This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
t/loc_tools.pl: 8859-16 is legit; 8859-0 isn't
[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;
645
646 /* If we have a valid LC_ALL value, just return it */
647 if (PL_curlocales[LC_ALL_INDEX]) {
648
649# ifdef DEBUGGING
650
651 if (DEBUG_Lv_TEST || debug_initialization) {
652 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
653 }
654
655# endif
656
657 return PL_curlocales[LC_ALL_INDEX];
658 }
659
660 /* Otherwise, we need to construct a string of name=value pairs.
661 * We use the glibc syntax, like
662 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
663 * First calculate the needed size. */
664 for (i = 0; i < LC_ALL_INDEX; i++) {
665
666# ifdef DEBUGGING
667
668 if (DEBUG_Lv_TEST || debug_initialization) {
669 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]);
670 }
671
672# endif
673
674 names_len += strlen(category_names[i])
675 + 1 /* '=' */
676 + strlen(PL_curlocales[i])
677 + 1; /* ';' */
678 }
679 names_len++; /* Trailing '\0' */
680 SAVEFREEPV(Newx(all_string, names_len, char));
681 *all_string = '\0';
682
683 /* Then fill in the string */
684 for (i = 0; i < LC_ALL_INDEX; i++) {
685
686# ifdef DEBUGGING
687
688 if (DEBUG_Lv_TEST || debug_initialization) {
689 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]);
690 }
691
692# endif
693
694 my_strlcat(all_string, category_names[i], names_len);
695 my_strlcat(all_string, "=", names_len);
696 my_strlcat(all_string, PL_curlocales[i], names_len);
697 my_strlcat(all_string, ";", names_len);
698 }
699
700# ifdef DEBUGGING
701
702 if (DEBUG_L_TEST || debug_initialization) {
703 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
704 }
705
706 #endif
707
708 return all_string;
709 }
710
711# ifdef EINVAL
712
713 SETERRNO(EINVAL, LIB_INVARG);
714
715# endif
716
717 return NULL;
718
d2b24094
KW
719# endif
720
e9bc6d6b
KW
721 }
722
723 assert(PL_C_locale_obj);
724
725 /* Otherwise, we are switching locales. This will generally entail freeing
726 * the current one's space (at the C library's discretion). We need to
727 * stop using that locale before the switch. So switch to a known locale
728 * object that we don't otherwise mess with. This returns the locale
729 * object in effect at the time of the switch. */
730 old_obj = uselocale(PL_C_locale_obj);
731
732# ifdef DEBUGGING
733
734 if (DEBUG_Lv_TEST || debug_initialization) {
735 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
736 }
737
738# endif
739
740 if (! old_obj) {
741
742# ifdef DEBUGGING
743
744 if (DEBUG_L_TEST || debug_initialization) {
745 dSAVE_ERRNO;
746 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
747 RESTORE_ERRNO;
748 }
749
750# endif
751
752 return NULL;
753 }
754
755# ifdef DEBUGGING
756
757 if (DEBUG_Lv_TEST || debug_initialization) {
758 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
759 }
760
761# endif
762
763 /* If we weren't in a thread safe locale, set so that newlocale() below
764 which uses 'old_obj', uses an empty one. Same for our reserved C object.
765 The latter is defensive coding, so that, even if there is some bug, we
766 will never end up trying to modify either of these, as if passed to
767 newlocale(), they can be. */
768 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
769 old_obj = (locale_t) 0;
770 }
771
772 /* Create the new locale (it may actually modify the current one). */
773
774# ifndef HAS_QUERYLOCALE
775
776 if (strEQ(locale, "")) {
777
778 /* For non-querylocale() systems, we do the setting of "" ourselves to
779 * be sure that we really know what's going on. We follow the Linux
780 * documented behavior (but if that differs from the actual behavior,
781 * this won't work exactly as the OS implements). We go out and
782 * examine the environment based on our understanding of how the system
783 * works, and use that to figure things out */
784
785 const char * const lc_all = PerlEnv_getenv("LC_ALL");
786
787 /* Use any "LC_ALL" environment variable, as it overrides everything
788 * else. */
789 if (lc_all && strNE(lc_all, "")) {
790 locale = lc_all;
791 }
792 else {
793
794 /* Otherwise, we need to dig deeper. Unless overridden, the
795 * default is the LANG environment variable; if it doesn't exist,
796 * then "C" */
797
798 const char * default_name;
799
800 /* To minimize other threads messing with the environment, we copy
801 * the variable, making it a temporary. But this doesn't work upon
802 * program initialization before any scopes are created, and at
803 * this time, there's nothing else going on that would interfere.
804 * So skip the copy in that case */
805 if (PL_scopestack_ix == 0) {
806 default_name = PerlEnv_getenv("LANG");
807 }
808 else {
809 default_name = savepv(PerlEnv_getenv("LANG"));
810 }
811
812 if (! default_name || strEQ(default_name, "")) {
813 default_name = "C";
814 }
815 else if (PL_scopestack_ix != 0) {
816 SAVEFREEPV(default_name);
817 }
818
819 if (category != LC_ALL) {
820 const char * const name = PerlEnv_getenv(category_names[index]);
821
822 /* Here we are setting a single category. Assume will have the
823 * default name */
824 locale = default_name;
825
826 /* But then look for an overriding environment variable */
827 if (name && strNE(name, "")) {
828 locale = name;
829 }
830 }
831 else {
832 bool did_override = FALSE;
833 unsigned int i;
834
835 /* Here, we are getting LC_ALL. Any categories that don't have
836 * a corresponding environment variable set should be set to
837 * LANG, or to "C" if there is no LANG. If no individual
838 * categories differ from this, we can just set LC_ALL. This
839 * is buggy on systems that have extra categories that we don't
840 * know about. If there is an environment variable that sets
841 * that category, we won't know to look for it, and so our use
842 * of LANG or "C" improperly overrides it. On the other hand,
843 * if we don't do what is done here, and there is no
844 * environment variable, the category's locale should be set to
845 * LANG or "C". So there is no good solution. khw thinks the
846 * best is to look at systems to see what categories they have,
847 * and include them, and then to assume that we know the
848 * complete set */
849
850 for (i = 0; i < LC_ALL_INDEX; i++) {
851 const char * const env_override
852 = savepv(PerlEnv_getenv(category_names[i]));
853 const char * this_locale = ( env_override
854 && strNE(env_override, ""))
855 ? env_override
856 : default_name;
857 emulate_setlocale(categories[i], this_locale, i, TRUE);
858
859 if (strNE(this_locale, default_name)) {
860 did_override = TRUE;
861 }
862
863 Safefree(env_override);
864 }
865
866 /* If all the categories are the same, we can set LC_ALL to
867 * that */
868 if (! did_override) {
869 locale = default_name;
870 }
871 else {
872
873 /* Here, LC_ALL is no longer valid, as some individual
874 * categories don't match it. We call ourselves
875 * recursively, as that will execute the code that
876 * generates the proper locale string for this situation.
877 * We don't do the remainder of this function, as that is
878 * to update our records, and we've just done that for the
879 * individual categories in the loop above, and doing so
880 * would cause LC_ALL to be done as well */
881 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
882 }
883 }
884 }
885 }
886 else if (strchr(locale, ';')) {
887
888 /* LC_ALL may actually incude a conglomeration of various categories.
889 * Without querylocale, this code uses the glibc (as of this writing)
890 * syntax for representing that, but that is not a stable API, and
891 * other platforms do it differently, so we have to handle all cases
892 * ourselves */
893
894 const char * s = locale;
895 const char * e = locale + strlen(locale);
896 const char * p = s;
897 const char * category_end;
898 const char * name_start;
899 const char * name_end;
900
901 while (s < e) {
902 unsigned int i;
903
904 /* Parse through the category */
905 while (isWORDCHAR(*p)) {
906 p++;
907 }
908 category_end = p;
909
910 if (*p++ != '=') {
911 Perl_croak(aTHX_
912 "panic: %s: %d: Unexpected character in locale name '%02X",
913 __FILE__, __LINE__, *(p-1));
914 }
915
916 /* Parse through the locale name */
917 name_start = p;
918 while (isGRAPH(*p) && *p != ';') {
919 p++;
920 }
921 name_end = p;
922
923 if (*p++ != ';') {
924 Perl_croak(aTHX_
925 "panic: %s: %d: Unexpected character in locale name '%02X",
926 __FILE__, __LINE__, *(p-1));
927 }
928
929 /* Find the index of the category name in our lists */
930 for (i = 0; i < LC_ALL_INDEX; i++) {
931
932 /* Keep going if this isn't the index. The strnNE() avoids a
933 * Perl_form(), but would fail if ever a category name could be
934 * a substring of another one, like if there were a
935 * "LC_TIME_DATE" */
936 if strnNE(s, category_names[i], category_end - s) {
937 continue;
938 }
939
940 /* If this index is for the single category we're changing, we
941 * have found the locale to set it to. */
942 if (category == categories[i]) {
943 locale = Perl_form(aTHX_ "%.*s",
944 (int) (name_end - name_start),
945 name_start);
946 goto ready_to_set;
947 }
948
949 if (category == LC_ALL) {
950 char * individ_locale = Perl_form(aTHX_ "%.*s", (int) (p - s), s);
951 emulate_setlocale(categories[i], individ_locale, i, TRUE);
952 Safefree(individ_locale);
953 }
954 }
955
956 s = p;
957 }
958
959 /* Here we have set all the individual categories by recursive calls.
960 * These collectively should have fixed up LC_ALL, so can just query
961 * what that now is */
962 assert(category == LC_ALL);
963
964 return do_setlocale_c(LC_ALL, NULL);
965 }
966
967 ready_to_set: ;
968
969# endif /* end of ! querylocale */
970
971 /* Ready to create a new locale by modification of the exising one */
972 new_obj = newlocale(mask, locale, old_obj);
973
974 if (! new_obj) {
975 dSAVE_ERRNO;
976
977# ifdef DEBUGGING
978
979 if (DEBUG_L_TEST || debug_initialization) {
980 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
981 }
982
983# endif
984
985 if (! uselocale(old_obj)) {
986 SAVE_ERRNO;
987
988# ifdef DEBUGGING
989
990 if (DEBUG_L_TEST || debug_initialization) {
991 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
992 }
993
994# endif
995
996 }
997 RESTORE_ERRNO;
998 return NULL;
999 }
1000
1001# ifdef DEBUGGING
1002
1003 if (DEBUG_Lv_TEST || debug_initialization) {
1004 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p\n", __FILE__, __LINE__, new_obj);
1005 }
1006
1007# endif
1008
1009 /* And switch into it */
1010 if (! uselocale(new_obj)) {
1011 dSAVE_ERRNO;
1012
1013# ifdef DEBUGGING
1014
1015 if (DEBUG_L_TEST || debug_initialization) {
1016 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1017 }
1018
1019# endif
1020
1021 if (! uselocale(old_obj)) {
1022
1023# ifdef DEBUGGING
1024
1025 if (DEBUG_L_TEST || debug_initialization) {
1026 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1027 }
1028
1029# endif
1030
1031 }
1032 freelocale(new_obj);
1033 RESTORE_ERRNO;
1034 return NULL;
1035 }
1036
1037# ifdef DEBUGGING
1038
1039 if (DEBUG_Lv_TEST || debug_initialization) {
1040 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1041 }
1042
1043# endif
1044
1045 /* We are done, except for updating our records (if the system doesn't keep
1046 * them) and in the case of locale "", we don't actually know what the
1047 * locale that got switched to is, as it came from the environment. So
1048 * have to find it */
1049
1050# ifdef HAS_QUERYLOCALE
1051
1052 if (strEQ(locale, "")) {
1053 locale = querylocale(mask, new_obj);
1054 }
1055
1056# else
1057
1058 /* Here, 'locale' is the return value */
1059
1060 /* Without querylocale(), we have to update our records */
1061
1062 if (category == LC_ALL) {
1063 unsigned int i;
1064
1065 /* For LC_ALL, we change all individual categories to correspond */
1066 /* PL_curlocales is a parallel array, so has same
1067 * length as 'categories' */
1068 for (i = 0; i <= LC_ALL_INDEX; i++) {
1069 Safefree(PL_curlocales[i]);
1070 PL_curlocales[i] = savepv(locale);
1071 }
1072 }
1073 else {
1074
1075 /* For a single category, if it's not the same as the one in LC_ALL, we
1076 * nullify LC_ALL */
1077
1078 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1079 Safefree(PL_curlocales[LC_ALL_INDEX]);
1080 PL_curlocales[LC_ALL_INDEX] = NULL;
1081 }
1082
1083 /* Then update the category's record */
1084 Safefree(PL_curlocales[index]);
1085 PL_curlocales[index] = savepv(locale);
1086 }
1087
1088# endif
1089
1090 return locale;
1091}
1092
1093#endif /* USE_POSIX_2008_LOCALE */
1094
1095#if 0 /* Code that was to emulate thread-safe locales on platforms that
1096 didn't natively support them */
1097
1098/* The way this would work is that we would keep a per-thread list of the
1099 * correct locale for that thread. Any operation that was locale-sensitive
1100 * would have to be changed so that it would look like this:
1101 *
1102 * LOCALE_LOCK;
1103 * setlocale to the correct locale for this operation
1104 * do operation
1105 * LOCALE_UNLOCK
1106 *
1107 * This leaves the global locale in the most recently used operation's, but it
1108 * was locked long enough to get the result. If that result is static, it
1109 * needs to be copied before the unlock.
1110 *
1111 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1112 * the setup, but are no-ops when not needed, and similarly,
1113 * END_LOCALE_DEPENDENT_OP for the tear-down
1114 *
1115 * But every call to a locale-sensitive function would have to be changed, and
1116 * if a module didn't cooperate by using the mutex, things would break.
1117 *
1118 * This code was abandoned before being completed or tested, and is left as-is
1119*/
1120
1121# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1122# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1123
1124STATIC char *
1125S_locking_setlocale(pTHX_
1126 const int category,
1127 const char * locale,
1128 int index,
1129 const bool is_index_valid
1130 )
1131{
1132 /* This function kind of performs a setlocale() on just the current thread;
1133 * thus it is kind of thread-safe. It does this by keeping a thread-level
1134 * array of the current locales for each category. Every time a locale is
1135 * switched to, it does the switch globally, but updates the thread's
1136 * array. A query as to what the current locale is just returns the
1137 * appropriate element from the array, and doesn't actually call the system
1138 * setlocale(). The saving into the array is done in an uninterruptible
1139 * section of code, so is unaffected by whatever any other threads might be
1140 * doing.
1141 *
1142 * All locale-sensitive operations must work by first starting a critical
1143 * section, then switching to the thread's locale as kept by this function,
1144 * and then doing the operation, then ending the critical section. Thus,
1145 * each gets done in the appropriate locale. simulating thread-safety.
1146 *
1147 * This function takes the same parameters, 'category' and 'locale', that
1148 * the regular setlocale() function does, but it also takes two additional
1149 * ones. This is because as described earlier. If we know on input the
1150 * index corresponding to the category into the array where we store the
1151 * current locales, we don't have to calculate it. If the caller knows at
1152 * compile time what the index is, it it can pass it, setting
1153 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1154 *
1155 */
1156
1157 /* If the input index might be incorrect, calculate the correct one */
1158 if (! is_index_valid) {
1159 unsigned int i;
1160
1161 if (DEBUG_Lv_TEST || debug_initialization) {
1162 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1163 }
1164
1165 for (i = 0; i <= LC_ALL_INDEX; i++) {
1166 if (category == categories[i]) {
1167 index = i;
1168 goto found_index;
1169 }
1170 }
1171
1172 /* Here, we don't know about this category, so can't handle it.
1173 * XXX best we can do is to unsafely set this
1174 * XXX warning */
1175
1176 return my_setlocale(category, locale);
1177
1178 found_index: ;
1179
1180 if (DEBUG_Lv_TEST || debug_initialization) {
1181 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1182 }
1183 }
1184
1185 /* For a query, just return what's in our records */
1186 if (new_locale == NULL) {
1187 return curlocales[index];
1188 }
1189
1190
1191 /* Otherwise, we need to do the switch, and save the result, all in a
1192 * critical section */
1193
1194 Safefree(curlocales[[index]]);
1195
1196 /* It might be that this is called from an already-locked section of code.
1197 * We would have to detect and skip the LOCK/UNLOCK if so */
1198 LOCALE_LOCK;
1199
1200 curlocales[index] = savepv(my_setlocale(category, new_locale));
1201
1202 if (strEQ(new_locale, "")) {
1203
1204#ifdef LC_ALL
1205
1206 /* The locale values come from the environment, and may not all be the
1207 * same, so for LC_ALL, we have to update all the others, while the
1208 * mutex is still locked */
1209
1210 if (category == LC_ALL) {
1211 unsigned int i;
1212 for (i = 0; i < LC_ALL_INDEX) {
1213 curlocales[i] = my_setlocale(categories[i], NULL);
1214 }
1215 }
1216 }
1217
1218#endif
1219
1220 LOCALE_UNLOCK;
1221
1222 return curlocales[index];
1223}
1224
1225#endif
837ce802 1226
a4f00dcc 1227STATIC void
86799d2d 1228S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1229{
86799d2d
KW
1230 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1231 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1232
86799d2d
KW
1233#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1234 || defined(HAS_NL_LANGINFO))
98994639 1235
87f8e8e7
KW
1236 const char * radix = (use_locale)
1237 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
1238 /* FALSE => already in dest locale */
1239 : ".";
2213a3be 1240
87f8e8e7 1241 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1242
87f8e8e7
KW
1243 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1244 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1245 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1246 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1247 && _is_cur_LC_category_utf8(LC_NUMERIC))
1248 {
1249 SvUTF8_on(PL_numeric_radix_sv);
1250 }
86799d2d
KW
1251
1252# ifdef DEBUGGING
7d4bcc4a 1253
2fcc0ca9
KW
1254 if (DEBUG_L_TEST || debug_initialization) {
1255 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1256 SvPVX(PL_numeric_radix_sv),
1257 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1258 }
69014004 1259
86799d2d
KW
1260# endif
1261#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1262
98994639
HS
1263}
1264
398eeea9
KW
1265STATIC void
1266S_new_numeric(pTHX_ const char *newnum)
98994639 1267{
7d4bcc4a
KW
1268
1269#ifndef USE_LOCALE_NUMERIC
1270
1271 PERL_UNUSED_ARG(newnum);
1272
1273#else
0d071d52 1274
291a84fb 1275 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1276 * core Perl this and that 'newnum' is the name of the new locale.
1277 * It installs this locale as the current underlying default.
1278 *
1279 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1280 * set_numeric_underlying() and set_numeric_standard() functions, which
1281 * should probably not be called directly, but only via macros like
0d071d52
KW
1282 * SET_NUMERIC_STANDARD() in perl.h.
1283 *
1284 * The toggling is necessary mainly so that a non-dot radix decimal point
1285 * character can be output, while allowing internal calculations to use a
1286 * dot.
1287 *
1288 * This sets several interpreter-level variables:
bb304765 1289 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1290 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1291 * that the current locale is the program's underlying
1292 * locale
1293 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1294 * that the current locale is the C locale or
1295 * indistinguishable from the C locale. If non-zero, it
1296 * is in C; if > 1, it means it may not be toggled away
7738054c 1297 * from C.
4c68b815
KW
1298 * PL_numeric_underlying_is_standard A bool kept by this function
1299 * indicating that the underlying locale and the standard
1300 * C locale are indistinguishable for the purposes of
1301 * LC_NUMERIC. This happens when both of the above two
1302 * variables are true at the same time. (Toggling is a
1303 * no-op under these circumstances.) This variable is
1304 * used to avoid having to recalculate.
398eeea9 1305 */
0d071d52 1306
b03f34cf 1307 char *save_newnum;
98994639
HS
1308
1309 if (! newnum) {
43c5f42d
NC
1310 Safefree(PL_numeric_name);
1311 PL_numeric_name = NULL;
98994639 1312 PL_numeric_standard = TRUE;
892e6465 1313 PL_numeric_underlying = TRUE;
4c68b815 1314 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1315 return;
1316 }
1317
b03f34cf 1318 save_newnum = stdize_locale(savepv(newnum));
892e6465 1319 PL_numeric_underlying = TRUE;
4c68b815
KW
1320 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1321
1322 /* If its name isn't C nor POSIX, it could still be indistinguishable from
1323 * them */
1324 if (! PL_numeric_standard) {
1325 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
1326 FALSE /* Don't toggle locale */ ))
1327 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
1328 FALSE)));
1329 }
abe1abcf 1330
4c68b815 1331 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1332 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1333 Safefree(PL_numeric_name);
b03f34cf 1334 PL_numeric_name = save_newnum;
b03f34cf 1335 }
abe1abcf
KW
1336 else {
1337 Safefree(save_newnum);
1338 }
4c28b29c 1339
4c68b815
KW
1340 PL_numeric_underlying_is_standard = PL_numeric_standard;
1341
7e5377f7 1342# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1343
1344 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1345 PL_numeric_name,
1346 PL_underlying_numeric_obj);
1347
1348#endif
1349
4c68b815
KW
1350 if (DEBUG_L_TEST || debug_initialization) {
1351 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1352 }
1353
4c28b29c
KW
1354 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1355 * have to worry about the radix being a non-dot. (Core operations that
1356 * need the underlying locale change to it temporarily). */
1357 set_numeric_standard();
1358
98994639 1359#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1360
98994639
HS
1361}
1362
1363void
1364Perl_set_numeric_standard(pTHX)
1365{
7d4bcc4a 1366
98994639 1367#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1368
28c1bf33
KW
1369 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1370 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1371 * macro avoids calling this routine if toggling isn't necessary according
1372 * to our records (which could be wrong if some XS code has changed the
1373 * locale behind our back) */
0d071d52 1374
837ce802 1375 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1376 PL_numeric_standard = TRUE;
4c68b815 1377 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1378 set_numeric_radix(0);
7d4bcc4a
KW
1379
1380# ifdef DEBUGGING
1381
2fcc0ca9
KW
1382 if (DEBUG_L_TEST || debug_initialization) {
1383 PerlIO_printf(Perl_debug_log,
58e4a467 1384 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1385 }
98994639 1386
7d4bcc4a 1387# endif
98994639 1388#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1389
98994639
HS
1390}
1391
1392void
5792c642 1393Perl_set_numeric_underlying(pTHX)
98994639 1394{
7d4bcc4a 1395
98994639 1396#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1397
28c1bf33 1398 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1399 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1400 * instead of calling this directly. The macro avoids calling this routine
1401 * if toggling isn't necessary according to our records (which could be
1402 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1403
837ce802 1404 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1405 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1406 PL_numeric_underlying = TRUE;
3c62fd46 1407 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1408
1409# ifdef DEBUGGING
1410
2fcc0ca9
KW
1411 if (DEBUG_L_TEST || debug_initialization) {
1412 PerlIO_printf(Perl_debug_log,
58e4a467 1413 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1414 PL_numeric_name);
1415 }
98994639 1416
7d4bcc4a 1417# endif
98994639 1418#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1419
98994639
HS
1420}
1421
1422/*
1423 * Set up for a new ctype locale.
1424 */
a4f00dcc
KW
1425STATIC void
1426S_new_ctype(pTHX_ const char *newctype)
98994639 1427{
7d4bcc4a
KW
1428
1429#ifndef USE_LOCALE_CTYPE
1430
1431 PERL_ARGS_ASSERT_NEW_CTYPE;
1432 PERL_UNUSED_ARG(newctype);
1433 PERL_UNUSED_CONTEXT;
1434
1435#else
0d071d52 1436
291a84fb 1437 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1438 * core Perl this and that 'newctype' is the name of the new locale.
1439 *
1440 * This function sets up the folding arrays for all 256 bytes, assuming
1441 * that tofold() is tolc() since fold case is not a concept in POSIX,
1442 *
1443 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1444 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1445 * this function should be called directly only from this file and from
0d071d52
KW
1446 * POSIX::setlocale() */
1447
27da23d5 1448 dVAR;
68067e4e 1449 UV i;
98994639 1450
8b7358b9
KW
1451 /* Don't check for problems if we are suppressing the warnings */
1452 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1453
7918f24d
NC
1454 PERL_ARGS_ASSERT_NEW_CTYPE;
1455
215c5139
KW
1456 /* We will replace any bad locale warning with 1) nothing if the new one is
1457 * ok; or 2) a new warning for the bad new locale */
1458 if (PL_warn_locale) {
1459 SvREFCNT_dec_NN(PL_warn_locale);
1460 PL_warn_locale = NULL;
1461 }
1462
c1284011 1463 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1464
1465 /* A UTF-8 locale gets standard rules. But note that code still has to
1466 * handle this specially because of the three problematic code points */
1467 if (PL_in_utf8_CTYPE_locale) {
1468 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1469 }
8b7358b9
KW
1470
1471 /* We don't populate the other lists if a UTF-8 locale, but do check that
1472 * everything works as expected, unless checking turned off */
1473 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1474 /* Assume enough space for every character being bad. 4 spaces each
1475 * for the 94 printable characters that are output like "'x' "; and 5
1476 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1477 * NUL */
8b7358b9 1478 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1479 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1480 to start */
1481 unsigned int bad_count = 0; /* Count of bad characters */
1482
baa60164 1483 for (i = 0; i < 256; i++) {
8b7358b9 1484 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1485 if (isupper(i))
1486 PL_fold_locale[i] = (U8) tolower(i);
1487 else if (islower(i))
1488 PL_fold_locale[i] = (U8) toupper(i);
1489 else
1490 PL_fold_locale[i] = (U8) i;
8b7358b9 1491 }
8c6180a9
KW
1492
1493 /* If checking for locale problems, see if the native ASCII-range
1494 * printables plus \n and \t are in their expected categories in
1495 * the new locale. If not, this could mean big trouble, upending
1496 * Perl's and most programs' assumptions, like having a
1497 * metacharacter with special meaning become a \w. Fortunately,
1498 * it's very rare to find locales that aren't supersets of ASCII
1499 * nowadays. It isn't a problem for most controls to be changed
1500 * into something else; we check only \n and \t, though perhaps \r
1501 * could be an issue as well. */
7d4bcc4a 1502 if ( check_for_problems
8c6180a9
KW
1503 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1504 {
8b7358b9
KW
1505 bool is_bad = FALSE;
1506 char name[3] = { '\0' };
1507
1508 /* Convert the name into a string */
1509 if (isPRINT_A(i)) {
1510 name[0] = i;
1511 name[1] = '\0';
1512 }
1513 else if (i == '\n') {
1514 my_strlcpy(name, "\n", sizeof(name));
1515 }
1516 else {
1517 my_strlcpy(name, "\t", sizeof(name));
1518 }
1519
1520 /* Check each possibe class */
1521 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1522 is_bad = TRUE;
1523 DEBUG_L(PerlIO_printf(Perl_debug_log,
1524 "isalnum('%s') unexpectedly is %d\n",
1525 name, cBOOL(isalnum(i))));
1526 }
1527 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1528 is_bad = TRUE;
1529 DEBUG_L(PerlIO_printf(Perl_debug_log,
1530 "isalpha('%s') unexpectedly is %d\n",
1531 name, cBOOL(isalpha(i))));
1532 }
1533 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1534 is_bad = TRUE;
1535 DEBUG_L(PerlIO_printf(Perl_debug_log,
1536 "isdigit('%s') unexpectedly is %d\n",
1537 name, cBOOL(isdigit(i))));
1538 }
1539 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1540 is_bad = TRUE;
1541 DEBUG_L(PerlIO_printf(Perl_debug_log,
1542 "isgraph('%s') unexpectedly is %d\n",
1543 name, cBOOL(isgraph(i))));
1544 }
1545 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1546 is_bad = TRUE;
1547 DEBUG_L(PerlIO_printf(Perl_debug_log,
1548 "islower('%s') unexpectedly is %d\n",
1549 name, cBOOL(islower(i))));
1550 }
1551 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1552 is_bad = TRUE;
1553 DEBUG_L(PerlIO_printf(Perl_debug_log,
1554 "isprint('%s') unexpectedly is %d\n",
1555 name, cBOOL(isprint(i))));
1556 }
1557 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1558 is_bad = TRUE;
1559 DEBUG_L(PerlIO_printf(Perl_debug_log,
1560 "ispunct('%s') unexpectedly is %d\n",
1561 name, cBOOL(ispunct(i))));
1562 }
1563 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1564 is_bad = TRUE;
1565 DEBUG_L(PerlIO_printf(Perl_debug_log,
1566 "isspace('%s') unexpectedly is %d\n",
1567 name, cBOOL(isspace(i))));
1568 }
1569 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1570 is_bad = TRUE;
1571 DEBUG_L(PerlIO_printf(Perl_debug_log,
1572 "isupper('%s') unexpectedly is %d\n",
1573 name, cBOOL(isupper(i))));
1574 }
1575 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1576 is_bad = TRUE;
1577 DEBUG_L(PerlIO_printf(Perl_debug_log,
1578 "isxdigit('%s') unexpectedly is %d\n",
1579 name, cBOOL(isxdigit(i))));
1580 }
1581 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1582 is_bad = TRUE;
1583 DEBUG_L(PerlIO_printf(Perl_debug_log,
1584 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1585 name, tolower(i), (int) toLOWER_A(i)));
1586 }
1587 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1588 is_bad = TRUE;
1589 DEBUG_L(PerlIO_printf(Perl_debug_log,
1590 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1591 name, toupper(i), (int) toUPPER_A(i)));
1592 }
1593 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1594 is_bad = TRUE;
1595 DEBUG_L(PerlIO_printf(Perl_debug_log,
1596 "'\\n' (=%02X) is not a control\n", (int) i));
1597 }
1598
1599 /* Add to the list; Separate multiple entries with a blank */
1600 if (is_bad) {
1601 if (bad_count) {
1602 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1603 }
8b7358b9
KW
1604 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1605 bad_count++;
8c6180a9
KW
1606 }
1607 }
1608 }
1609
7d4bcc4a
KW
1610# ifdef MB_CUR_MAX
1611
8c6180a9 1612 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1613 * this locale requires more than one byte, there are going to be
8c6180a9 1614 * problems. */
9c8a6dc2
KW
1615 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1616 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1617 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1618
8b7358b9
KW
1619 if ( check_for_problems && MB_CUR_MAX > 1
1620 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1621
1622 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1623 * locale. Just assume that the implementation for them (plus
1624 * for POSIX) is correct and the > 1 value is spurious. (Since
1625 * these are specially handled to never be considered UTF-8
1626 * locales, as long as this is the only problem, everything
1627 * should work fine */
1628 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1629 {
8c6180a9
KW
1630 multi_byte_locale = TRUE;
1631 }
7d4bcc4a
KW
1632
1633# endif
8c6180a9 1634
8b7358b9
KW
1635 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1636 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1637 PL_warn_locale = Perl_newSVpvf(aTHX_
1638 "Locale '%s' contains (at least) the following characters"
1639 " which have\nnon-standard meanings: %s\nThe Perl program"
1640 " will use the standard meanings",
1641 newctype, bad_chars_list);
1642
1643 }
1644 else {
1645 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1646 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1647 newctype,
8c6180a9
KW
1648 (multi_byte_locale)
1649 ? " Some characters in it are not recognized by"
1650 " Perl."
1651 : "",
1652 (bad_count)
1653 ? "\nThe following characters (and maybe others)"
1654 " may not have the same meaning as the Perl"
1655 " program expects:\n"
1656 : "",
1657 (bad_count)
1658 ? bad_chars_list
1659 : ""
1660 );
8b7358b9
KW
1661 }
1662
b119c2be
KW
1663# ifdef HAS_NL_LANGINFO
1664
1665 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1666 /* parameter FALSE is a don't care here */
1667 my_nl_langinfo(PERL_CODESET, FALSE));
1668
1669# endif
1670
8b7358b9
KW
1671 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1672
cc9eaeb0 1673 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1674 * output the message now. If not in that scope, we save the
1675 * message to be output at the first operation using this locale,
1676 * if that actually happens. Most programs don't use locales, so
1677 * they are immune to bad ones. */
cc9eaeb0 1678 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1679
780fcc9f
KW
1680 /* The '0' below suppresses a bogus gcc compiler warning */
1681 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1682
bddebb56
KW
1683 if (IN_LC(LC_CTYPE)) {
1684 SvREFCNT_dec_NN(PL_warn_locale);
1685 PL_warn_locale = NULL;
1686 }
780fcc9f 1687 }
baa60164 1688 }
31f05a37 1689 }
98994639
HS
1690
1691#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1692
98994639
HS
1693}
1694
98994639 1695void
2726666d
KW
1696Perl__warn_problematic_locale()
1697{
2726666d
KW
1698
1699#ifdef USE_LOCALE_CTYPE
1700
5f04a188
KW
1701 dTHX;
1702
1703 /* Internal-to-core function that outputs the message in PL_warn_locale,
1704 * and then NULLS it. Should be called only through the macro
1705 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1706
2726666d 1707 if (PL_warn_locale) {
2726666d
KW
1708 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1709 SvPVX(PL_warn_locale),
1710 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1711 SvREFCNT_dec_NN(PL_warn_locale);
1712 PL_warn_locale = NULL;
1713 }
1714
1715#endif
1716
1717}
1718
a4f00dcc
KW
1719STATIC void
1720S_new_collate(pTHX_ const char *newcoll)
98994639 1721{
7d4bcc4a
KW
1722
1723#ifndef USE_LOCALE_COLLATE
1724
1725 PERL_UNUSED_ARG(newcoll);
1726 PERL_UNUSED_CONTEXT;
1727
1728#else
0d071d52 1729
291a84fb 1730 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1731 * core Perl this and that 'newcoll' is the name of the new locale.
1732 *
d35fca5f
KW
1733 * The design of locale collation is that every locale change is given an
1734 * index 'PL_collation_ix'. The first time a string particpates in an
1735 * operation that requires collation while locale collation is active, it
1736 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1737 * magic includes the collation index, and the transformation of the string
1738 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1739 * instead of the string itself. If a string changes, the magic is
1740 * cleared. The next time the locale changes, the index is incremented,
1741 * and so we know during a comparison that the transformation is not
1742 * necessarily still valid, and so is recomputed. Note that if the locale
1743 * changes enough times, the index could wrap (a U32), and it is possible
1744 * that a transformation would improperly be considered valid, leading to
1745 * an unlikely bug */
0d071d52 1746
98994639
HS
1747 if (! newcoll) {
1748 if (PL_collation_name) {
1749 ++PL_collation_ix;
1750 Safefree(PL_collation_name);
1751 PL_collation_name = NULL;
1752 }
1753 PL_collation_standard = TRUE;
00bf60ca 1754 is_standard_collation:
98994639
HS
1755 PL_collxfrm_base = 0;
1756 PL_collxfrm_mult = 2;
165a1c52 1757 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1758 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1759 PL_strxfrm_max_cp = 0;
98994639
HS
1760 return;
1761 }
1762
d35fca5f 1763 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1764 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1765 ++PL_collation_ix;
1766 Safefree(PL_collation_name);
1767 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1768 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1769 if (PL_collation_standard) {
1770 goto is_standard_collation;
1771 }
98994639 1772
165a1c52 1773 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1774 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1775 PL_strxfrm_max_cp = 0;
165a1c52 1776
59c018b9
KW
1777 /* A locale collation definition includes primary, secondary, tertiary,
1778 * etc. weights for each character. To sort, the primary weights are
1779 * used, and only if they compare equal, then the secondary weights are
1780 * used, and only if they compare equal, then the tertiary, etc.
1781 *
1782 * strxfrm() works by taking the input string, say ABC, and creating an
1783 * output transformed string consisting of first the primary weights,
1784 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1785 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1786 * may not have weights at every level. In our example, let's say B
1787 * doesn't have a tertiary weight, and A doesn't have a secondary
1788 * weight. The constructed string is then going to be
1789 * A¹B¹C¹ B²C² A³C³ ....
1790 * This has the desired effect that strcmp() will look at the secondary
1791 * or tertiary weights only if the strings compare equal at all higher
1792 * priority weights. The spaces shown here, like in
c342d20e 1793 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1794 * are not just for readability. In the general case, these must
1795 * actually be bytes, which we will call here 'separator weights'; and
1796 * they must be smaller than any other weight value, but since these
1797 * are C strings, only the terminating one can be a NUL (some
1798 * implementations may include a non-NUL separator weight just before
1799 * the NUL). Implementations tend to reserve 01 for the separator
1800 * weights. They are needed so that a shorter string's secondary
1801 * weights won't be misconstrued as primary weights of a longer string,
1802 * etc. By making them smaller than any other weight, the shorter
1803 * string will sort first. (Actually, if all secondary weights are
1804 * smaller than all primary ones, there is no need for a separator
1805 * weight between those two levels, etc.)
1806 *
1807 * The length of the transformed string is roughly a linear function of
1808 * the input string. It's not exactly linear because some characters
1809 * don't have weights at all levels. When we call strxfrm() we have to
1810 * allocate some memory to hold the transformed string. The
1811 * calculations below try to find coefficients 'm' and 'b' for this
1812 * locale so that m*x + b equals how much space we need, given the size
1813 * of the input string in 'x'. If we calculate too small, we increase
1814 * the size as needed, and call strxfrm() again, but it is better to
1815 * get it right the first time to avoid wasted expensive string
1816 * transformations. */
1817
98994639 1818 {
79f120c8
KW
1819 /* We use the string below to find how long the tranformation of it
1820 * is. Almost all locales are supersets of ASCII, or at least the
1821 * ASCII letters. We use all of them, half upper half lower,
1822 * because if we used fewer, we might hit just the ones that are
1823 * outliers in a particular locale. Most of the strings being
1824 * collated will contain a preponderance of letters, and even if
1825 * they are above-ASCII, they are likely to have the same number of
1826 * weight levels as the ASCII ones. It turns out that digits tend
1827 * to have fewer levels, and some punctuation has more, but those
1828 * are relatively sparse in text, and khw believes this gives a
1829 * reasonable result, but it could be changed if experience so
1830 * dictates. */
1831 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1832 char * x_longer; /* Transformed 'longer' */
1833 Size_t x_len_longer; /* Length of 'x_longer' */
1834
1835 char * x_shorter; /* We also transform a substring of 'longer' */
1836 Size_t x_len_shorter;
1837
a4a439fb 1838 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1839 * are interested only in its length). It is used because it has
1840 * the intelligence to handle all cases, but to work, it needs some
1841 * values of 'm' and 'b' to get it started. For the purposes of
1842 * this calculation we use a very conservative estimate of 'm' and
1843 * 'b'. This assumes a weight can be multiple bytes, enough to
1844 * hold any UV on the platform, and there are 5 levels, 4 weight
1845 * bytes, and a trailing NUL. */
1846 PL_collxfrm_base = 5;
1847 PL_collxfrm_mult = 5 * sizeof(UV);
1848
1849 /* Find out how long the transformation really is */
a4a439fb
KW
1850 x_longer = _mem_collxfrm(longer,
1851 sizeof(longer) - 1,
1852 &x_len_longer,
1853
1854 /* We avoid converting to UTF-8 in the
1855 * called function by telling it the
1856 * string is in UTF-8 if the locale is a
1857 * UTF-8 one. Since the string passed
1858 * here is invariant under UTF-8, we can
1859 * claim it's UTF-8 even though it isn't.
1860 * */
1861 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1862 Safefree(x_longer);
1863
1864 /* Find out how long the transformation of a substring of 'longer'
1865 * is. Together the lengths of these transformations are
1866 * sufficient to calculate 'm' and 'b'. The substring is all of
1867 * 'longer' except the first character. This minimizes the chances
1868 * of being swayed by outliers */
a4a439fb 1869 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1870 sizeof(longer) - 2,
a4a439fb
KW
1871 &x_len_shorter,
1872 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1873 Safefree(x_shorter);
1874
1875 /* If the results are nonsensical for this simple test, the whole
1876 * locale definition is suspect. Mark it so that locale collation
1877 * is not active at all for it. XXX Should we warn? */
1878 if ( x_len_shorter == 0
1879 || x_len_longer == 0
1880 || x_len_shorter >= x_len_longer)
1881 {
1882 PL_collxfrm_mult = 0;
1883 PL_collxfrm_base = 0;
1884 }
1885 else {
1886 SSize_t base; /* Temporary */
1887
1888 /* We have both: m * strlen(longer) + b = x_len_longer
1889 * m * strlen(shorter) + b = x_len_shorter;
1890 * subtracting yields:
1891 * m * (strlen(longer) - strlen(shorter))
1892 * = x_len_longer - x_len_shorter
1893 * But we have set things up so that 'shorter' is 1 byte smaller
1894 * than 'longer'. Hence:
1895 * m = x_len_longer - x_len_shorter
1896 *
1897 * But if something went wrong, make sure the multiplier is at
1898 * least 1.
1899 */
1900 if (x_len_longer > x_len_shorter) {
1901 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1902 }
1903 else {
1904 PL_collxfrm_mult = 1;
1905 }
1906
1907 /* mx + b = len
1908 * so: b = len - mx
1909 * but in case something has gone wrong, make sure it is
1910 * non-negative */
1911 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1912 if (base < 0) {
1913 base = 0;
1914 }
1915
1916 /* Add 1 for the trailing NUL */
1917 PL_collxfrm_base = base + 1;
1918 }
58eebef2 1919
7d4bcc4a
KW
1920# ifdef DEBUGGING
1921
58eebef2
KW
1922 if (DEBUG_L_TEST || debug_initialization) {
1923 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1924 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1925 "x_len_longer=%zu,"
1926 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1927 __FILE__, __LINE__,
1928 PL_in_utf8_COLLATE_locale,
1929 x_len_shorter, x_len_longer,
1930 PL_collxfrm_mult, PL_collxfrm_base);
1931 }
7d4bcc4a
KW
1932# endif
1933
98994639
HS
1934 }
1935 }
1936
1937#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1938
98994639
HS
1939}
1940
d2b24094 1941#ifdef WIN32
b8cc575c 1942
a4f00dcc 1943STATIC char *
b8cc575c 1944S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1945{
1946 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1947 * difference between the two unless the input locale is "", which normally
1948 * means on Windows to get the machine default, which is set via the
1949 * computer's "Regional and Language Options" (or its current equivalent).
1950 * In POSIX, it instead means to find the locale from the user's
1951 * environment. This routine changes the Windows behavior to first look in
1952 * the environment, and, if anything is found, use that instead of going to
1953 * the machine default. If there is no environment override, the machine
1954 * default is used, by calling the real setlocale() with "".
1955 *
1956 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1957 * use the particular category's variable if set; otherwise to use the LANG
1958 * variable. */
b385bb4d 1959
175c4cf9 1960 bool override_LC_ALL = FALSE;
89f7b9aa 1961 char * result;
e5f10d49 1962 unsigned int i;
89f7b9aa 1963
b385bb4d 1964 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1965
1966# ifdef LC_ALL
1967
b385bb4d
KW
1968 locale = PerlEnv_getenv("LC_ALL");
1969 if (! locale) {
e5f10d49
KW
1970 if (category == LC_ALL) {
1971 override_LC_ALL = TRUE;
1972 }
1973 else {
7d4bcc4a
KW
1974
1975# endif
7d4bcc4a 1976
e5f10d49
KW
1977 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1978 if (category == categories[i]) {
1979 locale = PerlEnv_getenv(category_names[i]);
1980 goto found_locale;
1981 }
1982 }
7d4bcc4a 1983
b385bb4d 1984 locale = PerlEnv_getenv("LANG");
481465ea 1985 if (! locale) {
b385bb4d
KW
1986 locale = "";
1987 }
e5f10d49
KW
1988
1989 found_locale: ;
7d4bcc4a
KW
1990
1991# ifdef LC_ALL
1992
e5f10d49 1993 }
b385bb4d 1994 }
7d4bcc4a
KW
1995
1996# endif
1997
b385bb4d
KW
1998 }
1999
89f7b9aa 2000 result = setlocale(category, locale);
48015184
KW
2001 DEBUG_L(STMT_START {
2002 dSAVE_ERRNO;
2003 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2004 setlocale_debug_string(category, locale, result));
2005 RESTORE_ERRNO;
2006 } STMT_END);
89f7b9aa 2007
481465ea 2008 if (! override_LC_ALL) {
89f7b9aa
KW
2009 return result;
2010 }
2011
dfd77d7a 2012 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2013 * LANG variable or the system default if there is no LANG. But these have
2014 * lower priority than the other LC_foo variables, so override it for each
2015 * one that is set. (If they are set to "", it means to use the same thing
2016 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2017
948523db 2018 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2019 result = PerlEnv_getenv(category_names[i]);
2020 if (result && strNE(result, "")) {
2021 setlocale(categories[i], result);
2022 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2023 __FILE__, __LINE__,
2024 setlocale_debug_string(categories[i], result, "not captured")));
2025 }
89f7b9aa 2026 }
7d4bcc4a 2027
bbc98134 2028 result = setlocale(LC_ALL, NULL);
48015184
KW
2029 DEBUG_L(STMT_START {
2030 dSAVE_ERRNO;
2031 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2032 __FILE__, __LINE__,
48015184
KW
2033 setlocale_debug_string(LC_ALL, NULL, result));
2034 RESTORE_ERRNO;
2035 } STMT_END);
89f7b9aa 2036
bbc98134 2037 return result;
b385bb4d
KW
2038}
2039
2040#endif
2041
9aac5db8
KW
2042/*
2043
2044=head1 Locale-related functions and macros
2045
2046=for apidoc Perl_setlocale
2047
2048This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2049taking the same parameters, and returning the same information, except that it
2050returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2051perl keeps that locale category as C<C>, changing it briefly during the
2052operations where the underlying one is required.
2053
e9bc6d6b 2054Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2055declared to return S<C<const char *>>, whereas the system setlocale omits the
2056C<const>. (If it were being written today, plain setlocale would be declared
2057const, since it is illegal to change the information it returns; doing so leads
2058to segfaults.)
2059
e9bc6d6b
KW
2060Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2061C<setlocale> can be completely ineffective on some platforms under some
2062configurations.
2063
9aac5db8 2064C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2065where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2066the system C<setlocale()> is ineffective, returning the wrong information, and
2067failing to actually change the locale. C<Perl_setlocale>, however works
2068properly in all circumstances.
9aac5db8
KW
2069
2070The return points to a per-thread static buffer, which is overwritten the next
2071time C<Perl_setlocale> is called from the same thread.
2072
2073=cut
2074
2075*/
2076
2077const char *
2078Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2079{
2080 /* This wraps POSIX::setlocale() */
2081
9aac5db8
KW
2082 const char * retval;
2083 const char * newlocale;
2084 dSAVEDERRNO;
2085 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2086 dTHX;
2087
a4f00dcc
KW
2088#ifdef USE_LOCALE_NUMERIC
2089
291a84fb
KW
2090 /* A NULL locale means only query what the current one is. We have the
2091 * LC_NUMERIC name saved, because we are normally switched into the C
2092 * locale for it. For an LC_ALL query, switch back to get the correct
2093 * results. All other categories don't require special handling */
a4f00dcc
KW
2094 if (locale == NULL) {
2095 if (category == LC_NUMERIC) {
9aac5db8
KW
2096
2097 /* We don't have to copy this return value, as it is a per-thread
2098 * variable, and won't change until a future setlocale */
2099 return PL_numeric_name;
a4f00dcc
KW
2100 }
2101
7d4bcc4a 2102# ifdef LC_ALL
a4f00dcc 2103
9aac5db8
KW
2104 else if (category == LC_ALL) {
2105 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2106 }
2107
7d4bcc4a 2108# endif
a4f00dcc
KW
2109
2110 }
2111
2112#endif
2113
9aac5db8
KW
2114 retval = do_setlocale_r(category, locale);
2115 SAVE_ERRNO;
2116
2117#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2118
2119 if (locale == NULL && category == LC_ALL) {
2120 RESTORE_LC_NUMERIC();
2121 }
2122
2123#endif
a4f00dcc
KW
2124
2125 DEBUG_L(PerlIO_printf(Perl_debug_log,
2126 "%s:%d: %s\n", __FILE__, __LINE__,
2127 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2128
9aac5db8
KW
2129 RESTORE_ERRNO;
2130
2131 if (! retval) {
a4f00dcc
KW
2132 return NULL;
2133 }
2134
9aac5db8
KW
2135 save_to_buffer(retval, &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2136 retval = PL_setlocale_buf;
2137
2138 /* If locale == NULL, we are just querying the state */
a4f00dcc 2139 if (locale == NULL) {
a4f00dcc
KW
2140 return retval;
2141 }
a4f00dcc 2142
1159483a
KW
2143 /* Now that have switched locales, we have to update our records to
2144 * correspond. */
a4f00dcc 2145
1159483a 2146 switch (category) {
a4f00dcc 2147
1159483a 2148#ifdef USE_LOCALE_CTYPE
a4f00dcc 2149
1159483a
KW
2150 case LC_CTYPE:
2151 new_ctype(retval);
2152 break;
a4f00dcc 2153
1159483a 2154#endif
a4f00dcc
KW
2155#ifdef USE_LOCALE_COLLATE
2156
1159483a
KW
2157 case LC_COLLATE:
2158 new_collate(retval);
2159 break;
a4f00dcc 2160
1159483a 2161#endif
a4f00dcc
KW
2162#ifdef USE_LOCALE_NUMERIC
2163
1159483a
KW
2164 case LC_NUMERIC:
2165 new_numeric(retval);
2166 break;
a4f00dcc 2167
1159483a
KW
2168#endif
2169#ifdef LC_ALL
a4f00dcc 2170
1159483a 2171 case LC_ALL:
a4f00dcc 2172
1159483a
KW
2173 /* LC_ALL updates all the things we care about. The values may not
2174 * be the same as 'retval', as the locale "" may have set things
2175 * individually */
a4f00dcc 2176
1159483a 2177# ifdef USE_LOCALE_CTYPE
a4f00dcc 2178
1159483a
KW
2179 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2180 new_ctype(newlocale);
a4f00dcc 2181
1159483a
KW
2182# endif /* USE_LOCALE_CTYPE */
2183# ifdef USE_LOCALE_COLLATE
2184
2185 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2186 new_collate(newlocale);
a4f00dcc 2187
7d4bcc4a 2188# endif
1159483a 2189# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2190
1159483a
KW
2191 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2192 new_numeric(newlocale);
a4f00dcc 2193
1159483a
KW
2194# endif /* USE_LOCALE_NUMERIC */
2195#endif /* LC_ALL */
a4f00dcc 2196
1159483a
KW
2197 default:
2198 break;
a4f00dcc
KW
2199 }
2200
2201 return retval;
2202
f7416781
KW
2203}
2204
2205PERL_STATIC_INLINE const char *
2206S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2207{
2208 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2209 * growing it if necessary */
2210
2211 const Size_t string_size = strlen(string) + offset + 1;
2212
2213 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2214
2215 if (*buf_size == 0) {
2216 Newx(*buf, string_size, char);
2217 *buf_size = string_size;
2218 }
2219 else if (string_size > *buf_size) {
2220 Renew(*buf, string_size, char);
2221 *buf_size = string_size;
2222 }
2223
2224 Copy(string, *buf + offset, string_size - offset, char);
2225 return *buf;
2226}
2227
2228/*
2229
f7416781
KW
2230=for apidoc Perl_langinfo
2231
ce181bf6 2232This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2233taking the same C<item> parameter values, and returning the same information.
2234But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2235of Perl's locale handling from your code, and can be used on systems that lack
2236a native C<nl_langinfo>.
2237
2238Expanding on these:
2239
2240=over
2241
2242=item *
2243
ce181bf6
KW
2244The reason it isn't quite a drop-in replacement is actually an advantage. The
2245only difference is that it returns S<C<const char *>>, whereas plain
2246C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2247forbidden to write into the buffer. By declaring this C<const>, the compiler
2248enforces this restriction, so if it is violated, you know at compilation time,
2249rather than getting segfaults at runtime.
2250
2251=item *
2252
f7416781
KW
2253It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
2254without you having to write extra code. The reason for the extra code would be
2255because these are from the C<LC_NUMERIC> locale category, which is normally
2256kept set to the C locale by Perl, no matter what the underlying locale is
2257supposed to be, and so to get the expected results, you have to temporarily
ce181bf6
KW
2258toggle into the underlying locale, and later toggle back. (You could use plain
2259C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2260then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2261C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2262radix (decimal point) character to be a dot.)
2263
2264=item *
2265
2266The system function it replaces can have its static return buffer trashed,
2267not only by a subesequent call to that function, but by a C<freelocale>,
2268C<setlocale>, or other locale change. The returned buffer of this function is
2269not changed until the next call to it, so the buffer is never in a trashed
2270state.
f7416781
KW
2271
2272=item *
2273
ce181bf6
KW
2274Its return buffer is per-thread, so it also is never overwritten by a call to
2275this function from another thread; unlike the function it replaces.
2276
2277=item *
2278
2279But most importantly, it works on systems that don't have C<nl_langinfo>, such
2280as Windows, hence makes your code more portable. Of the fifty-some possible
2281items specified by the POSIX 2008 standard,
f7416781 2282L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
ce181bf6
KW
2283only two are completely unimplemented (though the loss of one of these is
2284significant). It uses various techniques to recover the other items, including
2285calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2286in C89, so should be always be available. Later C<strftime()> versions have
2287additional capabilities; C<""> is returned for those not available on your
2288system.
2289
2290It is important to note that when called with an item that is recovered by
2291using C<localeconv>, the buffer from any previous explicit call to
2292C<localeconv> will be overwritten. This means you must save that buffer's
2293contents if you need to access them after a call to this function.
9f0bc911 2294
f7416781
KW
2295The details for those items which may differ from what this emulation returns
2296and what a native C<nl_langinfo()> would return are:
2297
2298=over
2299
2300=item C<CODESET>
2301
2302=item C<ERA>
2303
2304Unimplemented, so returns C<"">.
2305
2306=item C<YESEXPR>
2307
c1566110
KW
2308=item C<YESSTR>
2309
f7416781
KW
2310=item C<NOEXPR>
2311
c1566110
KW
2312=item C<NOSTR>
2313
2314Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
ce181bf6
KW
2315removed from POSIX 2008, and are retained here for backwards compatibility.
2316Your platform's C<nl_langinfo> may not support them.
f7416781
KW
2317
2318=item C<D_FMT>
2319
2320Always evaluates to C<%x>, the locale's appropriate date representation.
2321
2322=item C<T_FMT>
2323
2324Always evaluates to C<%X>, the locale's appropriate time representation.
2325
2326=item C<D_T_FMT>
2327
2328Always evaluates to C<%c>, the locale's appropriate date and time
2329representation.
2330
2331=item C<CRNCYSTR>
2332
2333The return may be incorrect for those rare locales where the currency symbol
2334replaces the radix character.
2335Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2336to work differently.
2337
2338=item C<ALT_DIGITS>
2339
2340Currently this gives the same results as Linux does.
2341Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2342to work differently.
2343
2344=item C<ERA_D_FMT>
2345
2346=item C<ERA_T_FMT>
2347
2348=item C<ERA_D_T_FMT>
2349
2350=item C<T_FMT_AMPM>
2351
2352These are derived by using C<strftime()>, and not all versions of that function
2353know about them. C<""> is returned for these on such systems.
2354
2355=back
2356
ce181bf6
KW
2357=back
2358
f7416781
KW
2359When using C<Perl_langinfo> on systems that don't have a native
2360C<nl_langinfo()>, you must
2361
2362 #include "perl_langinfo.h"
2363
2364before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2365C<#include> with this one. (Doing it this way keeps out the symbols that plain
2366C<langinfo.h> imports into the namespace for code that doesn't need it.)
2367
2368You also should not use the bare C<langinfo.h> item names, but should preface
2369them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2370The C<PERL_I<foo>> versions will also work for this function on systems that do
2371have a native C<nl_langinfo>.
2372
f7416781
KW
2373The original impetus for C<Perl_langinfo()> was so that code that needs to
2374find out the current currency symbol, floating point radix character, or digit
2375grouping separator can use, on all systems, the simpler and more
2376thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2377pain to make thread-friendly. For other fields returned by C<localeconv>, it
2378is better to use the methods given in L<perlcall> to call
2379L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2380
2381=cut
2382
2383*/
2384
2385const char *
2386#ifdef HAS_NL_LANGINFO
2387Perl_langinfo(const nl_item item)
2388#else
2389Perl_langinfo(const int item)
2390#endif
2391{
f61748ac
KW
2392 return my_nl_langinfo(item, TRUE);
2393}
2394
2395const char *
2396#ifdef HAS_NL_LANGINFO
2397S_my_nl_langinfo(const nl_item item, bool toggle)
2398#else
2399S_my_nl_langinfo(const int item, bool toggle)
2400#endif
2401{
ae74815b 2402 dTHX;
f7416781 2403
5a854ab3
KW
2404 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2405 * two items, and only if not already there */
2406 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2407 || PL_numeric_underlying))
2408 {
2409 toggle = FALSE;
2410 }
2411
ab340fff 2412#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d
KW
2413# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2414 || ! defined(HAS_POSIX_2008_LOCALE)
f7416781 2415
ab340fff 2416 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2417 * for those items dependent on it. This must be copied to a buffer before
2418 * switching back, as some systems destroy the buffer when setlocale() is
2419 * called */
f7416781 2420
038d3702
KW
2421 {
2422 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2423
b2fee59c 2424 if (toggle) {
038d3702 2425 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2426 }
f7416781 2427
5acc4454
KW
2428 LOCALE_LOCK; /* Prevent interference from another thread executing
2429 this code section (the only call to nl_langinfo in
2430 the core) */
2431
ee90eb2d 2432
628ff75a
KW
2433 /* Copy to a per-thread buffer, which is also one that won't be
2434 * destroyed by a subsequent setlocale(), such as the
2435 * RESTORE_LC_NUMERIC may do just below. */
2436 save_to_buffer(nl_langinfo(item),
2437 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2438
5acc4454
KW
2439 LOCALE_UNLOCK;
2440
b2fee59c 2441 if (toggle) {
038d3702 2442 RESTORE_LC_NUMERIC();
b2fee59c 2443 }
f7416781
KW
2444 }
2445
ab340fff
KW
2446# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2447
5a854ab3 2448 {
b2fee59c
KW
2449 bool do_free = FALSE;
2450 locale_t cur = uselocale((locale_t) 0);
ab340fff 2451
b2fee59c
KW
2452 if (cur == LC_GLOBAL_LOCALE) {
2453 cur = duplocale(LC_GLOBAL_LOCALE);
2454 do_free = TRUE;
2455 }
ab340fff 2456
b2fee59c 2457 if (toggle) {
e1aa2579
KW
2458 if (PL_underlying_numeric_obj) {
2459 cur = PL_underlying_numeric_obj;
2460 }
2461 else {
2462 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2463 do_free = TRUE;
2464 }
b2fee59c 2465 }
ab340fff 2466
628ff75a
KW
2467 /* We have to save it to a buffer, because the freelocale() just below
2468 * can invalidate the internal one */
2469 save_to_buffer(nl_langinfo_l(item, cur),
2470 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2471
b2fee59c
KW
2472 if (do_free) {
2473 freelocale(cur);
2474 }
5a854ab3 2475 }
ab340fff 2476
c1566110
KW
2477# endif
2478
628ff75a 2479 if (strEQ(PL_langinfo_buf, "")) {
c1566110
KW
2480 if (item == PERL_YESSTR) {
2481 return "yes";
2482 }
2483 if (item == PERL_NOSTR) {
2484 return "no";
2485 }
2486 }
2487
628ff75a 2488 return PL_langinfo_buf;
ab340fff 2489
f7416781 2490#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2491
2492 {
2493
f7416781
KW
2494# ifdef HAS_LOCALECONV
2495
43dd6b15 2496 const struct lconv* lc;
628ff75a 2497 const char * temp;
038d3702 2498 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
2499
2500# endif
2501# ifdef HAS_STRFTIME
2502
43dd6b15
KW
2503 struct tm tm;
2504 bool return_format = FALSE; /* Return the %format, not the value */
2505 const char * format;
f7416781
KW
2506
2507# endif
2508
43dd6b15
KW
2509 /* We copy the results to a per-thread buffer, even if not
2510 * multi-threaded. This is in part to simplify this code, and partly
2511 * because we need a buffer anyway for strftime(), and partly because a
2512 * call of localeconv() could otherwise wipe out the buffer, and the
2513 * programmer would not be expecting this, as this is a nl_langinfo()
2514 * substitute after all, so s/he might be thinking their localeconv()
2515 * is safe until another localeconv() call. */
f7416781 2516
43dd6b15
KW
2517 switch (item) {
2518 Size_t len;
f7416781 2519
43dd6b15
KW
2520 /* These 2 are unimplemented */
2521 case PERL_CODESET:
2522 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 2523
43dd6b15
KW
2524 default:
2525 return "";
f7416781 2526
43dd6b15
KW
2527 /* We use only an English set, since we don't know any more */
2528 case PERL_YESEXPR: return "^[+1yY]";
2529 case PERL_YESSTR: return "yes";
2530 case PERL_NOEXPR: return "^[-0nN]";
2531 case PERL_NOSTR: return "no";
f7416781
KW
2532
2533# ifdef HAS_LOCALECONV
2534
43dd6b15 2535 case PERL_CRNCYSTR:
f7416781 2536
43dd6b15
KW
2537 /* We don't bother with localeconv_l() because any system that
2538 * has it is likely to also have nl_langinfo() */
291a84fb 2539
5acc4454
KW
2540 LOCALE_LOCK; /* Prevent interference with other threads
2541 using localeconv() */
2542
43dd6b15
KW
2543 lc = localeconv();
2544 if ( ! lc
2545 || ! lc->currency_symbol
2546 || strEQ("", lc->currency_symbol))
2547 {
2548 LOCALE_UNLOCK;
2549 return "";
2550 }
f7416781 2551
43dd6b15
KW
2552 /* Leave the first spot empty to be filled in below */
2553 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2554 &PL_langinfo_bufsize, 1);
2555 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2556 { /* khw couldn't figure out how the localedef specifications
2557 would show that the $ should replace the radix; this is
2558 just a guess as to how it might work.*/
2559 *PL_langinfo_buf = '.';
2560 }
2561 else if (lc->p_cs_precedes) {
2562 *PL_langinfo_buf = '-';
2563 }
2564 else {
2565 *PL_langinfo_buf = '+';
2566 }
f7416781 2567
43dd6b15
KW
2568 LOCALE_UNLOCK;
2569 break;
f7416781 2570
43dd6b15
KW
2571 case PERL_RADIXCHAR:
2572 case PERL_THOUSEP:
f7416781 2573
43dd6b15 2574 if (toggle) {
038d3702 2575 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2576 }
f7416781 2577
5acc4454
KW
2578 LOCALE_LOCK; /* Prevent interference with other threads
2579 using localeconv() */
2580
43dd6b15
KW
2581 lc = localeconv();
2582 if (! lc) {
628ff75a 2583 temp = "";
33394adc 2584 }
43dd6b15 2585 else {
628ff75a 2586 temp = (item == PERL_RADIXCHAR)
43dd6b15
KW
2587 ? lc->decimal_point
2588 : lc->thousands_sep;
628ff75a
KW
2589 if (! temp) {
2590 temp = "";
43dd6b15
KW
2591 }
2592 }
f7416781 2593
628ff75a 2594 save_to_buffer(temp, &PL_langinfo_buf,
43dd6b15 2595 &PL_langinfo_bufsize, 0);
f7416781 2596
5acc4454
KW
2597 LOCALE_UNLOCK;
2598
43dd6b15 2599 if (toggle) {
038d3702 2600 RESTORE_LC_NUMERIC();
43dd6b15 2601 }
f7416781 2602
43dd6b15 2603 break;
f7416781
KW
2604
2605# endif
2606# ifdef HAS_STRFTIME
2607
43dd6b15
KW
2608 /* These are defined by C89, so we assume that strftime supports
2609 * them, and so are returned unconditionally; they may not be what
2610 * the locale actually says, but should give good enough results
2611 * for someone using them as formats (as opposed to trying to parse
2612 * them to figure out what the locale says). The other format
2613 * items are actually tested to verify they work on the platform */
2614 case PERL_D_FMT: return "%x";
2615 case PERL_T_FMT: return "%X";
2616 case PERL_D_T_FMT: return "%c";
2617
2618 /* These formats are only available in later strfmtime's */
2619 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2620 case PERL_T_FMT_AMPM:
2621
2622 /* The rest can be gotten from most versions of strftime(). */
2623 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2624 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2625 case PERL_ABDAY_7:
2626 case PERL_ALT_DIGITS:
2627 case PERL_AM_STR: case PERL_PM_STR:
2628 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2629 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2630 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2631 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2632 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2633 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2634 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2635 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2636 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2637 case PERL_MON_12:
2638
2639 LOCALE_LOCK;
2640
2641 init_tm(&tm); /* Precaution against core dumps */
2642 tm.tm_sec = 30;
2643 tm.tm_min = 30;
2644 tm.tm_hour = 6;
2645 tm.tm_year = 2017 - 1900;
2646 tm.tm_wday = 0;
2647 tm.tm_mon = 0;
2648 switch (item) {
2649 default:
2650 LOCALE_UNLOCK;
2651 Perl_croak(aTHX_
2652 "panic: %s: %d: switch case: %d problem",
2653 __FILE__, __LINE__, item);
2654 NOT_REACHED; /* NOTREACHED */
2655
2656 case PERL_PM_STR: tm.tm_hour = 18;
2657 case PERL_AM_STR:
2658 format = "%p";
2659 break;
2660
2661 case PERL_ABDAY_7: tm.tm_wday++;
2662 case PERL_ABDAY_6: tm.tm_wday++;
2663 case PERL_ABDAY_5: tm.tm_wday++;
2664 case PERL_ABDAY_4: tm.tm_wday++;
2665 case PERL_ABDAY_3: tm.tm_wday++;
2666 case PERL_ABDAY_2: tm.tm_wday++;
2667 case PERL_ABDAY_1:
2668 format = "%a";
2669 break;
2670
2671 case PERL_DAY_7: tm.tm_wday++;
2672 case PERL_DAY_6: tm.tm_wday++;
2673 case PERL_DAY_5: tm.tm_wday++;
2674 case PERL_DAY_4: tm.tm_wday++;
2675 case PERL_DAY_3: tm.tm_wday++;
2676 case PERL_DAY_2: tm.tm_wday++;
2677 case PERL_DAY_1:
2678 format = "%A";
2679 break;
2680
2681 case PERL_ABMON_12: tm.tm_mon++;
2682 case PERL_ABMON_11: tm.tm_mon++;
2683 case PERL_ABMON_10: tm.tm_mon++;
2684 case PERL_ABMON_9: tm.tm_mon++;
2685 case PERL_ABMON_8: tm.tm_mon++;
2686 case PERL_ABMON_7: tm.tm_mon++;
2687 case PERL_ABMON_6: tm.tm_mon++;
2688 case PERL_ABMON_5: tm.tm_mon++;
2689 case PERL_ABMON_4: tm.tm_mon++;
2690 case PERL_ABMON_3: tm.tm_mon++;
2691 case PERL_ABMON_2: tm.tm_mon++;
2692 case PERL_ABMON_1:
2693 format = "%b";
2694 break;
2695
2696 case PERL_MON_12: tm.tm_mon++;
2697 case PERL_MON_11: tm.tm_mon++;
2698 case PERL_MON_10: tm.tm_mon++;
2699 case PERL_MON_9: tm.tm_mon++;
2700 case PERL_MON_8: tm.tm_mon++;
2701 case PERL_MON_7: tm.tm_mon++;
2702 case PERL_MON_6: tm.tm_mon++;
2703 case PERL_MON_5: tm.tm_mon++;
2704 case PERL_MON_4: tm.tm_mon++;
2705 case PERL_MON_3: tm.tm_mon++;
2706 case PERL_MON_2: tm.tm_mon++;
2707 case PERL_MON_1:
2708 format = "%B";
2709 break;
2710
2711 case PERL_T_FMT_AMPM:
2712 format = "%r";
2713 return_format = TRUE;
2714 break;
2715
2716 case PERL_ERA_D_FMT:
2717 format = "%Ex";
2718 return_format = TRUE;
2719 break;
2720
2721 case PERL_ERA_T_FMT:
2722 format = "%EX";
2723 return_format = TRUE;
2724 break;
2725
2726 case PERL_ERA_D_T_FMT:
2727 format = "%Ec";
2728 return_format = TRUE;
2729 break;
2730
2731 case PERL_ALT_DIGITS:
2732 tm.tm_wday = 0;
2733 format = "%Ow"; /* Find the alternate digit for 0 */
2734 break;
2735 }
f7416781 2736
43dd6b15
KW
2737 /* We can't use my_strftime() because it doesn't look at
2738 * tm_wday */
2739 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2740 format, &tm))
2741 {
2742 /* A zero return means one of:
2743 * a) there wasn't enough space in PL_langinfo_buf
2744 * b) the format, like a plain %p, returns empty
2745 * c) it was an illegal format, though some
2746 * implementations of strftime will just return the
2747 * illegal format as a plain character sequence.
2748 *
2749 * To quickly test for case 'b)', try again but precede
2750 * the format with a plain character. If that result is
2751 * still empty, the problem is either 'a)' or 'c)' */
2752
2753 Size_t format_size = strlen(format) + 1;
2754 Size_t mod_size = format_size + 1;
2755 char * mod_format;
2756 char * temp_result;
2757
2758 Newx(mod_format, mod_size, char);
2759 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2760 *mod_format = ' ';
43dd6b15
KW
2761 my_strlcpy(mod_format + 1, format, mod_size);
2762 len = strftime(temp_result,
2763 PL_langinfo_bufsize,
2764 mod_format, &tm);
2765 Safefree(mod_format);
2766 Safefree(temp_result);
2767
2768 /* If 'len' is non-zero, it means that we had a case like
2769 * %p which means the current locale doesn't use a.m. or
2770 * p.m., and that is valid */
2771 if (len == 0) {
2772
2773 /* Here, still didn't work. If we get well beyond a
2774 * reasonable size, bail out to prevent an infinite
2775 * loop. */
2776
2777 if (PL_langinfo_bufsize > 100 * format_size) {
2778 *PL_langinfo_buf = '\0';
2779 }
2780 else {
2781 /* Double the buffer size to retry; Add 1 in case
2782 * original was 0, so we aren't stuck at 0. */
2783 PL_langinfo_bufsize *= 2;
2784 PL_langinfo_bufsize++;
2785 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2786 continue;
2787 }
2788 }
f7416781 2789
f7416781 2790 break;
43dd6b15 2791 }
f7416781 2792
43dd6b15
KW
2793 /* Here, we got a result.
2794 *
2795 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2796 * alternate format for wday 0. If the value is the same as
2797 * the normal 0, there isn't an alternate, so clear the buffer.
2798 * */
2799 if ( item == PERL_ALT_DIGITS
2800 && strEQ(PL_langinfo_buf, "0"))
2801 {
2802 *PL_langinfo_buf = '\0';
2803 }
f7416781 2804
43dd6b15
KW
2805 /* ALT_DIGITS is problematic. Experiments on it showed that
2806 * strftime() did not always work properly when going from
2807 * alt-9 to alt-10. Only a few locales have this item defined,
2808 * and in all of them on Linux that khw was able to find,
2809 * nl_langinfo() merely returned the alt-0 character, possibly
2810 * doubled. Most Unicode digits are in blocks of 10
2811 * consecutive code points, so that is sufficient information
2812 * for those scripts, as we can infer alt-1, alt-2, .... But
2813 * for a Japanese locale, a CJK ideographic 0 is returned, and
2814 * the CJK digits are not in code point order, so you can't
2815 * really infer anything. The localedef for this locale did
2816 * specify the succeeding digits, so that strftime() works
2817 * properly on them, without needing to infer anything. But
2818 * the nl_langinfo() return did not give sufficient information
2819 * for the caller to understand what's going on. So until
2820 * there is evidence that it should work differently, this
2821 * returns the alt-0 string for ALT_DIGITS.
2822 *
2823 * wday was chosen because its range is all a single digit.
2824 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 2825
43dd6b15 2826 LOCALE_UNLOCK;
f7416781 2827
43dd6b15
KW
2828 /* If to return the format, not the value, overwrite the buffer
2829 * with it. But some strftime()s will keep the original format
2830 * if illegal, so change those to "" */
2831 if (return_format) {
2832 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
2833 *PL_langinfo_buf = '\0';
2834 }
43dd6b15
KW
2835 else {
2836 save_to_buffer(format, &PL_langinfo_buf,
2837 &PL_langinfo_bufsize, 0);
f7416781
KW
2838 }
2839 }
2840
2841 break;
f7416781
KW
2842
2843# endif
2844
43dd6b15 2845 }
f7416781
KW
2846 }
2847
2848 return PL_langinfo_buf;
2849
2850#endif
2851
a4f00dcc 2852}
b385bb4d 2853
98994639
HS
2854/*
2855 * Initialize locale awareness.
2856 */
2857int
2858Perl_init_i18nl10n(pTHX_ int printwarn)
2859{
0e92a118
KW
2860 /* printwarn is
2861 *
2862 * 0 if not to output warning when setup locale is bad
2863 * 1 if to output warning based on value of PERL_BADLANG
2864 * >1 if to output regardless of PERL_BADLANG
2865 *
2866 * returns
98994639 2867 * 1 = set ok or not applicable,
0e92a118
KW
2868 * 0 = fallback to a locale of lower priority
2869 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2870 *
2871 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2872 * set, debugging information is output.
2873 *
2874 * This looks more complicated than it is, mainly due to the #ifdefs.
2875 *
2876 * We try to set LC_ALL to the value determined by the environment. If
2877 * there is no LC_ALL on this platform, we try the individual categories we
2878 * know about. If this works, we are done.
2879 *
2880 * But if it doesn't work, we have to do something else. We search the
2881 * environment variables ourselves instead of relying on the system to do
2882 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2883 * think there is one), and the ultimate fallback "C". This is all done in
2884 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2885 * more complex. The 'trial_locales' array is initialized with just one
2886 * element; it causes the behavior described in the paragraph above this to
2887 * happen. If that fails, we add elements to 'trial_locales', and do extra
2888 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2889 *
2890 * On Ultrix, the locale MUST come from the environment, so there is
2891 * preliminary code to set it. I (khw) am not sure that it is necessary,
2892 * and that this couldn't be folded into the loop, but barring any real
2893 * platforms to test on, it's staying as-is
2894 *
2895 * A slight complication is that in embedded Perls, the locale may already
2896 * be set-up, and we don't want to get it from the normal environment
2897 * variables. This is handled by having a special environment variable
2898 * indicate we're in this situation. We simply set setlocale's 2nd
2899 * parameter to be a NULL instead of "". That indicates to setlocale that
2900 * it is not to change anything, but to return the current value,
2901 * effectively initializing perl's db to what the locale already is.
2902 *
2903 * We play the same trick with NULL if a LC_ALL succeeds. We call
2904 * setlocale() on the individual categores with NULL to get their existing
2905 * values for our db, instead of trying to change them.
2906 * */
98994639 2907
0e92a118
KW
2908 int ok = 1;
2909
7d4bcc4a
KW
2910#ifndef USE_LOCALE
2911
2912 PERL_UNUSED_ARG(printwarn);
2913
2914#else /* USE_LOCALE */
7d4bcc4a
KW
2915# ifdef __GLIBC__
2916
175c4cf9 2917 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2918
2919# endif
65ebb059 2920
ccd65d51
KW
2921 /* NULL uses the existing already set up locale */
2922 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2923 ? NULL
2924 : "";
c3fcd832
KW
2925 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2926 unsigned int trial_locales_count;
175c4cf9
KW
2927 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2928 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2929 bool setlocale_failure = FALSE;
65ebb059 2930 unsigned int i;
175c4cf9
KW
2931
2932 /* A later getenv() could zap this, so only use here */
2933 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2934
2935 const bool locwarn = (printwarn > 1
e5f10d49
KW
2936 || ( printwarn
2937 && ( ! bad_lang_use_once
22ff3130 2938 || (
e5f10d49
KW
2939 /* disallow with "" or "0" */
2940 *bad_lang_use_once
2941 && strNE("0", bad_lang_use_once)))));
ea92aad8 2942
291a84fb 2943 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2944 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2945
2946 /* current locale for given category; should have been copied so aren't
2947 * volatile */
8de4332b 2948 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2949
7d4bcc4a
KW
2950# ifdef WIN32
2951
6bce99ee
JH
2952 /* In some systems you can find out the system default locale
2953 * and use that as the fallback locale. */
7d4bcc4a
KW
2954# define SYSTEM_DEFAULT_LOCALE
2955# endif
2956# ifdef SYSTEM_DEFAULT_LOCALE
2957
65ebb059 2958 const char *system_default_locale = NULL;
98994639 2959
7d4bcc4a 2960# endif
948523db
KW
2961
2962# ifndef DEBUGGING
2963# define DEBUG_LOCALE_INIT(a,b,c)
2964# else
7d4bcc4a 2965
8298454c 2966 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2967
2968# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2969 STMT_START { \
2970 if (debug_initialization) { \
2971 PerlIO_printf(Perl_debug_log, \
2972 "%s:%d: %s\n", \
2973 __FILE__, __LINE__, \
a4f00dcc 2974 setlocale_debug_string(category, \
2fcc0ca9
KW
2975 locale, \
2976 result)); \
2977 } \
2978 } STMT_END
2fcc0ca9 2979
948523db
KW
2980/* Make sure the parallel arrays are properly set up */
2981# ifdef USE_LOCALE_NUMERIC
2982 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2983 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
2984# ifdef USE_POSIX_2008_LOCALE
2985 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
2986# endif
948523db
KW
2987# endif
2988# ifdef USE_LOCALE_CTYPE
2989 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2990 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
2991# ifdef USE_POSIX_2008_LOCALE
2992 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
2993# endif
948523db
KW
2994# endif
2995# ifdef USE_LOCALE_COLLATE
2996 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2997 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
2998# ifdef USE_POSIX_2008_LOCALE
2999 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3000# endif
948523db
KW
3001# endif
3002# ifdef USE_LOCALE_TIME
3003 assert(categories[LC_TIME_INDEX] == LC_TIME);
3004 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3005# ifdef USE_POSIX_2008_LOCALE
3006 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3007# endif
948523db
KW
3008# endif
3009# ifdef USE_LOCALE_MESSAGES
3010 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3011 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3012# ifdef USE_POSIX_2008_LOCALE
3013 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3014# endif
948523db
KW
3015# endif
3016# ifdef USE_LOCALE_MONETARY
3017 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3018 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3019# ifdef USE_POSIX_2008_LOCALE
3020 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3021# endif
948523db 3022# endif
9821811f
KW
3023# ifdef USE_LOCALE_ADDRESS
3024 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3025 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3026# ifdef USE_POSIX_2008_LOCALE
3027 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3028# endif
9821811f
KW
3029# endif
3030# ifdef USE_LOCALE_IDENTIFICATION
3031 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3032 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3033# ifdef USE_POSIX_2008_LOCALE
3034 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3035# endif
9821811f
KW
3036# endif
3037# ifdef USE_LOCALE_MEASUREMENT
3038 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3039 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3040# ifdef USE_POSIX_2008_LOCALE
3041 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3042# endif
9821811f
KW
3043# endif
3044# ifdef USE_LOCALE_PAPER
3045 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3046 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3047# ifdef USE_POSIX_2008_LOCALE
3048 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3049# endif
9821811f
KW
3050# endif
3051# ifdef USE_LOCALE_TELEPHONE
3052 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3053 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3054# ifdef USE_POSIX_2008_LOCALE
3055 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3056# endif
9821811f 3057# endif
948523db
KW
3058# ifdef LC_ALL
3059 assert(categories[LC_ALL_INDEX] == LC_ALL);
3060 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3061 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3062# ifdef USE_POSIX_2008_LOCALE
3063 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3064# endif
948523db
KW
3065# endif
3066# endif /* DEBUGGING */
47280b20
KW
3067
3068 /* Initialize the cache of the program's UTF-8ness for the always known
3069 * locales C and POSIX */
3070 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3071 sizeof(PL_locale_utf8ness));
3072
e9bc6d6b
KW
3073# ifdef USE_THREAD_SAFE_LOCALE
3074# ifdef WIN32
3075
3076 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3077
3078# endif
3079# endif
3080# if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3081
3082 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3083 if (! PL_C_locale_obj) {
3084 Perl_croak_nocontext(
3085 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3086 }
3087 if (DEBUG_Lv_TEST || debug_initialization) {
3088 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3089 }
3090
3091# endif
3092
3ca88433
KW
3093 PL_numeric_radix_sv = newSVpvs(".");
3094
e9bc6d6b
KW
3095# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3096
3097 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3098 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3099
3100# endif
b56c4436 3101# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3102
3103 /*
3104 * Ultrix setlocale(..., "") fails if there are no environment
3105 * variables from which to get a locale name.
3106 */
3107
b56c4436
KW
3108# ifndef LC_ALL
3109# error Ultrix without LC_ALL not implemented
3110# else
7d4bcc4a 3111
b56c4436
KW
3112 {
3113 bool done = FALSE;
ea92aad8
KW
3114 if (lang) {
3115 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3116 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3117 if (sl_result[LC_ALL_INDEX])
3118 done = TRUE;
3119 else
e5f10d49 3120 setlocale_failure = TRUE;
ea92aad8
KW
3121 }
3122 if (! setlocale_failure) {
3123 const char * locale_param;
3124 for (i = 0; i < LC_ALL_INDEX; i++) {
3125 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3126 ? setlocale_init
3127 : NULL;
3128 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3129 if (! sl_result[i]) {
3130 setlocale_failure = TRUE;
3131 }
3132 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3133 }
c835d6be 3134 }
7d4bcc4a
KW
3135 }
3136
3137# endif /* LC_ALL */
e5f10d49 3138# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3139
65ebb059 3140 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3141 * the list. Normally the loop is executed just once. But if setting the
3142 * locale fails, inside the loop we add fallback trials to the array and so
3143 * will execute the loop multiple times */
c3fcd832
KW
3144 trial_locales[0] = setlocale_init;
3145 trial_locales_count = 1;
7d4bcc4a 3146
65ebb059
KW
3147 for (i= 0; i < trial_locales_count; i++) {
3148 const char * trial_locale = trial_locales[i];
3149
3150 if (i > 0) {
3151
3152 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3153 * when i==0, but I (khw) don't think that behavior makes much
3154 * sense */
3155 setlocale_failure = FALSE;
3156
7d4bcc4a 3157# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3158# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3159
65ebb059
KW
3160 /* On Windows machines, an entry of "" after the 0th means to use
3161 * the system default locale, which we now proceed to get. */
3162 if (strEQ(trial_locale, "")) {
3163 unsigned int j;
3164
3165 /* Note that this may change the locale, but we are going to do
3166 * that anyway just below */
837ce802 3167 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3168 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3169
7d4bcc4a 3170 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3171 * try */
3172 if (! system_default_locale) {
3173 goto next_iteration;
3174 }
3175 for (j = 0; j < trial_locales_count; j++) {
3176 if (strEQ(system_default_locale, trial_locales[j])) {
3177 goto next_iteration;
3178 }
3179 }
3180
3181 trial_locale = system_default_locale;
3182 }
ec0202b5
KW
3183# else
3184# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3185# endif
7d4bcc4a 3186# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3187
3188 } /* For i > 0 */
65ebb059 3189
7d4bcc4a
KW
3190# ifdef LC_ALL
3191
948523db
KW
3192 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3193 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3194 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3195 setlocale_failure = TRUE;
7cd8b568
KW
3196 }
3197 else {
3198 /* Since LC_ALL succeeded, it should have changed all the other
3199 * categories it can to its value; so we massage things so that the
3200 * setlocales below just return their category's current values.
3201 * This adequately handles the case in NetBSD where LC_COLLATE may
3202 * not be defined for a locale, and setting it individually will
7d4bcc4a 3203 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3204 * the POSIX locale. */
3205 trial_locale = NULL;
3206 }
7d4bcc4a
KW
3207
3208# endif /* LC_ALL */
98994639 3209
e5f10d49
KW
3210 if (! setlocale_failure) {
3211 unsigned int j;
3212 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3213 curlocales[j]
3214 = savepv(do_setlocale_r(categories[j], trial_locale));
3215 if (! curlocales[j]) {
3216 setlocale_failure = TRUE;
3217 }
3218 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3219 }
c835d6be 3220
e5f10d49
KW
3221 if (! setlocale_failure) { /* All succeeded */
3222 break; /* Exit trial_locales loop */
49c85077 3223 }
65ebb059 3224 }
98994639 3225
49c85077
KW
3226 /* Here, something failed; will need to try a fallback. */
3227 ok = 0;
65ebb059 3228
49c85077
KW
3229 if (i == 0) {
3230 unsigned int j;
98994639 3231
65ebb059 3232 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3233
3234# ifdef LC_ALL
98994639 3235
49c85077
KW
3236 PerlIO_printf(Perl_error_log,
3237 "perl: warning: Setting locale failed.\n");
98994639 3238
7d4bcc4a 3239# else /* !LC_ALL */
98994639 3240
49c85077
KW
3241 PerlIO_printf(Perl_error_log,
3242 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3243
e5f10d49
KW
3244 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3245 if (! curlocales[j]) {
3246 PerlIO_printf(Perl_error_log, category_names[j]);
3247 }
3248 else {
3249 Safefree(curlocales[j]);
3250 }
3251 }
7d4bcc4a 3252
7d4bcc4a 3253# endif /* LC_ALL */
98994639 3254
49c85077
KW
3255 PerlIO_printf(Perl_error_log,
3256 "perl: warning: Please check that your locale settings:\n");
98994639 3257
7d4bcc4a
KW
3258# ifdef __GLIBC__
3259
49c85077
KW
3260 PerlIO_printf(Perl_error_log,
3261 "\tLANGUAGE = %c%s%c,\n",
3262 language ? '"' : '(',
3263 language ? language : "unset",
3264 language ? '"' : ')');
7d4bcc4a 3265# endif
98994639 3266
49c85077
KW
3267 PerlIO_printf(Perl_error_log,
3268 "\tLC_ALL = %c%s%c,\n",
3269 lc_all ? '"' : '(',
3270 lc_all ? lc_all : "unset",
3271 lc_all ? '"' : ')');
98994639 3272
7d4bcc4a
KW
3273# if defined(USE_ENVIRON_ARRAY)
3274
49c85077 3275 {
cd999af9 3276 char **e;
d5e32b93
KW
3277
3278 /* Look through the environment for any variables of the
3279 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3280 * already handled above. These are assumed to be locale
3281 * settings. Output them and their values. */
cd999af9 3282 for (e = environ; *e; e++) {
d5e32b93
KW
3283 const STRLEN prefix_len = sizeof("LC_") - 1;
3284 STRLEN uppers_len;
3285
cd999af9 3286 if ( strBEGINs(*e, "LC_")
c8b388b0 3287 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3288 && (uppers_len = strspn(*e + prefix_len,
3289 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3290 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3291 {
3292 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3293 (int) (prefix_len + uppers_len), *e,
3294 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3295 }
3296 }
49c85077 3297 }
7d4bcc4a
KW
3298
3299# else
3300
49c85077
KW
3301 PerlIO_printf(Perl_error_log,
3302 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3303
3304# endif
98994639 3305
49c85077
KW
3306 PerlIO_printf(Perl_error_log,
3307 "\tLANG = %c%s%c\n",
3308 lang ? '"' : '(',
3309 lang ? lang : "unset",
3310 lang ? '"' : ')');
98994639 3311
49c85077
KW
3312 PerlIO_printf(Perl_error_log,
3313 " are supported and installed on your system.\n");
3314 }
98994639 3315
65ebb059 3316 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3317 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3318 * usually change the upper bound of the loop we are in.
3319 *
3320 * Since the system's default way of setting the locale has not
3321 * found one that works, We use Perl's defined ordering: LC_ALL,
3322 * LANG, and the C locale. We don't try the same locale twice, so
3323 * don't add to the list if already there. (On POSIX systems, the
3324 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3325 * but there's no harm done by doing it explicitly.
3326 *
3327 * Note that this tries the LC_ALL environment variable even on
3328 * systems which have no LC_ALL locale setting. This may or may
3329 * not have been originally intentional, but there's no real need
3330 * to change the behavior. */
65ebb059
KW
3331 if (lc_all) {
3332 for (j = 0; j < trial_locales_count; j++) {
3333 if (strEQ(lc_all, trial_locales[j])) {
3334 goto done_lc_all;
3335 }
3336 }
3337 trial_locales[trial_locales_count++] = lc_all;
3338 }
3339 done_lc_all:
98994639 3340
65ebb059
KW
3341 if (lang) {
3342 for (j = 0; j < trial_locales_count; j++) {
3343 if (strEQ(lang, trial_locales[j])) {
3344 goto done_lang;
3345 }
3346 }
3347 trial_locales[trial_locales_count++] = lang;
3348 }
3349 done_lang:
3350
7d4bcc4a
KW
3351# if defined(WIN32) && defined(LC_ALL)
3352
65ebb059
KW
3353 /* For Windows, we also try the system default locale before "C".
3354 * (If there exists a Windows without LC_ALL we skip this because
3355 * it gets too complicated. For those, the "C" is the next
3356 * fallback possibility). The "" is the same as the 0th element of
3357 * the array, but the code at the loop above knows to treat it
3358 * differently when not the 0th */
3359 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3360
3361# endif
65ebb059
KW
3362
3363 for (j = 0; j < trial_locales_count; j++) {
3364 if (strEQ("C", trial_locales[j])) {
3365 goto done_C;
3366 }
3367 }
3368 trial_locales[trial_locales_count++] = "C";
98994639 3369
65ebb059
KW
3370 done_C: ;
3371 } /* end of first time through the loop */
98994639 3372
7d4bcc4a
KW
3373# ifdef WIN32
3374
65ebb059 3375 next_iteration: ;
7d4bcc4a
KW
3376
3377# endif
65ebb059
KW
3378
3379 } /* end of looping through the trial locales */
3380
3381 if (ok < 1) { /* If we tried to fallback */
3382 const char* msg;
3383 if (! setlocale_failure) { /* fallback succeeded */
3384 msg = "Falling back to";
3385 }
3386 else { /* fallback failed */
e5f10d49 3387 unsigned int j;
98994639 3388
65ebb059
KW
3389 /* We dropped off the end of the loop, so have to decrement i to
3390 * get back to the value the last time through */
3391 i--;
98994639 3392
65ebb059
KW
3393 ok = -1;
3394 msg = "Failed to fall back to";
3395
3396 /* To continue, we should use whatever values we've got */
7d4bcc4a 3397
e5f10d49
KW
3398 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3399 Safefree(curlocales[j]);
3400 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3401 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3402 }
65ebb059
KW
3403 }
3404
3405 if (locwarn) {
3406 const char * description;
3407 const char * name = "";
3408 if (strEQ(trial_locales[i], "C")) {
3409 description = "the standard locale";
3410 name = "C";
3411 }
7d4bcc4a
KW
3412
3413# ifdef SYSTEM_DEFAULT_LOCALE
3414
65ebb059
KW
3415 else if (strEQ(trial_locales[i], "")) {
3416 description = "the system default locale";
3417 if (system_default_locale) {
3418 name = system_default_locale;
3419 }
3420 }
7d4bcc4a
KW
3421
3422# endif /* SYSTEM_DEFAULT_LOCALE */
3423
65ebb059
KW
3424 else {
3425 description = "a fallback locale";
3426 name = trial_locales[i];
3427 }
3428 if (name && strNE(name, "")) {
3429 PerlIO_printf(Perl_error_log,
3430 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3431 }
3432 else {
3433 PerlIO_printf(Perl_error_log,
3434 "perl: warning: %s %s.\n", msg, description);
3435 }
3436 }
3437 } /* End of tried to fallback */
98994639 3438
e5f10d49
KW
3439 /* Done with finding the locales; update our records */
3440
7d4bcc4a
KW
3441# ifdef USE_LOCALE_CTYPE
3442
948523db 3443 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3444
e5f10d49 3445# endif
7d4bcc4a
KW
3446# ifdef USE_LOCALE_COLLATE
3447
948523db 3448 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3449
e5f10d49 3450# endif
7d4bcc4a
KW
3451# ifdef USE_LOCALE_NUMERIC
3452
948523db 3453 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3454
3455# endif
3456
948523db 3457 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3458
e9bc6d6b 3459# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3460
3461 /* This caches whether each category's locale is UTF-8 or not. This
3462 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3463 * initialization time before any threads have started, but not later
3464 * unless thread-safe operations are used.
47280b20
KW
3465 * Caching means that if the program heeds our dictate not to change
3466 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3467 * it may get queried without having to change locales. If the
3468 * environment is such that all categories have the same locale, this
3469 * isn't needed, as the code will not change the locale; but this
3470 * handles the uncommon case where the environment has disparate
3471 * locales for the categories */
47280b20
KW
3472 (void) _is_cur_LC_category_utf8(categories[i]);
3473
3474# endif
3475
e5f10d49
KW
3476 Safefree(curlocales[i]);
3477 }
b310b053 3478
7d4bcc4a
KW
3479# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3480
49c85077 3481 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3482 * locale is UTF-8. The call to new_ctype() just above has already
3483 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3484 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3485 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3486 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3487 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3488
a05d7ebb 3489 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3490 This is an alternative to using the -C command line switch
3491 (the -C if present will override this). */
3492 {
dd374669 3493 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3494 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3495 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3496 PL_utf8cache = -1;
b310b053
JH
3497 }
3498
7d4bcc4a 3499# endif
7d4bcc4a
KW
3500# ifdef __GLIBC__
3501
175c4cf9 3502 Safefree(language);
7d4bcc4a
KW
3503
3504# endif
175c4cf9
KW
3505
3506 Safefree(lc_all);
3507 Safefree(lang);
3508
e3305790 3509#endif /* USE_LOCALE */
2fcc0ca9 3510#ifdef DEBUGGING
7d4bcc4a 3511
2fcc0ca9 3512 /* So won't continue to output stuff */
27cdc72e 3513 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3514
2fcc0ca9
KW
3515#endif
3516
98994639
HS
3517 return ok;
3518}
3519
98994639
HS
3520#ifdef USE_LOCALE_COLLATE
3521
a4a439fb 3522char *
a4a439fb
KW
3523Perl__mem_collxfrm(pTHX_ const char *input_string,
3524 STRLEN len, /* Length of 'input_string' */
3525 STRLEN *xlen, /* Set to length of returned string
3526 (not including the collation index
3527 prefix) */
3528 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3529 )
98994639 3530{
a4a439fb
KW
3531
3532 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3533 * differences. First, it handles embedded NULs. Second, it allocates a bit
3534 * more memory than needed for the transformed data itself. The real
55e5378d 3535 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3536 * the length of that, and doesn't include the collation index size.
3537 * Please see sv_collxfrm() to see how this is used. */
3538
55e5378d
KW
3539#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3540
6696cfa7
KW
3541 char * s = (char *) input_string;
3542 STRLEN s_strlen = strlen(input_string);
79f120c8 3543 char *xbuf = NULL;
55e5378d 3544 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3545 STRLEN length_in_chars;
c664130f 3546 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3547
a4a439fb
KW
3548 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3549
3550 /* Must be NUL-terminated */
3551 assert(*(input_string + len) == '\0');
7918f24d 3552
79f120c8
KW
3553 /* If this locale has defective collation, skip */
3554 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3555 DEBUG_L(PerlIO_printf(Perl_debug_log,
3556 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3557 goto bad;
3558 }
3559
6696cfa7
KW
3560 /* Replace any embedded NULs with the control that sorts before any others.
3561 * This will give as good as possible results on strings that don't
3562 * otherwise contain that character, but otherwise there may be
3563 * less-than-perfect results with that character and NUL. This is
fdc080f3 3564 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3565 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3566 NUL */
6696cfa7
KW
3567 char * e = s + len;
3568 char * sans_nuls;
fdc080f3 3569 STRLEN sans_nuls_len;
94762aa0 3570 int try_non_controls;
afc4976f
KW
3571 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3572 making sure 2nd byte is NUL.
3573 */
3574 STRLEN this_replacement_len;
3575
1e4c9676
KW
3576 /* If we don't know what non-NUL control character sorts lowest for
3577 * this locale, find it */
f28f4d2a 3578 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3579 int j;
afc4976f 3580 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3581 includes the collation index
3582 prefixed. */
3583
91c0e2e0 3584 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3585
3586 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3587 * NUL, in which case we instead look for any character. Controls
3588 * are preferred because collation order is, in general, context
3589 * sensitive, with adjoining characters affecting the order, and
3590 * controls are less likely to have such interactions, allowing the
3591 * NUL-replacement to stand on its own. (Another way to look at it
3592 * is to imagine what would happen if the NUL were replaced by a
3593 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3594 for (try_non_controls = 0;
3595 try_non_controls < 2;
3596 try_non_controls++)
3597 {
d4ff9586
KW
3598 /* Look through all legal code points (NUL isn't) */
3599 for (j = 1; j < 256; j++) {
3600 char * x; /* j's xfrm plus collation index */
3601 STRLEN x_len; /* length of 'x' */
3602 STRLEN trial_len = 1;
736a4fed 3603 char cur_source[] = { '\0', '\0' };
d4ff9586 3604
736a4fed
KW
3605 /* Skip non-controls the first time through the loop. The
3606 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3607 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3608 ? ! isCNTRL_L1(j)
3609 : ! isCNTRL_LC(j))
3610 {
d4ff9586 3611 continue;
6696cfa7 3612 }
6696cfa7 3613
736a4fed
KW
3614 /* Create a 1-char string of the current code point */
3615 cur_source[0] = (char) j;
3616
d4ff9586
KW
3617 /* Then transform it */
3618 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3619 0 /* The string is not in UTF-8 */);
6696cfa7 3620
1e4c9676 3621 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3622 * */
3623 if (! x) {
3624 continue;
3625 }
6696cfa7 3626
d4ff9586
KW
3627 /* If this character's transformation is lower than
3628 * the current lowest, this one becomes the lowest */
3629 if ( cur_min_x == NULL
3630 || strLT(x + COLLXFRM_HDR_LEN,
3631 cur_min_x + COLLXFRM_HDR_LEN))
3632 {
f28f4d2a 3633 PL_strxfrm_NUL_replacement = j;
d4ff9586 3634 cur_min_x = x;
d4ff9586
KW
3635 }
3636 else {
3637 Safefree(x);
3638 }
1e4c9676 3639 } /* end of loop through all 255 characters */
6696cfa7 3640
1e4c9676 3641 /* Stop looking if found */
94762aa0
KW
3642 if (cur_min_x) {
3643 break;
3644 }
3645
3646 /* Unlikely, but possible, if there aren't any controls that
3647 * work in the locale, repeat the loop, looking for any
3648 * character that works */
3649 DEBUG_L(PerlIO_printf(Perl_debug_log,
3650 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3651 } /* End of loop to try first the controls, then any char */
6696cfa7 3652
94762aa0
KW
3653 if (! cur_min_x) {
3654 DEBUG_L(PerlIO_printf(Perl_debug_log,
3655 "_mem_collxfrm: Couldn't find any character to replace"
3656 " embedded NULs in locale %s with", PL_collation_name));
3657 goto bad;
58eebef2
KW
3658 }
3659
94762aa0
KW
3660 DEBUG_L(PerlIO_printf(Perl_debug_log,
3661 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3662 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3663
6696cfa7 3664 Safefree(cur_min_x);
1e4c9676 3665 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3666
3667 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3668 * UTF8-ness of the original */
f28f4d2a
KW
3669 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3670 this_replacement_char[0] =
3671 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3672 this_replacement_char[1] =
3673 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3674 this_replacement_len = 2;
3675 }
3676 else {
f28f4d2a 3677 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3678 /* this_replacement_char[1] = '\0' was done at initialization */
3679 this_replacement_len = 1;
6696cfa7
KW
3680 }
3681
3682 /* The worst case length for the replaced string would be if every
3683 * character in it is NUL. Multiply that by the length of each
3684 * replacement, and allow for a trailing NUL */
afc4976f 3685 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3686 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3687 *sans_nuls = '\0';
3688
6696cfa7
KW
3689 /* Replace each NUL with the lowest collating control. Loop until have
3690 * exhausted all the NULs */
3691 while (s + s_strlen < e) {
6069d6c5 3692 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3693
3694 /* Do the actual replacement */
6069d6c5 3695 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
3696
3697 /* Move past the input NUL */
3698 s += s_strlen + 1;
3699 s_strlen = strlen(s);
3700 }
3701
3702 /* And add anything that trails the final NUL */
6069d6c5 3703 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3704
3705 /* Switch so below we transform this modified string */
3706 s = sans_nuls;
3707 len = strlen(s);
1e4c9676 3708 } /* End of replacing NULs */
6696cfa7 3709
a4a439fb
KW
3710 /* Make sure the UTF8ness of the string and locale match */
3711 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 3712 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
3713 const char * const t = s; /* Temporary so we can later find where the
3714 input was */
3715
3716 /* Here they don't match. Change the string's to be what the locale is
3717 * expecting */
3718
3719 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3720 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3721 utf8 = TRUE;
3722 }
3723 else { /* locale is not UTF-8; but input is; downgrade the input */
3724
3725 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3726
3727 /* If the downgrade was successful we are done, but if the input
3728 * contains things that require UTF-8 to represent, have to do
3729 * damage control ... */
3730 if (UNLIKELY(utf8)) {
3731
3732 /* What we do is construct a non-UTF-8 string with
3733 * 1) the characters representable by a single byte converted
3734 * to be so (if necessary);
3735 * 2) and the rest converted to collate the same as the
3736 * highest collating representable character. That makes
3737 * them collate at the end. This is similar to how we
3738 * handle embedded NULs, but we use the highest collating
3739 * code point instead of the smallest. Like the NUL case,
3740 * this isn't perfect, but is the best we can reasonably
3741 * do. Every above-255 code point will sort the same as
3742 * the highest-sorting 0-255 code point. If that code
3743 * point can combine in a sequence with some other code
3744 * points for weight calculations, us changing something to
3745 * be it can adversely affect the results. But in most
3746 * cases, it should work reasonably. And note that this is
3747 * really an illegal situation: using code points above 255
3748 * on a locale where only 0-255 are valid. If two strings
3749 * sort entirely equal, then the sort order for the
3750 * above-255 code points will be in code point order. */
3751
3752 utf8 = FALSE;
3753
3754 /* If we haven't calculated the code point with the maximum
3755 * collating order for this locale, do so now */
3756 if (! PL_strxfrm_max_cp) {
3757 int j;
3758
3759 /* The current transformed string that collates the
3760 * highest (except it also includes the prefixed collation
3761 * index. */
3762 char * cur_max_x = NULL;
3763
3764 /* Look through all legal code points (NUL isn't) */
3765 for (j = 1; j < 256; j++) {
3766 char * x;
3767 STRLEN x_len;
736a4fed 3768 char cur_source[] = { '\0', '\0' };
a4a439fb 3769
736a4fed
KW
3770 /* Create a 1-char string of the current code point */
3771 cur_source[0] = (char) j;
a4a439fb
KW
3772
3773 /* Then transform it */
3774 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3775
3776 /* If something went wrong (which it shouldn't), just
3777 * ignore this code point */
94762aa0 3778 if (! x) {
a4a439fb
KW
3779 continue;
3780 }
3781
3782 /* If this character's transformation is higher than
3783 * the current highest, this one becomes the highest */
3784 if ( cur_max_x == NULL
55e5378d
KW
3785 || strGT(x + COLLXFRM_HDR_LEN,
3786 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
3787 {
3788 PL_strxfrm_max_cp = j;
3789 cur_max_x = x;
3790 }
3791 else {
3792 Safefree(x);
3793 }
3794 }
3795
94762aa0
KW
3796 if (! cur_max_x) {
3797 DEBUG_L(PerlIO_printf(Perl_debug_log,
3798 "_mem_collxfrm: Couldn't find any character to"
3799 " replace above-Latin1 chars in locale %s with",
3800 PL_collation_name));
3801 goto bad;
3802 }
3803
58eebef2
KW
3804 DEBUG_L(PerlIO_printf(Perl_debug_log,
3805 "_mem_collxfrm: highest 1-byte collating character"
3806 " in locale %s is 0x%02X\n",
3807 PL_collation_name,
3808 PL_strxfrm_max_cp));
58eebef2 3809
a4a439fb
KW
3810 Safefree(cur_max_x);
3811 }
3812
3813 /* Here we know which legal code point collates the highest.
3814 * We are ready to construct the non-UTF-8 string. The length
3815 * will be at least 1 byte smaller than the input string
3816 * (because we changed at least one 2-byte character into a
3817 * single byte), but that is eaten up by the trailing NUL */
3818 Newx(s, len, char);
3819
3820 {
3821 STRLEN i;
3822 STRLEN d= 0;
042d9e50 3823 char * e = (char *) t + len;
a4a439fb
KW
3824
3825 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3826 U8 cur_char = t[i];
3827 if (UTF8_IS_INVARIANT(cur_char)) {
3828 s[d++] = cur_char;
3829 }
042d9e50 3830 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
3831 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3832 }
3833 else { /* Replace illegal cp with highest collating
3834 one */
3835 s[d++] = PL_strxfrm_max_cp;
3836 }
3837 }
3838 s[d++] = '\0';
3839 Renew(s, d, char); /* Free up unused space */
3840 }
3841 }
3842 }
3843
3844 /* Here, we have constructed a modified version of the input. It could
3845 * be that we already had a modified copy before we did this version.
3846 * If so, that copy is no longer needed */
3847 if (t != input_string) {
3848 Safefree(t);
3849 }
3850 }
3851
17f41037
KW
3852 length_in_chars = (utf8)
3853 ? utf8_length((U8 *) s, (U8 *) s + len)
3854 : len;
3855
59c018b9
KW
3856 /* The first element in the output is the collation id, used by
3857 * sv_collxfrm(); then comes the space for the transformed string. The
3858 * equation should give us a good estimate as to how much is needed */
55e5378d 3859 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 3860 + PL_collxfrm_base
17f41037 3861 + (PL_collxfrm_mult * length_in_chars);
a02a5408 3862 Newx(xbuf, xAlloc, char);
c7202dee
KW
3863 if (UNLIKELY(! xbuf)) {
3864 DEBUG_L(PerlIO_printf(Perl_debug_log,
3865 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 3866 goto bad;
c7202dee 3867 }
98994639 3868
d35fca5f 3869 /* Store the collation id */
98994639 3870 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
3871
3872 /* Then the transformation of the input. We loop until successful, or we
3873 * give up */
4ebeff16 3874 for (;;) {
1adab0a7 3875
55e5378d 3876 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
3877
3878 /* If the transformed string occupies less space than we told strxfrm()
3879 * was available, it means it successfully transformed the whole
3880 * string. */
55e5378d 3881 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 3882
1adab0a7
KW
3883 /* Some systems include a trailing NUL in the returned length.
3884 * Ignore it, using a loop in case multiple trailing NULs are
3885 * returned. */
3886 while ( (*xlen) > 0
3887 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3888 {
3889 (*xlen)--;
3890 }
3891
17f41037
KW
3892 /* If the first try didn't get it, it means our prediction was low.
3893 * Modify the coefficients so that we predict a larger value in any
3894 * future transformations */
3895 if (! first_time) {
3896 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3897 STRLEN computed_guess = PL_collxfrm_base
3898 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
3899
3900 /* On zero-length input, just keep current slope instead of
3901 * dividing by 0 */
3902 const STRLEN new_m = (length_in_chars != 0)
3903 ? needed / length_in_chars
3904 : PL_collxfrm_mult;
17f41037
KW
3905
3906 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
3907 "%s: %d: initial size of %zu bytes for a length "
3908 "%zu string was insufficient, %zu needed\n",
17f41037 3909 __FILE__, __LINE__,
b07929e4 3910 computed_guess, length_in_chars, needed));
17f41037
KW
3911
3912 /* If slope increased, use it, but discard this result for
3913 * length 1 strings, as we can't be sure that it's a real slope
3914 * change */
3915 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
<