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