2481# ifdef HAS_STRFTIME
2482
43dd6b15
KW
2483 struct tm tm;
2484 bool return_format = FALSE; /* Return the %format, not the value */
2485 const char * format;
f7416781
KW
2486
2487# endif
2488
43dd6b15
KW
2489 /* We copy the results to a per-thread buffer, even if not
2490 * multi-threaded. This is in part to simplify this code, and partly
2491 * because we need a buffer anyway for strftime(), and partly because a
2492 * call of localeconv() could otherwise wipe out the buffer, and the
2493 * programmer would not be expecting this, as this is a nl_langinfo()
2494 * substitute after all, so s/he might be thinking their localeconv()
2495 * is safe until another localeconv() call. */
f7416781 2496
43dd6b15
KW
2497 switch (item) {
2498 Size_t len;
2499 const char * retval;
f7416781 2500
43dd6b15
KW
2501 /* These 2 are unimplemented */
2502 case PERL_CODESET:
2503 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 2504
43dd6b15
KW
2505 default:
2506 return "";
f7416781 2507
43dd6b15
KW
2508 /* We use only an English set, since we don't know any more */
2509 case PERL_YESEXPR: return "^[+1yY]";
2510 case PERL_YESSTR: return "yes";
2511 case PERL_NOEXPR: return "^[-0nN]";
2512 case PERL_NOSTR: return "no";
f7416781
KW
2513
2514# ifdef HAS_LOCALECONV
2515
43dd6b15 2516 case PERL_CRNCYSTR:
f7416781 2517
43dd6b15
KW
2518 /* We don't bother with localeconv_l() because any system that
2519 * has it is likely to also have nl_langinfo() */
291a84fb 2520
5acc4454
KW
2521 LOCALE_LOCK; /* Prevent interference with other threads
2522 using localeconv() */
2523
43dd6b15
KW
2524 lc = localeconv();
2525 if ( ! lc
2526 || ! lc->currency_symbol
2527 || strEQ("", lc->currency_symbol))
2528 {
2529 LOCALE_UNLOCK;
2530 return "";
2531 }
f7416781 2532
43dd6b15
KW
2533 /* Leave the first spot empty to be filled in below */
2534 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2535 &PL_langinfo_bufsize, 1);
2536 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2537 { /* khw couldn't figure out how the localedef specifications
2538 would show that the $ should replace the radix; this is
2539 just a guess as to how it might work.*/
2540 *PL_langinfo_buf = '.';
2541 }
2542 else if (lc->p_cs_precedes) {
2543 *PL_langinfo_buf = '-';
2544 }
2545 else {
2546 *PL_langinfo_buf = '+';
2547 }
f7416781 2548
43dd6b15
KW
2549 LOCALE_UNLOCK;
2550 break;
f7416781 2551
43dd6b15
KW
2552 case PERL_RADIXCHAR:
2553 case PERL_THOUSEP:
f7416781 2554
43dd6b15 2555 if (toggle) {
038d3702 2556 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2557 }
f7416781 2558
5acc4454
KW
2559 LOCALE_LOCK; /* Prevent interference with other threads
2560 using localeconv() */
2561
43dd6b15
KW
2562 lc = localeconv();
2563 if (! lc) {
33394adc
KW
2564 retval = "";
2565 }
43dd6b15
KW
2566 else {
2567 retval = (item == PERL_RADIXCHAR)
2568 ? lc->decimal_point
2569 : lc->thousands_sep;
2570 if (! retval) {
2571 retval = "";
2572 }
2573 }
f7416781 2574
43dd6b15
KW
2575 save_to_buffer(retval, &PL_langinfo_buf,
2576 &PL_langinfo_bufsize, 0);
f7416781 2577
5acc4454
KW
2578 LOCALE_UNLOCK;
2579
43dd6b15 2580 if (toggle) {
038d3702 2581 RESTORE_LC_NUMERIC();
43dd6b15 2582 }
f7416781 2583
43dd6b15 2584 break;
f7416781
KW
2585
2586# endif
2587# ifdef HAS_STRFTIME
2588
43dd6b15
KW
2589 /* These are defined by C89, so we assume that strftime supports
2590 * them, and so are returned unconditionally; they may not be what
2591 * the locale actually says, but should give good enough results
2592 * for someone using them as formats (as opposed to trying to parse
2593 * them to figure out what the locale says). The other format
2594 * items are actually tested to verify they work on the platform */
2595 case PERL_D_FMT: return "%x";
2596 case PERL_T_FMT: return "%X";
2597 case PERL_D_T_FMT: return "%c";
2598
2599 /* These formats are only available in later strfmtime's */
2600 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2601 case PERL_T_FMT_AMPM:
2602
2603 /* The rest can be gotten from most versions of strftime(). */
2604 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2605 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2606 case PERL_ABDAY_7:
2607 case PERL_ALT_DIGITS:
2608 case PERL_AM_STR: case PERL_PM_STR:
2609 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2610 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2611 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2612 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2613 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2614 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2615 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2616 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2617 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2618 case PERL_MON_12:
2619
2620 LOCALE_LOCK;
2621
2622 init_tm(&tm); /* Precaution against core dumps */
2623 tm.tm_sec = 30;
2624 tm.tm_min = 30;
2625 tm.tm_hour = 6;
2626 tm.tm_year = 2017 - 1900;
2627 tm.tm_wday = 0;
2628 tm.tm_mon = 0;
2629 switch (item) {
2630 default:
2631 LOCALE_UNLOCK;
2632 Perl_croak(aTHX_
2633 "panic: %s: %d: switch case: %d problem",
2634 __FILE__, __LINE__, item);
2635 NOT_REACHED; /* NOTREACHED */
2636
2637 case PERL_PM_STR: tm.tm_hour = 18;
2638 case PERL_AM_STR:
2639 format = "%p";
2640 break;
2641
2642 case PERL_ABDAY_7: tm.tm_wday++;
2643 case PERL_ABDAY_6: tm.tm_wday++;
2644 case PERL_ABDAY_5: tm.tm_wday++;
2645 case PERL_ABDAY_4: tm.tm_wday++;
2646 case PERL_ABDAY_3: tm.tm_wday++;
2647 case PERL_ABDAY_2: tm.tm_wday++;
2648 case PERL_ABDAY_1:
2649 format = "%a";
2650 break;
2651
2652 case PERL_DAY_7: tm.tm_wday++;
2653 case PERL_DAY_6: tm.tm_wday++;
2654 case PERL_DAY_5: tm.tm_wday++;
2655 case PERL_DAY_4: tm.tm_wday++;
2656 case PERL_DAY_3: tm.tm_wday++;
2657 case PERL_DAY_2: tm.tm_wday++;
2658 case PERL_DAY_1:
2659 format = "%A";
2660 break;
2661
2662 case PERL_ABMON_12: tm.tm_mon++;
2663 case PERL_ABMON_11: tm.tm_mon++;
2664 case PERL_ABMON_10: tm.tm_mon++;
2665 case PERL_ABMON_9: tm.tm_mon++;
2666 case PERL_ABMON_8: tm.tm_mon++;
2667 case PERL_ABMON_7: tm.tm_mon++;
2668 case PERL_ABMON_6: tm.tm_mon++;
2669 case PERL_ABMON_5: tm.tm_mon++;
2670 case PERL_ABMON_4: tm.tm_mon++;
2671 case PERL_ABMON_3: tm.tm_mon++;
2672 case PERL_ABMON_2: tm.tm_mon++;
2673 case PERL_ABMON_1:
2674 format = "%b";
2675 break;
2676
2677 case PERL_MON_12: tm.tm_mon++;
2678 case PERL_MON_11: tm.tm_mon++;
2679 case PERL_MON_10: tm.tm_mon++;
2680 case PERL_MON_9: tm.tm_mon++;
2681 case PERL_MON_8: tm.tm_mon++;
2682 case PERL_MON_7: tm.tm_mon++;
2683 case PERL_MON_6: tm.tm_mon++;
2684 case PERL_MON_5: tm.tm_mon++;
2685 case PERL_MON_4: tm.tm_mon++;
2686 case PERL_MON_3: tm.tm_mon++;
2687 case PERL_MON_2: tm.tm_mon++;
2688 case PERL_MON_1:
2689 format = "%B";
2690 break;
2691
2692 case PERL_T_FMT_AMPM:
2693 format = "%r";
2694 return_format = TRUE;
2695 break;
2696
2697 case PERL_ERA_D_FMT:
2698 format = "%Ex";
2699 return_format = TRUE;
2700 break;
2701
2702 case PERL_ERA_T_FMT:
2703 format = "%EX";
2704 return_format = TRUE;
2705 break;
2706
2707 case PERL_ERA_D_T_FMT:
2708 format = "%Ec";
2709 return_format = TRUE;
2710 break;
2711
2712 case PERL_ALT_DIGITS:
2713 tm.tm_wday = 0;
2714 format = "%Ow"; /* Find the alternate digit for 0 */
2715 break;
2716 }
f7416781 2717
43dd6b15
KW
2718 /* We can't use my_strftime() because it doesn't look at
2719 * tm_wday */
2720 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2721 format, &tm))
2722 {
2723 /* A zero return means one of:
2724 * a) there wasn't enough space in PL_langinfo_buf
2725 * b) the format, like a plain %p, returns empty
2726 * c) it was an illegal format, though some
2727 * implementations of strftime will just return the
2728 * illegal format as a plain character sequence.
2729 *
2730 * To quickly test for case 'b)', try again but precede
2731 * the format with a plain character. If that result is
2732 * still empty, the problem is either 'a)' or 'c)' */
2733
2734 Size_t format_size = strlen(format) + 1;
2735 Size_t mod_size = format_size + 1;
2736 char * mod_format;
2737 char * temp_result;
2738
2739 Newx(mod_format, mod_size, char);
2740 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2741 *mod_format = ' ';
43dd6b15
KW
2742 my_strlcpy(mod_format + 1, format, mod_size);
2743 len = strftime(temp_result,
2744 PL_langinfo_bufsize,
2745 mod_format, &tm);
2746 Safefree(mod_format);
2747 Safefree(temp_result);
2748
2749 /* If 'len' is non-zero, it means that we had a case like
2750 * %p which means the current locale doesn't use a.m. or
2751 * p.m., and that is valid */
2752 if (len == 0) {
2753
2754 /* Here, still didn't work. If we get well beyond a
2755 * reasonable size, bail out to prevent an infinite
2756 * loop. */
2757
2758 if (PL_langinfo_bufsize > 100 * format_size) {
2759 *PL_langinfo_buf = '\0';
2760 }
2761 else {
2762 /* Double the buffer size to retry; Add 1 in case
2763 * original was 0, so we aren't stuck at 0. */
2764 PL_langinfo_bufsize *= 2;
2765 PL_langinfo_bufsize++;
2766 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2767 continue;
2768 }
2769 }
f7416781 2770
f7416781 2771 break;
43dd6b15 2772 }
f7416781 2773
43dd6b15
KW
2774 /* Here, we got a result.
2775 *
2776 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2777 * alternate format for wday 0. If the value is the same as
2778 * the normal 0, there isn't an alternate, so clear the buffer.
2779 * */
2780 if ( item == PERL_ALT_DIGITS
2781 && strEQ(PL_langinfo_buf, "0"))
2782 {
2783 *PL_langinfo_buf = '\0';
2784 }
f7416781 2785
43dd6b15
KW
2786 /* ALT_DIGITS is problematic. Experiments on it showed that
2787 * strftime() did not always work properly when going from
2788 * alt-9 to alt-10. Only a few locales have this item defined,
2789 * and in all of them on Linux that khw was able to find,
2790 * nl_langinfo() merely returned the alt-0 character, possibly
2791 * doubled. Most Unicode digits are in blocks of 10
2792 * consecutive code points, so that is sufficient information
2793 * for those scripts, as we can infer alt-1, alt-2, .... But
2794 * for a Japanese locale, a CJK ideographic 0 is returned, and
2795 * the CJK digits are not in code point order, so you can't
2796 * really infer anything. The localedef for this locale did
2797 * specify the succeeding digits, so that strftime() works
2798 * properly on them, without needing to infer anything. But
2799 * the nl_langinfo() return did not give sufficient information
2800 * for the caller to understand what's going on. So until
2801 * there is evidence that it should work differently, this
2802 * returns the alt-0 string for ALT_DIGITS.
2803 *
2804 * wday was chosen because its range is all a single digit.
2805 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 2806
43dd6b15 2807 LOCALE_UNLOCK;
f7416781 2808
43dd6b15
KW
2809 /* If to return the format, not the value, overwrite the buffer
2810 * with it. But some strftime()s will keep the original format
2811 * if illegal, so change those to "" */
2812 if (return_format) {
2813 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
2814 *PL_langinfo_buf = '\0';
2815 }
43dd6b15
KW
2816 else {
2817 save_to_buffer(format, &PL_langinfo_buf,
2818 &PL_langinfo_bufsize, 0);
f7416781
KW
2819 }
2820 }
2821
2822 break;
f7416781
KW
2823
2824# endif
2825
43dd6b15 2826 }
f7416781
KW
2827 }
2828
2829 return PL_langinfo_buf;
2830
2831#endif
2832
a4f00dcc 2833}
b385bb4d 2834
98994639
HS
2835/*
2836 * Initialize locale awareness.
2837 */
2838int
2839Perl_init_i18nl10n(pTHX_ int printwarn)
2840{
0e92a118
KW
2841 /* printwarn is
2842 *
2843 * 0 if not to output warning when setup locale is bad
2844 * 1 if to output warning based on value of PERL_BADLANG
2845 * >1 if to output regardless of PERL_BADLANG
2846 *
2847 * returns
98994639 2848 * 1 = set ok or not applicable,
0e92a118
KW
2849 * 0 = fallback to a locale of lower priority
2850 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2851 *
2852 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2853 * set, debugging information is output.
2854 *
2855 * This looks more complicated than it is, mainly due to the #ifdefs.
2856 *
2857 * We try to set LC_ALL to the value determined by the environment. If
2858 * there is no LC_ALL on this platform, we try the individual categories we
2859 * know about. If this works, we are done.
2860 *
2861 * But if it doesn't work, we have to do something else. We search the
2862 * environment variables ourselves instead of relying on the system to do
2863 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2864 * think there is one), and the ultimate fallback "C". This is all done in
2865 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2866 * more complex. The 'trial_locales' array is initialized with just one
2867 * element; it causes the behavior described in the paragraph above this to
2868 * happen. If that fails, we add elements to 'trial_locales', and do extra
2869 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2870 *
2871 * On Ultrix, the locale MUST come from the environment, so there is
2872 * preliminary code to set it. I (khw) am not sure that it is necessary,
2873 * and that this couldn't be folded into the loop, but barring any real
2874 * platforms to test on, it's staying as-is
2875 *
2876 * A slight complication is that in embedded Perls, the locale may already
2877 * be set-up, and we don't want to get it from the normal environment
2878 * variables. This is handled by having a special environment variable
2879 * indicate we're in this situation. We simply set setlocale's 2nd
2880 * parameter to be a NULL instead of "". That indicates to setlocale that
2881 * it is not to change anything, but to return the current value,
2882 * effectively initializing perl's db to what the locale already is.
2883 *
2884 * We play the same trick with NULL if a LC_ALL succeeds. We call
2885 * setlocale() on the individual categores with NULL to get their existing
2886 * values for our db, instead of trying to change them.
2887 * */
98994639 2888
0e92a118
KW
2889 int ok = 1;
2890
7d4bcc4a
KW
2891#ifndef USE_LOCALE
2892
2893 PERL_UNUSED_ARG(printwarn);
2894
2895#else /* USE_LOCALE */
7d4bcc4a
KW
2896# ifdef __GLIBC__
2897
175c4cf9 2898 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2899
2900# endif
65ebb059 2901
ccd65d51
KW
2902 /* NULL uses the existing already set up locale */
2903 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2904 ? NULL
2905 : "";
c3fcd832
KW
2906 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2907 unsigned int trial_locales_count;
175c4cf9
KW
2908 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2909 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2910 bool setlocale_failure = FALSE;
65ebb059 2911 unsigned int i;
175c4cf9
KW
2912
2913 /* A later getenv() could zap this, so only use here */
2914 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2915
2916 const bool locwarn = (printwarn > 1
e5f10d49
KW
2917 || ( printwarn
2918 && ( ! bad_lang_use_once
22ff3130 2919 || (
e5f10d49
KW
2920 /* disallow with "" or "0" */
2921 *bad_lang_use_once
2922 && strNE("0", bad_lang_use_once)))));
ea92aad8 2923
291a84fb 2924 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2925 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2926
2927 /* current locale for given category; should have been copied so aren't
2928 * volatile */
8de4332b 2929 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2930
7d4bcc4a
KW
2931# ifdef WIN32
2932
6bce99ee
JH
2933 /* In some systems you can find out the system default locale
2934 * and use that as the fallback locale. */
7d4bcc4a
KW
2935# define SYSTEM_DEFAULT_LOCALE
2936# endif
2937# ifdef SYSTEM_DEFAULT_LOCALE
2938
65ebb059 2939 const char *system_default_locale = NULL;
98994639 2940
7d4bcc4a 2941# endif
948523db
KW
2942
2943# ifndef DEBUGGING
2944# define DEBUG_LOCALE_INIT(a,b,c)
2945# else
7d4bcc4a 2946
8298454c 2947 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2948
2949# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2950 STMT_START { \
2951 if (debug_initialization) { \
2952 PerlIO_printf(Perl_debug_log, \
2953 "%s:%d: %s\n", \
2954 __FILE__, __LINE__, \
a4f00dcc 2955 setlocale_debug_string(category, \
2fcc0ca9
KW
2956 locale, \
2957 result)); \
2958 } \
2959 } STMT_END
2fcc0ca9 2960
948523db
KW
2961/* Make sure the parallel arrays are properly set up */
2962# ifdef USE_LOCALE_NUMERIC
2963 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2964 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
2965# ifdef USE_POSIX_2008_LOCALE
2966 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
2967# endif
948523db
KW
2968# endif
2969# ifdef USE_LOCALE_CTYPE
2970 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2971 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
2972# ifdef USE_POSIX_2008_LOCALE
2973 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
2974# endif
948523db
KW
2975# endif
2976# ifdef USE_LOCALE_COLLATE
2977 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2978 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
2979# ifdef USE_POSIX_2008_LOCALE
2980 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
2981# endif
948523db
KW
2982# endif
2983# ifdef USE_LOCALE_TIME
2984 assert(categories[LC_TIME_INDEX] == LC_TIME);
2985 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
2986# ifdef USE_POSIX_2008_LOCALE
2987 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
2988# endif
948523db
KW
2989# endif
2990# ifdef USE_LOCALE_MESSAGES
2991 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2992 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
2993# ifdef USE_POSIX_2008_LOCALE
2994 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
2995# endif
948523db
KW
2996# endif
2997# ifdef USE_LOCALE_MONETARY
2998 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2999 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3000# ifdef USE_POSIX_2008_LOCALE
3001 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3002# endif
948523db 3003# endif
9821811f
KW
3004# ifdef USE_LOCALE_ADDRESS
3005 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3006 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3007# ifdef USE_POSIX_2008_LOCALE
3008 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3009# endif
9821811f
KW
3010# endif
3011# ifdef USE_LOCALE_IDENTIFICATION
3012 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3013 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3014# ifdef USE_POSIX_2008_LOCALE
3015 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3016# endif
9821811f
KW
3017# endif
3018# ifdef USE_LOCALE_MEASUREMENT
3019 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3020 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3021# ifdef USE_POSIX_2008_LOCALE
3022 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3023# endif
9821811f
KW
3024# endif
3025# ifdef USE_LOCALE_PAPER
3026 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3027 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3028# ifdef USE_POSIX_2008_LOCALE
3029 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3030# endif
9821811f
KW
3031# endif
3032# ifdef USE_LOCALE_TELEPHONE
3033 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3034 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3035# ifdef USE_POSIX_2008_LOCALE
3036 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3037# endif
9821811f 3038# endif
948523db
KW
3039# ifdef LC_ALL
3040 assert(categories[LC_ALL_INDEX] == LC_ALL);
3041 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3042 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3043# ifdef USE_POSIX_2008_LOCALE
3044 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3045# endif
948523db
KW
3046# endif
3047# endif /* DEBUGGING */
47280b20
KW
3048
3049 /* Initialize the cache of the program's UTF-8ness for the always known
3050 * locales C and POSIX */
3051 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3052 sizeof(PL_locale_utf8ness));
3053
e9bc6d6b
KW
3054# ifdef USE_THREAD_SAFE_LOCALE
3055# ifdef WIN32
3056
3057 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3058
3059# endif
3060# endif
3061# if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3062
3063 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3064 if (! PL_C_locale_obj) {
3065 Perl_croak_nocontext(
3066 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3067 }
3068 if (DEBUG_Lv_TEST || debug_initialization) {
3069 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3070 }
3071
3072# endif
3073
3ca88433
KW
3074 PL_numeric_radix_sv = newSVpvs(".");
3075
e9bc6d6b
KW
3076# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3077
3078 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3079 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3080
3081# endif
b56c4436 3082# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3083
3084 /*
3085 * Ultrix setlocale(..., "") fails if there are no environment
3086 * variables from which to get a locale name.
3087 */
3088
b56c4436
KW
3089# ifndef LC_ALL
3090# error Ultrix without LC_ALL not implemented
3091# else
7d4bcc4a 3092
b56c4436
KW
3093 {
3094 bool done = FALSE;
ea92aad8
KW
3095 if (lang) {
3096 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3097 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3098 if (sl_result[LC_ALL_INDEX])
3099 done = TRUE;
3100 else
e5f10d49 3101 setlocale_failure = TRUE;
ea92aad8
KW
3102 }
3103 if (! setlocale_failure) {
3104 const char * locale_param;
3105 for (i = 0; i < LC_ALL_INDEX; i++) {
3106 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3107 ? setlocale_init
3108 : NULL;
3109 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3110 if (! sl_result[i]) {
3111 setlocale_failure = TRUE;
3112 }
3113 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3114 }
c835d6be 3115 }
7d4bcc4a
KW
3116 }
3117
3118# endif /* LC_ALL */
e5f10d49 3119# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3120
65ebb059 3121 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3122 * the list. Normally the loop is executed just once. But if setting the
3123 * locale fails, inside the loop we add fallback trials to the array and so
3124 * will execute the loop multiple times */
c3fcd832
KW
3125 trial_locales[0] = setlocale_init;
3126 trial_locales_count = 1;
7d4bcc4a 3127
65ebb059
KW
3128 for (i= 0; i < trial_locales_count; i++) {
3129 const char * trial_locale = trial_locales[i];
3130
3131 if (i > 0) {
3132
3133 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3134 * when i==0, but I (khw) don't think that behavior makes much
3135 * sense */
3136 setlocale_failure = FALSE;
3137
7d4bcc4a 3138# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3139# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3140
65ebb059
KW
3141 /* On Windows machines, an entry of "" after the 0th means to use
3142 * the system default locale, which we now proceed to get. */
3143 if (strEQ(trial_locale, "")) {
3144 unsigned int j;
3145
3146 /* Note that this may change the locale, but we are going to do
3147 * that anyway just below */
837ce802 3148 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3149 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3150
7d4bcc4a 3151 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3152 * try */
3153 if (! system_default_locale) {
3154 goto next_iteration;
3155 }
3156 for (j = 0; j < trial_locales_count; j++) {
3157 if (strEQ(system_default_locale, trial_locales[j])) {
3158 goto next_iteration;
3159 }
3160 }
3161
3162 trial_locale = system_default_locale;
3163 }
ec0202b5
KW
3164# else
3165# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3166# endif
7d4bcc4a 3167# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3168
3169 } /* For i > 0 */
65ebb059 3170
7d4bcc4a
KW
3171# ifdef LC_ALL
3172
948523db
KW
3173 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3174 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3175 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3176 setlocale_failure = TRUE;
7cd8b568
KW
3177 }
3178 else {
3179 /* Since LC_ALL succeeded, it should have changed all the other
3180 * categories it can to its value; so we massage things so that the
3181 * setlocales below just return their category's current values.
3182 * This adequately handles the case in NetBSD where LC_COLLATE may
3183 * not be defined for a locale, and setting it individually will
7d4bcc4a 3184 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3185 * the POSIX locale. */
3186 trial_locale = NULL;
3187 }
7d4bcc4a
KW
3188
3189# endif /* LC_ALL */
98994639 3190
e5f10d49
KW
3191 if (! setlocale_failure) {
3192 unsigned int j;
3193 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3194 curlocales[j]
3195 = savepv(do_setlocale_r(categories[j], trial_locale));
3196 if (! curlocales[j]) {
3197 setlocale_failure = TRUE;
3198 }
3199 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3200 }
c835d6be 3201
e5f10d49
KW
3202 if (! setlocale_failure) { /* All succeeded */
3203 break; /* Exit trial_locales loop */
49c85077 3204 }
65ebb059 3205 }
98994639 3206
49c85077
KW
3207 /* Here, something failed; will need to try a fallback. */
3208 ok = 0;
65ebb059 3209
49c85077
KW
3210 if (i == 0) {
3211 unsigned int j;
98994639 3212
65ebb059 3213 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3214
3215# ifdef LC_ALL
98994639 3216
49c85077
KW
3217 PerlIO_printf(Perl_error_log,
3218 "perl: warning: Setting locale failed.\n");
98994639 3219
7d4bcc4a 3220# else /* !LC_ALL */
98994639 3221
49c85077
KW
3222 PerlIO_printf(Perl_error_log,
3223 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3224
e5f10d49
KW
3225 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3226 if (! curlocales[j]) {
3227 PerlIO_printf(Perl_error_log, category_names[j]);
3228 }
3229 else {
3230 Safefree(curlocales[j]);
3231 }
3232 }
7d4bcc4a 3233
7d4bcc4a 3234# endif /* LC_ALL */
98994639 3235
49c85077
KW
3236 PerlIO_printf(Perl_error_log,
3237 "perl: warning: Please check that your locale settings:\n");
98994639 3238
7d4bcc4a
KW
3239# ifdef __GLIBC__
3240
49c85077
KW
3241 PerlIO_printf(Perl_error_log,
3242 "\tLANGUAGE = %c%s%c,\n",
3243 language ? '"' : '(',
3244 language ? language : "unset",
3245 language ? '"' : ')');
7d4bcc4a 3246# endif
98994639 3247
49c85077
KW
3248 PerlIO_printf(Perl_error_log,
3249 "\tLC_ALL = %c%s%c,\n",
3250 lc_all ? '"' : '(',
3251 lc_all ? lc_all : "unset",
3252 lc_all ? '"' : ')');
98994639 3253
7d4bcc4a
KW
3254# if defined(USE_ENVIRON_ARRAY)
3255
49c85077 3256 {
cd999af9 3257 char **e;
d5e32b93
KW
3258
3259 /* Look through the environment for any variables of the
3260 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3261 * already handled above. These are assumed to be locale
3262 * settings. Output them and their values. */
cd999af9 3263 for (e = environ; *e; e++) {
d5e32b93
KW
3264 const STRLEN prefix_len = sizeof("LC_") - 1;
3265 STRLEN uppers_len;
3266
cd999af9 3267 if ( strBEGINs(*e, "LC_")
c8b388b0 3268 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3269 && (uppers_len = strspn(*e + prefix_len,
3270 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3271 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3272 {
3273 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3274 (int) (prefix_len + uppers_len), *e,
3275 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3276 }
3277 }
49c85077 3278 }
7d4bcc4a
KW
3279
3280# else
3281
49c85077
KW
3282 PerlIO_printf(Perl_error_log,
3283 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3284
3285# endif
98994639 3286
49c85077
KW
3287 PerlIO_printf(Perl_error_log,
3288 "\tLANG = %c%s%c\n",
3289 lang ? '"' : '(',
3290 lang ? lang : "unset",
3291 lang ? '"' : ')');
98994639 3292
49c85077
KW
3293 PerlIO_printf(Perl_error_log,
3294 " are supported and installed on your system.\n");
3295 }
98994639 3296
65ebb059 3297 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3298 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3299 * usually change the upper bound of the loop we are in.
3300 *
3301 * Since the system's default way of setting the locale has not
3302 * found one that works, We use Perl's defined ordering: LC_ALL,
3303 * LANG, and the C locale. We don't try the same locale twice, so
3304 * don't add to the list if already there. (On POSIX systems, the
3305 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3306 * but there's no harm done by doing it explicitly.
3307 *
3308 * Note that this tries the LC_ALL environment variable even on
3309 * systems which have no LC_ALL locale setting. This may or may
3310 * not have been originally intentional, but there's no real need
3311 * to change the behavior. */
65ebb059
KW
3312 if (lc_all) {
3313 for (j = 0; j < trial_locales_count; j++) {
3314 if (strEQ(lc_all, trial_locales[j])) {
3315 goto done_lc_all;
3316 }
3317 }
3318 trial_locales[trial_locales_count++] = lc_all;
3319 }
3320 done_lc_all:
98994639 3321
65ebb059
KW
3322 if (lang) {
3323 for (j = 0; j < trial_locales_count; j++) {
3324 if (strEQ(lang, trial_locales[j])) {
3325 goto done_lang;
3326 }
3327 }
3328 trial_locales[trial_locales_count++] = lang;
3329 }
3330 done_lang:
3331
7d4bcc4a
KW
3332# if defined(WIN32) && defined(LC_ALL)
3333
65ebb059
KW
3334 /* For Windows, we also try the system default locale before "C".
3335 * (If there exists a Windows without LC_ALL we skip this because
3336 * it gets too complicated. For those, the "C" is the next
3337 * fallback possibility). The "" is the same as the 0th element of
3338 * the array, but the code at the loop above knows to treat it
3339 * differently when not the 0th */
3340 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3341
3342# endif
65ebb059
KW
3343
3344 for (j = 0; j < trial_locales_count; j++) {
3345 if (strEQ("C", trial_locales[j])) {
3346 goto done_C;
3347 }
3348 }
3349 trial_locales[trial_locales_count++] = "C";
98994639 3350
65ebb059
KW
3351 done_C: ;
3352 } /* end of first time through the loop */
98994639 3353
7d4bcc4a
KW
3354# ifdef WIN32
3355
65ebb059 3356 next_iteration: ;
7d4bcc4a
KW
3357
3358# endif
65ebb059
KW
3359
3360 } /* end of looping through the trial locales */
3361
3362 if (ok < 1) { /* If we tried to fallback */
3363 const char* msg;
3364 if (! setlocale_failure) { /* fallback succeeded */
3365 msg = "Falling back to";
3366 }
3367 else { /* fallback failed */
e5f10d49 3368 unsigned int j;
98994639 3369
65ebb059
KW
3370 /* We dropped off the end of the loop, so have to decrement i to
3371 * get back to the value the last time through */
3372 i--;
98994639 3373
65ebb059
KW
3374 ok = -1;
3375 msg = "Failed to fall back to";
3376
3377 /* To continue, we should use whatever values we've got */
7d4bcc4a 3378
e5f10d49
KW
3379 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3380 Safefree(curlocales[j]);
3381 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3382 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3383 }
65ebb059
KW
3384 }
3385
3386 if (locwarn) {
3387 const char * description;
3388 const char * name = "";
3389 if (strEQ(trial_locales[i], "C")) {
3390 description = "the standard locale";
3391 name = "C";
3392 }
7d4bcc4a
KW
3393
3394# ifdef SYSTEM_DEFAULT_LOCALE
3395
65ebb059
KW
3396 else if (strEQ(trial_locales[i], "")) {
3397 description = "the system default locale";
3398 if (system_default_locale) {
3399 name = system_default_locale;
3400 }
3401 }
7d4bcc4a
KW
3402
3403# endif /* SYSTEM_DEFAULT_LOCALE */
3404
65ebb059
KW
3405 else {
3406 description = "a fallback locale";
3407 name = trial_locales[i];
3408 }
3409 if (name && strNE(name, "")) {
3410 PerlIO_printf(Perl_error_log,
3411 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3412 }
3413 else {
3414 PerlIO_printf(Perl_error_log,
3415 "perl: warning: %s %s.\n", msg, description);
3416 }
3417 }
3418 } /* End of tried to fallback */
98994639 3419
e5f10d49
KW
3420 /* Done with finding the locales; update our records */
3421
7d4bcc4a
KW
3422# ifdef USE_LOCALE_CTYPE
3423
948523db 3424 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3425
e5f10d49 3426# endif
7d4bcc4a
KW
3427# ifdef USE_LOCALE_COLLATE
3428
948523db 3429 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3430
e5f10d49 3431# endif
7d4bcc4a
KW
3432# ifdef USE_LOCALE_NUMERIC
3433
948523db 3434 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3435
3436# endif
3437
948523db 3438 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3439
e9bc6d6b 3440# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3441
3442 /* This caches whether each category's locale is UTF-8 or not. This
3443 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3444 * initialization time before any threads have started, but not later
3445 * unless thread-safe operations are used.
47280b20
KW
3446 * Caching means that if the program heeds our dictate not to change
3447 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3448 * it may get queried without having to change locales. If the
3449 * environment is such that all categories have the same locale, this
3450 * isn't needed, as the code will not change the locale; but this
3451 * handles the uncommon case where the environment has disparate
3452 * locales for the categories */
47280b20
KW
3453 (void) _is_cur_LC_category_utf8(categories[i]);
3454
3455# endif
3456
e5f10d49
KW
3457 Safefree(curlocales[i]);
3458 }
b310b053 3459
7d4bcc4a
KW
3460# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3461
49c85077 3462 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3463 * locale is UTF-8. The call to new_ctype() just above has already
3464 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3465 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3466 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3467 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3468 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3469
a05d7ebb 3470 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3471 This is an alternative to using the -C command line switch
3472 (the -C if present will override this). */
3473 {
dd374669 3474 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3475 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3476 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3477 PL_utf8cache = -1;
b310b053
JH
3478 }
3479
7d4bcc4a 3480# endif
7d4bcc4a
KW
3481# ifdef __GLIBC__
3482
175c4cf9 3483 Safefree(language);
7d4bcc4a
KW
3484
3485# endif
175c4cf9
KW
3486
3487 Safefree(lc_all);
3488 Safefree(lang);
3489
e3305790 3490#endif /* USE_LOCALE */
2fcc0ca9 3491#ifdef DEBUGGING
7d4bcc4a 3492
2fcc0ca9 3493 /* So won't continue to output stuff */
27cdc72e 3494 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3495
2fcc0ca9
KW
3496#endif
3497
98994639
HS
3498 return ok;
3499}
3500
98994639
HS
3501#ifdef USE_LOCALE_COLLATE
3502
a4a439fb 3503char *
a4a439fb
KW
3504Perl__mem_collxfrm(pTHX_ const char *input_string,
3505 STRLEN len, /* Length of 'input_string' */
3506 STRLEN *xlen, /* Set to length of returned string
3507 (not including the collation index
3508 prefix) */
3509 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3510 )
98994639 3511{
a4a439fb
KW
3512
3513 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3514 * differences. First, it handles embedded NULs. Second, it allocates a bit
3515 * more memory than needed for the transformed data itself. The real
55e5378d 3516 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3517 * the length of that, and doesn't include the collation index size.
3518 * Please see sv_collxfrm() to see how this is used. */
3519
55e5378d
KW
3520#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3521
6696cfa7
KW
3522 char * s = (char *) input_string;
3523 STRLEN s_strlen = strlen(input_string);
79f120c8 3524 char *xbuf = NULL;
55e5378d 3525 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3526 STRLEN length_in_chars;
c664130f 3527 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3528
a4a439fb
KW
3529 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3530
3531 /* Must be NUL-terminated */
3532 assert(*(input_string + len) == '\0');
7918f24d 3533
79f120c8
KW
3534 /* If this locale has defective collation, skip */
3535 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3536 DEBUG_L(PerlIO_printf(Perl_debug_log,
3537 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3538 goto bad;
3539 }
3540
6696cfa7
KW
3541 /* Replace any embedded NULs with the control that sorts before any others.
3542 * This will give as good as possible results on strings that don't
3543 * otherwise contain that character, but otherwise there may be
3544 * less-than-perfect results with that character and NUL. This is
fdc080f3 3545 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3546 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3547 NUL */
6696cfa7
KW
3548 char * e = s + len;
3549 char * sans_nuls;
fdc080f3 3550 STRLEN sans_nuls_len;
94762aa0 3551 int try_non_controls;
afc4976f
KW
3552 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3553 making sure 2nd byte is NUL.
3554 */
3555 STRLEN this_replacement_len;
3556
1e4c9676
KW
3557 /* If we don't know what non-NUL control character sorts lowest for
3558 * this locale, find it */
f28f4d2a 3559 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3560 int j;
afc4976f 3561 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3562 includes the collation index
3563 prefixed. */
3564
91c0e2e0 3565 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3566
3567 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3568 * NUL, in which case we instead look for any character. Controls
3569 * are preferred because collation order is, in general, context
3570 * sensitive, with adjoining characters affecting the order, and
3571 * controls are less likely to have such interactions, allowing the
3572 * NUL-replacement to stand on its own. (Another way to look at it
3573 * is to imagine what would happen if the NUL were replaced by a
3574 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3575 for (try_non_controls = 0;
3576 try_non_controls < 2;
3577 try_non_controls++)
3578 {
d4ff9586
KW
3579 /* Look through all legal code points (NUL isn't) */
3580 for (j = 1; j < 256; j++) {
3581 char * x; /* j's xfrm plus collation index */
3582 STRLEN x_len; /* length of 'x' */
3583 STRLEN trial_len = 1;
736a4fed 3584 char cur_source[] = { '\0', '\0' };
d4ff9586 3585
736a4fed
KW
3586 /* Skip non-controls the first time through the loop. The
3587 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3588 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3589 ? ! isCNTRL_L1(j)
3590 : ! isCNTRL_LC(j))
3591 {
d4ff9586 3592 continue;
6696cfa7 3593 }
6696cfa7 3594
736a4fed
KW
3595 /* Create a 1-char string of the current code point */
3596 cur_source[0] = (char) j;
3597
d4ff9586
KW
3598 /* Then transform it */
3599 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3600 0 /* The string is not in UTF-8 */);
6696cfa7 3601
1e4c9676 3602 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3603 * */
3604 if (! x) {
3605 continue;
3606 }
6696cfa7 3607
d4ff9586
KW
3608 /* If this character's transformation is lower than
3609 * the current lowest, this one becomes the lowest */
3610 if ( cur_min_x == NULL
3611 || strLT(x + COLLXFRM_HDR_LEN,
3612 cur_min_x + COLLXFRM_HDR_LEN))
3613 {
f28f4d2a 3614 PL_strxfrm_NUL_replacement = j;
d4ff9586 3615 cur_min_x = x;
d4ff9586
KW
3616 }
3617 else {
3618 Safefree(x);
3619 }
1e4c9676 3620 } /* end of loop through all 255 characters */
6696cfa7 3621
1e4c9676 3622 /* Stop looking if found */
94762aa0
KW
3623 if (cur_min_x) {
3624 break;
3625 }
3626
3627 /* Unlikely, but possible, if there aren't any controls that
3628 * work in the locale, repeat the loop, looking for any
3629 * character that works */
3630 DEBUG_L(PerlIO_printf(Perl_debug_log,
3631 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3632 } /* End of loop to try first the controls, then any char */
6696cfa7 3633
94762aa0
KW
3634 if (! cur_min_x) {
3635 DEBUG_L(PerlIO_printf(Perl_debug_log,
3636 "_mem_collxfrm: Couldn't find any character to replace"
3637 " embedded NULs in locale %s with", PL_collation_name));
3638 goto bad;
58eebef2
KW
3639 }
3640
94762aa0
KW
3641 DEBUG_L(PerlIO_printf(Perl_debug_log,
3642 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3643 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3644
6696cfa7 3645 Safefree(cur_min_x);
1e4c9676 3646 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3647
3648 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3649 * UTF8-ness of the original */
f28f4d2a
KW
3650 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3651 this_replacement_char[0] =
3652 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3653 this_replacement_char[1] =
3654 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3655 this_replacement_len = 2;
3656 }
3657 else {
f28f4d2a 3658 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3659 /* this_replacement_char[1] = '\0' was done at initialization */
3660 this_replacement_len = 1;
6696cfa7
KW
3661 }
3662
3663 /* The worst case length for the replaced string would be if every
3664 * character in it is NUL. Multiply that by the length of each
3665 * replacement, and allow for a trailing NUL */
afc4976f 3666 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3667 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3668 *sans_nuls = '\0';
3669
6696cfa7
KW
3670 /* Replace each NUL with the lowest collating control. Loop until have
3671 * exhausted all the NULs */
3672 while (s + s_strlen < e) {
6069d6c5 3673 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3674
3675 /* Do the actual replacement */
6069d6c5 3676 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
3677
3678 /* Move past the input NUL */
3679 s += s_strlen + 1;
3680 s_strlen = strlen(s);
3681 }
3682
3683 /* And add anything that trails the final NUL */
6069d6c5 3684 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3685
3686 /* Switch so below we transform this modified string */
3687 s = sans_nuls;
3688 len = strlen(s);
1e4c9676 3689 } /* End of replacing NULs */
6696cfa7 3690
a4a439fb
KW
3691 /* Make sure the UTF8ness of the string and locale match */
3692 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 3693 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
3694 const char * const t = s; /* Temporary so we can later find where the
3695 input was */
3696
3697 /* Here they don't match. Change the string's to be what the locale is
3698 * expecting */
3699
3700 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3701 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3702 utf8 = TRUE;
3703 }
3704 else { /* locale is not UTF-8; but input is; downgrade the input */
3705
3706 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3707
3708 /* If the downgrade was successful we are done, but if the input
3709 * contains things that require UTF-8 to represent, have to do
3710 * damage control ... */
3711 if (UNLIKELY(utf8)) {
3712
3713 /* What we do is construct a non-UTF-8 string with
3714 * 1) the characters representable by a single byte converted
3715 * to be so (if necessary);
3716 * 2) and the rest converted to collate the same as the
3717 * highest collating representable character. That makes
3718 * them collate at the end. This is similar to how we
3719 * handle embedded NULs, but we use the highest collating
3720 * code point instead of the smallest. Like the NUL case,
3721 * this isn't perfect, but is the best we can reasonably
3722 * do. Every above-255 code point will sort the same as
3723 * the highest-sorting 0-255 code point. If that code
3724 * point can combine in a sequence with some other code
3725 * points for weight calculations, us changing something to
3726 * be it can adversely affect the results. But in most
3727 * cases, it should work reasonably. And note that this is
3728 * really an illegal situation: using code points above 255
3729 * on a locale where only 0-255 are valid. If two strings
3730 * sort entirely equal, then the sort order for the
3731 * above-255 code points will be in code point order. */
3732
3733 utf8 = FALSE;
3734
3735 /* If we haven't calculated the code point with the maximum
3736 * collating order for this locale, do so now */
3737 if (! PL_strxfrm_max_cp) {
3738 int j;
3739
3740 /* The current transformed string that collates the
3741 * highest (except it also includes the prefixed collation
3742 * index. */
3743 char * cur_max_x = NULL;
3744
3745 /* Look through all legal code points (NUL isn't) */
3746 for (j = 1; j < 256; j++) {
3747 char * x;
3748 STRLEN x_len;
736a4fed 3749 char cur_source[] = { '\0', '\0' };
a4a439fb 3750
736a4fed
KW
3751 /* Create a 1-char string of the current code point */
3752 cur_source[0] = (char) j;
a4a439fb
KW
3753
3754 /* Then transform it */
3755 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3756
3757 /* If something went wrong (which it shouldn't), just
3758 * ignore this code point */
94762aa0 3759 if (! x) {
a4a439fb
KW
3760 continue;
3761 }
3762
3763 /* If this character's transformation is higher than
3764 * the current highest, this one becomes the highest */
3765 if ( cur_max_x == NULL
55e5378d
KW
3766 || strGT(x + COLLXFRM_HDR_LEN,
3767 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
3768 {
3769 PL_strxfrm_max_cp = j;
3770 cur_max_x = x;
3771 }
3772 else {
3773 Safefree(x);
3774 }
3775 }
3776
94762aa0
KW
3777 if (! cur_max_x) {
3778 DEBUG_L(PerlIO_printf(Perl_debug_log,
3779 "_mem_collxfrm: Couldn't find any character to"
3780 " replace above-Latin1 chars in locale %s with",
3781 PL_collation_name));
3782 goto bad;
3783 }
3784
58eebef2
KW
3785 DEBUG_L(PerlIO_printf(Perl_debug_log,
3786 "_mem_collxfrm: highest 1-byte collating character"
3787 " in locale %s is 0x%02X\n",
3788 PL_collation_name,
3789 PL_strxfrm_max_cp));
58eebef2 3790
a4a439fb
KW
3791 Safefree(cur_max_x);
3792 }
3793
3794 /* Here we know which legal code point collates the highest.
3795 * We are ready to construct the non-UTF-8 string. The length
3796 * will be at least 1 byte smaller than the input string
3797 * (because we changed at least one 2-byte character into a
3798 * single byte), but that is eaten up by the trailing NUL */
3799 Newx(s, len, char);
3800
3801 {
3802 STRLEN i;
3803 STRLEN d= 0;
042d9e50 3804 char * e = (char *) t + len;
a4a439fb
KW
3805
3806 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3807 U8 cur_char = t[i];
3808 if (UTF8_IS_INVARIANT(cur_char)) {
3809 s[d++] = cur_char;
3810 }
042d9e50 3811 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
3812 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3813 }
3814 else { /* Replace illegal cp with highest collating
3815 one */
3816 s[d++] = PL_strxfrm_max_cp;
3817 }
3818 }
3819 s[d++] = '\0';
3820 Renew(s, d, char); /* Free up unused space */
3821 }
3822 }
3823 }
3824
3825 /* Here, we have constructed a modified version of the input. It could
3826 * be that we already had a modified copy before we did this version.
3827 * If so, that copy is no longer needed */
3828 if (t != input_string) {
3829 Safefree(t);
3830 }
3831 }
3832
17f41037
KW
3833 length_in_chars = (utf8)
3834 ? utf8_length((U8 *) s, (U8 *) s + len)
3835 : len;
3836
59c018b9
KW
3837 /* The first element in the output is the collation id, used by
3838 * sv_collxfrm(); then comes the space for the transformed string. The
3839 * equation should give us a good estimate as to how much is needed */
55e5378d 3840 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 3841 + PL_collxfrm_base
17f41037 3842 + (PL_collxfrm_mult * length_in_chars);
a02a5408 3843 Newx(xbuf, xAlloc, char);
c7202dee
KW
3844 if (UNLIKELY(! xbuf)) {
3845 DEBUG_L(PerlIO_printf(Perl_debug_log,
3846 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 3847 goto bad;
c7202dee 3848 }
98994639 3849
d35fca5f 3850 /* Store the collation id */
98994639 3851 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
3852
3853 /* Then the transformation of the input. We loop until successful, or we
3854 * give up */
4ebeff16 3855 for (;;) {
1adab0a7 3856
55e5378d 3857 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
3858
3859 /* If the transformed string occupies less space than we told strxfrm()
3860 * was available, it means it successfully transformed the whole
3861 * string. */
55e5378d 3862 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 3863
1adab0a7
KW
3864 /* Some systems include a trailing NUL in the returned length.
3865 * Ignore it, using a loop in case multiple trailing NULs are
3866 * returned. */
3867 while ( (*xlen) > 0
3868 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3869 {
3870 (*xlen)--;
3871 }
3872
17f41037
KW
3873 /* If the first try didn't get it, it means our prediction was low.
3874 * Modify the coefficients so that we predict a larger value in any
3875 * future transformations */
3876 if (! first_time) {
3877 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3878 STRLEN computed_guess = PL_collxfrm_base
3879 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
3880
3881 /* On zero-length input, just keep current slope instead of
3882 * dividing by 0 */
3883 const STRLEN new_m = (length_in_chars != 0)
3884 ? needed / length_in_chars
3885 : PL_collxfrm_mult;
17f41037
KW
3886
3887 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
3888 "%s: %d: initial size of %zu bytes for a length "
3889 "%zu string was insufficient, %zu needed\n",
17f41037 3890 __FILE__, __LINE__,
b07929e4 3891 computed_guess, length_in_chars, needed));
17f41037
KW
3892
3893 /* If slope increased, use it, but discard this result for
3894 * length 1 strings, as we can't be sure that it's a real slope
3895 * change */
3896 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
3897
3898# ifdef DEBUGGING
3899
17f41037
KW
3900 STRLEN old_m = PL_collxfrm_mult;
3901 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
3902
3903# endif
3904
17f41037
KW
3905 PL_collxfrm_mult = new_m;
3906 PL_collxfrm_base = 1; /* +1 For trailing NUL */
3907 computed_guess = PL_collxfrm_base
3908 + (PL_collxfrm_mult * length_in_chars);
3909 if (computed_guess < needed) {
3910 PL_collxfrm_base += needed - computed_guess;
3911 }
3912
3913 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
3914 "%s: %d: slope is now %zu; was %zu, base "
3915 "is now %zu; was %zu\n",
17f41037 3916 __FILE__, __LINE__,
b07929e4
KW
3917 PL_collxfrm_mult, old_m,
3918 PL_collxfrm_base, old_b));
17f41037
KW
3919 }
3920 else { /* Slope didn't change, but 'b' did */
3921 const STRLEN new_b = needed
3922 - computed_guess
3923 + PL_collxfrm_base;
3924 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 3925 "%s: %d: base is now %zu; was %zu\n",
17f41037 3926 __FILE__, __LINE__,
b07929e4 3927 new_b, PL_collxfrm_base));