This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Fix dangling set of LC_NUMERIC to C
[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). */
84f1d29f
KW
1357 if (PL_numeric_standard) {
1358 set_numeric_radix(0);
1359 }
1360 else {
1361 set_numeric_standard();
1362 }
4c28b29c 1363
98994639 1364#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1365
98994639
HS
1366}
1367
1368void
1369Perl_set_numeric_standard(pTHX)
1370{
7d4bcc4a 1371
98994639 1372#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1373
28c1bf33
KW
1374 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1375 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1376 * macro avoids calling this routine if toggling isn't necessary according
1377 * to our records (which could be wrong if some XS code has changed the
1378 * locale behind our back) */
0d071d52 1379
837ce802 1380 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 1381 PL_numeric_standard = TRUE;
4c68b815 1382 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 1383 set_numeric_radix(0);
7d4bcc4a
KW
1384
1385# ifdef DEBUGGING
1386
2fcc0ca9
KW
1387 if (DEBUG_L_TEST || debug_initialization) {
1388 PerlIO_printf(Perl_debug_log,
58e4a467 1389 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 1390 }
98994639 1391
7d4bcc4a 1392# endif
98994639 1393#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1394
98994639
HS
1395}
1396
1397void
5792c642 1398Perl_set_numeric_underlying(pTHX)
98994639 1399{
7d4bcc4a 1400
98994639 1401#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1402
28c1bf33 1403 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1404 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1405 * instead of calling this directly. The macro avoids calling this routine
1406 * if toggling isn't necessary according to our records (which could be
1407 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1408
837ce802 1409 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 1410 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 1411 PL_numeric_underlying = TRUE;
3c62fd46 1412 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
1413
1414# ifdef DEBUGGING
1415
2fcc0ca9
KW
1416 if (DEBUG_L_TEST || debug_initialization) {
1417 PerlIO_printf(Perl_debug_log,
58e4a467 1418 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
1419 PL_numeric_name);
1420 }
98994639 1421
7d4bcc4a 1422# endif
98994639 1423#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1424
98994639
HS
1425}
1426
1427/*
1428 * Set up for a new ctype locale.
1429 */
a4f00dcc
KW
1430STATIC void
1431S_new_ctype(pTHX_ const char *newctype)
98994639 1432{
7d4bcc4a
KW
1433
1434#ifndef USE_LOCALE_CTYPE
1435
1436 PERL_ARGS_ASSERT_NEW_CTYPE;
1437 PERL_UNUSED_ARG(newctype);
1438 PERL_UNUSED_CONTEXT;
1439
1440#else
0d071d52 1441
291a84fb 1442 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1443 * core Perl this and that 'newctype' is the name of the new locale.
1444 *
1445 * This function sets up the folding arrays for all 256 bytes, assuming
1446 * that tofold() is tolc() since fold case is not a concept in POSIX,
1447 *
1448 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1449 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1450 * this function should be called directly only from this file and from
0d071d52
KW
1451 * POSIX::setlocale() */
1452
27da23d5 1453 dVAR;
68067e4e 1454 UV i;
98994639 1455
8b7358b9
KW
1456 /* Don't check for problems if we are suppressing the warnings */
1457 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
1458
7918f24d
NC
1459 PERL_ARGS_ASSERT_NEW_CTYPE;
1460
215c5139
KW
1461 /* We will replace any bad locale warning with 1) nothing if the new one is
1462 * ok; or 2) a new warning for the bad new locale */
1463 if (PL_warn_locale) {
1464 SvREFCNT_dec_NN(PL_warn_locale);
1465 PL_warn_locale = NULL;
1466 }
1467
c1284011 1468 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1469
1470 /* A UTF-8 locale gets standard rules. But note that code still has to
1471 * handle this specially because of the three problematic code points */
1472 if (PL_in_utf8_CTYPE_locale) {
1473 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
1474 }
8b7358b9
KW
1475
1476 /* We don't populate the other lists if a UTF-8 locale, but do check that
1477 * everything works as expected, unless checking turned off */
1478 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1479 /* Assume enough space for every character being bad. 4 spaces each
1480 * for the 94 printable characters that are output like "'x' "; and 5
1481 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1482 * NUL */
8b7358b9 1483 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1484 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1485 to start */
1486 unsigned int bad_count = 0; /* Count of bad characters */
1487
baa60164 1488 for (i = 0; i < 256; i++) {
8b7358b9 1489 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1490 if (isupper(i))
1491 PL_fold_locale[i] = (U8) tolower(i);
1492 else if (islower(i))
1493 PL_fold_locale[i] = (U8) toupper(i);
1494 else
1495 PL_fold_locale[i] = (U8) i;
8b7358b9 1496 }
8c6180a9
KW
1497
1498 /* If checking for locale problems, see if the native ASCII-range
1499 * printables plus \n and \t are in their expected categories in
1500 * the new locale. If not, this could mean big trouble, upending
1501 * Perl's and most programs' assumptions, like having a
1502 * metacharacter with special meaning become a \w. Fortunately,
1503 * it's very rare to find locales that aren't supersets of ASCII
1504 * nowadays. It isn't a problem for most controls to be changed
1505 * into something else; we check only \n and \t, though perhaps \r
1506 * could be an issue as well. */
7d4bcc4a 1507 if ( check_for_problems
8c6180a9
KW
1508 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1509 {
8b7358b9
KW
1510 bool is_bad = FALSE;
1511 char name[3] = { '\0' };
1512
1513 /* Convert the name into a string */
1514 if (isPRINT_A(i)) {
1515 name[0] = i;
1516 name[1] = '\0';
1517 }
1518 else if (i == '\n') {
1519 my_strlcpy(name, "\n", sizeof(name));
1520 }
1521 else {
1522 my_strlcpy(name, "\t", sizeof(name));
1523 }
1524
1525 /* Check each possibe class */
1526 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1527 is_bad = TRUE;
1528 DEBUG_L(PerlIO_printf(Perl_debug_log,
1529 "isalnum('%s') unexpectedly is %d\n",
1530 name, cBOOL(isalnum(i))));
1531 }
1532 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1533 is_bad = TRUE;
1534 DEBUG_L(PerlIO_printf(Perl_debug_log,
1535 "isalpha('%s') unexpectedly is %d\n",
1536 name, cBOOL(isalpha(i))));
1537 }
1538 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1539 is_bad = TRUE;
1540 DEBUG_L(PerlIO_printf(Perl_debug_log,
1541 "isdigit('%s') unexpectedly is %d\n",
1542 name, cBOOL(isdigit(i))));
1543 }
1544 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1545 is_bad = TRUE;
1546 DEBUG_L(PerlIO_printf(Perl_debug_log,
1547 "isgraph('%s') unexpectedly is %d\n",
1548 name, cBOOL(isgraph(i))));
1549 }
1550 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1551 is_bad = TRUE;
1552 DEBUG_L(PerlIO_printf(Perl_debug_log,
1553 "islower('%s') unexpectedly is %d\n",
1554 name, cBOOL(islower(i))));
1555 }
1556 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1557 is_bad = TRUE;
1558 DEBUG_L(PerlIO_printf(Perl_debug_log,
1559 "isprint('%s') unexpectedly is %d\n",
1560 name, cBOOL(isprint(i))));
1561 }
1562 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1563 is_bad = TRUE;
1564 DEBUG_L(PerlIO_printf(Perl_debug_log,
1565 "ispunct('%s') unexpectedly is %d\n",
1566 name, cBOOL(ispunct(i))));
1567 }
1568 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1569 is_bad = TRUE;
1570 DEBUG_L(PerlIO_printf(Perl_debug_log,
1571 "isspace('%s') unexpectedly is %d\n",
1572 name, cBOOL(isspace(i))));
1573 }
1574 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1575 is_bad = TRUE;
1576 DEBUG_L(PerlIO_printf(Perl_debug_log,
1577 "isupper('%s') unexpectedly is %d\n",
1578 name, cBOOL(isupper(i))));
1579 }
1580 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1581 is_bad = TRUE;
1582 DEBUG_L(PerlIO_printf(Perl_debug_log,
1583 "isxdigit('%s') unexpectedly is %d\n",
1584 name, cBOOL(isxdigit(i))));
1585 }
1586 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1587 is_bad = TRUE;
1588 DEBUG_L(PerlIO_printf(Perl_debug_log,
1589 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1590 name, tolower(i), (int) toLOWER_A(i)));
1591 }
1592 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1593 is_bad = TRUE;
1594 DEBUG_L(PerlIO_printf(Perl_debug_log,
1595 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1596 name, toupper(i), (int) toUPPER_A(i)));
1597 }
1598 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1599 is_bad = TRUE;
1600 DEBUG_L(PerlIO_printf(Perl_debug_log,
1601 "'\\n' (=%02X) is not a control\n", (int) i));
1602 }
1603
1604 /* Add to the list; Separate multiple entries with a blank */
1605 if (is_bad) {
1606 if (bad_count) {
1607 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1608 }
8b7358b9
KW
1609 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1610 bad_count++;
8c6180a9
KW
1611 }
1612 }
1613 }
1614
7d4bcc4a
KW
1615# ifdef MB_CUR_MAX
1616
8c6180a9 1617 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1618 * this locale requires more than one byte, there are going to be
8c6180a9 1619 * problems. */
9c8a6dc2
KW
1620 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1621 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1622 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1623
8b7358b9
KW
1624 if ( check_for_problems && MB_CUR_MAX > 1
1625 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1626
1627 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1628 * locale. Just assume that the implementation for them (plus
1629 * for POSIX) is correct and the > 1 value is spurious. (Since
1630 * these are specially handled to never be considered UTF-8
1631 * locales, as long as this is the only problem, everything
1632 * should work fine */
1633 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1634 {
8c6180a9
KW
1635 multi_byte_locale = TRUE;
1636 }
7d4bcc4a
KW
1637
1638# endif
8c6180a9 1639
8b7358b9
KW
1640 if (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale)) {
1641 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1642 PL_warn_locale = Perl_newSVpvf(aTHX_
1643 "Locale '%s' contains (at least) the following characters"
1644 " which have\nnon-standard meanings: %s\nThe Perl program"
1645 " will use the standard meanings",
1646 newctype, bad_chars_list);
1647
1648 }
1649 else {
1650 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1651 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1652 newctype,
8c6180a9
KW
1653 (multi_byte_locale)
1654 ? " Some characters in it are not recognized by"
1655 " Perl."
1656 : "",
1657 (bad_count)
1658 ? "\nThe following characters (and maybe others)"
1659 " may not have the same meaning as the Perl"
1660 " program expects:\n"
1661 : "",
1662 (bad_count)
1663 ? bad_chars_list
1664 : ""
1665 );
8b7358b9
KW
1666 }
1667
b119c2be
KW
1668# ifdef HAS_NL_LANGINFO
1669
1670 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1671 /* parameter FALSE is a don't care here */
1672 my_nl_langinfo(PERL_CODESET, FALSE));
1673
1674# endif
1675
8b7358b9
KW
1676 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1677
cc9eaeb0 1678 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1679 * output the message now. If not in that scope, we save the
1680 * message to be output at the first operation using this locale,
1681 * if that actually happens. Most programs don't use locales, so
1682 * they are immune to bad ones. */
cc9eaeb0 1683 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1684
780fcc9f
KW
1685 /* The '0' below suppresses a bogus gcc compiler warning */
1686 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1687
bddebb56
KW
1688 if (IN_LC(LC_CTYPE)) {
1689 SvREFCNT_dec_NN(PL_warn_locale);
1690 PL_warn_locale = NULL;
1691 }
780fcc9f 1692 }
baa60164 1693 }
31f05a37 1694 }
98994639
HS
1695
1696#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1697
98994639
HS
1698}
1699
98994639 1700void
2726666d
KW
1701Perl__warn_problematic_locale()
1702{
2726666d
KW
1703
1704#ifdef USE_LOCALE_CTYPE
1705
5f04a188
KW
1706 dTHX;
1707
1708 /* Internal-to-core function that outputs the message in PL_warn_locale,
1709 * and then NULLS it. Should be called only through the macro
1710 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1711
2726666d 1712 if (PL_warn_locale) {
2726666d
KW
1713 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1714 SvPVX(PL_warn_locale),
1715 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1716 SvREFCNT_dec_NN(PL_warn_locale);
1717 PL_warn_locale = NULL;
1718 }
1719
1720#endif
1721
1722}
1723
a4f00dcc
KW
1724STATIC void
1725S_new_collate(pTHX_ const char *newcoll)
98994639 1726{
7d4bcc4a
KW
1727
1728#ifndef USE_LOCALE_COLLATE
1729
1730 PERL_UNUSED_ARG(newcoll);
1731 PERL_UNUSED_CONTEXT;
1732
1733#else
0d071d52 1734
291a84fb 1735 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1736 * core Perl this and that 'newcoll' is the name of the new locale.
1737 *
d35fca5f
KW
1738 * The design of locale collation is that every locale change is given an
1739 * index 'PL_collation_ix'. The first time a string particpates in an
1740 * operation that requires collation while locale collation is active, it
1741 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1742 * magic includes the collation index, and the transformation of the string
1743 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1744 * instead of the string itself. If a string changes, the magic is
1745 * cleared. The next time the locale changes, the index is incremented,
1746 * and so we know during a comparison that the transformation is not
1747 * necessarily still valid, and so is recomputed. Note that if the locale
1748 * changes enough times, the index could wrap (a U32), and it is possible
1749 * that a transformation would improperly be considered valid, leading to
1750 * an unlikely bug */
0d071d52 1751
98994639
HS
1752 if (! newcoll) {
1753 if (PL_collation_name) {
1754 ++PL_collation_ix;
1755 Safefree(PL_collation_name);
1756 PL_collation_name = NULL;
1757 }
1758 PL_collation_standard = TRUE;
00bf60ca 1759 is_standard_collation:
98994639
HS
1760 PL_collxfrm_base = 0;
1761 PL_collxfrm_mult = 2;
165a1c52 1762 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1763 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1764 PL_strxfrm_max_cp = 0;
98994639
HS
1765 return;
1766 }
1767
d35fca5f 1768 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1769 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1770 ++PL_collation_ix;
1771 Safefree(PL_collation_name);
1772 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1773 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1774 if (PL_collation_standard) {
1775 goto is_standard_collation;
1776 }
98994639 1777
165a1c52 1778 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1779 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1780 PL_strxfrm_max_cp = 0;
165a1c52 1781
59c018b9
KW
1782 /* A locale collation definition includes primary, secondary, tertiary,
1783 * etc. weights for each character. To sort, the primary weights are
1784 * used, and only if they compare equal, then the secondary weights are
1785 * used, and only if they compare equal, then the tertiary, etc.
1786 *
1787 * strxfrm() works by taking the input string, say ABC, and creating an
1788 * output transformed string consisting of first the primary weights,
1789 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1790 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1791 * may not have weights at every level. In our example, let's say B
1792 * doesn't have a tertiary weight, and A doesn't have a secondary
1793 * weight. The constructed string is then going to be
1794 * A¹B¹C¹ B²C² A³C³ ....
1795 * This has the desired effect that strcmp() will look at the secondary
1796 * or tertiary weights only if the strings compare equal at all higher
1797 * priority weights. The spaces shown here, like in
c342d20e 1798 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1799 * are not just for readability. In the general case, these must
1800 * actually be bytes, which we will call here 'separator weights'; and
1801 * they must be smaller than any other weight value, but since these
1802 * are C strings, only the terminating one can be a NUL (some
1803 * implementations may include a non-NUL separator weight just before
1804 * the NUL). Implementations tend to reserve 01 for the separator
1805 * weights. They are needed so that a shorter string's secondary
1806 * weights won't be misconstrued as primary weights of a longer string,
1807 * etc. By making them smaller than any other weight, the shorter
1808 * string will sort first. (Actually, if all secondary weights are
1809 * smaller than all primary ones, there is no need for a separator
1810 * weight between those two levels, etc.)
1811 *
1812 * The length of the transformed string is roughly a linear function of
1813 * the input string. It's not exactly linear because some characters
1814 * don't have weights at all levels. When we call strxfrm() we have to
1815 * allocate some memory to hold the transformed string. The
1816 * calculations below try to find coefficients 'm' and 'b' for this
1817 * locale so that m*x + b equals how much space we need, given the size
1818 * of the input string in 'x'. If we calculate too small, we increase
1819 * the size as needed, and call strxfrm() again, but it is better to
1820 * get it right the first time to avoid wasted expensive string
1821 * transformations. */
1822
98994639 1823 {
79f120c8
KW
1824 /* We use the string below to find how long the tranformation of it
1825 * is. Almost all locales are supersets of ASCII, or at least the
1826 * ASCII letters. We use all of them, half upper half lower,
1827 * because if we used fewer, we might hit just the ones that are
1828 * outliers in a particular locale. Most of the strings being
1829 * collated will contain a preponderance of letters, and even if
1830 * they are above-ASCII, they are likely to have the same number of
1831 * weight levels as the ASCII ones. It turns out that digits tend
1832 * to have fewer levels, and some punctuation has more, but those
1833 * are relatively sparse in text, and khw believes this gives a
1834 * reasonable result, but it could be changed if experience so
1835 * dictates. */
1836 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1837 char * x_longer; /* Transformed 'longer' */
1838 Size_t x_len_longer; /* Length of 'x_longer' */
1839
1840 char * x_shorter; /* We also transform a substring of 'longer' */
1841 Size_t x_len_shorter;
1842
a4a439fb 1843 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1844 * are interested only in its length). It is used because it has
1845 * the intelligence to handle all cases, but to work, it needs some
1846 * values of 'm' and 'b' to get it started. For the purposes of
1847 * this calculation we use a very conservative estimate of 'm' and
1848 * 'b'. This assumes a weight can be multiple bytes, enough to
1849 * hold any UV on the platform, and there are 5 levels, 4 weight
1850 * bytes, and a trailing NUL. */
1851 PL_collxfrm_base = 5;
1852 PL_collxfrm_mult = 5 * sizeof(UV);
1853
1854 /* Find out how long the transformation really is */
a4a439fb
KW
1855 x_longer = _mem_collxfrm(longer,
1856 sizeof(longer) - 1,
1857 &x_len_longer,
1858
1859 /* We avoid converting to UTF-8 in the
1860 * called function by telling it the
1861 * string is in UTF-8 if the locale is a
1862 * UTF-8 one. Since the string passed
1863 * here is invariant under UTF-8, we can
1864 * claim it's UTF-8 even though it isn't.
1865 * */
1866 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1867 Safefree(x_longer);
1868
1869 /* Find out how long the transformation of a substring of 'longer'
1870 * is. Together the lengths of these transformations are
1871 * sufficient to calculate 'm' and 'b'. The substring is all of
1872 * 'longer' except the first character. This minimizes the chances
1873 * of being swayed by outliers */
a4a439fb 1874 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1875 sizeof(longer) - 2,
a4a439fb
KW
1876 &x_len_shorter,
1877 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1878 Safefree(x_shorter);
1879
1880 /* If the results are nonsensical for this simple test, the whole
1881 * locale definition is suspect. Mark it so that locale collation
1882 * is not active at all for it. XXX Should we warn? */
1883 if ( x_len_shorter == 0
1884 || x_len_longer == 0
1885 || x_len_shorter >= x_len_longer)
1886 {
1887 PL_collxfrm_mult = 0;
1888 PL_collxfrm_base = 0;
1889 }
1890 else {
1891 SSize_t base; /* Temporary */
1892
1893 /* We have both: m * strlen(longer) + b = x_len_longer
1894 * m * strlen(shorter) + b = x_len_shorter;
1895 * subtracting yields:
1896 * m * (strlen(longer) - strlen(shorter))
1897 * = x_len_longer - x_len_shorter
1898 * But we have set things up so that 'shorter' is 1 byte smaller
1899 * than 'longer'. Hence:
1900 * m = x_len_longer - x_len_shorter
1901 *
1902 * But if something went wrong, make sure the multiplier is at
1903 * least 1.
1904 */
1905 if (x_len_longer > x_len_shorter) {
1906 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
1907 }
1908 else {
1909 PL_collxfrm_mult = 1;
1910 }
1911
1912 /* mx + b = len
1913 * so: b = len - mx
1914 * but in case something has gone wrong, make sure it is
1915 * non-negative */
1916 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1917 if (base < 0) {
1918 base = 0;
1919 }
1920
1921 /* Add 1 for the trailing NUL */
1922 PL_collxfrm_base = base + 1;
1923 }
58eebef2 1924
7d4bcc4a
KW
1925# ifdef DEBUGGING
1926
58eebef2
KW
1927 if (DEBUG_L_TEST || debug_initialization) {
1928 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1929 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1930 "x_len_longer=%zu,"
1931 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1932 __FILE__, __LINE__,
1933 PL_in_utf8_COLLATE_locale,
1934 x_len_shorter, x_len_longer,
1935 PL_collxfrm_mult, PL_collxfrm_base);
1936 }
7d4bcc4a
KW
1937# endif
1938
98994639
HS
1939 }
1940 }
1941
1942#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1943
98994639
HS
1944}
1945
d2b24094 1946#ifdef WIN32
b8cc575c 1947
a4f00dcc 1948STATIC char *
b8cc575c 1949S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1950{
1951 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1952 * difference between the two unless the input locale is "", which normally
1953 * means on Windows to get the machine default, which is set via the
1954 * computer's "Regional and Language Options" (or its current equivalent).
1955 * In POSIX, it instead means to find the locale from the user's
1956 * environment. This routine changes the Windows behavior to first look in
1957 * the environment, and, if anything is found, use that instead of going to
1958 * the machine default. If there is no environment override, the machine
1959 * default is used, by calling the real setlocale() with "".
1960 *
1961 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1962 * use the particular category's variable if set; otherwise to use the LANG
1963 * variable. */
b385bb4d 1964
175c4cf9 1965 bool override_LC_ALL = FALSE;
89f7b9aa 1966 char * result;
e5f10d49 1967 unsigned int i;
89f7b9aa 1968
b385bb4d 1969 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1970
1971# ifdef LC_ALL
1972
b385bb4d
KW
1973 locale = PerlEnv_getenv("LC_ALL");
1974 if (! locale) {
e5f10d49
KW
1975 if (category == LC_ALL) {
1976 override_LC_ALL = TRUE;
1977 }
1978 else {
7d4bcc4a
KW
1979
1980# endif
7d4bcc4a 1981
e5f10d49
KW
1982 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1983 if (category == categories[i]) {
1984 locale = PerlEnv_getenv(category_names[i]);
1985 goto found_locale;
1986 }
1987 }
7d4bcc4a 1988
b385bb4d 1989 locale = PerlEnv_getenv("LANG");
481465ea 1990 if (! locale) {
b385bb4d
KW
1991 locale = "";
1992 }
e5f10d49
KW
1993
1994 found_locale: ;
7d4bcc4a
KW
1995
1996# ifdef LC_ALL
1997
e5f10d49 1998 }
b385bb4d 1999 }
7d4bcc4a
KW
2000
2001# endif
2002
b385bb4d
KW
2003 }
2004
89f7b9aa 2005 result = setlocale(category, locale);
48015184
KW
2006 DEBUG_L(STMT_START {
2007 dSAVE_ERRNO;
2008 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2009 setlocale_debug_string(category, locale, result));
2010 RESTORE_ERRNO;
2011 } STMT_END);
89f7b9aa 2012
481465ea 2013 if (! override_LC_ALL) {
89f7b9aa
KW
2014 return result;
2015 }
2016
dfd77d7a 2017 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2018 * LANG variable or the system default if there is no LANG. But these have
2019 * lower priority than the other LC_foo variables, so override it for each
2020 * one that is set. (If they are set to "", it means to use the same thing
2021 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2022
948523db 2023 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2024 result = PerlEnv_getenv(category_names[i]);
2025 if (result && strNE(result, "")) {
2026 setlocale(categories[i], result);
2027 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2028 __FILE__, __LINE__,
2029 setlocale_debug_string(categories[i], result, "not captured")));
2030 }
89f7b9aa 2031 }
7d4bcc4a 2032
bbc98134 2033 result = setlocale(LC_ALL, NULL);
48015184
KW
2034 DEBUG_L(STMT_START {
2035 dSAVE_ERRNO;
2036 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2037 __FILE__, __LINE__,
48015184
KW
2038 setlocale_debug_string(LC_ALL, NULL, result));
2039 RESTORE_ERRNO;
2040 } STMT_END);
89f7b9aa 2041
bbc98134 2042 return result;
b385bb4d
KW
2043}
2044
2045#endif
2046
9aac5db8
KW
2047/*
2048
2049=head1 Locale-related functions and macros
2050
2051=for apidoc Perl_setlocale
2052
2053This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2054taking the same parameters, and returning the same information, except that it
2055returns the correct underlying C<LC_NUMERIC> locale, instead of C<C> always, as
2056perl keeps that locale category as C<C>, changing it briefly during the
2057operations where the underlying one is required.
2058
e9bc6d6b 2059Another reason it isn't completely a drop-in replacement is that it is
9aac5db8
KW
2060declared to return S<C<const char *>>, whereas the system setlocale omits the
2061C<const>. (If it were being written today, plain setlocale would be declared
2062const, since it is illegal to change the information it returns; doing so leads
2063to segfaults.)
2064
e9bc6d6b
KW
2065Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2066C<setlocale> can be completely ineffective on some platforms under some
2067configurations.
2068
9aac5db8 2069C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2070where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2071the system C<setlocale()> is ineffective, returning the wrong information, and
2072failing to actually change the locale. C<Perl_setlocale>, however works
2073properly in all circumstances.
9aac5db8
KW
2074
2075The return points to a per-thread static buffer, which is overwritten the next
2076time C<Perl_setlocale> is called from the same thread.
2077
2078=cut
2079
2080*/
2081
2082const char *
2083Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2084{
2085 /* This wraps POSIX::setlocale() */
2086
9aac5db8
KW
2087 const char * retval;
2088 const char * newlocale;
2089 dSAVEDERRNO;
2090 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc
KW
2091 dTHX;
2092
a4f00dcc
KW
2093#ifdef USE_LOCALE_NUMERIC
2094
291a84fb
KW
2095 /* A NULL locale means only query what the current one is. We have the
2096 * LC_NUMERIC name saved, because we are normally switched into the C
2097 * locale for it. For an LC_ALL query, switch back to get the correct
2098 * results. All other categories don't require special handling */
a4f00dcc
KW
2099 if (locale == NULL) {
2100 if (category == LC_NUMERIC) {
9aac5db8
KW
2101
2102 /* We don't have to copy this return value, as it is a per-thread
2103 * variable, and won't change until a future setlocale */
2104 return PL_numeric_name;
a4f00dcc
KW
2105 }
2106
7d4bcc4a 2107# ifdef LC_ALL
a4f00dcc 2108
9aac5db8
KW
2109 else if (category == LC_ALL) {
2110 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2111 }
2112
7d4bcc4a 2113# endif
a4f00dcc
KW
2114
2115 }
2116
2117#endif
2118
9aac5db8
KW
2119 retval = do_setlocale_r(category, locale);
2120 SAVE_ERRNO;
2121
2122#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2123
2124 if (locale == NULL && category == LC_ALL) {
2125 RESTORE_LC_NUMERIC();
2126 }
2127
2128#endif
a4f00dcc
KW
2129
2130 DEBUG_L(PerlIO_printf(Perl_debug_log,
2131 "%s:%d: %s\n", __FILE__, __LINE__,
2132 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2133
9aac5db8
KW
2134 RESTORE_ERRNO;
2135
2136 if (! retval) {
a4f00dcc
KW
2137 return NULL;
2138 }
2139
9aac5db8
KW
2140 save_to_buffer(retval, &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
2141 retval = PL_setlocale_buf;
2142
2143 /* If locale == NULL, we are just querying the state */
a4f00dcc 2144 if (locale == NULL) {
a4f00dcc
KW
2145 return retval;
2146 }
a4f00dcc 2147
1159483a
KW
2148 /* Now that have switched locales, we have to update our records to
2149 * correspond. */
a4f00dcc 2150
1159483a 2151 switch (category) {
a4f00dcc 2152
1159483a 2153#ifdef USE_LOCALE_CTYPE
a4f00dcc 2154
1159483a
KW
2155 case LC_CTYPE:
2156 new_ctype(retval);
2157 break;
a4f00dcc 2158
1159483a 2159#endif
a4f00dcc
KW
2160#ifdef USE_LOCALE_COLLATE
2161
1159483a
KW
2162 case LC_COLLATE:
2163 new_collate(retval);
2164 break;
a4f00dcc 2165
1159483a 2166#endif
a4f00dcc
KW
2167#ifdef USE_LOCALE_NUMERIC
2168
1159483a
KW
2169 case LC_NUMERIC:
2170 new_numeric(retval);
2171 break;
a4f00dcc 2172
1159483a
KW
2173#endif
2174#ifdef LC_ALL
a4f00dcc 2175
1159483a 2176 case LC_ALL:
a4f00dcc 2177
1159483a
KW
2178 /* LC_ALL updates all the things we care about. The values may not
2179 * be the same as 'retval', as the locale "" may have set things
2180 * individually */
a4f00dcc 2181
1159483a 2182# ifdef USE_LOCALE_CTYPE
a4f00dcc 2183
1159483a
KW
2184 newlocale = do_setlocale_c(LC_CTYPE, NULL);
2185 new_ctype(newlocale);
a4f00dcc 2186
1159483a
KW
2187# endif /* USE_LOCALE_CTYPE */
2188# ifdef USE_LOCALE_COLLATE
2189
2190 newlocale = do_setlocale_c(LC_COLLATE, NULL);
2191 new_collate(newlocale);
a4f00dcc 2192
7d4bcc4a 2193# endif
1159483a 2194# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2195
1159483a
KW
2196 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
2197 new_numeric(newlocale);
a4f00dcc 2198
1159483a
KW
2199# endif /* USE_LOCALE_NUMERIC */
2200#endif /* LC_ALL */
a4f00dcc 2201
1159483a
KW
2202 default:
2203 break;
a4f00dcc
KW
2204 }
2205
2206 return retval;
2207
f7416781
KW
2208}
2209
2210PERL_STATIC_INLINE const char *
2211S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2212{
2213 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2214 * growing it if necessary */
2215
2216 const Size_t string_size = strlen(string) + offset + 1;
2217
2218 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2219
2220 if (*buf_size == 0) {
2221 Newx(*buf, string_size, char);
2222 *buf_size = string_size;
2223 }
2224 else if (string_size > *buf_size) {
2225 Renew(*buf, string_size, char);
2226 *buf_size = string_size;
2227 }
2228
2229 Copy(string, *buf + offset, string_size - offset, char);
2230 return *buf;
2231}
2232
2233/*
2234
f7416781
KW
2235=for apidoc Perl_langinfo
2236
ce181bf6 2237This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2238taking the same C<item> parameter values, and returning the same information.
2239But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2240of Perl's locale handling from your code, and can be used on systems that lack
2241a native C<nl_langinfo>.
2242
2243Expanding on these:
2244
2245=over
2246
2247=item *
2248
ce181bf6
KW
2249The reason it isn't quite a drop-in replacement is actually an advantage. The
2250only difference is that it returns S<C<const char *>>, whereas plain
2251C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2252forbidden to write into the buffer. By declaring this C<const>, the compiler
2253enforces this restriction, so if it is violated, you know at compilation time,
2254rather than getting segfaults at runtime.
2255
2256=item *
2257
f7416781
KW
2258It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
2259without you having to write extra code. The reason for the extra code would be
2260because these are from the C<LC_NUMERIC> locale category, which is normally
2261kept set to the C locale by Perl, no matter what the underlying locale is
2262supposed to be, and so to get the expected results, you have to temporarily
ce181bf6
KW
2263toggle into the underlying locale, and later toggle back. (You could use plain
2264C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but
2265then you wouldn't get the other advantages of C<Perl_langinfo()>; not keeping
2266C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is expecting the
2267radix (decimal point) character to be a dot.)
2268
2269=item *
2270
2271The system function it replaces can have its static return buffer trashed,
2272not only by a subesequent call to that function, but by a C<freelocale>,
2273C<setlocale>, or other locale change. The returned buffer of this function is
2274not changed until the next call to it, so the buffer is never in a trashed
2275state.
f7416781
KW
2276
2277=item *
2278
ce181bf6
KW
2279Its return buffer is per-thread, so it also is never overwritten by a call to
2280this function from another thread; unlike the function it replaces.
2281
2282=item *
2283
2284But most importantly, it works on systems that don't have C<nl_langinfo>, such
2285as Windows, hence makes your code more portable. Of the fifty-some possible
2286items specified by the POSIX 2008 standard,
f7416781 2287L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
ce181bf6
KW
2288only two are completely unimplemented (though the loss of one of these is
2289significant). It uses various techniques to recover the other items, including
2290calling C<L<localeconv(3)>>, and C<L<strftime(3)>>, both of which are specified
2291in C89, so should be always be available. Later C<strftime()> versions have
2292additional capabilities; C<""> is returned for those not available on your
2293system.
2294
2295It is important to note that when called with an item that is recovered by
2296using C<localeconv>, the buffer from any previous explicit call to
2297C<localeconv> will be overwritten. This means you must save that buffer's
2298contents if you need to access them after a call to this function.
9f0bc911 2299
f7416781
KW
2300The details for those items which may differ from what this emulation returns
2301and what a native C<nl_langinfo()> would return are:
2302
2303=over
2304
2305=item C<CODESET>
2306
2307=item C<ERA>
2308
2309Unimplemented, so returns C<"">.
2310
2311=item C<YESEXPR>
2312
c1566110
KW
2313=item C<YESSTR>
2314
f7416781
KW
2315=item C<NOEXPR>
2316
c1566110
KW
2317=item C<NOSTR>
2318
2319Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
ce181bf6
KW
2320removed from POSIX 2008, and are retained here for backwards compatibility.
2321Your platform's C<nl_langinfo> may not support them.
f7416781
KW
2322
2323=item C<D_FMT>
2324
2325Always evaluates to C<%x>, the locale's appropriate date representation.
2326
2327=item C<T_FMT>
2328
2329Always evaluates to C<%X>, the locale's appropriate time representation.
2330
2331=item C<D_T_FMT>
2332
2333Always evaluates to C<%c>, the locale's appropriate date and time
2334representation.
2335
2336=item C<CRNCYSTR>
2337
2338The return may be incorrect for those rare locales where the currency symbol
2339replaces the radix character.
2340Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2341to work differently.
2342
2343=item C<ALT_DIGITS>
2344
2345Currently this gives the same results as Linux does.
2346Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
2347to work differently.
2348
2349=item C<ERA_D_FMT>
2350
2351=item C<ERA_T_FMT>
2352
2353=item C<ERA_D_T_FMT>
2354
2355=item C<T_FMT_AMPM>
2356
2357These are derived by using C<strftime()>, and not all versions of that function
2358know about them. C<""> is returned for these on such systems.
2359
2360=back
2361
ce181bf6
KW
2362=back
2363
f7416781
KW
2364When using C<Perl_langinfo> on systems that don't have a native
2365C<nl_langinfo()>, you must
2366
2367 #include "perl_langinfo.h"
2368
2369before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2370C<#include> with this one. (Doing it this way keeps out the symbols that plain
2371C<langinfo.h> imports into the namespace for code that doesn't need it.)
2372
2373You also should not use the bare C<langinfo.h> item names, but should preface
2374them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
2375The C<PERL_I<foo>> versions will also work for this function on systems that do
2376have a native C<nl_langinfo>.
2377
f7416781
KW
2378The original impetus for C<Perl_langinfo()> was so that code that needs to
2379find out the current currency symbol, floating point radix character, or digit
2380grouping separator can use, on all systems, the simpler and more
2381thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2382pain to make thread-friendly. For other fields returned by C<localeconv>, it
2383is better to use the methods given in L<perlcall> to call
2384L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2385
2386=cut
2387
2388*/
2389
2390const char *
2391#ifdef HAS_NL_LANGINFO
2392Perl_langinfo(const nl_item item)
2393#else
2394Perl_langinfo(const int item)
2395#endif
2396{
f61748ac
KW
2397 return my_nl_langinfo(item, TRUE);
2398}
2399
2400const char *
2401#ifdef HAS_NL_LANGINFO
2402S_my_nl_langinfo(const nl_item item, bool toggle)
2403#else
2404S_my_nl_langinfo(const int item, bool toggle)
2405#endif
2406{
ae74815b 2407 dTHX;
f7416781 2408
5a854ab3
KW
2409 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2410 * two items, and only if not already there */
2411 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
2412 || PL_numeric_underlying))
2413 {
2414 toggle = FALSE;
2415 }
2416
ab340fff 2417#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d
KW
2418# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
2419 || ! defined(HAS_POSIX_2008_LOCALE)
f7416781 2420
ab340fff 2421 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2422 * for those items dependent on it. This must be copied to a buffer before
2423 * switching back, as some systems destroy the buffer when setlocale() is
2424 * called */
f7416781 2425
038d3702
KW
2426 {
2427 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2428
b2fee59c 2429 if (toggle) {
038d3702 2430 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2431 }
f7416781 2432
5acc4454
KW
2433 LOCALE_LOCK; /* Prevent interference from another thread executing
2434 this code section (the only call to nl_langinfo in
2435 the core) */
2436
ee90eb2d 2437
628ff75a
KW
2438 /* Copy to a per-thread buffer, which is also one that won't be
2439 * destroyed by a subsequent setlocale(), such as the
2440 * RESTORE_LC_NUMERIC may do just below. */
2441 save_to_buffer(nl_langinfo(item),
2442 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2443
5acc4454
KW
2444 LOCALE_UNLOCK;
2445
b2fee59c 2446 if (toggle) {
038d3702 2447 RESTORE_LC_NUMERIC();
b2fee59c 2448 }
f7416781
KW
2449 }
2450
ab340fff
KW
2451# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2452
5a854ab3 2453 {
b2fee59c
KW
2454 bool do_free = FALSE;
2455 locale_t cur = uselocale((locale_t) 0);
ab340fff 2456
b2fee59c
KW
2457 if (cur == LC_GLOBAL_LOCALE) {
2458 cur = duplocale(LC_GLOBAL_LOCALE);
2459 do_free = TRUE;
2460 }
ab340fff 2461
b2fee59c 2462 if (toggle) {
e1aa2579
KW
2463 if (PL_underlying_numeric_obj) {
2464 cur = PL_underlying_numeric_obj;
2465 }
2466 else {
2467 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2468 do_free = TRUE;
2469 }
b2fee59c 2470 }
ab340fff 2471
628ff75a
KW
2472 /* We have to save it to a buffer, because the freelocale() just below
2473 * can invalidate the internal one */
2474 save_to_buffer(nl_langinfo_l(item, cur),
2475 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2476
b2fee59c
KW
2477 if (do_free) {
2478 freelocale(cur);
2479 }
5a854ab3 2480 }
ab340fff 2481
c1566110
KW
2482# endif
2483
628ff75a 2484 if (strEQ(PL_langinfo_buf, "")) {
c1566110
KW
2485 if (item == PERL_YESSTR) {
2486 return "yes";
2487 }
2488 if (item == PERL_NOSTR) {
2489 return "no";
2490 }
2491 }
2492
628ff75a 2493 return PL_langinfo_buf;
ab340fff 2494
f7416781 2495#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2496
2497 {
2498
f7416781
KW
2499# ifdef HAS_LOCALECONV
2500
43dd6b15 2501 const struct lconv* lc;
628ff75a 2502 const char * temp;
038d3702 2503 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
2504
2505# endif
2506# ifdef HAS_STRFTIME
2507
43dd6b15
KW
2508 struct tm tm;
2509 bool return_format = FALSE; /* Return the %format, not the value */
2510 const char * format;
f7416781
KW
2511
2512# endif
2513
43dd6b15
KW
2514 /* We copy the results to a per-thread buffer, even if not
2515 * multi-threaded. This is in part to simplify this code, and partly
2516 * because we need a buffer anyway for strftime(), and partly because a
2517 * call of localeconv() could otherwise wipe out the buffer, and the
2518 * programmer would not be expecting this, as this is a nl_langinfo()
2519 * substitute after all, so s/he might be thinking their localeconv()
2520 * is safe until another localeconv() call. */
f7416781 2521
43dd6b15
KW
2522 switch (item) {
2523 Size_t len;
f7416781 2524
43dd6b15
KW
2525 /* These 2 are unimplemented */
2526 case PERL_CODESET:
2527 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 2528
43dd6b15
KW
2529 default:
2530 return "";
f7416781 2531
43dd6b15
KW
2532 /* We use only an English set, since we don't know any more */
2533 case PERL_YESEXPR: return "^[+1yY]";
2534 case PERL_YESSTR: return "yes";
2535 case PERL_NOEXPR: return "^[-0nN]";
2536 case PERL_NOSTR: return "no";
f7416781
KW
2537
2538# ifdef HAS_LOCALECONV
2539
43dd6b15 2540 case PERL_CRNCYSTR:
f7416781 2541
43dd6b15
KW
2542 /* We don't bother with localeconv_l() because any system that
2543 * has it is likely to also have nl_langinfo() */
291a84fb 2544
5acc4454
KW
2545 LOCALE_LOCK; /* Prevent interference with other threads
2546 using localeconv() */
2547
43dd6b15
KW
2548 lc = localeconv();
2549 if ( ! lc
2550 || ! lc->currency_symbol
2551 || strEQ("", lc->currency_symbol))
2552 {
2553 LOCALE_UNLOCK;
2554 return "";
2555 }
f7416781 2556
43dd6b15
KW
2557 /* Leave the first spot empty to be filled in below */
2558 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2559 &PL_langinfo_bufsize, 1);
2560 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2561 { /* khw couldn't figure out how the localedef specifications
2562 would show that the $ should replace the radix; this is
2563 just a guess as to how it might work.*/
2564 *PL_langinfo_buf = '.';
2565 }
2566 else if (lc->p_cs_precedes) {
2567 *PL_langinfo_buf = '-';
2568 }
2569 else {
2570 *PL_langinfo_buf = '+';
2571 }
f7416781 2572
43dd6b15
KW
2573 LOCALE_UNLOCK;
2574 break;
f7416781 2575
43dd6b15
KW
2576 case PERL_RADIXCHAR:
2577 case PERL_THOUSEP:
f7416781 2578
43dd6b15 2579 if (toggle) {
038d3702 2580 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2581 }
f7416781 2582
5acc4454
KW
2583 LOCALE_LOCK; /* Prevent interference with other threads
2584 using localeconv() */
2585
43dd6b15
KW
2586 lc = localeconv();
2587 if (! lc) {
628ff75a 2588 temp = "";
33394adc 2589 }
43dd6b15 2590 else {
628ff75a 2591 temp = (item == PERL_RADIXCHAR)
43dd6b15
KW
2592 ? lc->decimal_point
2593 : lc->thousands_sep;
628ff75a
KW
2594 if (! temp) {
2595 temp = "";
43dd6b15
KW
2596 }
2597 }
f7416781 2598
628ff75a 2599 save_to_buffer(temp, &PL_langinfo_buf,
43dd6b15 2600 &PL_langinfo_bufsize, 0);
f7416781 2601
5acc4454
KW
2602 LOCALE_UNLOCK;
2603
43dd6b15 2604 if (toggle) {
038d3702 2605 RESTORE_LC_NUMERIC();
43dd6b15 2606 }
f7416781 2607
43dd6b15 2608 break;
f7416781
KW
2609
2610# endif
2611# ifdef HAS_STRFTIME
2612
43dd6b15
KW
2613 /* These are defined by C89, so we assume that strftime supports
2614 * them, and so are returned unconditionally; they may not be what
2615 * the locale actually says, but should give good enough results
2616 * for someone using them as formats (as opposed to trying to parse
2617 * them to figure out what the locale says). The other format
2618 * items are actually tested to verify they work on the platform */
2619 case PERL_D_FMT: return "%x";
2620 case PERL_T_FMT: return "%X";
2621 case PERL_D_T_FMT: return "%c";
2622
2623 /* These formats are only available in later strfmtime's */
2624 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
2625 case PERL_T_FMT_AMPM:
2626
2627 /* The rest can be gotten from most versions of strftime(). */
2628 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
2629 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
2630 case PERL_ABDAY_7:
2631 case PERL_ALT_DIGITS:
2632 case PERL_AM_STR: case PERL_PM_STR:
2633 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
2634 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
2635 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
2636 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
2637 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
2638 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
2639 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
2640 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
2641 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
2642 case PERL_MON_12:
2643
2644 LOCALE_LOCK;
2645
2646 init_tm(&tm); /* Precaution against core dumps */
2647 tm.tm_sec = 30;
2648 tm.tm_min = 30;
2649 tm.tm_hour = 6;
2650 tm.tm_year = 2017 - 1900;
2651 tm.tm_wday = 0;
2652 tm.tm_mon = 0;
2653 switch (item) {
2654 default:
2655 LOCALE_UNLOCK;
2656 Perl_croak(aTHX_
2657 "panic: %s: %d: switch case: %d problem",
2658 __FILE__, __LINE__, item);
2659 NOT_REACHED; /* NOTREACHED */
2660
2661 case PERL_PM_STR: tm.tm_hour = 18;
2662 case PERL_AM_STR:
2663 format = "%p";
2664 break;
2665
2666 case PERL_ABDAY_7: tm.tm_wday++;
2667 case PERL_ABDAY_6: tm.tm_wday++;
2668 case PERL_ABDAY_5: tm.tm_wday++;
2669 case PERL_ABDAY_4: tm.tm_wday++;
2670 case PERL_ABDAY_3: tm.tm_wday++;
2671 case PERL_ABDAY_2: tm.tm_wday++;
2672 case PERL_ABDAY_1:
2673 format = "%a";
2674 break;
2675
2676 case PERL_DAY_7: tm.tm_wday++;
2677 case PERL_DAY_6: tm.tm_wday++;
2678 case PERL_DAY_5: tm.tm_wday++;
2679 case PERL_DAY_4: tm.tm_wday++;
2680 case PERL_DAY_3: tm.tm_wday++;
2681 case PERL_DAY_2: tm.tm_wday++;
2682 case PERL_DAY_1:
2683 format = "%A";
2684 break;
2685
2686 case PERL_ABMON_12: tm.tm_mon++;
2687 case PERL_ABMON_11: tm.tm_mon++;
2688 case PERL_ABMON_10: tm.tm_mon++;
2689 case PERL_ABMON_9: tm.tm_mon++;
2690 case PERL_ABMON_8: tm.tm_mon++;
2691 case PERL_ABMON_7: tm.tm_mon++;
2692 case PERL_ABMON_6: tm.tm_mon++;
2693 case PERL_ABMON_5: tm.tm_mon++;
2694 case PERL_ABMON_4: tm.tm_mon++;
2695 case PERL_ABMON_3: tm.tm_mon++;
2696 case PERL_ABMON_2: tm.tm_mon++;
2697 case PERL_ABMON_1:
2698 format = "%b";
2699 break;
2700
2701 case PERL_MON_12: tm.tm_mon++;
2702 case PERL_MON_11: tm.tm_mon++;
2703 case PERL_MON_10: tm.tm_mon++;
2704 case PERL_MON_9: tm.tm_mon++;
2705 case PERL_MON_8: tm.tm_mon++;
2706 case PERL_MON_7: tm.tm_mon++;
2707 case PERL_MON_6: tm.tm_mon++;
2708 case PERL_MON_5: tm.tm_mon++;
2709 case PERL_MON_4: tm.tm_mon++;
2710 case PERL_MON_3: tm.tm_mon++;
2711 case PERL_MON_2: tm.tm_mon++;
2712 case PERL_MON_1:
2713 format = "%B";
2714 break;
2715
2716 case PERL_T_FMT_AMPM:
2717 format = "%r";
2718 return_format = TRUE;
2719 break;
2720
2721 case PERL_ERA_D_FMT:
2722 format = "%Ex";
2723 return_format = TRUE;
2724 break;
2725
2726 case PERL_ERA_T_FMT:
2727 format = "%EX";
2728 return_format = TRUE;
2729 break;
2730
2731 case PERL_ERA_D_T_FMT:
2732 format = "%Ec";
2733 return_format = TRUE;
2734 break;
2735
2736 case PERL_ALT_DIGITS:
2737 tm.tm_wday = 0;
2738 format = "%Ow"; /* Find the alternate digit for 0 */
2739 break;
2740 }
f7416781 2741
43dd6b15
KW
2742 /* We can't use my_strftime() because it doesn't look at
2743 * tm_wday */
2744 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
2745 format, &tm))
2746 {
2747 /* A zero return means one of:
2748 * a) there wasn't enough space in PL_langinfo_buf
2749 * b) the format, like a plain %p, returns empty
2750 * c) it was an illegal format, though some
2751 * implementations of strftime will just return the
2752 * illegal format as a plain character sequence.
2753 *
2754 * To quickly test for case 'b)', try again but precede
2755 * the format with a plain character. If that result is
2756 * still empty, the problem is either 'a)' or 'c)' */
2757
2758 Size_t format_size = strlen(format) + 1;
2759 Size_t mod_size = format_size + 1;
2760 char * mod_format;
2761 char * temp_result;
2762
2763 Newx(mod_format, mod_size, char);
2764 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 2765 *mod_format = ' ';
43dd6b15
KW
2766 my_strlcpy(mod_format + 1, format, mod_size);
2767 len = strftime(temp_result,
2768 PL_langinfo_bufsize,
2769 mod_format, &tm);
2770 Safefree(mod_format);
2771 Safefree(temp_result);
2772
2773 /* If 'len' is non-zero, it means that we had a case like
2774 * %p which means the current locale doesn't use a.m. or
2775 * p.m., and that is valid */
2776 if (len == 0) {
2777
2778 /* Here, still didn't work. If we get well beyond a
2779 * reasonable size, bail out to prevent an infinite
2780 * loop. */
2781
2782 if (PL_langinfo_bufsize > 100 * format_size) {
2783 *PL_langinfo_buf = '\0';
2784 }
2785 else {
2786 /* Double the buffer size to retry; Add 1 in case
2787 * original was 0, so we aren't stuck at 0. */
2788 PL_langinfo_bufsize *= 2;
2789 PL_langinfo_bufsize++;
2790 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2791 continue;
2792 }
2793 }
f7416781 2794
f7416781 2795 break;
43dd6b15 2796 }
f7416781 2797
43dd6b15
KW
2798 /* Here, we got a result.
2799 *
2800 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
2801 * alternate format for wday 0. If the value is the same as
2802 * the normal 0, there isn't an alternate, so clear the buffer.
2803 * */
2804 if ( item == PERL_ALT_DIGITS
2805 && strEQ(PL_langinfo_buf, "0"))
2806 {
2807 *PL_langinfo_buf = '\0';
2808 }
f7416781 2809
43dd6b15
KW
2810 /* ALT_DIGITS is problematic. Experiments on it showed that
2811 * strftime() did not always work properly when going from
2812 * alt-9 to alt-10. Only a few locales have this item defined,
2813 * and in all of them on Linux that khw was able to find,
2814 * nl_langinfo() merely returned the alt-0 character, possibly
2815 * doubled. Most Unicode digits are in blocks of 10
2816 * consecutive code points, so that is sufficient information
2817 * for those scripts, as we can infer alt-1, alt-2, .... But
2818 * for a Japanese locale, a CJK ideographic 0 is returned, and
2819 * the CJK digits are not in code point order, so you can't
2820 * really infer anything. The localedef for this locale did
2821 * specify the succeeding digits, so that strftime() works
2822 * properly on them, without needing to infer anything. But
2823 * the nl_langinfo() return did not give sufficient information
2824 * for the caller to understand what's going on. So until
2825 * there is evidence that it should work differently, this
2826 * returns the alt-0 string for ALT_DIGITS.
2827 *
2828 * wday was chosen because its range is all a single digit.
2829 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 2830
43dd6b15 2831 LOCALE_UNLOCK;
f7416781 2832
43dd6b15
KW
2833 /* If to return the format, not the value, overwrite the buffer
2834 * with it. But some strftime()s will keep the original format
2835 * if illegal, so change those to "" */
2836 if (return_format) {
2837 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
2838 *PL_langinfo_buf = '\0';
2839 }
43dd6b15
KW
2840 else {
2841 save_to_buffer(format, &PL_langinfo_buf,
2842 &PL_langinfo_bufsize, 0);
f7416781
KW
2843 }
2844 }
2845
2846 break;
f7416781
KW
2847
2848# endif
2849
43dd6b15 2850 }
f7416781
KW
2851 }
2852
2853 return PL_langinfo_buf;
2854
2855#endif
2856
a4f00dcc 2857}
b385bb4d 2858
98994639
HS
2859/*
2860 * Initialize locale awareness.
2861 */
2862int
2863Perl_init_i18nl10n(pTHX_ int printwarn)
2864{
0e92a118
KW
2865 /* printwarn is
2866 *
2867 * 0 if not to output warning when setup locale is bad
2868 * 1 if to output warning based on value of PERL_BADLANG
2869 * >1 if to output regardless of PERL_BADLANG
2870 *
2871 * returns
98994639 2872 * 1 = set ok or not applicable,
0e92a118
KW
2873 * 0 = fallback to a locale of lower priority
2874 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
2875 *
2876 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
2877 * set, debugging information is output.
2878 *
2879 * This looks more complicated than it is, mainly due to the #ifdefs.
2880 *
2881 * We try to set LC_ALL to the value determined by the environment. If
2882 * there is no LC_ALL on this platform, we try the individual categories we
2883 * know about. If this works, we are done.
2884 *
2885 * But if it doesn't work, we have to do something else. We search the
2886 * environment variables ourselves instead of relying on the system to do
2887 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
2888 * think there is one), and the ultimate fallback "C". This is all done in
2889 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
2890 * more complex. The 'trial_locales' array is initialized with just one
2891 * element; it causes the behavior described in the paragraph above this to
2892 * happen. If that fails, we add elements to 'trial_locales', and do extra
2893 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
2894 *
2895 * On Ultrix, the locale MUST come from the environment, so there is
2896 * preliminary code to set it. I (khw) am not sure that it is necessary,
2897 * and that this couldn't be folded into the loop, but barring any real
2898 * platforms to test on, it's staying as-is
2899 *
2900 * A slight complication is that in embedded Perls, the locale may already
2901 * be set-up, and we don't want to get it from the normal environment
2902 * variables. This is handled by having a special environment variable
2903 * indicate we're in this situation. We simply set setlocale's 2nd
2904 * parameter to be a NULL instead of "". That indicates to setlocale that
2905 * it is not to change anything, but to return the current value,
2906 * effectively initializing perl's db to what the locale already is.
2907 *
2908 * We play the same trick with NULL if a LC_ALL succeeds. We call
2909 * setlocale() on the individual categores with NULL to get their existing
2910 * values for our db, instead of trying to change them.
2911 * */
98994639 2912
0e92a118
KW
2913 int ok = 1;
2914
7d4bcc4a
KW
2915#ifndef USE_LOCALE
2916
2917 PERL_UNUSED_ARG(printwarn);
2918
2919#else /* USE_LOCALE */
7d4bcc4a
KW
2920# ifdef __GLIBC__
2921
175c4cf9 2922 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
2923
2924# endif
65ebb059 2925
ccd65d51
KW
2926 /* NULL uses the existing already set up locale */
2927 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
2928 ? NULL
2929 : "";
c3fcd832
KW
2930 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
2931 unsigned int trial_locales_count;
175c4cf9
KW
2932 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
2933 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 2934 bool setlocale_failure = FALSE;
65ebb059 2935 unsigned int i;
175c4cf9
KW
2936
2937 /* A later getenv() could zap this, so only use here */
2938 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
2939
2940 const bool locwarn = (printwarn > 1
e5f10d49
KW
2941 || ( printwarn
2942 && ( ! bad_lang_use_once
22ff3130 2943 || (
e5f10d49
KW
2944 /* disallow with "" or "0" */
2945 *bad_lang_use_once
2946 && strNE("0", bad_lang_use_once)))));
ea92aad8 2947
291a84fb 2948 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 2949 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
2950
2951 /* current locale for given category; should have been copied so aren't
2952 * volatile */
8de4332b 2953 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 2954
7d4bcc4a
KW
2955# ifdef WIN32
2956
6bce99ee
JH
2957 /* In some systems you can find out the system default locale
2958 * and use that as the fallback locale. */
7d4bcc4a
KW
2959# define SYSTEM_DEFAULT_LOCALE
2960# endif
2961# ifdef SYSTEM_DEFAULT_LOCALE
2962
65ebb059 2963 const char *system_default_locale = NULL;
98994639 2964
7d4bcc4a 2965# endif
948523db
KW
2966
2967# ifndef DEBUGGING
2968# define DEBUG_LOCALE_INIT(a,b,c)
2969# else
7d4bcc4a 2970
8298454c 2971 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
2972
2973# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
2974 STMT_START { \
2975 if (debug_initialization) { \
2976 PerlIO_printf(Perl_debug_log, \
2977 "%s:%d: %s\n", \
2978 __FILE__, __LINE__, \
a4f00dcc 2979 setlocale_debug_string(category, \
2fcc0ca9
KW
2980 locale, \
2981 result)); \
2982 } \
2983 } STMT_END
2fcc0ca9 2984
948523db
KW
2985/* Make sure the parallel arrays are properly set up */
2986# ifdef USE_LOCALE_NUMERIC
2987 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
2988 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
2989# ifdef USE_POSIX_2008_LOCALE
2990 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
2991# endif
948523db
KW
2992# endif
2993# ifdef USE_LOCALE_CTYPE
2994 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
2995 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
2996# ifdef USE_POSIX_2008_LOCALE
2997 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
2998# endif
948523db
KW
2999# endif
3000# ifdef USE_LOCALE_COLLATE
3001 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3002 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3003# ifdef USE_POSIX_2008_LOCALE
3004 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3005# endif
948523db
KW
3006# endif
3007# ifdef USE_LOCALE_TIME
3008 assert(categories[LC_TIME_INDEX] == LC_TIME);
3009 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3010# ifdef USE_POSIX_2008_LOCALE
3011 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3012# endif
948523db
KW
3013# endif
3014# ifdef USE_LOCALE_MESSAGES
3015 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3016 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3017# ifdef USE_POSIX_2008_LOCALE
3018 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3019# endif
948523db
KW
3020# endif
3021# ifdef USE_LOCALE_MONETARY
3022 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3023 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3024# ifdef USE_POSIX_2008_LOCALE
3025 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3026# endif
948523db 3027# endif
9821811f
KW
3028# ifdef USE_LOCALE_ADDRESS
3029 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3030 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3031# ifdef USE_POSIX_2008_LOCALE
3032 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3033# endif
9821811f
KW
3034# endif
3035# ifdef USE_LOCALE_IDENTIFICATION
3036 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3037 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3038# ifdef USE_POSIX_2008_LOCALE
3039 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3040# endif
9821811f
KW
3041# endif
3042# ifdef USE_LOCALE_MEASUREMENT
3043 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3044 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3045# ifdef USE_POSIX_2008_LOCALE
3046 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3047# endif
9821811f
KW
3048# endif
3049# ifdef USE_LOCALE_PAPER
3050 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3051 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3052# ifdef USE_POSIX_2008_LOCALE
3053 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3054# endif
9821811f
KW
3055# endif
3056# ifdef USE_LOCALE_TELEPHONE
3057 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3058 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3059# ifdef USE_POSIX_2008_LOCALE
3060 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3061# endif
9821811f 3062# endif
948523db
KW
3063# ifdef LC_ALL
3064 assert(categories[LC_ALL_INDEX] == LC_ALL);
3065 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3066 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3067# ifdef USE_POSIX_2008_LOCALE
3068 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3069# endif
948523db
KW
3070# endif
3071# endif /* DEBUGGING */
47280b20
KW
3072
3073 /* Initialize the cache of the program's UTF-8ness for the always known
3074 * locales C and POSIX */
3075 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3076 sizeof(PL_locale_utf8ness));
3077
e9bc6d6b
KW
3078# ifdef USE_THREAD_SAFE_LOCALE
3079# ifdef WIN32
3080
3081 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3082
3083# endif
3084# endif
3085# if defined(LC_ALL_MASK) && defined(HAS_POSIX_2008_LOCALE)
3086
3087 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3088 if (! PL_C_locale_obj) {
3089 Perl_croak_nocontext(
3090 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3091 }
3092 if (DEBUG_Lv_TEST || debug_initialization) {
3093 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3094 }
3095
3096# endif
3097
3ca88433
KW
3098 PL_numeric_radix_sv = newSVpvs(".");
3099
e9bc6d6b
KW
3100# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3101
3102 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3103 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3104
3105# endif
b56c4436 3106# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3107
3108 /*
3109 * Ultrix setlocale(..., "") fails if there are no environment
3110 * variables from which to get a locale name.
3111 */
3112
b56c4436
KW
3113# ifndef LC_ALL
3114# error Ultrix without LC_ALL not implemented
3115# else
7d4bcc4a 3116
b56c4436
KW
3117 {
3118 bool done = FALSE;
ea92aad8
KW
3119 if (lang) {
3120 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3121 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3122 if (sl_result[LC_ALL_INDEX])
3123 done = TRUE;
3124 else
e5f10d49 3125 setlocale_failure = TRUE;
ea92aad8
KW
3126 }
3127 if (! setlocale_failure) {
3128 const char * locale_param;
3129 for (i = 0; i < LC_ALL_INDEX; i++) {
3130 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3131 ? setlocale_init
3132 : NULL;
3133 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3134 if (! sl_result[i]) {
3135 setlocale_failure = TRUE;
3136 }
3137 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3138 }
c835d6be 3139 }
7d4bcc4a
KW
3140 }
3141
3142# endif /* LC_ALL */
e5f10d49 3143# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3144
65ebb059 3145 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3146 * the list. Normally the loop is executed just once. But if setting the
3147 * locale fails, inside the loop we add fallback trials to the array and so
3148 * will execute the loop multiple times */
c3fcd832
KW
3149 trial_locales[0] = setlocale_init;
3150 trial_locales_count = 1;
7d4bcc4a 3151
65ebb059
KW
3152 for (i= 0; i < trial_locales_count; i++) {
3153 const char * trial_locale = trial_locales[i];
3154
3155 if (i > 0) {
3156
3157 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3158 * when i==0, but I (khw) don't think that behavior makes much
3159 * sense */
3160 setlocale_failure = FALSE;
3161
7d4bcc4a 3162# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3163# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3164
65ebb059
KW
3165 /* On Windows machines, an entry of "" after the 0th means to use
3166 * the system default locale, which we now proceed to get. */
3167 if (strEQ(trial_locale, "")) {
3168 unsigned int j;
3169
3170 /* Note that this may change the locale, but we are going to do
3171 * that anyway just below */
837ce802 3172 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3173 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3174
7d4bcc4a 3175 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3176 * try */
3177 if (! system_default_locale) {
3178 goto next_iteration;
3179 }
3180 for (j = 0; j < trial_locales_count; j++) {
3181 if (strEQ(system_default_locale, trial_locales[j])) {
3182 goto next_iteration;
3183 }
3184 }
3185
3186 trial_locale = system_default_locale;
3187 }
ec0202b5
KW
3188# else
3189# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3190# endif
7d4bcc4a 3191# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3192
3193 } /* For i > 0 */
65ebb059 3194
7d4bcc4a
KW
3195# ifdef LC_ALL
3196
948523db
KW
3197 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3198 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3199 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3200 setlocale_failure = TRUE;
7cd8b568
KW
3201 }
3202 else {
3203 /* Since LC_ALL succeeded, it should have changed all the other
3204 * categories it can to its value; so we massage things so that the
3205 * setlocales below just return their category's current values.
3206 * This adequately handles the case in NetBSD where LC_COLLATE may
3207 * not be defined for a locale, and setting it individually will
7d4bcc4a 3208 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3209 * the POSIX locale. */
3210 trial_locale = NULL;
3211 }
7d4bcc4a
KW
3212
3213# endif /* LC_ALL */
98994639 3214
e5f10d49
KW
3215 if (! setlocale_failure) {
3216 unsigned int j;
3217 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3218 curlocales[j]
3219 = savepv(do_setlocale_r(categories[j], trial_locale));
3220 if (! curlocales[j]) {
3221 setlocale_failure = TRUE;
3222 }
3223 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3224 }
c835d6be 3225
e5f10d49
KW
3226 if (! setlocale_failure) { /* All succeeded */
3227 break; /* Exit trial_locales loop */
49c85077 3228 }
65ebb059 3229 }
98994639 3230
49c85077
KW
3231 /* Here, something failed; will need to try a fallback. */
3232 ok = 0;
65ebb059 3233
49c85077
KW
3234 if (i == 0) {
3235 unsigned int j;
98994639 3236
65ebb059 3237 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3238
3239# ifdef LC_ALL
98994639 3240
49c85077
KW
3241 PerlIO_printf(Perl_error_log,
3242 "perl: warning: Setting locale failed.\n");
98994639 3243
7d4bcc4a 3244# else /* !LC_ALL */
98994639 3245
49c85077
KW
3246 PerlIO_printf(Perl_error_log,
3247 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3248
e5f10d49
KW
3249 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3250 if (! curlocales[j]) {
3251 PerlIO_printf(Perl_error_log, category_names[j]);
3252 }
3253 else {
3254 Safefree(curlocales[j]);
3255 }
3256 }
7d4bcc4a 3257
7d4bcc4a 3258# endif /* LC_ALL */
98994639 3259
49c85077
KW
3260 PerlIO_printf(Perl_error_log,
3261 "perl: warning: Please check that your locale settings:\n");
98994639 3262
7d4bcc4a
KW
3263# ifdef __GLIBC__
3264
49c85077
KW
3265 PerlIO_printf(Perl_error_log,
3266 "\tLANGUAGE = %c%s%c,\n",
3267 language ? '"' : '(',
3268 language ? language : "unset",
3269 language ? '"' : ')');
7d4bcc4a 3270# endif
98994639 3271
49c85077
KW
3272 PerlIO_printf(Perl_error_log,
3273 "\tLC_ALL = %c%s%c,\n",
3274 lc_all ? '"' : '(',
3275 lc_all ? lc_all : "unset",
3276 lc_all ? '"' : ')');
98994639 3277
7d4bcc4a
KW
3278# if defined(USE_ENVIRON_ARRAY)
3279
49c85077 3280 {
cd999af9 3281 char **e;
d5e32b93
KW
3282
3283 /* Look through the environment for any variables of the
3284 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3285 * already handled above. These are assumed to be locale
3286 * settings. Output them and their values. */
cd999af9 3287 for (e = environ; *e; e++) {
d5e32b93
KW
3288 const STRLEN prefix_len = sizeof("LC_") - 1;
3289 STRLEN uppers_len;
3290
cd999af9 3291 if ( strBEGINs(*e, "LC_")
c8b388b0 3292 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3293 && (uppers_len = strspn(*e + prefix_len,
3294 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3295 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3296 {
3297 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3298 (int) (prefix_len + uppers_len), *e,
3299 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3300 }
3301 }
49c85077 3302 }
7d4bcc4a
KW
3303
3304# else
3305
49c85077
KW
3306 PerlIO_printf(Perl_error_log,
3307 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3308
3309# endif
98994639 3310
49c85077
KW
3311 PerlIO_printf(Perl_error_log,
3312 "\tLANG = %c%s%c\n",
3313 lang ? '"' : '(',
3314 lang ? lang : "unset",
3315 lang ? '"' : ')');
98994639 3316
49c85077
KW
3317 PerlIO_printf(Perl_error_log,
3318 " are supported and installed on your system.\n");
3319 }
98994639 3320
65ebb059 3321 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3322 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3323 * usually change the upper bound of the loop we are in.
3324 *
3325 * Since the system's default way of setting the locale has not
3326 * found one that works, We use Perl's defined ordering: LC_ALL,
3327 * LANG, and the C locale. We don't try the same locale twice, so
3328 * don't add to the list if already there. (On POSIX systems, the
3329 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3330 * but there's no harm done by doing it explicitly.
3331 *
3332 * Note that this tries the LC_ALL environment variable even on
3333 * systems which have no LC_ALL locale setting. This may or may
3334 * not have been originally intentional, but there's no real need
3335 * to change the behavior. */
65ebb059
KW
3336 if (lc_all) {
3337 for (j = 0; j < trial_locales_count; j++) {
3338 if (strEQ(lc_all, trial_locales[j])) {
3339 goto done_lc_all;
3340 }
3341 }
3342 trial_locales[trial_locales_count++] = lc_all;
3343 }
3344 done_lc_all:
98994639 3345
65ebb059
KW
3346 if (lang) {
3347 for (j = 0; j < trial_locales_count; j++) {
3348 if (strEQ(lang, trial_locales[j])) {
3349 goto done_lang;
3350 }
3351 }
3352 trial_locales[trial_locales_count++] = lang;
3353 }
3354 done_lang:
3355
7d4bcc4a
KW
3356# if defined(WIN32) && defined(LC_ALL)
3357
65ebb059
KW
3358 /* For Windows, we also try the system default locale before "C".
3359 * (If there exists a Windows without LC_ALL we skip this because
3360 * it gets too complicated. For those, the "C" is the next
3361 * fallback possibility). The "" is the same as the 0th element of
3362 * the array, but the code at the loop above knows to treat it
3363 * differently when not the 0th */
3364 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3365
3366# endif
65ebb059
KW
3367
3368 for (j = 0; j < trial_locales_count; j++) {
3369 if (strEQ("C", trial_locales[j])) {
3370 goto done_C;
3371 }
3372 }
3373 trial_locales[trial_locales_count++] = "C";
98994639 3374
65ebb059
KW
3375 done_C: ;
3376 } /* end of first time through the loop */
98994639 3377
7d4bcc4a
KW
3378# ifdef WIN32
3379
65ebb059 3380 next_iteration: ;
7d4bcc4a
KW
3381
3382# endif
65ebb059
KW
3383
3384 } /* end of looping through the trial locales */
3385
3386 if (ok < 1) { /* If we tried to fallback */
3387 const char* msg;
3388 if (! setlocale_failure) { /* fallback succeeded */
3389 msg = "Falling back to";
3390 }
3391 else { /* fallback failed */
e5f10d49 3392 unsigned int j;
98994639 3393
65ebb059
KW
3394 /* We dropped off the end of the loop, so have to decrement i to
3395 * get back to the value the last time through */
3396 i--;
98994639 3397
65ebb059
KW
3398 ok = -1;
3399 msg = "Failed to fall back to";
3400
3401 /* To continue, we should use whatever values we've got */
7d4bcc4a 3402
e5f10d49
KW
3403 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3404 Safefree(curlocales[j]);
3405 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3406 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3407 }
65ebb059
KW
3408 }
3409
3410 if (locwarn) {
3411 const char * description;
3412 const char * name = "";
3413 if (strEQ(trial_locales[i], "C")) {
3414 description = "the standard locale";
3415 name = "C";
3416 }
7d4bcc4a
KW
3417
3418# ifdef SYSTEM_DEFAULT_LOCALE
3419
65ebb059
KW
3420 else if (strEQ(trial_locales[i], "")) {
3421 description = "the system default locale";
3422 if (system_default_locale) {
3423 name = system_default_locale;
3424 }
3425 }
7d4bcc4a
KW
3426
3427# endif /* SYSTEM_DEFAULT_LOCALE */
3428
65ebb059
KW
3429 else {
3430 description = "a fallback locale";
3431 name = trial_locales[i];
3432 }
3433 if (name && strNE(name, "")) {
3434 PerlIO_printf(Perl_error_log,
3435 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3436 }
3437 else {
3438 PerlIO_printf(Perl_error_log,
3439 "perl: warning: %s %s.\n", msg, description);
3440 }
3441 }
3442 } /* End of tried to fallback */
98994639 3443
e5f10d49
KW
3444 /* Done with finding the locales; update our records */
3445
7d4bcc4a
KW
3446# ifdef USE_LOCALE_CTYPE
3447
948523db 3448 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3449
e5f10d49 3450# endif
7d4bcc4a
KW
3451# ifdef USE_LOCALE_COLLATE
3452
948523db 3453 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3454
e5f10d49 3455# endif
7d4bcc4a
KW
3456# ifdef USE_LOCALE_NUMERIC
3457
948523db 3458 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3459
3460# endif
3461
948523db 3462 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3463
e9bc6d6b 3464# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3465
3466 /* This caches whether each category's locale is UTF-8 or not. This
3467 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3468 * initialization time before any threads have started, but not later
3469 * unless thread-safe operations are used.
47280b20
KW
3470 * Caching means that if the program heeds our dictate not to change
3471 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3472 * it may get queried without having to change locales. If the
3473 * environment is such that all categories have the same locale, this
3474 * isn't needed, as the code will not change the locale; but this
3475 * handles the uncommon case where the environment has disparate
3476 * locales for the categories */
47280b20
KW
3477 (void) _is_cur_LC_category_utf8(categories[i]);
3478
3479# endif
3480
e5f10d49
KW
3481 Safefree(curlocales[i]);
3482 }
b310b053 3483
7d4bcc4a
KW
3484# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3485
49c85077 3486 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3487 * locale is UTF-8. The call to new_ctype() just above has already
3488 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3489 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3490 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3491 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3492 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3493
a05d7ebb 3494 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3495 This is an alternative to using the -C command line switch
3496 (the -C if present will override this). */
3497 {
dd374669 3498 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3499 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3500 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3501 PL_utf8cache = -1;
b310b053
JH
3502 }
3503
7d4bcc4a 3504# endif
7d4bcc4a
KW
3505# ifdef __GLIBC__
3506
175c4cf9 3507 Safefree(language);
7d4bcc4a
KW
3508
3509# endif
175c4cf9
KW
3510
3511 Safefree(lc_all);
3512 Safefree(lang);
3513
e3305790 3514#endif /* USE_LOCALE */
2fcc0ca9 3515#ifdef DEBUGGING
7d4bcc4a 3516
2fcc0ca9 3517 /* So won't continue to output stuff */
27cdc72e 3518 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3519
2fcc0ca9
KW
3520#endif
3521
98994639
HS
3522 return ok;
3523}
3524
98994639
HS
3525#ifdef USE_LOCALE_COLLATE
3526
a4a439fb 3527char *
a4a439fb
KW
3528Perl__mem_collxfrm(pTHX_ const char *input_string,
3529 STRLEN len, /* Length of 'input_string' */
3530 STRLEN *xlen, /* Set to length of returned string
3531 (not including the collation index
3532 prefix) */
3533 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3534 )
98994639 3535{
a4a439fb
KW
3536
3537 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3538 * differences. First, it handles embedded NULs. Second, it allocates a bit
3539 * more memory than needed for the transformed data itself. The real
55e5378d 3540 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3541 * the length of that, and doesn't include the collation index size.
3542 * Please see sv_collxfrm() to see how this is used. */
3543
55e5378d
KW
3544#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3545
6696cfa7
KW
3546 char * s = (char *) input_string;
3547 STRLEN s_strlen = strlen(input_string);
79f120c8 3548 char *xbuf = NULL;
55e5378d 3549 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3550 STRLEN length_in_chars;
c664130f 3551 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3552
a4a439fb
KW
3553 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3554
3555 /* Must be NUL-terminated */
3556 assert(*(input_string + len) == '\0');
7918f24d 3557
79f120c8
KW
3558 /* If this locale has defective collation, skip */
3559 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3560 DEBUG_L(PerlIO_printf(Perl_debug_log,
3561 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3562 goto bad;
3563 }
3564
6696cfa7
KW
3565 /* Replace any embedded NULs with the control that sorts before any others.
3566 * This will give as good as possible results on strings that don't
3567 * otherwise contain that character, but otherwise there may be
3568 * less-than-perfect results with that character and NUL. This is
fdc080f3 3569 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3570 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3571 NUL */
6696cfa7
KW
3572 char * e = s + len;
3573 char * sans_nuls;
fdc080f3 3574 STRLEN sans_nuls_len;
94762aa0 3575 int try_non_controls;
afc4976f
KW
3576 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3577 making sure 2nd byte is NUL.
3578 */
3579 STRLEN this_replacement_len;
3580
1e4c9676
KW
3581 /* If we don't know what non-NUL control character sorts lowest for
3582 * this locale, find it */
f28f4d2a 3583 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3584 int j;
afc4976f 3585 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
3586 includes the collation index
3587 prefixed. */
3588
91c0e2e0 3589 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
3590
3591 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
3592 * NUL, in which case we instead look for any character. Controls
3593 * are preferred because collation order is, in general, context
3594 * sensitive, with adjoining characters affecting the order, and
3595 * controls are less likely to have such interactions, allowing the
3596 * NUL-replacement to stand on its own. (Another way to look at it
3597 * is to imagine what would happen if the NUL were replaced by a
3598 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
3599 for (try_non_controls = 0;
3600 try_non_controls < 2;
3601 try_non_controls++)
3602 {
d4ff9586
KW
3603 /* Look through all legal code points (NUL isn't) */
3604 for (j = 1; j < 256; j++) {
3605 char * x; /* j's xfrm plus collation index */
3606 STRLEN x_len; /* length of 'x' */
3607 STRLEN trial_len = 1;
736a4fed 3608 char cur_source[] = { '\0', '\0' };
d4ff9586 3609
736a4fed
KW
3610 /* Skip non-controls the first time through the loop. The
3611 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
3612 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
3613 ? ! isCNTRL_L1(j)
3614 : ! isCNTRL_LC(j))
3615 {
d4ff9586 3616 continue;
6696cfa7 3617 }
6696cfa7 3618
736a4fed
KW
3619 /* Create a 1-char string of the current code point */
3620 cur_source[0] = (char) j;
3621
d4ff9586
KW
3622 /* Then transform it */
3623 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 3624 0 /* The string is not in UTF-8 */);
6696cfa7 3625
1e4c9676 3626 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
3627 * */
3628 if (! x) {
3629 continue;
3630 }
6696cfa7 3631
d4ff9586
KW
3632 /* If this character's transformation is lower than
3633 * the current lowest, this one becomes the lowest */
3634 if ( cur_min_x == NULL
3635 || strLT(x + COLLXFRM_HDR_LEN,
3636 cur_min_x + COLLXFRM_HDR_LEN))
3637 {
f28f4d2a 3638 PL_strxfrm_NUL_replacement = j;
d4ff9586 3639 cur_min_x = x;
d4ff9586
KW
3640 }
3641 else {
3642 Safefree(x);
3643 }
1e4c9676 3644 } /* end of loop through all 255 characters */
6696cfa7 3645
1e4c9676 3646 /* Stop looking if found */
94762aa0
KW
3647 if (cur_min_x) {
3648 break;
3649 }
3650
3651 /* Unlikely, but possible, if there aren't any controls that
3652 * work in the locale, repeat the loop, looking for any
3653 * character that works */
3654 DEBUG_L(PerlIO_printf(Perl_debug_log,
3655 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 3656 } /* End of loop to try first the controls, then any char */
6696cfa7 3657
94762aa0
KW
3658 if (! cur_min_x) {
3659 DEBUG_L(PerlIO_printf(Perl_debug_log,
3660 "_mem_collxfrm: Couldn't find any character to replace"
3661 " embedded NULs in locale %s with", PL_collation_name));
3662 goto bad;
58eebef2
KW
3663 }
3664
94762aa0
KW
3665 DEBUG_L(PerlIO_printf(Perl_debug_log,
3666 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 3667 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 3668
6696cfa7 3669 Safefree(cur_min_x);
1e4c9676 3670 } /* End of determining the character that is to replace NULs */
afc4976f
KW
3671
3672 /* If the replacement is variant under UTF-8, it must match the
291a84fb 3673 * UTF8-ness of the original */
f28f4d2a
KW
3674 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
3675 this_replacement_char[0] =
3676 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
3677 this_replacement_char[1] =
3678 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
3679 this_replacement_len = 2;
3680 }
3681 else {
f28f4d2a 3682 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
3683 /* this_replacement_char[1] = '\0' was done at initialization */
3684 this_replacement_len = 1;
6696cfa7
KW
3685 }
3686
3687 /* The worst case length for the replaced string would be if every
3688 * character in it is NUL. Multiply that by the length of each
3689 * replacement, and allow for a trailing NUL */
afc4976f 3690 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 3691 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
3692 *sans_nuls = '\0';
3693
6696cfa7
KW
3694 /* Replace each NUL with the lowest collating control. Loop until have
3695 * exhausted all the NULs */
3696 while (s + s_strlen < e) {
6069d6c5 3697 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3698
3699 /* Do the actual replacement */
6069d6c5 3700 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
3701
3702 /* Move past the input NUL */
3703 s += s_strlen + 1;
3704 s_strlen = strlen(s);
3705 }
3706
3707 /* And add anything that trails the final NUL */
6069d6c5 3708 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
3709
3710 /* Switch so below we transform this modified string */
3711 s = sans_nuls;
3712 len = strlen(s);
1e4c9676 3713 } /* End of replacing NULs */
6696cfa7 3714
a4a439fb
KW
3715 /* Make sure the UTF8ness of the string and locale match */
3716 if (utf8 != PL_in_utf8_COLLATE_locale) {
9de6fe47 3717 /* XXX convert above Unicode to 10FFFF? */
a4a439fb
KW
3718 const char * const t = s; /* Temporary so we can later find where the
3719 input was */
3720
3721 /* Here they don't match. Change the string's to be what the locale is
3722 * expecting */
3723
3724 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
3725 s = (char *) bytes_to_utf8((const U8 *) s, &len);
3726 utf8 = TRUE;
3727 }
3728 else { /* locale is not UTF-8; but input is; downgrade the input */
3729
3730 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
3731
3732 /* If the downgrade was successful we are done, but if the input
3733 * contains things that require UTF-8 to represent, have to do
3734 * damage control ... */
3735 if (UNLIKELY(utf8)) {
3736
3737 /* What we do is construct a non-UTF-8 string with
3738 * 1) the characters representable by a single byte converted
3739 * to be so (if necessary);
3740 * 2) and the rest converted to collate the same as the
3741 * highest collating representable character. That makes
3742 * them collate at the end. This is similar to how we
3743 * handle embedded NULs, but we use the highest collating
3744 * code point instead of the smallest. Like the NUL case,
3745 * this isn't perfect, but is the best we can reasonably
3746 * do. Every above-255 code point will sort the same as
3747 * the highest-sorting 0-255 code point. If that code
3748 * point can combine in a sequence with some other code
3749 * points for weight calculations, us changing something to
3750 * be it can adversely affect the results. But in most
3751 * cases, it should work reasonably. And note that this is
3752 * really an illegal situation: using code points above 255
3753 * on a locale where only 0-255 are valid. If two strings
3754 * sort entirely equal, then the sort order for the
3755 * above-255 code points will be in code point order. */
3756
3757 utf8 = FALSE;
3758
3759 /* If we haven't calculated the code point with the maximum
3760 * collating order for this locale, do so now */
3761 if (! PL_strxfrm_max_cp) {
3762 int j;
3763
3764 /* The current transformed string that collates the
3765 * highest (except it also includes the prefixed collation
3766 * index. */
3767 char * cur_max_x = NULL;
3768
3769 /* Look through all legal code points (NUL isn't) */
3770 for (j = 1; j < 256; j++) {
3771 char * x;
3772 STRLEN x_len;
736a4fed 3773 char cur_source[] = { '\0', '\0' };
a4a439fb 3774
736a4fed
KW
3775 /* Create a 1-char string of the current code point */
3776 cur_source[0] = (char) j;
a4a439fb
KW
3777
3778 /* Then transform it */
3779 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
3780
3781 /* If something went wrong (which it shouldn't), just
3782 * ignore this code point */
94762aa0 3783 if (! x) {
a4a439fb
KW
3784 continue;
3785 }
3786
3787 /* If this character's transformation is higher than
3788 * the current highest, this one becomes the highest */
3789 if ( cur_max_x == NULL
55e5378d
KW
3790 || strGT(x + COLLXFRM_HDR_LEN,
3791 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
3792 {
3793 PL_strxfrm_max_cp = j;
3794 cur_max_x = x;
3795 }
3796 else {
3797 Safefree(x);
3798 }
3799 }
3800
94762aa0
KW
3801 if (! cur_max_x) {
3802 DEBUG_L(PerlIO_printf(Perl_debug_log,
3803 "_mem_collxfrm: Couldn't find any character to"
3804 " replace above-Latin1 chars in locale %s with",
3805 PL_collation_name));
3806 goto bad;
3807 }
3808
58eebef2
KW
3809 DEBUG_L(PerlIO_printf(Perl_debug_log,
3810 "_mem_collxfrm: highest 1-byte collating character"
3811 " in locale %s is 0x%02X\n",
3812 PL_collation_name,
3813 PL_strxfrm_max_cp));
58eebef2 3814
a4a439fb
KW
3815 Safefree(cur_max_x);
3816 }
3817
3818 /* Here we know which legal code point collates the highest.
3819 * We are ready to construct the non-UTF-8 string. The length
3820 * will be at least 1 byte smaller than the input string
3821 * (because we changed at least one 2-byte character into a
3822 * single byte), but that is eaten up by the trailing NUL */
3823 Newx(s, len, char);
3824
3825 {
3826 STRLEN i;
3827 STRLEN d= 0;
042d9e50 3828 char * e = (char *) t + len;
a4a439fb
KW
3829
3830 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
3831 U8 cur_char = t[i];
3832 if (UTF8_IS_INVARIANT(cur_char)) {
3833 s[d++] = cur_char;
3834 }
042d9e50 3835 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
3836 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
3837 }
3838 else { /* Replace illegal cp with highest collating
3839 one */
3840 s[d++] = PL_strxfrm_max_cp;
3841 }
3842 }
3843 s[d++] = '\0';
3844 Renew(s, d, char); /* Free up unused space */
3845 }
3846 }
3847 }
3848
3849 /* Here, we have constructed a modified version of the input. It could
3850 * be that we already had a modified copy before we did this version.
3851 * If so, that copy is no longer needed */
3852 if (t != input_string) {
3853 Safefree(t);
3854 }
3855 }
3856
17f41037
KW
3857 length_in_chars = (utf8)
3858 ? utf8_length((U8 *) s, (U8 *) s + len)
3859 : len;
3860
59c018b9
KW
3861 /* The first element in the output is the collation id, used by
3862 * sv_collxfrm(); then comes the space for the transformed string. The
3863 * equation should give us a good estimate as to how much is needed */
55e5378d 3864 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 3865 + PL_collxfrm_base
17f41037 3866 + (PL_collxfrm_mult * length_in_chars);
a02a5408 3867 Newx(xbuf, xAlloc, char);
c7202dee
KW
3868 if (UNLIKELY(! xbuf)) {
3869 DEBUG_L(PerlIO_printf(Perl_debug_log,
3870 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 3871 goto bad;
c7202dee 3872 }
98994639 3873
d35fca5f 3874 /* Store the collation id */
98994639 3875 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
3876
3877 /* Then the transformation of the input. We loop until successful, or we
3878 * give up */
4ebeff16 3879 for (;;) {
1adab0a7 3880
55e5378d 3881 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
3882
3883 /* If the transformed string occupies less space than we told strxfrm()
3884 * was available, it means it successfully transformed the whole
3885 * string. */
55e5378d 3886 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 3887
1adab0a7
KW
3888 /* Some systems include a trailing NUL in the returned length.
3889 * Ignore it, using a loop in case multiple trailing NULs are
3890 * returned. */
3891 while ( (*xlen) > 0
3892 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
3893 {
3894 (*xlen)--;
3895 }
3896
17f41037
KW
3897 /* If the first try didn't get it, it means our prediction was low.
3898 * Modify the coefficients so that we predict a larger value in any
3899 * future transformations */
3900 if (! first_time) {
3901 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
3902 STRLEN computed_guess = PL_collxfrm_base
3903 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
3904
3905 /* On zero-length input, just keep current slope instead of
3906 * dividing by 0 */
3907 const STRLEN new_m = (length_in_chars != 0)
3908 ? needed / length_in_chars
3909 : PL_collxfrm_mult;
17f41037
KW
3910
3911 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
3912 "%s: %d: initial size of %zu bytes for a length "
3913 "%zu string was insufficient, %zu needed\n",
17f41037 3914 __FILE__, __LINE__,
b07929e4 3915 computed_guess, length_in_chars, needed));