This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add, improve some debugging stmts for -DL (locales)
[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
5b64f24c
KW
56#ifdef I_WCTYPE
57# include <wctype.h>
58#endif
0dec74cd 59
2fcc0ca9
KW
60/* If the environment says to, we can output debugging information during
61 * initialization. This is done before option parsing, and before any thread
62 * creation, so can be a file-level static */
5a4b0634
KW
63#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
64# define debug_initialization 0
65# define DEBUG_INITIALIZATION_set(v)
66#else
2fcc0ca9 67static bool debug_initialization = FALSE;
5a4b0634 68# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
69#endif
70
48015184
KW
71
72/* Returns the Unix errno portion; ignoring any others. This is a macro here
73 * instead of putting it into perl.h, because unclear to khw what should be
74 * done generally. */
75#define GET_ERRNO saved_errno
76
291a84fb
KW
77/* strlen() of a literal string constant. We might want this more general,
78 * but using it in just this file for now. A problem with more generality is
79 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
80#define STRLENs(s) (sizeof("" s "") - 1)
81
63e5b0d7
KW
82/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
83 * return of setlocale(), then this is extremely likely to be the C or POSIX
84 * locale. However, the output of setlocale() is documented to be opaque, but
85 * the odds are extremely small that it would return these two strings for some
86 * other locale. Note that VMS in these two locales includes many non-ASCII
87 * characters as controls and punctuation (below are hex bytes):
88 * cntrl: 84-97 9B-9F
89 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
90 * Oddly, none there are listed as alphas, though some represent alphabetics
91 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
92#define isNAME_C_OR_POSIX(name) \
93 ( (name) != NULL \
94 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
95 || strEQ((name), "POSIX")))
96
8ef6e574
KW
97#ifdef USE_LOCALE
98
47280b20
KW
99/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
100 * looked up. This is in the form of a C string: */
101
102#define UTF8NESS_SEP "\v"
103#define UTF8NESS_PREFIX "\f"
104
105/* So, the string looks like:
98994639 106 *
47280b20 107 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 108 *
47280b20
KW
109 * where the digit 0 after the \a indicates that the locale starting just
110 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
111
112STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
113STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
114
115#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
116 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
117
118/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
119 * kept there always. The remining portion of the cache is LRU, with the
120 * oldest looked-up locale at the tail end */
121
98994639
HS
122STATIC char *
123S_stdize_locale(pTHX_ char *locs)
124{
47280b20
KW
125 /* Standardize the locale name from a string returned by 'setlocale',
126 * possibly modifying that string.
127 *
128 * The typical return value of setlocale() is either
129 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
130 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
131 * (the space-separated values represent the various sublocales,
132 * in some unspecified order). This is not handled by this function.
133 *
134 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
135 * which is harmful for further use of the string in setlocale(). This
136 * function removes the trailing new line and everything up through the '='
137 * */
138
7452cf6a 139 const char * const s = strchr(locs, '=');
98994639
HS
140 bool okay = TRUE;
141
7918f24d
NC
142 PERL_ARGS_ASSERT_STDIZE_LOCALE;
143
8772537c
AL
144 if (s) {
145 const char * const t = strchr(s, '.');
98994639 146 okay = FALSE;
8772537c
AL
147 if (t) {
148 const char * const u = strchr(t, '\n');
149 if (u && (u[1] == 0)) {
150 const STRLEN len = u - s;
151 Move(s + 1, locs, len, char);
152 locs[len] = 0;
153 okay = TRUE;
98994639
HS
154 }
155 }
156 }
157
158 if (!okay)
159 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
160
161 return locs;
162}
163
e5f10d49
KW
164/* Two parallel arrays; first the locale categories Perl uses on this system;
165 * the second array is their names. These arrays are in mostly arbitrary
166 * order. */
167
168const int categories[] = {
169
170# ifdef USE_LOCALE_NUMERIC
171 LC_NUMERIC,
172# endif
173# ifdef USE_LOCALE_CTYPE
174 LC_CTYPE,
175# endif
176# ifdef USE_LOCALE_COLLATE
177 LC_COLLATE,
178# endif
179# ifdef USE_LOCALE_TIME
180 LC_TIME,
181# endif
182# ifdef USE_LOCALE_MESSAGES
183 LC_MESSAGES,
184# endif
185# ifdef USE_LOCALE_MONETARY
186 LC_MONETARY,
187# endif
9821811f
KW
188# ifdef USE_LOCALE_ADDRESS
189 LC_ADDRESS,
190# endif
191# ifdef USE_LOCALE_IDENTIFICATION
192 LC_IDENTIFICATION,
193# endif
194# ifdef USE_LOCALE_MEASUREMENT
195 LC_MEASUREMENT,
196# endif
197# ifdef USE_LOCALE_PAPER
198 LC_PAPER,
199# endif
200# ifdef USE_LOCALE_TELEPHONE
201 LC_TELEPHONE,
202# endif
e5f10d49
KW
203# ifdef LC_ALL
204 LC_ALL,
205# endif
206 -1 /* Placeholder because C doesn't allow a
207 trailing comma, and it would get complicated
208 with all the #ifdef's */
209};
210
211/* The top-most real element is LC_ALL */
212
4ef8bdf9 213const char * const category_names[] = {
e5f10d49
KW
214
215# ifdef USE_LOCALE_NUMERIC
216 "LC_NUMERIC",
217# endif
218# ifdef USE_LOCALE_CTYPE
219 "LC_CTYPE",
220# endif
221# ifdef USE_LOCALE_COLLATE
222 "LC_COLLATE",
223# endif
224# ifdef USE_LOCALE_TIME
225 "LC_TIME",
226# endif
227# ifdef USE_LOCALE_MESSAGES
228 "LC_MESSAGES",
229# endif
230# ifdef USE_LOCALE_MONETARY
231 "LC_MONETARY",
232# endif
9821811f
KW
233# ifdef USE_LOCALE_ADDRESS
234 "LC_ADDRESS",
235# endif
236# ifdef USE_LOCALE_IDENTIFICATION
237 "LC_IDENTIFICATION",
238# endif
239# ifdef USE_LOCALE_MEASUREMENT
240 "LC_MEASUREMENT",
241# endif
242# ifdef USE_LOCALE_PAPER
243 "LC_PAPER",
244# endif
245# ifdef USE_LOCALE_TELEPHONE
246 "LC_TELEPHONE",
247# endif
e5f10d49
KW
248# ifdef LC_ALL
249 "LC_ALL",
250# endif
251 NULL /* Placeholder */
252 };
253
254# ifdef LC_ALL
255
256 /* On systems with LC_ALL, it is kept in the highest index position. (-2
257 * to account for the final unused placeholder element.) */
258# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
259
260# else
261
262 /* On systems without LC_ALL, we pretend it is there, one beyond the real
263 * top element, hence in the unused placeholder element. */
264# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
265
266# endif
267
268/* Pretending there is an LC_ALL element just above allows us to avoid most
269 * special cases. Most loops through these arrays in the code below are
270 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
271 * on either type of system. But the code must be written to not access the
948523db
KW
272 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
273 * checked for at compile time by using the #define LC_ALL_INDEX which is only
274 * defined if we do have LC_ALL. */
e5f10d49 275
b09aaf40
KW
276STATIC const char *
277S_category_name(const int category)
278{
279 unsigned int i;
280
281#ifdef LC_ALL
282
283 if (category == LC_ALL) {
284 return "LC_ALL";
285 }
286
287#endif
288
289 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
290 if (category == categories[i]) {
291 return category_names[i];
292 }
293 }
294
295 {
296 const char suffix[] = " (unknown)";
297 int temp = category;
298 Size_t length = sizeof(suffix) + 1;
299 char * unknown;
300 dTHX;
301
302 if (temp < 0) {
303 length++;
304 temp = - temp;
305 }
306
307 /* Calculate the number of digits */
308 while (temp >= 10) {
309 temp /= 10;
310 length++;
311 }
312
313 Newx(unknown, length, char);
314 my_snprintf(unknown, length, "%d%s", category, suffix);
315 SAVEFREEPV(unknown);
316 return unknown;
317 }
318}
319
948523db
KW
320/* Now create LC_foo_INDEX #defines for just those categories on this system */
321# ifdef USE_LOCALE_NUMERIC
322# define LC_NUMERIC_INDEX 0
323# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
324# else
325# define _DUMMY_NUMERIC -1
326# endif
327# ifdef USE_LOCALE_CTYPE
328# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
329# define _DUMMY_CTYPE LC_CTYPE_INDEX
330# else
331# define _DUMMY_CTYPE _DUMMY_NUMERIC
332# endif
333# ifdef USE_LOCALE_COLLATE
334# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
335# define _DUMMY_COLLATE LC_COLLATE_INDEX
336# else
8203b3c3 337# define _DUMMY_COLLATE _DUMMY_CTYPE
948523db
KW
338# endif
339# ifdef USE_LOCALE_TIME
340# define LC_TIME_INDEX _DUMMY_COLLATE + 1
341# define _DUMMY_TIME LC_TIME_INDEX
342# else
343# define _DUMMY_TIME _DUMMY_COLLATE
344# endif
345# ifdef USE_LOCALE_MESSAGES
346# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
347# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
348# else
349# define _DUMMY_MESSAGES _DUMMY_TIME
350# endif
351# ifdef USE_LOCALE_MONETARY
352# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
353# define _DUMMY_MONETARY LC_MONETARY_INDEX
354# else
355# define _DUMMY_MONETARY _DUMMY_MESSAGES
356# endif
9821811f
KW
357# ifdef USE_LOCALE_ADDRESS
358# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
359# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
360# else
361# define _DUMMY_ADDRESS _DUMMY_MONETARY
362# endif
363# ifdef USE_LOCALE_IDENTIFICATION
364# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
365# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
366# else
367# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
368# endif
369# ifdef USE_LOCALE_MEASUREMENT
370# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
371# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
372# else
373# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
374# endif
375# ifdef USE_LOCALE_PAPER
376# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
377# define _DUMMY_PAPER LC_PAPER_INDEX
378# else
379# define _DUMMY_PAPER _DUMMY_MEASUREMENT
380# endif
381# ifdef USE_LOCALE_TELEPHONE
382# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
383# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
384# else
385# define _DUMMY_TELEPHONE _DUMMY_PAPER
386# endif
948523db 387# ifdef LC_ALL
9821811f 388# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
389# endif
390#endif /* ifdef USE_LOCALE */
8ef6e574 391
d2b24094 392/* Windows requres a customized base-level setlocale() */
e9bc6d6b
KW
393#ifdef WIN32
394# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
395#else
396# define my_setlocale(cat, locale) setlocale(cat, locale)
397#endif
398
399#ifndef USE_POSIX_2008_LOCALE
400
401/* "do_setlocale_c" is intended to be called when the category is a constant
402 * known at compile time; "do_setlocale_r", not known until run time */
403# define do_setlocale_c(cat, locale) my_setlocale(cat, locale)
404# define do_setlocale_r(cat, locale) my_setlocale(cat, locale)
405
406#else /* Below uses POSIX 2008 */
407
408/* We emulate setlocale with our own function. LC_foo is not valid for the
409 * POSIX 2008 functions. Instead LC_foo_MASK is used, which we use an array
410 * lookup to convert to. At compile time we have defined LC_foo_INDEX as the
411 * proper offset into the array 'category_masks[]'. At runtime, we have to
412 * search through the array (as the actual numbers may not be small contiguous
413 * positive integers which would lend themselves to array lookup). */
414# define do_setlocale_c(cat, locale) \
415 emulate_setlocale(cat, locale, cat ## _INDEX, TRUE)
416# define do_setlocale_r(cat, locale) emulate_setlocale(cat, locale, 0, FALSE)
417
418/* A third array, parallel to the ones above to map from category to its
419 * equivalent mask */
420const int category_masks[] = {
421# ifdef USE_LOCALE_NUMERIC
422 LC_NUMERIC_MASK,
423# endif
424# ifdef USE_LOCALE_CTYPE
425 LC_CTYPE_MASK,
426# endif
427# ifdef USE_LOCALE_COLLATE
428 LC_COLLATE_MASK,
429# endif
430# ifdef USE_LOCALE_TIME
431 LC_TIME_MASK,
432# endif
433# ifdef USE_LOCALE_MESSAGES
434 LC_MESSAGES_MASK,
435# endif
436# ifdef USE_LOCALE_MONETARY
437 LC_MONETARY_MASK,
438# endif
439# ifdef USE_LOCALE_ADDRESS
440 LC_ADDRESS_MASK,
441# endif
442# ifdef USE_LOCALE_IDENTIFICATION
443 LC_IDENTIFICATION_MASK,
444# endif
445# ifdef USE_LOCALE_MEASUREMENT
446 LC_MEASUREMENT_MASK,
447# endif
448# ifdef USE_LOCALE_PAPER
449 LC_PAPER_MASK,
450# endif
451# ifdef USE_LOCALE_TELEPHONE
452 LC_TELEPHONE_MASK,
453# endif
454 /* LC_ALL can't be turned off by a Configure
455 * option, and in Posix 2008, should always be
456 * here, so compile it in unconditionally.
457 * This could catch some glitches at compile
458 * time */
459 LC_ALL_MASK
460 };
461
462STATIC const char *
463S_emulate_setlocale(const int category,
464 const char * locale,
465 unsigned int index,
466 const bool is_index_valid
467 )
468{
469 /* This function effectively performs a setlocale() on just the current
470 * thread; thus it is thread-safe. It does this by using the POSIX 2008
471 * locale functions to emulate the behavior of setlocale(). Similar to
472 * regular setlocale(), the return from this function points to memory that
473 * can be overwritten by other system calls, so needs to be copied
474 * immediately if you need to retain it. The difference here is that
475 * system calls besides another setlocale() can overwrite it.
476 *
477 * By doing this, most locale-sensitive functions become thread-safe. The
478 * exceptions are mostly those that return a pointer to static memory.
479 *
480 * This function takes the same parameters, 'category' and 'locale', that
481 * the regular setlocale() function does, but it also takes two additional
482 * ones. This is because the 2008 functions don't use a category; instead
483 * they use a corresponding mask. Because this function operates in both
484 * worlds, it may need one or the other or both. This function can
485 * calculate the mask from the input category, but to avoid this
486 * calculation, if the caller knows at compile time what the mask is, it
487 * can pass it, setting 'is_index_valid' to TRUE; otherwise the mask
488 * parameter is ignored.
489 *
490 * POSIX 2008, for some sick reason, chose not to provide a method to find
491 * the category name of a locale. Some vendors have created a
492 * querylocale() function to do just that. This function is a lot simpler
493 * to implement on systems that have this. Otherwise, we have to keep
494 * track of what the locale has been set to, so that we can return its
495 * name to emulate setlocale(). It's also possible for C code in some
496 * library to change the locale without us knowing it, though as of
497 * September 2017, there are no occurrences in CPAN of uselocale(). Some
498 * libraries do use setlocale(), but that changes the global locale, and
499 * threads using per-thread locales will just ignore those changes.
500 * Another problem is that without querylocale(), we have to guess at what
501 * was meant by setting a locale of "". We handle this by not actually
502 * ever setting to "" (unless querylocale exists), but to emulate what we
503 * think should happen for "".
504 */
505
506 int mask;
507 locale_t old_obj;
508 locale_t new_obj;
509 dTHX;
510
511# ifdef DEBUGGING
512
513 if (DEBUG_Lv_TEST || debug_initialization) {
514 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);
515 }
516
517# endif
518
519 /* If the input mask might be incorrect, calculate the correct one */
520 if (! is_index_valid) {
521 unsigned int i;
522
523# ifdef DEBUGGING
524
525 if (DEBUG_Lv_TEST || debug_initialization) {
526 PerlIO_printf(Perl_debug_log, "%s:%d: finding index of category %d (%s)\n", __FILE__, __LINE__, category, category_name(category));
527 }
528
529# endif
530
531 for (i = 0; i <= LC_ALL_INDEX; i++) {
532 if (category == categories[i]) {
533 index = i;
534 goto found_index;
535 }
536 }
537
538 /* Here, we don't know about this category, so can't handle it.
539 * Fallback to the early POSIX usages */
540 Perl_warner(aTHX_ packWARN(WARN_LOCALE),
541 "Unknown locale category %d; can't set it to %s\n",
542 category, locale);
543 return NULL;
544
545 found_index: ;
546
547# ifdef DEBUGGING
548
549 if (DEBUG_Lv_TEST || debug_initialization) {
550 PerlIO_printf(Perl_debug_log, "%s:%d: index is %d for %s\n", __FILE__, __LINE__, index, category_name(category));
551 }
552
553# endif
554
555 }
556
557 mask = category_masks[index];
558
559# ifdef DEBUGGING
560
561 if (DEBUG_Lv_TEST || debug_initialization) {
562 PerlIO_printf(Perl_debug_log, "%s:%d: category name is %s; mask is 0x%x\n", __FILE__, __LINE__, category_names[index], mask);
563 }
564
565# endif
566
567 /* If just querying what the existing locale is ... */
568 if (locale == NULL) {
569 locale_t cur_obj = uselocale((locale_t) 0);
570
571# ifdef DEBUGGING
572
573 if (DEBUG_Lv_TEST || debug_initialization) {
574 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale querying %p\n", __FILE__, __LINE__, cur_obj);
575 }
576
577# endif
578
579 if (cur_obj == LC_GLOBAL_LOCALE) {
580 return my_setlocale(category, NULL);
581 }
582
583# ifdef HAS_QUERYLOCALE
584
585 return (char *) querylocale(mask, cur_obj);
586
d2b24094 587# else
ecdda939
KW
588
589 /* If this assert fails, adjust the size of curlocales in intrpvar.h */
590 STATIC_ASSERT_STMT(C_ARRAY_LENGTH(PL_curlocales) > LC_ALL_INDEX);
591
8e13243a
KW
592# if defined(_NL_LOCALE_NAME) \
593 && defined(DEBUGGING) \
594 && ! defined(SETLOCALE_ACCEPTS_ANY_LOCALE_NAME)
595 /* On systems that accept any locale name, the real underlying locale
596 * is often returned by this internal function, so we can't use it */
e9bc6d6b
KW
597 {
598 /* Internal glibc for querylocale(), but doesn't handle
599 * empty-string ("") locale properly; who knows what other
3a732259 600 * glitches. Check for it now, under debug. */
e9bc6d6b
KW
601
602 char * temp_name = nl_langinfo_l(_NL_LOCALE_NAME(category),
603 uselocale((locale_t) 0));
604 /*
605 PerlIO_printf(Perl_debug_log, "%s:%d: temp_name=%s\n", __FILE__, __LINE__, temp_name ? temp_name : "NULL");
606 PerlIO_printf(Perl_debug_log, "%s:%d: index=%d\n", __FILE__, __LINE__, index);
607 PerlIO_printf(Perl_debug_log, "%s:%d: PL_curlocales[index]=%s\n", __FILE__, __LINE__, PL_curlocales[index]);
608 */
609 if (temp_name && PL_curlocales[index] && strNE(temp_name, "")) {
610 if ( strNE(PL_curlocales[index], temp_name)
611 && ! ( isNAME_C_OR_POSIX(temp_name)
612 && isNAME_C_OR_POSIX(PL_curlocales[index]))) {
613
614# ifdef USE_C_BACKTRACE
615
616 dump_c_backtrace(Perl_debug_log, 20, 1);
617
618# endif
619
620 Perl_croak(aTHX_ "panic: Mismatch between what Perl thinks %s is"
621 " (%s) and what internal glibc thinks"
622 " (%s)\n", category_names[index],
623 PL_curlocales[index], temp_name);
624 }
625
626 return temp_name;
627 }
628 }
629
630# endif
631
632 /* Without querylocale(), we have to use our record-keeping we've
633 * done. */
634
635 if (category != LC_ALL) {
636
637# ifdef DEBUGGING
638
639 if (DEBUG_Lv_TEST || debug_initialization) {
640 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[index]);
641 }
642
643# endif
644
645 return PL_curlocales[index];
646 }
647 else { /* For LC_ALL */
648 unsigned int i;
649 Size_t names_len = 0;
650 char * all_string;
7c93a471 651 bool are_all_categories_the_same_locale = TRUE;
e9bc6d6b
KW
652
653 /* If we have a valid LC_ALL value, just return it */
654 if (PL_curlocales[LC_ALL_INDEX]) {
655
656# ifdef DEBUGGING
657
658 if (DEBUG_Lv_TEST || debug_initialization) {
659 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, PL_curlocales[LC_ALL_INDEX]);
660 }
661
662# endif
663
664 return PL_curlocales[LC_ALL_INDEX];
665 }
666
667 /* Otherwise, we need to construct a string of name=value pairs.
668 * We use the glibc syntax, like
669 * LC_NUMERIC=C;LC_TIME=en_US.UTF-8;...
7c93a471
KW
670 * First calculate the needed size. Along the way, check if all
671 * the locale names are the same */
e9bc6d6b
KW
672 for (i = 0; i < LC_ALL_INDEX; i++) {
673
674# ifdef DEBUGGING
675
676 if (DEBUG_Lv_TEST || debug_initialization) {
677 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]);
678 }
679
680# endif
681
682 names_len += strlen(category_names[i])
683 + 1 /* '=' */
684 + strlen(PL_curlocales[i])
685 + 1; /* ';' */
7c93a471
KW
686
687 if (i > 0 && strNE(PL_curlocales[i], PL_curlocales[i-1])) {
688 are_all_categories_the_same_locale = FALSE;
689 }
690 }
691
692 /* If they are the same, we don't actually have to construct the
693 * string; we just make the entry in LC_ALL_INDEX valid, and be
694 * that single name */
695 if (are_all_categories_the_same_locale) {
696 PL_curlocales[LC_ALL_INDEX] = savepv(PL_curlocales[0]);
697 return PL_curlocales[LC_ALL_INDEX];
e9bc6d6b 698 }
7c93a471 699
e9bc6d6b
KW
700 names_len++; /* Trailing '\0' */
701 SAVEFREEPV(Newx(all_string, names_len, char));
702 *all_string = '\0';
703
704 /* Then fill in the string */
705 for (i = 0; i < LC_ALL_INDEX; i++) {
706
707# ifdef DEBUGGING
708
709 if (DEBUG_Lv_TEST || debug_initialization) {
710 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]);
711 }
712
713# endif
714
715 my_strlcat(all_string, category_names[i], names_len);
716 my_strlcat(all_string, "=", names_len);
717 my_strlcat(all_string, PL_curlocales[i], names_len);
718 my_strlcat(all_string, ";", names_len);
719 }
720
721# ifdef DEBUGGING
722
723 if (DEBUG_L_TEST || debug_initialization) {
724 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale returning %s\n", __FILE__, __LINE__, all_string);
725 }
726
727 #endif
728
729 return all_string;
730 }
731
732# ifdef EINVAL
733
734 SETERRNO(EINVAL, LIB_INVARG);
735
736# endif
737
738 return NULL;
739
d2b24094
KW
740# endif
741
e9bc6d6b
KW
742 }
743
f8115d60 744 /* Here, we are switching locales. */
e9bc6d6b
KW
745
746# ifndef HAS_QUERYLOCALE
747
748 if (strEQ(locale, "")) {
749
750 /* For non-querylocale() systems, we do the setting of "" ourselves to
751 * be sure that we really know what's going on. We follow the Linux
752 * documented behavior (but if that differs from the actual behavior,
753 * this won't work exactly as the OS implements). We go out and
754 * examine the environment based on our understanding of how the system
755 * works, and use that to figure things out */
756
757 const char * const lc_all = PerlEnv_getenv("LC_ALL");
758
759 /* Use any "LC_ALL" environment variable, as it overrides everything
760 * else. */
761 if (lc_all && strNE(lc_all, "")) {
762 locale = lc_all;
763 }
764 else {
765
766 /* Otherwise, we need to dig deeper. Unless overridden, the
767 * default is the LANG environment variable; if it doesn't exist,
768 * then "C" */
769
770 const char * default_name;
771
772 /* To minimize other threads messing with the environment, we copy
773 * the variable, making it a temporary. But this doesn't work upon
774 * program initialization before any scopes are created, and at
775 * this time, there's nothing else going on that would interfere.
776 * So skip the copy in that case */
777 if (PL_scopestack_ix == 0) {
778 default_name = PerlEnv_getenv("LANG");
779 }
780 else {
781 default_name = savepv(PerlEnv_getenv("LANG"));
782 }
783
784 if (! default_name || strEQ(default_name, "")) {
929a7f8c 785 default_name = "C";
e9bc6d6b
KW
786 }
787 else if (PL_scopestack_ix != 0) {
788 SAVEFREEPV(default_name);
789 }
790
791 if (category != LC_ALL) {
792 const char * const name = PerlEnv_getenv(category_names[index]);
793
794 /* Here we are setting a single category. Assume will have the
795 * default name */
796 locale = default_name;
797
798 /* But then look for an overriding environment variable */
799 if (name && strNE(name, "")) {
800 locale = name;
801 }
802 }
803 else {
804 bool did_override = FALSE;
805 unsigned int i;
806
807 /* Here, we are getting LC_ALL. Any categories that don't have
808 * a corresponding environment variable set should be set to
809 * LANG, or to "C" if there is no LANG. If no individual
810 * categories differ from this, we can just set LC_ALL. This
811 * is buggy on systems that have extra categories that we don't
812 * know about. If there is an environment variable that sets
813 * that category, we won't know to look for it, and so our use
814 * of LANG or "C" improperly overrides it. On the other hand,
815 * if we don't do what is done here, and there is no
816 * environment variable, the category's locale should be set to
817 * LANG or "C". So there is no good solution. khw thinks the
818 * best is to look at systems to see what categories they have,
819 * and include them, and then to assume that we know the
820 * complete set */
821
822 for (i = 0; i < LC_ALL_INDEX; i++) {
823 const char * const env_override
824 = savepv(PerlEnv_getenv(category_names[i]));
825 const char * this_locale = ( env_override
826 && strNE(env_override, ""))
827 ? env_override
828 : default_name;
6435f98d
KW
829 if (! emulate_setlocale(categories[i], this_locale, i, TRUE))
830 {
831 Safefree(env_override);
832 return NULL;
833 }
e9bc6d6b
KW
834
835 if (strNE(this_locale, default_name)) {
836 did_override = TRUE;
837 }
838
839 Safefree(env_override);
840 }
841
842 /* If all the categories are the same, we can set LC_ALL to
843 * that */
844 if (! did_override) {
845 locale = default_name;
846 }
847 else {
848
849 /* Here, LC_ALL is no longer valid, as some individual
850 * categories don't match it. We call ourselves
851 * recursively, as that will execute the code that
852 * generates the proper locale string for this situation.
853 * We don't do the remainder of this function, as that is
854 * to update our records, and we've just done that for the
855 * individual categories in the loop above, and doing so
856 * would cause LC_ALL to be done as well */
857 return emulate_setlocale(LC_ALL, NULL, LC_ALL_INDEX, TRUE);
858 }
859 }
860 }
861 }
862 else if (strchr(locale, ';')) {
863
864 /* LC_ALL may actually incude a conglomeration of various categories.
865 * Without querylocale, this code uses the glibc (as of this writing)
866 * syntax for representing that, but that is not a stable API, and
867 * other platforms do it differently, so we have to handle all cases
868 * ourselves */
869
f0023d47 870 unsigned int i;
e9bc6d6b
KW
871 const char * s = locale;
872 const char * e = locale + strlen(locale);
873 const char * p = s;
874 const char * category_end;
875 const char * name_start;
876 const char * name_end;
877
f0023d47
KW
878 /* If the string that gives what to set doesn't include all categories,
879 * the omitted ones get set to "C". To get this behavior, first set
880 * all the individual categories to "C", and override the furnished
881 * ones below */
882 for (i = 0; i < LC_ALL_INDEX; i++) {
883 if (! emulate_setlocale(categories[i], "C", i, TRUE)) {
884 return NULL;
885 }
886 }
887
e9bc6d6b 888 while (s < e) {
e9bc6d6b
KW
889
890 /* Parse through the category */
891 while (isWORDCHAR(*p)) {
892 p++;
893 }
894 category_end = p;
895
896 if (*p++ != '=') {
897 Perl_croak(aTHX_
898 "panic: %s: %d: Unexpected character in locale name '%02X",
899 __FILE__, __LINE__, *(p-1));
900 }
901
902 /* Parse through the locale name */
903 name_start = p;
bee74f4b
KW
904 while (p < e && *p != ';') {
905 if (! isGRAPH(*p)) {
906 Perl_croak(aTHX_
907 "panic: %s: %d: Unexpected character in locale name '%02X",
908 __FILE__, __LINE__, *(p-1));
909 }
e9bc6d6b
KW
910 p++;
911 }
912 name_end = p;
913
bee74f4b
KW
914 /* Space past the semi-colon */
915 if (p < e) {
916 p++;
e9bc6d6b
KW
917 }
918
919 /* Find the index of the category name in our lists */
920 for (i = 0; i < LC_ALL_INDEX; i++) {
3c76a412 921 char * individ_locale;
e9bc6d6b
KW
922
923 /* Keep going if this isn't the index. The strnNE() avoids a
924 * Perl_form(), but would fail if ever a category name could be
925 * a substring of another one, like if there were a
926 * "LC_TIME_DATE" */
927 if strnNE(s, category_names[i], category_end - s) {
928 continue;
929 }
930
931 /* If this index is for the single category we're changing, we
932 * have found the locale to set it to. */
933 if (category == categories[i]) {
934 locale = Perl_form(aTHX_ "%.*s",
935 (int) (name_end - name_start),
936 name_start);
937 goto ready_to_set;
938 }
939
3c76a412 940 assert(category == LC_ALL);
bee74f4b
KW
941 individ_locale = Perl_form(aTHX_ "%.*s",
942 (int) (name_end - name_start), name_start);
6435f98d
KW
943 if (! emulate_setlocale(categories[i], individ_locale, i, TRUE))
944 {
945 return NULL;
946 }
e9bc6d6b
KW
947 }
948
949 s = p;
950 }
951
952 /* Here we have set all the individual categories by recursive calls.
953 * These collectively should have fixed up LC_ALL, so can just query
954 * what that now is */
955 assert(category == LC_ALL);
956
957 return do_setlocale_c(LC_ALL, NULL);
958 }
959
960 ready_to_set: ;
961
f8115d60
KW
962 /* Here at the end of having to deal with the absence of querylocale().
963 * Some cases have already been fully handled by recursive calls to this
964 * function. But at this point, we haven't dealt with those, but are now
965 * prepared to, knowing what the locale name to set this category to is.
966 * This would have come for free if this system had had querylocale() */
967
e9bc6d6b
KW
968# endif /* end of ! querylocale */
969
f8115d60
KW
970 assert(PL_C_locale_obj);
971
972 /* Switching locales generally entails freeing the current one's space (at
973 * the C library's discretion). We need to stop using that locale before
974 * the switch. So switch to a known locale object that we don't otherwise
975 * mess with. This returns the locale object in effect at the time of the
976 * switch. */
977 old_obj = uselocale(PL_C_locale_obj);
978
979# ifdef DEBUGGING
980
981 if (DEBUG_Lv_TEST || debug_initialization) {
982 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale was using %p\n", __FILE__, __LINE__, old_obj);
983 }
984
985# endif
986
987 if (! old_obj) {
988
989# ifdef DEBUGGING
990
991 if (DEBUG_L_TEST || debug_initialization) {
992 dSAVE_ERRNO;
993 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to C failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
994 RESTORE_ERRNO;
995 }
996
997# endif
998
999 return NULL;
1000 }
1001
1002# ifdef DEBUGGING
1003
1004 if (DEBUG_Lv_TEST || debug_initialization) {
1005 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, PL_C_locale_obj);
1006 }
1007
1008# endif
1009
1010 /* If we weren't in a thread safe locale, set so that newlocale() below
1011 which uses 'old_obj', uses an empty one. Same for our reserved C object.
1012 The latter is defensive coding, so that, even if there is some bug, we
1013 will never end up trying to modify either of these, as if passed to
1014 newlocale(), they can be. */
1015 if (old_obj == LC_GLOBAL_LOCALE || old_obj == PL_C_locale_obj) {
1016 old_obj = (locale_t) 0;
1017 }
1018
e9bc6d6b
KW
1019 /* Ready to create a new locale by modification of the exising one */
1020 new_obj = newlocale(mask, locale, old_obj);
1021
1022 if (! new_obj) {
1023 dSAVE_ERRNO;
1024
1025# ifdef DEBUGGING
1026
1027 if (DEBUG_L_TEST || debug_initialization) {
1028 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale creating new object failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1029 }
1030
1031# endif
1032
1033 if (! uselocale(old_obj)) {
e9bc6d6b
KW
1034
1035# ifdef DEBUGGING
1036
1037 if (DEBUG_L_TEST || debug_initialization) {
1038 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1039 }
1040
1041# endif
1042
1043 }
1044 RESTORE_ERRNO;
1045 return NULL;
1046 }
1047
1048# ifdef DEBUGGING
1049
1050 if (DEBUG_Lv_TEST || debug_initialization) {
19ee3daf
KW
1051 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale created %p", __FILE__, __LINE__, new_obj);
1052 if (old_obj) {
1053 PerlIO_printf(Perl_debug_log, "; should have freed %p", old_obj);
1054 }
1055 PerlIO_printf(Perl_debug_log, "\n");
e9bc6d6b
KW
1056 }
1057
1058# endif
1059
1060 /* And switch into it */
1061 if (! uselocale(new_obj)) {
1062 dSAVE_ERRNO;
1063
1064# ifdef DEBUGGING
1065
1066 if (DEBUG_L_TEST || debug_initialization) {
1067 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale switching to new object failed\n", __FILE__, __LINE__);
1068 }
1069
1070# endif
1071
1072 if (! uselocale(old_obj)) {
1073
1074# ifdef DEBUGGING
1075
1076 if (DEBUG_L_TEST || debug_initialization) {
1077 PerlIO_printf(Perl_debug_log, "%s:%d: switching back failed: %d\n", __FILE__, __LINE__, GET_ERRNO);
1078 }
1079
1080# endif
1081
1082 }
1083 freelocale(new_obj);
1084 RESTORE_ERRNO;
1085 return NULL;
1086 }
1087
1088# ifdef DEBUGGING
1089
1090 if (DEBUG_Lv_TEST || debug_initialization) {
1091 PerlIO_printf(Perl_debug_log, "%s:%d: emulate_setlocale now using %p\n", __FILE__, __LINE__, new_obj);
1092 }
1093
1094# endif
1095
1096 /* We are done, except for updating our records (if the system doesn't keep
1097 * them) and in the case of locale "", we don't actually know what the
1098 * locale that got switched to is, as it came from the environment. So
1099 * have to find it */
1100
1101# ifdef HAS_QUERYLOCALE
1102
1103 if (strEQ(locale, "")) {
1104 locale = querylocale(mask, new_obj);
1105 }
1106
1107# else
1108
1109 /* Here, 'locale' is the return value */
1110
1111 /* Without querylocale(), we have to update our records */
1112
1113 if (category == LC_ALL) {
1114 unsigned int i;
1115
1116 /* For LC_ALL, we change all individual categories to correspond */
1117 /* PL_curlocales is a parallel array, so has same
1118 * length as 'categories' */
1119 for (i = 0; i <= LC_ALL_INDEX; i++) {
1120 Safefree(PL_curlocales[i]);
1121 PL_curlocales[i] = savepv(locale);
1122 }
1123 }
1124 else {
1125
1126 /* For a single category, if it's not the same as the one in LC_ALL, we
1127 * nullify LC_ALL */
1128
1129 if (PL_curlocales[LC_ALL_INDEX] && strNE(PL_curlocales[LC_ALL_INDEX], locale)) {
1130 Safefree(PL_curlocales[LC_ALL_INDEX]);
1131 PL_curlocales[LC_ALL_INDEX] = NULL;
1132 }
1133
1134 /* Then update the category's record */
1135 Safefree(PL_curlocales[index]);
1136 PL_curlocales[index] = savepv(locale);
1137 }
1138
1139# endif
1140
1141 return locale;
1142}
1143
1144#endif /* USE_POSIX_2008_LOCALE */
1145
1146#if 0 /* Code that was to emulate thread-safe locales on platforms that
1147 didn't natively support them */
1148
1149/* The way this would work is that we would keep a per-thread list of the
1150 * correct locale for that thread. Any operation that was locale-sensitive
1151 * would have to be changed so that it would look like this:
1152 *
1153 * LOCALE_LOCK;
1154 * setlocale to the correct locale for this operation
1155 * do operation
1156 * LOCALE_UNLOCK
1157 *
1158 * This leaves the global locale in the most recently used operation's, but it
1159 * was locked long enough to get the result. If that result is static, it
1160 * needs to be copied before the unlock.
1161 *
1162 * Macros could be written like SETUP_LOCALE_DEPENDENT_OP(category) that did
1163 * the setup, but are no-ops when not needed, and similarly,
1164 * END_LOCALE_DEPENDENT_OP for the tear-down
1165 *
1166 * But every call to a locale-sensitive function would have to be changed, and
1167 * if a module didn't cooperate by using the mutex, things would break.
1168 *
1169 * This code was abandoned before being completed or tested, and is left as-is
1170*/
1171
1172# define do_setlocale_c(cat, locale) locking_setlocale(cat, locale, cat ## _INDEX, TRUE)
1173# define do_setlocale_r(cat, locale) locking_setlocale(cat, locale, 0, FALSE)
1174
1175STATIC char *
1176S_locking_setlocale(pTHX_
1177 const int category,
1178 const char * locale,
1179 int index,
1180 const bool is_index_valid
1181 )
1182{
1183 /* This function kind of performs a setlocale() on just the current thread;
1184 * thus it is kind of thread-safe. It does this by keeping a thread-level
1185 * array of the current locales for each category. Every time a locale is
1186 * switched to, it does the switch globally, but updates the thread's
1187 * array. A query as to what the current locale is just returns the
1188 * appropriate element from the array, and doesn't actually call the system
1189 * setlocale(). The saving into the array is done in an uninterruptible
1190 * section of code, so is unaffected by whatever any other threads might be
1191 * doing.
1192 *
1193 * All locale-sensitive operations must work by first starting a critical
1194 * section, then switching to the thread's locale as kept by this function,
1195 * and then doing the operation, then ending the critical section. Thus,
1196 * each gets done in the appropriate locale. simulating thread-safety.
1197 *
1198 * This function takes the same parameters, 'category' and 'locale', that
1199 * the regular setlocale() function does, but it also takes two additional
1200 * ones. This is because as described earlier. If we know on input the
1201 * index corresponding to the category into the array where we store the
1202 * current locales, we don't have to calculate it. If the caller knows at
1203 * compile time what the index is, it it can pass it, setting
1204 * 'is_index_valid' to TRUE; otherwise the index parameter is ignored.
1205 *
1206 */
1207
1208 /* If the input index might be incorrect, calculate the correct one */
1209 if (! is_index_valid) {
1210 unsigned int i;
1211
1212 if (DEBUG_Lv_TEST || debug_initialization) {
1213 PerlIO_printf(Perl_debug_log, "%s:%d: converting category %d to index\n", __FILE__, __LINE__, category);
1214 }
1215
1216 for (i = 0; i <= LC_ALL_INDEX; i++) {
1217 if (category == categories[i]) {
1218 index = i;
1219 goto found_index;
1220 }
1221 }
1222
1223 /* Here, we don't know about this category, so can't handle it.
1224 * XXX best we can do is to unsafely set this
1225 * XXX warning */
1226
1227 return my_setlocale(category, locale);
1228
1229 found_index: ;
1230
1231 if (DEBUG_Lv_TEST || debug_initialization) {
1232 PerlIO_printf(Perl_debug_log, "%s:%d: index is 0x%x\n", __FILE__, __LINE__, index);
1233 }
1234 }
1235
1236 /* For a query, just return what's in our records */
1237 if (new_locale == NULL) {
1238 return curlocales[index];
1239 }
1240
1241
1242 /* Otherwise, we need to do the switch, and save the result, all in a
1243 * critical section */
1244
1245 Safefree(curlocales[[index]]);
1246
1247 /* It might be that this is called from an already-locked section of code.
1248 * We would have to detect and skip the LOCK/UNLOCK if so */
1249 LOCALE_LOCK;
1250
1251 curlocales[index] = savepv(my_setlocale(category, new_locale));
1252
1253 if (strEQ(new_locale, "")) {
1254
1255#ifdef LC_ALL
1256
1257 /* The locale values come from the environment, and may not all be the
1258 * same, so for LC_ALL, we have to update all the others, while the
1259 * mutex is still locked */
1260
1261 if (category == LC_ALL) {
1262 unsigned int i;
1263 for (i = 0; i < LC_ALL_INDEX) {
1264 curlocales[i] = my_setlocale(categories[i], NULL);
1265 }
1266 }
1267 }
1268
1269#endif
1270
1271 LOCALE_UNLOCK;
1272
1273 return curlocales[index];
1274}
1275
1276#endif
d36adde0 1277#ifdef USE_LOCALE
837ce802 1278
a4f00dcc 1279STATIC void
86799d2d 1280S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 1281{
86799d2d
KW
1282 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
1283 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 1284
86799d2d
KW
1285#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
1286 || defined(HAS_NL_LANGINFO))
98994639 1287
87f8e8e7 1288 const char * radix = (use_locale)
4e6826bf 1289 ? my_nl_langinfo(RADIXCHAR, FALSE)
87f8e8e7
KW
1290 /* FALSE => already in dest locale */
1291 : ".";
2213a3be 1292
87f8e8e7 1293 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 1294
87f8e8e7
KW
1295 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
1296 * a UTF-8 locale, then mark the radix as being in UTF-8 */
1297 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 1298 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
1299 && _is_cur_LC_category_utf8(LC_NUMERIC))
1300 {
1301 SvUTF8_on(PL_numeric_radix_sv);
1302 }
86799d2d
KW
1303
1304# ifdef DEBUGGING
7d4bcc4a 1305
2fcc0ca9
KW
1306 if (DEBUG_L_TEST || debug_initialization) {
1307 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
1308 SvPVX(PL_numeric_radix_sv),
1309 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 1310 }
69014004 1311
86799d2d 1312# endif
d36adde0
KW
1313#else
1314
1315 PERL_UNUSED_ARG(use_locale);
1316
86799d2d 1317#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 1318
98994639
HS
1319}
1320
398eeea9
KW
1321STATIC void
1322S_new_numeric(pTHX_ const char *newnum)
98994639 1323{
7d4bcc4a
KW
1324
1325#ifndef USE_LOCALE_NUMERIC
1326
1327 PERL_UNUSED_ARG(newnum);
1328
1329#else
0d071d52 1330
291a84fb 1331 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
1332 * core Perl this and that 'newnum' is the name of the new locale.
1333 * It installs this locale as the current underlying default.
1334 *
1335 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
1336 * set_numeric_underlying() and set_numeric_standard() functions, which
1337 * should probably not be called directly, but only via macros like
0d071d52
KW
1338 * SET_NUMERIC_STANDARD() in perl.h.
1339 *
1340 * The toggling is necessary mainly so that a non-dot radix decimal point
1341 * character can be output, while allowing internal calculations to use a
1342 * dot.
1343 *
1344 * This sets several interpreter-level variables:
bb304765 1345 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 1346 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
1347 * that the current locale is the program's underlying
1348 * locale
1349 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
1350 * that the current locale is the C locale or
1351 * indistinguishable from the C locale. If non-zero, it
1352 * is in C; if > 1, it means it may not be toggled away
7738054c 1353 * from C.
4c68b815
KW
1354 * PL_numeric_underlying_is_standard A bool kept by this function
1355 * indicating that the underlying locale and the standard
1356 * C locale are indistinguishable for the purposes of
1357 * LC_NUMERIC. This happens when both of the above two
1358 * variables are true at the same time. (Toggling is a
1359 * no-op under these circumstances.) This variable is
1360 * used to avoid having to recalculate.
398eeea9 1361 */
0d071d52 1362
b03f34cf 1363 char *save_newnum;
98994639
HS
1364
1365 if (! newnum) {
43c5f42d
NC
1366 Safefree(PL_numeric_name);
1367 PL_numeric_name = NULL;
98994639 1368 PL_numeric_standard = TRUE;
892e6465 1369 PL_numeric_underlying = TRUE;
4c68b815 1370 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
1371 return;
1372 }
1373
b03f34cf 1374 save_newnum = stdize_locale(savepv(newnum));
892e6465 1375 PL_numeric_underlying = TRUE;
4c68b815
KW
1376 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
1377
69c5e0db
KW
1378#ifndef TS_W32_BROKEN_LOCALECONV
1379
4c68b815 1380 /* If its name isn't C nor POSIX, it could still be indistinguishable from
69c5e0db
KW
1381 * them. But on broken Windows systems calling my_nl_langinfo() for
1382 * THOUSEP can currently (but rarely) cause a race, so avoid doing that,
1383 * and just always change the locale if not C nor POSIX on those systems */
4c68b815 1384 if (! PL_numeric_standard) {
4e6826bf 1385 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(RADIXCHAR,
4c68b815 1386 FALSE /* Don't toggle locale */ ))
4e6826bf 1387 && strEQ("", my_nl_langinfo(THOUSEP, FALSE)));
4c68b815 1388 }
abe1abcf 1389
69c5e0db
KW
1390#endif
1391
4c68b815 1392 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 1393 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 1394 Safefree(PL_numeric_name);
b03f34cf 1395 PL_numeric_name = save_newnum;
b03f34cf 1396 }
abe1abcf
KW
1397 else {
1398 Safefree(save_newnum);
1399 }
4c28b29c 1400
4c68b815
KW
1401 PL_numeric_underlying_is_standard = PL_numeric_standard;
1402
7e5377f7 1403# ifdef HAS_POSIX_2008_LOCALE
e1aa2579
KW
1404
1405 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
1406 PL_numeric_name,
1407 PL_underlying_numeric_obj);
1408
1409#endif
1410
4c68b815
KW
1411 if (DEBUG_L_TEST || debug_initialization) {
1412 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
1413 }
1414
4c28b29c
KW
1415 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
1416 * have to worry about the radix being a non-dot. (Core operations that
1417 * need the underlying locale change to it temporarily). */
84f1d29f
KW
1418 if (PL_numeric_standard) {
1419 set_numeric_radix(0);
1420 }
1421 else {
1422 set_numeric_standard();
1423 }
4c28b29c 1424
98994639 1425#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1426
98994639
HS
1427}
1428
1429void
1430Perl_set_numeric_standard(pTHX)
1431{
7d4bcc4a 1432
98994639 1433#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1434
28c1bf33
KW
1435 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
1436 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
1437 * macro avoids calling this routine if toggling isn't necessary according
1438 * to our records (which could be wrong if some XS code has changed the
1439 * locale behind our back) */
0d071d52 1440
7d4bcc4a
KW
1441# ifdef DEBUGGING
1442
2fcc0ca9
KW
1443 if (DEBUG_L_TEST || debug_initialization) {
1444 PerlIO_printf(Perl_debug_log,
7bb8f702 1445 "Setting LC_NUMERIC locale to standard C\n");
2fcc0ca9 1446 }
98994639 1447
7d4bcc4a 1448# endif
7bb8f702
KW
1449
1450 do_setlocale_c(LC_NUMERIC, "C");
1451 PL_numeric_standard = TRUE;
1452 PL_numeric_underlying = PL_numeric_underlying_is_standard;
1453 set_numeric_radix(0);
1454
98994639 1455#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1456
98994639
HS
1457}
1458
1459void
5792c642 1460Perl_set_numeric_underlying(pTHX)
98994639 1461{
7d4bcc4a 1462
98994639 1463#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 1464
28c1bf33 1465 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
1466 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
1467 * instead of calling this directly. The macro avoids calling this routine
1468 * if toggling isn't necessary according to our records (which could be
1469 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 1470
7d4bcc4a
KW
1471# ifdef DEBUGGING
1472
2fcc0ca9
KW
1473 if (DEBUG_L_TEST || debug_initialization) {
1474 PerlIO_printf(Perl_debug_log,
7bb8f702 1475 "Setting LC_NUMERIC locale to %s\n",
2fcc0ca9
KW
1476 PL_numeric_name);
1477 }
98994639 1478
7d4bcc4a 1479# endif
7bb8f702
KW
1480
1481 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1482 PL_numeric_standard = PL_numeric_underlying_is_standard;
1483 PL_numeric_underlying = TRUE;
1484 set_numeric_radix(! PL_numeric_standard);
1485
98994639 1486#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 1487
98994639
HS
1488}
1489
1490/*
1491 * Set up for a new ctype locale.
1492 */
a4f00dcc
KW
1493STATIC void
1494S_new_ctype(pTHX_ const char *newctype)
98994639 1495{
7d4bcc4a
KW
1496
1497#ifndef USE_LOCALE_CTYPE
1498
7d4bcc4a
KW
1499 PERL_UNUSED_ARG(newctype);
1500 PERL_UNUSED_CONTEXT;
1501
1502#else
0d071d52 1503
291a84fb 1504 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
1505 * core Perl this and that 'newctype' is the name of the new locale.
1506 *
1507 * This function sets up the folding arrays for all 256 bytes, assuming
1508 * that tofold() is tolc() since fold case is not a concept in POSIX,
1509 *
1510 * Any code changing the locale (outside this file) should use
9aac5db8
KW
1511 * Perl_setlocale or POSIX::setlocale, which call this function. Therefore
1512 * this function should be called directly only from this file and from
0d071d52
KW
1513 * POSIX::setlocale() */
1514
27da23d5 1515 dVAR;
013f4e03 1516 unsigned int i;
98994639 1517
8b7358b9
KW
1518 /* Don't check for problems if we are suppressing the warnings */
1519 bool check_for_problems = ckWARN_d(WARN_LOCALE) || UNLIKELY(DEBUG_L_TEST);
30d8090d 1520 bool maybe_utf8_turkic = FALSE;
8b7358b9 1521
7918f24d
NC
1522 PERL_ARGS_ASSERT_NEW_CTYPE;
1523
215c5139
KW
1524 /* We will replace any bad locale warning with 1) nothing if the new one is
1525 * ok; or 2) a new warning for the bad new locale */
1526 if (PL_warn_locale) {
1527 SvREFCNT_dec_NN(PL_warn_locale);
1528 PL_warn_locale = NULL;
1529 }
1530
c1284011 1531 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
1532
1533 /* A UTF-8 locale gets standard rules. But note that code still has to
1534 * handle this specially because of the three problematic code points */
1535 if (PL_in_utf8_CTYPE_locale) {
1536 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
30d8090d
KW
1537
1538 /* UTF-8 locales can have special handling for 'I' and 'i' if they are
1539 * Turkic. Make sure these two are the only anomalies. (We don't use
1540 * towupper and towlower because they aren't in C89.) */
5b64f24c
KW
1541
1542#if defined(HAS_TOWUPPER) && defined (HAS_TOWLOWER)
1543
1544 if (towupper('i') == 0x130 && towlower('I') == 0x131) {
1545
1546#else
1547
30d8090d 1548 if (toupper('i') == 'i' && tolower('I') == 'I') {
5b64f24c
KW
1549
1550#endif
30d8090d
KW
1551 check_for_problems = TRUE;
1552 maybe_utf8_turkic = TRUE;
1553 }
31f05a37 1554 }
8b7358b9
KW
1555
1556 /* We don't populate the other lists if a UTF-8 locale, but do check that
1557 * everything works as expected, unless checking turned off */
1558 if (check_for_problems || ! PL_in_utf8_CTYPE_locale) {
8c6180a9
KW
1559 /* Assume enough space for every character being bad. 4 spaces each
1560 * for the 94 printable characters that are output like "'x' "; and 5
1561 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
1562 * NUL */
8b7358b9 1563 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ] = { '\0' };
8c6180a9
KW
1564 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
1565 to start */
1566 unsigned int bad_count = 0; /* Count of bad characters */
1567
baa60164 1568 for (i = 0; i < 256; i++) {
8b7358b9 1569 if (! PL_in_utf8_CTYPE_locale) {
bd6d0898
KW
1570 if (isupper(i))
1571 PL_fold_locale[i] = (U8) tolower(i);
1572 else if (islower(i))
1573 PL_fold_locale[i] = (U8) toupper(i);
1574 else
1575 PL_fold_locale[i] = (U8) i;
8b7358b9 1576 }
8c6180a9
KW
1577
1578 /* If checking for locale problems, see if the native ASCII-range
1579 * printables plus \n and \t are in their expected categories in
1580 * the new locale. If not, this could mean big trouble, upending
1581 * Perl's and most programs' assumptions, like having a
1582 * metacharacter with special meaning become a \w. Fortunately,
1583 * it's very rare to find locales that aren't supersets of ASCII
1584 * nowadays. It isn't a problem for most controls to be changed
1585 * into something else; we check only \n and \t, though perhaps \r
1586 * could be an issue as well. */
7d4bcc4a 1587 if ( check_for_problems
8c6180a9
KW
1588 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
1589 {
8b7358b9 1590 bool is_bad = FALSE;
6726b4f4 1591 char name[4] = { '\0' };
8b7358b9
KW
1592
1593 /* Convert the name into a string */
6726b4f4 1594 if (isGRAPH_A(i)) {
8b7358b9
KW
1595 name[0] = i;
1596 name[1] = '\0';
1597 }
1598 else if (i == '\n') {
6726b4f4
KW
1599 my_strlcpy(name, "\\n", sizeof(name));
1600 }
1601 else if (i == '\t') {
1602 my_strlcpy(name, "\\t", sizeof(name));
8b7358b9
KW
1603 }
1604 else {
6726b4f4
KW
1605 assert(i == ' ');
1606 my_strlcpy(name, "' '", sizeof(name));
8b7358b9
KW
1607 }
1608
1609 /* Check each possibe class */
1610 if (UNLIKELY(cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC_A(i)))) {
1611 is_bad = TRUE;
1612 DEBUG_L(PerlIO_printf(Perl_debug_log,
1613 "isalnum('%s') unexpectedly is %d\n",
1614 name, cBOOL(isalnum(i))));
1615 }
1616 if (UNLIKELY(cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i)))) {
1617 is_bad = TRUE;
1618 DEBUG_L(PerlIO_printf(Perl_debug_log,
1619 "isalpha('%s') unexpectedly is %d\n",
1620 name, cBOOL(isalpha(i))));
1621 }
1622 if (UNLIKELY(cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i)))) {
1623 is_bad = TRUE;
1624 DEBUG_L(PerlIO_printf(Perl_debug_log,
1625 "isdigit('%s') unexpectedly is %d\n",
1626 name, cBOOL(isdigit(i))));
1627 }
1628 if (UNLIKELY(cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i)))) {
1629 is_bad = TRUE;
1630 DEBUG_L(PerlIO_printf(Perl_debug_log,
1631 "isgraph('%s') unexpectedly is %d\n",
1632 name, cBOOL(isgraph(i))));
1633 }
1634 if (UNLIKELY(cBOOL(islower(i)) != cBOOL(isLOWER_A(i)))) {
1635 is_bad = TRUE;
1636 DEBUG_L(PerlIO_printf(Perl_debug_log,
1637 "islower('%s') unexpectedly is %d\n",
1638 name, cBOOL(islower(i))));
1639 }
1640 if (UNLIKELY(cBOOL(isprint(i)) != cBOOL(isPRINT_A(i)))) {
1641 is_bad = TRUE;
1642 DEBUG_L(PerlIO_printf(Perl_debug_log,
1643 "isprint('%s') unexpectedly is %d\n",
1644 name, cBOOL(isprint(i))));
1645 }
1646 if (UNLIKELY(cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i)))) {
1647 is_bad = TRUE;
1648 DEBUG_L(PerlIO_printf(Perl_debug_log,
1649 "ispunct('%s') unexpectedly is %d\n",
1650 name, cBOOL(ispunct(i))));
1651 }
1652 if (UNLIKELY(cBOOL(isspace(i)) != cBOOL(isSPACE_A(i)))) {
1653 is_bad = TRUE;
1654 DEBUG_L(PerlIO_printf(Perl_debug_log,
1655 "isspace('%s') unexpectedly is %d\n",
1656 name, cBOOL(isspace(i))));
1657 }
1658 if (UNLIKELY(cBOOL(isupper(i)) != cBOOL(isUPPER_A(i)))) {
1659 is_bad = TRUE;
1660 DEBUG_L(PerlIO_printf(Perl_debug_log,
1661 "isupper('%s') unexpectedly is %d\n",
1662 name, cBOOL(isupper(i))));
1663 }
1664 if (UNLIKELY(cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i)))) {
1665 is_bad = TRUE;
1666 DEBUG_L(PerlIO_printf(Perl_debug_log,
1667 "isxdigit('%s') unexpectedly is %d\n",
1668 name, cBOOL(isxdigit(i))));
1669 }
1670 if (UNLIKELY(tolower(i) != (int) toLOWER_A(i))) {
1671 is_bad = TRUE;
1672 DEBUG_L(PerlIO_printf(Perl_debug_log,
1673 "tolower('%s')=0x%x instead of the expected 0x%x\n",
1674 name, tolower(i), (int) toLOWER_A(i)));
1675 }
1676 if (UNLIKELY(toupper(i) != (int) toUPPER_A(i))) {
1677 is_bad = TRUE;
1678 DEBUG_L(PerlIO_printf(Perl_debug_log,
1679 "toupper('%s')=0x%x instead of the expected 0x%x\n",
1680 name, toupper(i), (int) toUPPER_A(i)));
1681 }
1682 if (UNLIKELY((i == '\n' && ! isCNTRL_LC(i)))) {
1683 is_bad = TRUE;
1684 DEBUG_L(PerlIO_printf(Perl_debug_log,
1685 "'\\n' (=%02X) is not a control\n", (int) i));
1686 }
1687
1688 /* Add to the list; Separate multiple entries with a blank */
1689 if (is_bad) {
1690 if (bad_count) {
1691 my_strlcat(bad_chars_list, " ", sizeof(bad_chars_list));
8c6180a9 1692 }
8b7358b9
KW
1693 my_strlcat(bad_chars_list, name, sizeof(bad_chars_list));
1694 bad_count++;
8c6180a9
KW
1695 }
1696 }
1697 }
1698
30d8090d
KW
1699 if (bad_count == 2 && maybe_utf8_turkic) {
1700 bad_count = 0;
1701 *bad_chars_list = '\0';
1702 PL_fold_locale['I'] = 'I';
1703 PL_fold_locale['i'] = 'i';
1704 PL_in_utf8_turkic_locale = TRUE;
1705 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s is turkic\n",
1706 __FILE__, __LINE__, newctype));
1707 }
1708 else {
b8df1494 1709 PL_in_utf8_turkic_locale = FALSE;
30d8090d 1710 }
b8df1494 1711
7d4bcc4a
KW
1712# ifdef MB_CUR_MAX
1713
8c6180a9 1714 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 1715 * this locale requires more than one byte, there are going to be
8c6180a9 1716 * problems. */
9c8a6dc2
KW
1717 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
1718 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
1719 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
1720
8b7358b9
KW
1721 if ( check_for_problems && MB_CUR_MAX > 1
1722 && ! PL_in_utf8_CTYPE_locale
ba1a4362
KW
1723
1724 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
1725 * locale. Just assume that the implementation for them (plus
1726 * for POSIX) is correct and the > 1 value is spurious. (Since
1727 * these are specially handled to never be considered UTF-8
1728 * locales, as long as this is the only problem, everything
1729 * should work fine */
1730 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
1731 {
8c6180a9
KW
1732 multi_byte_locale = TRUE;
1733 }
7d4bcc4a
KW
1734
1735# endif
8c6180a9 1736
30d8090d
KW
1737 /* If we found problems and we want them output, do so */
1738 if ( (UNLIKELY(bad_count) || UNLIKELY(multi_byte_locale))
1739 && (LIKELY(ckWARN_d(WARN_LOCALE)) || UNLIKELY(DEBUG_L_TEST)))
1740 {
8b7358b9
KW
1741 if (UNLIKELY(bad_count) && PL_in_utf8_CTYPE_locale) {
1742 PL_warn_locale = Perl_newSVpvf(aTHX_
1743 "Locale '%s' contains (at least) the following characters"
578a6a87
KW
1744 " which have\nunexpected meanings: %s\nThe Perl program"
1745 " will use the expected meanings",
8b7358b9 1746 newctype, bad_chars_list);
8b7358b9
KW
1747 }
1748 else {
1749 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 1750 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 1751 newctype,
8c6180a9
KW
1752 (multi_byte_locale)
1753 ? " Some characters in it are not recognized by"
1754 " Perl."
1755 : "",
1756 (bad_count)
1757 ? "\nThe following characters (and maybe others)"
1758 " may not have the same meaning as the Perl"
1759 " program expects:\n"
1760 : "",
1761 (bad_count)
1762 ? bad_chars_list
1763 : ""
1764 );
8b7358b9
KW
1765 }
1766
b119c2be
KW
1767# ifdef HAS_NL_LANGINFO
1768
1769 Perl_sv_catpvf(aTHX_ PL_warn_locale, "; codeset=%s",
1770 /* parameter FALSE is a don't care here */
4e6826bf 1771 my_nl_langinfo(CODESET, FALSE));
b119c2be
KW
1772
1773# endif
1774
8b7358b9
KW
1775 Perl_sv_catpvf(aTHX_ PL_warn_locale, "\n");
1776
cc9eaeb0 1777 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
1778 * output the message now. If not in that scope, we save the
1779 * message to be output at the first operation using this locale,
1780 * if that actually happens. Most programs don't use locales, so
1781 * they are immune to bad ones. */
cc9eaeb0 1782 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f 1783
780fcc9f
KW
1784 /* The '0' below suppresses a bogus gcc compiler warning */
1785 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 1786
bddebb56
KW
1787 if (IN_LC(LC_CTYPE)) {
1788 SvREFCNT_dec_NN(PL_warn_locale);
1789 PL_warn_locale = NULL;
1790 }
780fcc9f 1791 }
baa60164 1792 }
31f05a37 1793 }
98994639
HS
1794
1795#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 1796
98994639
HS
1797}
1798
98994639 1799void
2726666d
KW
1800Perl__warn_problematic_locale()
1801{
2726666d
KW
1802
1803#ifdef USE_LOCALE_CTYPE
1804
5f04a188
KW
1805 dTHX;
1806
1807 /* Internal-to-core function that outputs the message in PL_warn_locale,
1808 * and then NULLS it. Should be called only through the macro
1809 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
1810
2726666d 1811 if (PL_warn_locale) {
2726666d
KW
1812 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1813 SvPVX(PL_warn_locale),
1814 0 /* dummy to avoid compiler warning */ );
2726666d
KW
1815 SvREFCNT_dec_NN(PL_warn_locale);
1816 PL_warn_locale = NULL;
1817 }
1818
1819#endif
1820
1821}
1822
a4f00dcc
KW
1823STATIC void
1824S_new_collate(pTHX_ const char *newcoll)
98994639 1825{
7d4bcc4a
KW
1826
1827#ifndef USE_LOCALE_COLLATE
1828
1829 PERL_UNUSED_ARG(newcoll);
1830 PERL_UNUSED_CONTEXT;
1831
1832#else
0d071d52 1833
291a84fb 1834 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
1835 * core Perl this and that 'newcoll' is the name of the new locale.
1836 *
d35fca5f
KW
1837 * The design of locale collation is that every locale change is given an
1838 * index 'PL_collation_ix'. The first time a string particpates in an
1839 * operation that requires collation while locale collation is active, it
1840 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
1841 * magic includes the collation index, and the transformation of the string
1842 * by strxfrm(), q.v. That transformation is used when doing comparisons,
1843 * instead of the string itself. If a string changes, the magic is
1844 * cleared. The next time the locale changes, the index is incremented,
1845 * and so we know during a comparison that the transformation is not
1846 * necessarily still valid, and so is recomputed. Note that if the locale
1847 * changes enough times, the index could wrap (a U32), and it is possible
1848 * that a transformation would improperly be considered valid, leading to
1849 * an unlikely bug */
0d071d52 1850
98994639
HS
1851 if (! newcoll) {
1852 if (PL_collation_name) {
1853 ++PL_collation_ix;
1854 Safefree(PL_collation_name);
1855 PL_collation_name = NULL;
1856 }
1857 PL_collation_standard = TRUE;
00bf60ca 1858 is_standard_collation:
98994639
HS
1859 PL_collxfrm_base = 0;
1860 PL_collxfrm_mult = 2;
165a1c52 1861 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 1862 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1863 PL_strxfrm_max_cp = 0;
98994639
HS
1864 return;
1865 }
1866
d35fca5f 1867 /* If this is not the same locale as currently, set the new one up */
98994639
HS
1868 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
1869 ++PL_collation_ix;
1870 Safefree(PL_collation_name);
1871 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 1872 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
1873 if (PL_collation_standard) {
1874 goto is_standard_collation;
1875 }
98994639 1876
165a1c52 1877 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 1878 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 1879 PL_strxfrm_max_cp = 0;
165a1c52 1880
59c018b9
KW
1881 /* A locale collation definition includes primary, secondary, tertiary,
1882 * etc. weights for each character. To sort, the primary weights are
1883 * used, and only if they compare equal, then the secondary weights are
1884 * used, and only if they compare equal, then the tertiary, etc.
1885 *
1886 * strxfrm() works by taking the input string, say ABC, and creating an
1887 * output transformed string consisting of first the primary weights,
1888 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
1889 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
1890 * may not have weights at every level. In our example, let's say B
1891 * doesn't have a tertiary weight, and A doesn't have a secondary
1892 * weight. The constructed string is then going to be
1893 * A¹B¹C¹ B²C² A³C³ ....
1894 * This has the desired effect that strcmp() will look at the secondary
1895 * or tertiary weights only if the strings compare equal at all higher
1896 * priority weights. The spaces shown here, like in
c342d20e 1897 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
1898 * are not just for readability. In the general case, these must
1899 * actually be bytes, which we will call here 'separator weights'; and
1900 * they must be smaller than any other weight value, but since these
1901 * are C strings, only the terminating one can be a NUL (some
1902 * implementations may include a non-NUL separator weight just before
1903 * the NUL). Implementations tend to reserve 01 for the separator
1904 * weights. They are needed so that a shorter string's secondary
1905 * weights won't be misconstrued as primary weights of a longer string,
1906 * etc. By making them smaller than any other weight, the shorter
1907 * string will sort first. (Actually, if all secondary weights are
1908 * smaller than all primary ones, there is no need for a separator
1909 * weight between those two levels, etc.)
1910 *
1911 * The length of the transformed string is roughly a linear function of
1912 * the input string. It's not exactly linear because some characters
1913 * don't have weights at all levels. When we call strxfrm() we have to
1914 * allocate some memory to hold the transformed string. The
1915 * calculations below try to find coefficients 'm' and 'b' for this
1916 * locale so that m*x + b equals how much space we need, given the size
1917 * of the input string in 'x'. If we calculate too small, we increase
1918 * the size as needed, and call strxfrm() again, but it is better to
1919 * get it right the first time to avoid wasted expensive string
1920 * transformations. */
1921
98994639 1922 {
79f120c8
KW
1923 /* We use the string below to find how long the tranformation of it
1924 * is. Almost all locales are supersets of ASCII, or at least the
1925 * ASCII letters. We use all of them, half upper half lower,
1926 * because if we used fewer, we might hit just the ones that are
1927 * outliers in a particular locale. Most of the strings being
1928 * collated will contain a preponderance of letters, and even if
1929 * they are above-ASCII, they are likely to have the same number of
1930 * weight levels as the ASCII ones. It turns out that digits tend
1931 * to have fewer levels, and some punctuation has more, but those
1932 * are relatively sparse in text, and khw believes this gives a
1933 * reasonable result, but it could be changed if experience so
1934 * dictates. */
1935 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
1936 char * x_longer; /* Transformed 'longer' */
1937 Size_t x_len_longer; /* Length of 'x_longer' */
1938
1939 char * x_shorter; /* We also transform a substring of 'longer' */
1940 Size_t x_len_shorter;
1941
a4a439fb 1942 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
1943 * are interested only in its length). It is used because it has
1944 * the intelligence to handle all cases, but to work, it needs some
1945 * values of 'm' and 'b' to get it started. For the purposes of
1946 * this calculation we use a very conservative estimate of 'm' and
1947 * 'b'. This assumes a weight can be multiple bytes, enough to
1948 * hold any UV on the platform, and there are 5 levels, 4 weight
1949 * bytes, and a trailing NUL. */
1950 PL_collxfrm_base = 5;
1951 PL_collxfrm_mult = 5 * sizeof(UV);
1952
1953 /* Find out how long the transformation really is */
a4a439fb
KW
1954 x_longer = _mem_collxfrm(longer,
1955 sizeof(longer) - 1,
1956 &x_len_longer,
1957
1958 /* We avoid converting to UTF-8 in the
1959 * called function by telling it the
1960 * string is in UTF-8 if the locale is a
1961 * UTF-8 one. Since the string passed
1962 * here is invariant under UTF-8, we can
1963 * claim it's UTF-8 even though it isn't.
1964 * */
1965 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1966 Safefree(x_longer);
1967
1968 /* Find out how long the transformation of a substring of 'longer'
1969 * is. Together the lengths of these transformations are
1970 * sufficient to calculate 'm' and 'b'. The substring is all of
1971 * 'longer' except the first character. This minimizes the chances
1972 * of being swayed by outliers */
a4a439fb 1973 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 1974 sizeof(longer) - 2,
a4a439fb
KW
1975 &x_len_shorter,
1976 PL_in_utf8_COLLATE_locale);
79f120c8
KW
1977 Safefree(x_shorter);
1978
1979 /* If the results are nonsensical for this simple test, the whole
1980 * locale definition is suspect. Mark it so that locale collation
1981 * is not active at all for it. XXX Should we warn? */
1982 if ( x_len_shorter == 0
1983 || x_len_longer == 0
1984 || x_len_shorter >= x_len_longer)
1985 {
1986 PL_collxfrm_mult = 0;
1987 PL_collxfrm_base = 0;
1988 }
1989 else {
1990 SSize_t base; /* Temporary */
1991
1992 /* We have both: m * strlen(longer) + b = x_len_longer
1993 * m * strlen(shorter) + b = x_len_shorter;
1994 * subtracting yields:
1995 * m * (strlen(longer) - strlen(shorter))
1996 * = x_len_longer - x_len_shorter
1997 * But we have set things up so that 'shorter' is 1 byte smaller
1998 * than 'longer'. Hence:
1999 * m = x_len_longer - x_len_shorter
2000 *
2001 * But if something went wrong, make sure the multiplier is at
2002 * least 1.
2003 */
2004 if (x_len_longer > x_len_shorter) {
2005 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
2006 }
2007 else {
2008 PL_collxfrm_mult = 1;
2009 }
2010
2011 /* mx + b = len
2012 * so: b = len - mx
2013 * but in case something has gone wrong, make sure it is
2014 * non-negative */
2015 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
2016 if (base < 0) {
2017 base = 0;
2018 }
2019
2020 /* Add 1 for the trailing NUL */
2021 PL_collxfrm_base = base + 1;
2022 }
58eebef2 2023
7d4bcc4a
KW
2024# ifdef DEBUGGING
2025
58eebef2
KW
2026 if (DEBUG_L_TEST || debug_initialization) {
2027 PerlIO_printf(Perl_debug_log,
b07929e4
KW
2028 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
2029 "x_len_longer=%zu,"
2030 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
2031 __FILE__, __LINE__,
2032 PL_in_utf8_COLLATE_locale,
2033 x_len_shorter, x_len_longer,
2034 PL_collxfrm_mult, PL_collxfrm_base);
2035 }
7d4bcc4a
KW
2036# endif
2037
98994639
HS
2038 }
2039 }
2040
2041#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 2042
98994639
HS
2043}
2044
d36adde0
KW
2045#endif
2046
d2b24094 2047#ifdef WIN32
b8cc575c 2048
a4f00dcc 2049STATIC char *
b8cc575c 2050S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
2051{
2052 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
2053 * difference between the two unless the input locale is "", which normally
2054 * means on Windows to get the machine default, which is set via the
2055 * computer's "Regional and Language Options" (or its current equivalent).
2056 * In POSIX, it instead means to find the locale from the user's
2057 * environment. This routine changes the Windows behavior to first look in
2058 * the environment, and, if anything is found, use that instead of going to
2059 * the machine default. If there is no environment override, the machine
2060 * default is used, by calling the real setlocale() with "".
2061 *
2062 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
2063 * use the particular category's variable if set; otherwise to use the LANG
2064 * variable. */
b385bb4d 2065
175c4cf9 2066 bool override_LC_ALL = FALSE;
89f7b9aa 2067 char * result;
e5f10d49 2068 unsigned int i;
89f7b9aa 2069
b385bb4d 2070 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
2071
2072# ifdef LC_ALL
2073
b385bb4d
KW
2074 locale = PerlEnv_getenv("LC_ALL");
2075 if (! locale) {
e5f10d49
KW
2076 if (category == LC_ALL) {
2077 override_LC_ALL = TRUE;
2078 }
2079 else {
7d4bcc4a
KW
2080
2081# endif
7d4bcc4a 2082
e5f10d49
KW
2083 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
2084 if (category == categories[i]) {
2085 locale = PerlEnv_getenv(category_names[i]);
2086 goto found_locale;
2087 }
2088 }
7d4bcc4a 2089
b385bb4d 2090 locale = PerlEnv_getenv("LANG");
481465ea 2091 if (! locale) {
b385bb4d
KW
2092 locale = "";
2093 }
e5f10d49
KW
2094
2095 found_locale: ;
7d4bcc4a
KW
2096
2097# ifdef LC_ALL
2098
e5f10d49 2099 }
b385bb4d 2100 }
7d4bcc4a
KW
2101
2102# endif
2103
b385bb4d
KW
2104 }
2105
89f7b9aa 2106 result = setlocale(category, locale);
48015184
KW
2107 DEBUG_L(STMT_START {
2108 dSAVE_ERRNO;
2109 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
2110 setlocale_debug_string(category, locale, result));
2111 RESTORE_ERRNO;
2112 } STMT_END);
89f7b9aa 2113
481465ea 2114 if (! override_LC_ALL) {
89f7b9aa
KW
2115 return result;
2116 }
2117
dfd77d7a 2118 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
2119 * LANG variable or the system default if there is no LANG. But these have
2120 * lower priority than the other LC_foo variables, so override it for each
2121 * one that is set. (If they are set to "", it means to use the same thing
2122 * we just set LC_ALL to, so can skip) */
7d4bcc4a 2123
948523db 2124 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
2125 result = PerlEnv_getenv(category_names[i]);
2126 if (result && strNE(result, "")) {
2127 setlocale(categories[i], result);
2128 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
2129 __FILE__, __LINE__,
2130 setlocale_debug_string(categories[i], result, "not captured")));
2131 }
89f7b9aa 2132 }
7d4bcc4a 2133
bbc98134 2134 result = setlocale(LC_ALL, NULL);
48015184
KW
2135 DEBUG_L(STMT_START {
2136 dSAVE_ERRNO;
2137 PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
bbc98134 2138 __FILE__, __LINE__,
48015184
KW
2139 setlocale_debug_string(LC_ALL, NULL, result));
2140 RESTORE_ERRNO;
2141 } STMT_END);
89f7b9aa 2142
bbc98134 2143 return result;
b385bb4d
KW
2144}
2145
2146#endif
2147
9aac5db8
KW
2148/*
2149
2150=head1 Locale-related functions and macros
2151
2152=for apidoc Perl_setlocale
2153
2154This is an (almost) drop-in replacement for the system L<C<setlocale(3)>>,
2155taking the same parameters, and returning the same information, except that it
9487427b
KW
2156returns the correct underlying C<LC_NUMERIC> locale. Regular C<setlocale> will
2157instead return C<C> if the underlying locale has a non-dot decimal point
2158character, or a non-empty thousands separator for displaying floating point
2159numbers. This is because perl keeps that locale category such that it has a
2160dot and empty separator, changing the locale briefly during the operations
2161where the underlying one is required. C<Perl_setlocale> knows about this, and
2162compensates; regular C<setlocale> doesn't.
9aac5db8 2163
e9bc6d6b 2164Another reason it isn't completely a drop-in replacement is that it is
9aac5db8 2165declared to return S<C<const char *>>, whereas the system setlocale omits the
163deff3
KW
2166C<const> (presumably because its API was specified long ago, and can't be
2167updated; it is illegal to change the information C<setlocale> returns; doing
2168so leads to segfaults.)
9aac5db8 2169
e9bc6d6b
KW
2170Finally, C<Perl_setlocale> works under all circumstances, whereas plain
2171C<setlocale> can be completely ineffective on some platforms under some
2172configurations.
2173
9aac5db8 2174C<Perl_setlocale> should not be used to change the locale except on systems
e9bc6d6b
KW
2175where the predefined variable C<${^SAFE_LOCALES}> is 1. On some such systems,
2176the system C<setlocale()> is ineffective, returning the wrong information, and
2177failing to actually change the locale. C<Perl_setlocale>, however works
2178properly in all circumstances.
9aac5db8
KW
2179
2180The return points to a per-thread static buffer, which is overwritten the next
2181time C<Perl_setlocale> is called from the same thread.
2182
2183=cut
2184
2185*/
2186
2187const char *
2188Perl_setlocale(const int category, const char * locale)
a4f00dcc
KW
2189{
2190 /* This wraps POSIX::setlocale() */
2191
d36adde0
KW
2192#ifdef NO_LOCALE
2193
2194 PERL_UNUSED_ARG(category);
2195 PERL_UNUSED_ARG(locale);
2196
2197 return "C";
2198
2199#else
2200
9aac5db8
KW
2201 const char * retval;
2202 const char * newlocale;
2203 dSAVEDERRNO;
a4f00dcc 2204 dTHX;
d36adde0 2205 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
a4f00dcc 2206
a4f00dcc
KW
2207#ifdef USE_LOCALE_NUMERIC
2208
291a84fb
KW
2209 /* A NULL locale means only query what the current one is. We have the
2210 * LC_NUMERIC name saved, because we are normally switched into the C
9487427b
KW
2211 * (or equivalent) locale for it. For an LC_ALL query, switch back to get
2212 * the correct results. All other categories don't require special
2213 * handling */
a4f00dcc
KW
2214 if (locale == NULL) {
2215 if (category == LC_NUMERIC) {
9aac5db8
KW
2216
2217 /* We don't have to copy this return value, as it is a per-thread
2218 * variable, and won't change until a future setlocale */
2219 return PL_numeric_name;
a4f00dcc
KW
2220 }
2221
7d4bcc4a 2222# ifdef LC_ALL
a4f00dcc 2223
9aac5db8
KW
2224 else if (category == LC_ALL) {
2225 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
a4f00dcc
KW
2226 }
2227
7d4bcc4a 2228# endif
a4f00dcc
KW
2229
2230 }
2231
2232#endif
2233
33e5a354
KW
2234 retval = save_to_buffer(do_setlocale_r(category, locale),
2235 &PL_setlocale_buf, &PL_setlocale_bufsize, 0);
9aac5db8
KW
2236 SAVE_ERRNO;
2237
2238#if defined(USE_LOCALE_NUMERIC) && defined(LC_ALL)
2239
2240 if (locale == NULL && category == LC_ALL) {
2241 RESTORE_LC_NUMERIC();
2242 }
2243
2244#endif
a4f00dcc
KW
2245
2246 DEBUG_L(PerlIO_printf(Perl_debug_log,
2247 "%s:%d: %s\n", __FILE__, __LINE__,
2248 setlocale_debug_string(category, locale, retval)));
7d4bcc4a 2249
9aac5db8
KW
2250 RESTORE_ERRNO;
2251
2252 if (! retval) {
a4f00dcc
KW
2253 return NULL;
2254 }
2255
9aac5db8 2256 /* If locale == NULL, we are just querying the state */
a4f00dcc 2257 if (locale == NULL) {
a4f00dcc
KW
2258 return retval;
2259 }
a4f00dcc 2260
1159483a
KW
2261 /* Now that have switched locales, we have to update our records to
2262 * correspond. */
a4f00dcc 2263
1159483a 2264 switch (category) {
a4f00dcc 2265
1159483a 2266#ifdef USE_LOCALE_CTYPE
a4f00dcc 2267
1159483a
KW
2268 case LC_CTYPE:
2269 new_ctype(retval);
2270 break;
a4f00dcc 2271
1159483a 2272#endif
a4f00dcc
KW
2273#ifdef USE_LOCALE_COLLATE
2274
1159483a
KW
2275 case LC_COLLATE:
2276 new_collate(retval);
2277 break;
a4f00dcc 2278
1159483a 2279#endif
a4f00dcc
KW
2280#ifdef USE_LOCALE_NUMERIC
2281
1159483a
KW
2282 case LC_NUMERIC:
2283 new_numeric(retval);
2284 break;
a4f00dcc 2285
1159483a
KW
2286#endif
2287#ifdef LC_ALL
a4f00dcc 2288
1159483a 2289 case LC_ALL:
a4f00dcc 2290
1159483a
KW
2291 /* LC_ALL updates all the things we care about. The values may not
2292 * be the same as 'retval', as the locale "" may have set things
2293 * individually */
a4f00dcc 2294
1159483a 2295# ifdef USE_LOCALE_CTYPE
a4f00dcc 2296
b0bde642 2297 newlocale = savepv(do_setlocale_c(LC_CTYPE, NULL));
1159483a 2298 new_ctype(newlocale);
b0bde642 2299 Safefree(newlocale);
a4f00dcc 2300
1159483a
KW
2301# endif /* USE_LOCALE_CTYPE */
2302# ifdef USE_LOCALE_COLLATE
2303
b0bde642 2304 newlocale = savepv(do_setlocale_c(LC_COLLATE, NULL));
1159483a 2305 new_collate(newlocale);
b0bde642 2306 Safefree(newlocale);
a4f00dcc 2307
7d4bcc4a 2308# endif
1159483a 2309# ifdef USE_LOCALE_NUMERIC
a4f00dcc 2310
b0bde642 2311 newlocale = savepv(do_setlocale_c(LC_NUMERIC, NULL));
1159483a 2312 new_numeric(newlocale);
b0bde642 2313 Safefree(newlocale);
a4f00dcc 2314
1159483a
KW
2315# endif /* USE_LOCALE_NUMERIC */
2316#endif /* LC_ALL */
a4f00dcc 2317
1159483a
KW
2318 default:
2319 break;
a4f00dcc
KW
2320 }
2321
2322 return retval;
2323
d36adde0
KW
2324#endif
2325
f7416781
KW
2326}
2327
2328PERL_STATIC_INLINE const char *
2329S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
2330{
2331 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
2332 * growing it if necessary */
2333
0b6697c5 2334 Size_t string_size;
f7416781
KW
2335
2336 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
2337
0b6697c5
KW
2338 if (! string) {
2339 return NULL;
2340 }
2341
2342 string_size = strlen(string) + offset + 1;
2343
f7416781
KW
2344 if (*buf_size == 0) {
2345 Newx(*buf, string_size, char);
2346 *buf_size = string_size;
2347 }
2348 else if (string_size > *buf_size) {
2349 Renew(*buf, string_size, char);
2350 *buf_size = string_size;
2351 }
2352
2353 Copy(string, *buf + offset, string_size - offset, char);
2354 return *buf;
2355}
2356
2357/*
2358
f7416781
KW
2359=for apidoc Perl_langinfo
2360
ce181bf6 2361This is an (almost) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
2362taking the same C<item> parameter values, and returning the same information.
2363But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
2364of Perl's locale handling from your code, and can be used on systems that lack
2365a native C<nl_langinfo>.
2366
2367Expanding on these:
2368
2369=over
2370
2371=item *
2372
ce181bf6
KW
2373The reason it isn't quite a drop-in replacement is actually an advantage. The
2374only difference is that it returns S<C<const char *>>, whereas plain
2375C<nl_langinfo()> returns S<C<char *>>, but you are (only by documentation)
2376forbidden to write into the buffer. By declaring this C<const>, the compiler
2377enforces this restriction, so if it is violated, you know at compilation time,
2378rather than getting segfaults at runtime.
2379
2380=item *
2381
69e2275e 2382It delivers the correct results for the C<RADIXCHAR> and C<THOUSEP> items,
f7416781
KW
2383without you having to write extra code. The reason for the extra code would be
2384because these are from the C<LC_NUMERIC> locale category, which is normally
9487427b
KW
2385kept set by Perl so that the radix is a dot, and the separator is the empty
2386string, no matter what the underlying locale is supposed to be, and so to get
2387the expected results, you have to temporarily toggle into the underlying
2388locale, and later toggle back. (You could use plain C<nl_langinfo> and
2389C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this but then you wouldn't get
2390the other advantages of C<Perl_langinfo()>; not keeping C<LC_NUMERIC> in the C
2391(or equivalent) locale would break a lot of CPAN, which is expecting the radix
2392(decimal point) character to be a dot.)
ce181bf6
KW
2393
2394=item *
2395
2396The system function it replaces can have its static return buffer trashed,
2397not only by a subesequent call to that function, but by a C<freelocale>,
2398C<setlocale>, or other locale change. The returned buffer of this function is
2399not changed until the next call to it, so the buffer is never in a trashed
2400state.
f7416781
KW
2401
2402=item *
2403
ce181bf6
KW
2404Its return buffer is per-thread, so it also is never overwritten by a call to
2405this function from another thread; unlike the function it replaces.
2406
2407=item *
2408
2409But most importantly, it works on systems that don't have C<nl_langinfo>, such
2410as Windows, hence makes your code more portable. Of the fifty-some possible
2411items specified by the POSIX 2008 standard,
f7416781 2412L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
8d72e74e
KW
2413only one is completely unimplemented, though on non-Windows platforms, another
2414significant one is also not implemented). It uses various techniques to
2415recover the other items, including calling C<L<localeconv(3)>>, and
2416C<L<strftime(3)>>, both of which are specified in C89, so should be always be
2417available. Later C<strftime()> versions have additional capabilities; C<""> is
2418returned for those not available on your system.
ce181bf6
KW
2419
2420It is important to note that when called with an item that is recovered by
2421using C<localeconv>, the buffer from any previous explicit call to
2422C<localeconv> will be overwritten. This means you must save that buffer's
c3997e39
KW
2423contents if you need to access them after a call to this function. (But note
2424that you might not want to be using C<localeconv()> directly anyway, because of
2425issues like the ones listed in the second item of this list (above) for
2426C<RADIXCHAR> and C<THOUSEP>. You can use the methods given in L<perlcall> to
2427call L<POSIX/localeconv> and avoid all the issues, but then you have a hash to
2428unpack).
9f0bc911 2429
6201c220
KW
2430The details for those items which may deviate from what this emulation returns
2431and what a native C<nl_langinfo()> would return are specified in
2432L<I18N::Langinfo>.
f7416781 2433
ce181bf6
KW
2434=back
2435
f7416781
KW
2436When using C<Perl_langinfo> on systems that don't have a native
2437C<nl_langinfo()>, you must
2438
2439 #include "perl_langinfo.h"
2440
2441before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
2442C<#include> with this one. (Doing it this way keeps out the symbols that plain
c3997e39
KW
2443C<langinfo.h> would try to import into the namespace for code that doesn't need
2444it.)
f7416781 2445
f7416781
KW
2446The original impetus for C<Perl_langinfo()> was so that code that needs to
2447find out the current currency symbol, floating point radix character, or digit
2448grouping separator can use, on all systems, the simpler and more
2449thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
2450pain to make thread-friendly. For other fields returned by C<localeconv>, it
2451is better to use the methods given in L<perlcall> to call
2452L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
2453
2454=cut
2455
2456*/
2457
2458const char *
2459#ifdef HAS_NL_LANGINFO
2460Perl_langinfo(const nl_item item)
2461#else
2462Perl_langinfo(const int item)
2463#endif
2464{
f61748ac
KW
2465 return my_nl_langinfo(item, TRUE);
2466}
2467
59a15af0 2468STATIC const char *
f61748ac
KW
2469#ifdef HAS_NL_LANGINFO
2470S_my_nl_langinfo(const nl_item item, bool toggle)
2471#else
2472S_my_nl_langinfo(const int item, bool toggle)
2473#endif
2474{
ae74815b 2475 dTHX;
0b6697c5 2476 const char * retval;
f7416781 2477
d36adde0
KW
2478#ifdef USE_LOCALE_NUMERIC
2479
5a854ab3
KW
2480 /* We only need to toggle into the underlying LC_NUMERIC locale for these
2481 * two items, and only if not already there */
4e6826bf 2482 if (toggle && (( item != RADIXCHAR && item != THOUSEP)
5a854ab3 2483 || PL_numeric_underlying))
d36adde0
KW
2484
2485#endif /* No toggling needed if not using LC_NUMERIC */
2486
5a854ab3 2487 toggle = FALSE;
5a854ab3 2488
ab340fff 2489#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
ee90eb2d 2490# if ! defined(HAS_THREAD_SAFE_NL_LANGINFO_L) \
80306bb6
KW
2491 || ! defined(HAS_POSIX_2008_LOCALE) \
2492 || ! defined(DUPLOCALE)
f7416781 2493
ab340fff 2494 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
2495 * for those items dependent on it. This must be copied to a buffer before
2496 * switching back, as some systems destroy the buffer when setlocale() is
2497 * called */
f7416781 2498
038d3702
KW
2499 {
2500 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
2501
b2fee59c 2502 if (toggle) {
038d3702 2503 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 2504 }
f7416781 2505
5acc4454
KW
2506 LOCALE_LOCK; /* Prevent interference from another thread executing
2507 this code section (the only call to nl_langinfo in
2508 the core) */
2509
ee90eb2d 2510
628ff75a
KW
2511 /* Copy to a per-thread buffer, which is also one that won't be
2512 * destroyed by a subsequent setlocale(), such as the
2513 * RESTORE_LC_NUMERIC may do just below. */
0b6697c5
KW
2514 retval = save_to_buffer(nl_langinfo(item),
2515 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
f7416781 2516
5acc4454
KW
2517 LOCALE_UNLOCK;
2518
b2fee59c 2519 if (toggle) {
038d3702 2520 RESTORE_LC_NUMERIC();
b2fee59c 2521 }
f7416781
KW
2522 }
2523
ab340fff
KW
2524# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
2525
5a854ab3 2526 {
b2fee59c
KW
2527 bool do_free = FALSE;
2528 locale_t cur = uselocale((locale_t) 0);
ab340fff 2529
b2fee59c
KW
2530 if (cur == LC_GLOBAL_LOCALE) {
2531 cur = duplocale(LC_GLOBAL_LOCALE);
2532 do_free = TRUE;
2533 }
ab340fff 2534
d36adde0
KW
2535# ifdef USE_LOCALE_NUMERIC
2536
b2fee59c 2537 if (toggle) {
e1aa2579
KW
2538 if (PL_underlying_numeric_obj) {
2539 cur = PL_underlying_numeric_obj;
2540 }
2541 else {
2542 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
2543 do_free = TRUE;
2544 }
b2fee59c 2545 }
ab340fff 2546
d36adde0
KW
2547# endif
2548
628ff75a
KW
2549 /* We have to save it to a buffer, because the freelocale() just below
2550 * can invalidate the internal one */
0b6697c5
KW
2551 retval = save_to_buffer(nl_langinfo_l(item, cur),
2552 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
ee90eb2d 2553
b2fee59c
KW
2554 if (do_free) {
2555 freelocale(cur);
2556 }
5a854ab3 2557 }
ab340fff 2558
c1566110
KW
2559# endif
2560
0b6697c5 2561 if (strEQ(retval, "")) {
4e6826bf 2562 if (item == YESSTR) {
c1566110
KW
2563 return "yes";
2564 }
4e6826bf 2565 if (item == NOSTR) {
c1566110
KW
2566 return "no";
2567 }
2568 }
2569
0b6697c5 2570 return retval;
ab340fff 2571
f7416781 2572#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
2573
2574 {
2575
f7416781
KW
2576# ifdef HAS_LOCALECONV
2577
43dd6b15 2578 const struct lconv* lc;
628ff75a 2579 const char * temp;
038d3702 2580 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781 2581
69c5e0db
KW
2582# ifdef TS_W32_BROKEN_LOCALECONV
2583
2584 const char * save_global;
2585 const char * save_thread;
2586 int needed_size;
2587 char * ptr;
2588 char * e;
2589 char * item_start;
2590
2591# endif
f7416781
KW
2592# endif
2593# ifdef HAS_STRFTIME
2594
43dd6b15
KW
2595 struct tm tm;
2596 bool return_format = FALSE; /* Return the %format, not the value */
2597 const char * format;
f7416781
KW
2598
2599# endif
2600
43dd6b15
KW
2601 /* We copy the results to a per-thread buffer, even if not
2602 * multi-threaded. This is in part to simplify this code, and partly
2603 * because we need a buffer anyway for strftime(), and partly because a
2604 * call of localeconv() could otherwise wipe out the buffer, and the
2605 * programmer would not be expecting this, as this is a nl_langinfo()
2606 * substitute after all, so s/he might be thinking their localeconv()
2607 * is safe until another localeconv() call. */
f7416781 2608
43dd6b15
KW
2609 switch (item) {
2610 Size_t len;
f7416781 2611
8d72e74e 2612 /* This is unimplemented */
4e6826bf 2613 case ERA: /* For use with strftime() %E modifier */
f7416781 2614
43dd6b15
KW
2615 default:
2616 return "";
f7416781 2617
43dd6b15 2618 /* We use only an English set, since we don't know any more */
4e6826bf
KW
2619 case YESEXPR: return "^[+1yY]";
2620 case YESSTR: return "yes";
2621 case NOEXPR: return "^[-0nN]";
2622 case NOSTR: return "no";
2623
8d72e74e
KW
2624 case CODESET:
2625
2626# ifndef WIN32
2627
2628 /* On non-windows, this is unimplemented, in part because of
2629 * inconsistencies between vendors. The Darwin native
2630 * nl_langinfo() implementation simply looks at everything past
2631 * any dot in the name, but that doesn't work for other
2632 * vendors. Many Linux locales that don't have UTF-8 in their
2633 * names really are UTF-8, for example; z/OS locales that do
2634 * have UTF-8 in their names, aren't really UTF-8 */
2635 return "";
2636
2637# else
2638
2639 { /* But on Windows, the name does seem to be consistent, so
2640 use that. */
2641 const char * p;
2642 const char * first;
2643 Size_t offset = 0;
2644 const char * name = my_setlocale(LC_CTYPE, NULL);
2645
2646 if (isNAME_C_OR_POSIX(name)) {
2647 return "ANSI_X3.4-1968";
2648 }
2649
2650 /* Find the dot in the locale name */
2651 first = (const char *) strchr(name, '.');
2652 if (! first) {
2653 first = name;
2654 goto has_nondigit;
2655 }
2656
2657 /* Look at everything past the dot */
2658 first++;
2659 p = first;
2660
2661 while (*p) {
2662 if (! isDIGIT(*p)) {
2663 goto has_nondigit;
2664 }
2665
2666 p++;
2667 }
2668
2669 /* Here everything past the dot is a digit. Treat it as a
2670 * code page */
18503cbd 2671 retval = save_to_buffer("CP", &PL_langinfo_buf,
32a62865 2672 &PL_langinfo_bufsize, 0);
8d72e74e 2673 offset = STRLENs("CP");
f7416781 2674
8d72e74e
KW
2675 has_nondigit:
2676
2677 retval = save_to_buffer(first, &PL_langinfo_buf,
2678 &PL_langinfo_bufsize, offset);
2679 }
2680
2681 break;
2682
2683# endif
f7416781
KW
2684# ifdef HAS_LOCALECONV
2685
4e6826bf 2686 case CRNCYSTR:
f7416781 2687
43dd6b15
KW
2688 /* We don't bother with localeconv_l() because any system that
2689 * has it is likely to also have nl_langinfo() */
291a84fb 2690
69c5e0db
KW
2691 LOCALE_LOCK_V; /* Prevent interference with other threads
2692 using localeconv() */
2693
2694# ifdef TS_W32_BROKEN_LOCALECONV
2695
2696 /* This is a workaround for a Windows bug prior to VS 15.
2697 * What we do here is, while locked, switch to the global
2698 * locale so localeconv() works; then switch back just before
2699 * the unlock. This can screw things up if some thread is
2700 * already using the global locale while assuming no other is.
2701 * A different workaround would be to call GetCurrencyFormat on
2702 * a known value, and parse it; patches welcome
2703 *
2704 * We have to use LC_ALL instead of LC_MONETARY because of
2705 * another bug in Windows */
2706
2707 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2708 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2709 save_global= savepv(my_setlocale(LC_ALL, NULL));
2710 my_setlocale(LC_ALL, save_thread);
2711
2712# endif
5acc4454 2713
43dd6b15
KW
2714 lc = localeconv();
2715 if ( ! lc
2716 || ! lc->currency_symbol
2717 || strEQ("", lc->currency_symbol))
2718 {
69c5e0db 2719 LOCALE_UNLOCK_V;
43dd6b15
KW
2720 return "";
2721 }
f7416781 2722
43dd6b15 2723 /* Leave the first spot empty to be filled in below */
0b6697c5
KW
2724 retval = save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
2725 &PL_langinfo_bufsize, 1);
43dd6b15
KW
2726 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
2727 { /* khw couldn't figure out how the localedef specifications
2728 would show that the $ should replace the radix; this is
2729 just a guess as to how it might work.*/
a34edee3 2730 PL_langinfo_buf[0] = '.';
43dd6b15
KW
2731 }
2732 else if (lc->p_cs_precedes) {
a34edee3 2733 PL_langinfo_buf[0] = '-';
43dd6b15
KW
2734 }
2735 else {
a34edee3 2736 PL_langinfo_buf[0] = '+';
43dd6b15 2737 }
f7416781 2738
69c5e0db
KW
2739# ifdef TS_W32_BROKEN_LOCALECONV
2740
2741 my_setlocale(LC_ALL, save_global);
2742 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2743 my_setlocale(LC_ALL, save_thread);
2744 Safefree(save_global);
2745 Safefree(save_thread);
2746
2747# endif
2748
2749 LOCALE_UNLOCK_V;
43dd6b15 2750 break;
f7416781 2751
69c5e0db
KW
2752# ifdef TS_W32_BROKEN_LOCALECONV
2753
4e6826bf 2754 case RADIXCHAR:
69c5e0db
KW
2755
2756 /* For this, we output a known simple floating point number to
2757 * a buffer, and parse it, looking for the radix */
2758
2759 if (toggle) {
2760 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
2761 }
2762
2763 if (PL_langinfo_bufsize < 10) {
2764 PL_langinfo_bufsize = 10;
2765 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2766 }
2767
2768 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2769 "%.1f", 1.5);
2770 if (needed_size >= (int) PL_langinfo_bufsize) {
2771 PL_langinfo_bufsize = needed_size + 1;
2772 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
2773 needed_size = my_snprintf(PL_langinfo_buf, PL_langinfo_bufsize,
2774 "%.1f", 1.5);
2775 assert(needed_size < (int) PL_langinfo_bufsize);
2776 }
2777
2778 ptr = PL_langinfo_buf;
2779 e = PL_langinfo_buf + PL_langinfo_bufsize;
2780
2781 /* Find the '1' */
2782 while (ptr < e && *ptr != '1') {
2783 ptr++;
2784 }
2785 ptr++;
2786
2787 /* Find the '5' */
2788 item_start = ptr;
2789 while (ptr < e && *ptr != '5') {
2790 ptr++;
2791 }
2792
2793 /* Everything in between is the radix string */
2794 if (ptr >= e) {
2795 PL_langinfo_buf[0] = '?';
2796 PL_langinfo_buf[1] = '\0';
2797 }
2798 else {
2799 *ptr = '\0';
2800 Move(item_start, PL_langinfo_buf, ptr - PL_langinfo_buf, char);
2801 }
2802
2803 if (toggle) {
2804 RESTORE_LC_NUMERIC();
2805 }
2806
2807 retval = PL_langinfo_buf;
2808 break;
2809
2810# else
2811
2812 case RADIXCHAR: /* No special handling needed */
2813
2814# endif
2815
4e6826bf 2816 case THOUSEP:
f7416781 2817
43dd6b15 2818 if (toggle) {
038d3702 2819 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 2820 }
f7416781 2821
69c5e0db
KW
2822 LOCALE_LOCK_V; /* Prevent interference with other threads
2823 using localeconv() */
2824
2825# ifdef TS_W32_BROKEN_LOCALECONV
2826
2827 /* This should only be for the thousands separator. A
2828 * different work around would be to use GetNumberFormat on a
2829 * known value and parse the result to find the separator */
2830 save_thread = savepv(my_setlocale(LC_ALL, NULL));
2831 _configthreadlocale(_DISABLE_PER_THREAD_LOCALE);
2832 save_global = savepv(my_setlocale(LC_ALL, NULL));
2833 my_setlocale(LC_ALL, save_thread);
2834# if 0
2835 /* This is the start of code that for broken Windows replaces
2836 * the above and below code, and instead calls
2837 * GetNumberFormat() and then would parse that to find the
2838 * thousands separator. It needs to handle UTF-16 vs -8
2839 * issues. */
2840
2841 needed_size = GetNumberFormatEx(PL_numeric_name, 0, "1234.5", NULL, PL_langinfo_buf, PL_langinfo_bufsize);
2842 DEBUG_L(PerlIO_printf(Perl_debug_log,
2843 "%s: %d: return from GetNumber, count=%d, val=%s\n",
2844 __FILE__, __LINE__, needed_size, PL_langinfo_buf));
2845
2846# endif
2847# endif
5acc4454 2848
43dd6b15
KW
2849 lc = localeconv();
2850 if (! lc) {
628ff75a 2851 temp = "";
33394adc 2852 }
43dd6b15 2853 else {
4e6826bf 2854 temp = (item == RADIXCHAR)
43dd6b15
KW
2855 ? lc->decimal_point
2856 : lc->thousands_sep;
628ff75a
KW
2857 if (! temp) {
2858 temp = "";
43dd6b15
KW
2859 }
2860 }
f7416781 2861
0b6697c5
KW
2862 retval = save_to_buffer(temp, &PL_langinfo_buf,
2863 &PL_langinfo_bufsize, 0);
f7416781 2864
69c5e0db
KW
2865# ifdef TS_W32_BROKEN_LOCALECONV
2866
2867 my_setlocale(LC_ALL, save_global);
2868 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
2869 my_setlocale(LC_ALL, save_thread);
2870 Safefree(save_global);
2871 Safefree(save_thread);
2872
2873# endif
2874
2875 LOCALE_UNLOCK_V;
5acc4454 2876
43dd6b15 2877 if (toggle) {
038d3702 2878 RESTORE_LC_NUMERIC();
43dd6b15 2879 }
f7416781 2880
43dd6b15 2881 break;
f7416781
KW
2882
2883# endif
2884# ifdef HAS_STRFTIME
2885
43dd6b15
KW
2886 /* These are defined by C89, so we assume that strftime supports
2887 * them, and so are returned unconditionally; they may not be what
2888 * the locale actually says, but should give good enough results
2889 * for someone using them as formats (as opposed to trying to parse
2890 * them to figure out what the locale says). The other format
2891 * items are actually tested to verify they work on the platform */
4e6826bf
KW
2892 case D_FMT: return "%x";
2893 case T_FMT: return "%X";
2894 case D_T_FMT: return "%c";
43dd6b15
KW
2895
2896 /* These formats are only available in later strfmtime's */
4e6826bf 2897 case ERA_D_FMT: case ERA_T_FMT: case ERA_D_T_FMT: case T_FMT_AMPM:
43dd6b15
KW
2898
2899 /* The rest can be gotten from most versions of strftime(). */
4e6826bf
KW
2900 case ABDAY_1: case ABDAY_2: case ABDAY_3:
2901 case ABDAY_4: case ABDAY_5: case ABDAY_6: case ABDAY_7:
2902 case ALT_DIGITS:
2903 case AM_STR: case PM_STR:
2904 case ABMON_1: case ABMON_2: case ABMON_3: case ABMON_4:
2905 case ABMON_5: case ABMON_6: case ABMON_7: case ABMON_8:
2906 case ABMON_9: case ABMON_10: case ABMON_11: case ABMON_12:
2907 case DAY_1: case DAY_2: case DAY_3: case DAY_4:
2908 case DAY_5: case DAY_6: case DAY_7:
2909 case MON_1: case MON_2: case MON_3: case MON_4:
2910 case MON_5: case MON_6: case MON_7: case MON_8:
2911 case MON_9: case MON_10: case MON_11: case MON_12:
43dd6b15
KW
2912
2913 LOCALE_LOCK;
2914
2915 init_tm(&tm); /* Precaution against core dumps */
2916 tm.tm_sec = 30;
2917 tm.tm_min = 30;
2918 tm.tm_hour = 6;
2919 tm.tm_year = 2017 - 1900;
2920 tm.tm_wday = 0;
2921 tm.tm_mon = 0;
2922 switch (item) {
2923 default:
2924 LOCALE_UNLOCK;
2925 Perl_croak(aTHX_
2926 "panic: %s: %d: switch case: %d problem",
2927 __FILE__, __LINE__, item);
2928 NOT_REACHED; /* NOTREACHED */
2929
4e6826bf
KW
2930 case PM_STR: tm.tm_hour = 18;
2931 case AM_STR:
43dd6b15
KW
2932 format = "%p";
2933 break;
2934
4e6826bf
KW
2935 case ABDAY_7: tm.tm_wday++;
2936 case ABDAY_6: tm.tm_wday++;
2937 case ABDAY_5: tm.tm_wday++;
2938 case ABDAY_4: tm.tm_wday++;
2939 case ABDAY_3: tm.tm_wday++;
2940 case ABDAY_2: tm.tm_wday++;
2941 case ABDAY_1:
43dd6b15
KW
2942 format = "%a";
2943 break;
2944
4e6826bf
KW
2945 case DAY_7: tm.tm_wday++;
2946 case DAY_6: tm.tm_wday++;
2947 case DAY_5: tm.tm_wday++;
2948 case DAY_4: tm.tm_wday++;
2949 case DAY_3: tm.tm_wday++;
2950 case DAY_2: tm.tm_wday++;
2951 case DAY_1:
43dd6b15
KW
2952 format = "%A";
2953 break;
2954
4e6826bf
KW
2955 case ABMON_12: tm.tm_mon++;
2956 case ABMON_11: tm.tm_mon++;
2957 case ABMON_10: tm.tm_mon++;
2958 case ABMON_9: tm.tm_mon++;
2959 case ABMON_8: tm.tm_mon++;
2960 case ABMON_7: tm.tm_mon++;
2961 case ABMON_6: tm.tm_mon++;
2962 case ABMON_5: tm.tm_mon++;
2963 case ABMON_4: tm.tm_mon++;
2964 case ABMON_3: tm.tm_mon++;
2965 case ABMON_2: tm.tm_mon++;
2966 case ABMON_1:
43dd6b15
KW
2967 format = "%b";
2968 break;
2969
4e6826bf
KW
2970 case MON_12: tm.tm_mon++;
2971 case MON_11: tm.tm_mon++;
2972 case MON_10: tm.tm_mon++;
2973 case MON_9: tm.tm_mon++;
2974 case MON_8: tm.tm_mon++;
2975 case MON_7: tm.tm_mon++;
2976 case MON_6: tm.tm_mon++;
2977 case MON_5: tm.tm_mon++;
2978 case MON_4: tm.tm_mon++;
2979 case MON_3: tm.tm_mon++;
2980 case MON_2: tm.tm_mon++;
2981 case MON_1:
43dd6b15
KW
2982 format = "%B";
2983 break;
2984
4e6826bf 2985 case T_FMT_AMPM:
43dd6b15
KW
2986 format = "%r";
2987 return_format = TRUE;
2988 break;
2989
4e6826bf 2990 case ERA_D_FMT:
43dd6b15
KW
2991 format = "%Ex";
2992 return_format = TRUE;
2993 break;
2994
4e6826bf 2995 case ERA_T_FMT:
43dd6b15
KW
2996 format = "%EX";
2997 return_format = TRUE;
2998 break;
2999
4e6826bf 3000 case ERA_D_T_FMT:
43dd6b15
KW
3001 format = "%Ec";
3002 return_format = TRUE;
3003 break;
3004
4e6826bf 3005 case ALT_DIGITS:
43dd6b15
KW
3006 tm.tm_wday = 0;
3007 format = "%Ow"; /* Find the alternate digit for 0 */
3008 break;
3009 }
f7416781 3010
43dd6b15
KW
3011 /* We can't use my_strftime() because it doesn't look at
3012 * tm_wday */
3013 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
3014 format, &tm))
3015 {
3016 /* A zero return means one of:
3017 * a) there wasn't enough space in PL_langinfo_buf
3018 * b) the format, like a plain %p, returns empty
3019 * c) it was an illegal format, though some
3020 * implementations of strftime will just return the
3021 * illegal format as a plain character sequence.
3022 *
3023 * To quickly test for case 'b)', try again but precede
3024 * the format with a plain character. If that result is
3025 * still empty, the problem is either 'a)' or 'c)' */
3026
3027 Size_t format_size = strlen(format) + 1;
3028 Size_t mod_size = format_size + 1;
3029 char * mod_format;
3030 char * temp_result;
3031
3032 Newx(mod_format, mod_size, char);
3033 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 3034 *mod_format = ' ';
43dd6b15
KW
3035 my_strlcpy(mod_format + 1, format, mod_size);
3036 len = strftime(temp_result,
3037 PL_langinfo_bufsize,
3038 mod_format, &tm);
3039 Safefree(mod_format);
3040 Safefree(temp_result);
3041
3042 /* If 'len' is non-zero, it means that we had a case like
3043 * %p which means the current locale doesn't use a.m. or
3044 * p.m., and that is valid */
3045 if (len == 0) {
3046
3047 /* Here, still didn't work. If we get well beyond a
3048 * reasonable size, bail out to prevent an infinite
3049 * loop. */
3050
3051 if (PL_langinfo_bufsize > 100 * format_size) {
3052 *PL_langinfo_buf = '\0';
3053 }
3054 else {
3055 /* Double the buffer size to retry; Add 1 in case
3056 * original was 0, so we aren't stuck at 0. */
3057 PL_langinfo_bufsize *= 2;
3058 PL_langinfo_bufsize++;
3059 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
3060 continue;
3061 }
3062 }
f7416781 3063
f7416781 3064 break;
43dd6b15 3065 }
f7416781 3066
43dd6b15
KW
3067 /* Here, we got a result.
3068 *
3069 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
3070 * alternate format for wday 0. If the value is the same as
3071 * the normal 0, there isn't an alternate, so clear the buffer.
3072 * */
4e6826bf 3073 if ( item == ALT_DIGITS
43dd6b15
KW
3074 && strEQ(PL_langinfo_buf, "0"))
3075 {
3076 *PL_langinfo_buf = '\0';
3077 }
f7416781 3078
43dd6b15
KW
3079 /* ALT_DIGITS is problematic. Experiments on it showed that
3080 * strftime() did not always work properly when going from
3081 * alt-9 to alt-10. Only a few locales have this item defined,
3082 * and in all of them on Linux that khw was able to find,
3083 * nl_langinfo() merely returned the alt-0 character, possibly
3084 * doubled. Most Unicode digits are in blocks of 10
3085 * consecutive code points, so that is sufficient information
3086 * for those scripts, as we can infer alt-1, alt-2, .... But
3087 * for a Japanese locale, a CJK ideographic 0 is returned, and
3088 * the CJK digits are not in code point order, so you can't
3089 * really infer anything. The localedef for this locale did
3090 * specify the succeeding digits, so that strftime() works
3091 * properly on them, without needing to infer anything. But
3092 * the nl_langinfo() return did not give sufficient information
3093 * for the caller to understand what's going on. So until
3094 * there is evidence that it should work differently, this
3095 * returns the alt-0 string for ALT_DIGITS.
3096 *
3097 * wday was chosen because its range is all a single digit.
3098 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 3099
43dd6b15 3100 LOCALE_UNLOCK;
f7416781 3101
d1b150d9
KW
3102 retval = PL_langinfo_buf;
3103
43dd6b15
KW
3104 /* If to return the format, not the value, overwrite the buffer
3105 * with it. But some strftime()s will keep the original format
3106 * if illegal, so change those to "" */
3107 if (return_format) {
3108 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
3109 *PL_langinfo_buf = '\0';
3110 }
43dd6b15 3111 else {
0b6697c5
KW
3112 retval = save_to_buffer(format, &PL_langinfo_buf,
3113 &PL_langinfo_bufsize, 0);
f7416781
KW
3114 }
3115 }
3116
3117 break;
f7416781
KW
3118
3119# endif
3120
43dd6b15 3121 }
f7416781
KW
3122 }
3123
0b6697c5 3124 return retval;
f7416781
KW
3125
3126#endif
3127
a4f00dcc 3128}
b385bb4d 3129
98994639
HS
3130/*
3131 * Initialize locale awareness.
3132 */
3133int
3134Perl_init_i18nl10n(pTHX_ int printwarn)
3135{
0e92a118
KW
3136 /* printwarn is
3137 *
3138 * 0 if not to output warning when setup locale is bad
3139 * 1 if to output warning based on value of PERL_BADLANG
3140 * >1 if to output regardless of PERL_BADLANG
3141 *
3142 * returns
98994639 3143 * 1 = set ok or not applicable,
0e92a118
KW
3144 * 0 = fallback to a locale of lower priority
3145 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
3146 *
3147 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
3148 * set, debugging information is output.
3149 *
3150 * This looks more complicated than it is, mainly due to the #ifdefs.
3151 *
3152 * We try to set LC_ALL to the value determined by the environment. If
3153 * there is no LC_ALL on this platform, we try the individual categories we
3154 * know about. If this works, we are done.
3155 *
3156 * But if it doesn't work, we have to do something else. We search the
3157 * environment variables ourselves instead of relying on the system to do
3158 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
3159 * think there is one), and the ultimate fallback "C". This is all done in
3160 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
3161 * more complex. The 'trial_locales' array is initialized with just one
3162 * element; it causes the behavior described in the paragraph above this to
3163 * happen. If that fails, we add elements to 'trial_locales', and do extra
3164 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
3165 *
3166 * On Ultrix, the locale MUST come from the environment, so there is
3167 * preliminary code to set it. I (khw) am not sure that it is necessary,
3168 * and that this couldn't be folded into the loop, but barring any real
3169 * platforms to test on, it's staying as-is
3170 *
3171 * A slight complication is that in embedded Perls, the locale may already
3172 * be set-up, and we don't want to get it from the normal environment
3173 * variables. This is handled by having a special environment variable
3174 * indicate we're in this situation. We simply set setlocale's 2nd
3175 * parameter to be a NULL instead of "". That indicates to setlocale that
3176 * it is not to change anything, but to return the current value,
3177 * effectively initializing perl's db to what the locale already is.
3178 *
3179 * We play the same trick with NULL if a LC_ALL succeeds. We call
3180 * setlocale() on the individual categores with NULL to get their existing
3181 * values for our db, instead of trying to change them.
3182 * */
98994639 3183
1565c085
DM
3184 dVAR;
3185
0e92a118
KW
3186 int ok = 1;
3187
7d4bcc4a
KW
3188#ifndef USE_LOCALE
3189
3190 PERL_UNUSED_ARG(printwarn);
3191
3192#else /* USE_LOCALE */
7d4bcc4a
KW
3193# ifdef __GLIBC__
3194
175c4cf9 3195 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
3196
3197# endif
65ebb059 3198
ccd65d51
KW
3199 /* NULL uses the existing already set up locale */
3200 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
3201 ? NULL
3202 : "";
c3fcd832
KW
3203 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
3204 unsigned int trial_locales_count;
175c4cf9
KW
3205 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
3206 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 3207 bool setlocale_failure = FALSE;
65ebb059 3208 unsigned int i;
175c4cf9
KW
3209
3210 /* A later getenv() could zap this, so only use here */
3211 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
3212
3213 const bool locwarn = (printwarn > 1
e5f10d49
KW
3214 || ( printwarn
3215 && ( ! bad_lang_use_once
22ff3130 3216 || (
e5f10d49
KW
3217 /* disallow with "" or "0" */
3218 *bad_lang_use_once
3219 && strNE("0", bad_lang_use_once)))));
ea92aad8 3220
291a84fb 3221 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 3222 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
3223
3224 /* current locale for given category; should have been copied so aren't
3225 * volatile */
8de4332b 3226 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 3227
7d4bcc4a
KW
3228# ifdef WIN32
3229
6bce99ee
JH
3230 /* In some systems you can find out the system default locale
3231 * and use that as the fallback locale. */
7d4bcc4a
KW
3232# define SYSTEM_DEFAULT_LOCALE
3233# endif
3234# ifdef SYSTEM_DEFAULT_LOCALE
3235
65ebb059 3236 const char *system_default_locale = NULL;
98994639 3237
7d4bcc4a 3238# endif
948523db
KW
3239
3240# ifndef DEBUGGING
3241# define DEBUG_LOCALE_INIT(a,b,c)
3242# else
7d4bcc4a 3243
8298454c 3244 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
3245
3246# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
3247 STMT_START { \
3248 if (debug_initialization) { \
3249 PerlIO_printf(Perl_debug_log, \
3250 "%s:%d: %s\n", \
3251 __FILE__, __LINE__, \
a4f00dcc 3252 setlocale_debug_string(category, \
2fcc0ca9
KW
3253 locale, \
3254 result)); \
3255 } \
3256 } STMT_END
2fcc0ca9 3257
948523db
KW
3258/* Make sure the parallel arrays are properly set up */
3259# ifdef USE_LOCALE_NUMERIC
3260 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
3261 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
e9bc6d6b
KW
3262# ifdef USE_POSIX_2008_LOCALE
3263 assert(category_masks[LC_NUMERIC_INDEX] == LC_NUMERIC_MASK);
3264# endif
948523db
KW
3265# endif
3266# ifdef USE_LOCALE_CTYPE
3267 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
3268 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
e9bc6d6b
KW
3269# ifdef USE_POSIX_2008_LOCALE
3270 assert(category_masks[LC_CTYPE_INDEX] == LC_CTYPE_MASK);
3271# endif
948523db
KW
3272# endif
3273# ifdef USE_LOCALE_COLLATE
3274 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
3275 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
e9bc6d6b
KW
3276# ifdef USE_POSIX_2008_LOCALE
3277 assert(category_masks[LC_COLLATE_INDEX] == LC_COLLATE_MASK);
3278# endif
948523db
KW
3279# endif
3280# ifdef USE_LOCALE_TIME
3281 assert(categories[LC_TIME_INDEX] == LC_TIME);
3282 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
e9bc6d6b
KW
3283# ifdef USE_POSIX_2008_LOCALE
3284 assert(category_masks[LC_TIME_INDEX] == LC_TIME_MASK);
3285# endif
948523db
KW
3286# endif
3287# ifdef USE_LOCALE_MESSAGES
3288 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
3289 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
e9bc6d6b
KW
3290# ifdef USE_POSIX_2008_LOCALE
3291 assert(category_masks[LC_MESSAGES_INDEX] == LC_MESSAGES_MASK);
3292# endif
948523db
KW
3293# endif
3294# ifdef USE_LOCALE_MONETARY
3295 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
3296 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
e9bc6d6b
KW
3297# ifdef USE_POSIX_2008_LOCALE
3298 assert(category_masks[LC_MONETARY_INDEX] == LC_MONETARY_MASK);
3299# endif
948523db 3300# endif
9821811f
KW
3301# ifdef USE_LOCALE_ADDRESS
3302 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
3303 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
e9bc6d6b
KW
3304# ifdef USE_POSIX_2008_LOCALE
3305 assert(category_masks[LC_ADDRESS_INDEX] == LC_ADDRESS_MASK);
3306# endif
9821811f
KW
3307# endif
3308# ifdef USE_LOCALE_IDENTIFICATION
3309 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
3310 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
e9bc6d6b
KW
3311# ifdef USE_POSIX_2008_LOCALE
3312 assert(category_masks[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION_MASK);
3313# endif
9821811f
KW
3314# endif
3315# ifdef USE_LOCALE_MEASUREMENT
3316 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
3317 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
e9bc6d6b
KW
3318# ifdef USE_POSIX_2008_LOCALE
3319 assert(category_masks[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT_MASK);
3320# endif
9821811f
KW
3321# endif
3322# ifdef USE_LOCALE_PAPER
3323 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
3324 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
e9bc6d6b
KW
3325# ifdef USE_POSIX_2008_LOCALE
3326 assert(category_masks[LC_PAPER_INDEX] == LC_PAPER_MASK);
3327# endif
9821811f
KW
3328# endif
3329# ifdef USE_LOCALE_TELEPHONE
3330 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
3331 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
e9bc6d6b
KW
3332# ifdef USE_POSIX_2008_LOCALE
3333 assert(category_masks[LC_TELEPHONE_INDEX] == LC_TELEPHONE_MASK);
3334# endif
9821811f 3335# endif
948523db
KW
3336# ifdef LC_ALL
3337 assert(categories[LC_ALL_INDEX] == LC_ALL);
3338 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
3339 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
e9bc6d6b
KW
3340# ifdef USE_POSIX_2008_LOCALE
3341 assert(category_masks[LC_ALL_INDEX] == LC_ALL_MASK);
3342# endif
948523db
KW
3343# endif
3344# endif /* DEBUGGING */
47280b20
KW
3345
3346 /* Initialize the cache of the program's UTF-8ness for the always known
3347 * locales C and POSIX */
3348 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
3349 sizeof(PL_locale_utf8ness));
3350
e9bc6d6b
KW
3351# ifdef USE_THREAD_SAFE_LOCALE
3352# ifdef WIN32
3353
3354 _configthreadlocale(_ENABLE_PER_THREAD_LOCALE);
3355
3356# endif
3357# endif
39e69e77 3358# ifdef USE_POSIX_2008_LOCALE
e9bc6d6b
KW
3359
3360 PL_C_locale_obj = newlocale(LC_ALL_MASK, "C", (locale_t) 0);
3361 if (! PL_C_locale_obj) {
3362 Perl_croak_nocontext(
3363 "panic: Cannot create POSIX 2008 C locale object; errno=%d", errno);
3364 }
3365 if (DEBUG_Lv_TEST || debug_initialization) {
3366 PerlIO_printf(Perl_debug_log, "%s:%d: created C object %p\n", __FILE__, __LINE__, PL_C_locale_obj);
3367 }
3368
3369# endif
3370
78f7c09f
KW
3371# ifdef USE_LOCALE_NUMERIC
3372
3ca88433
KW
3373 PL_numeric_radix_sv = newSVpvs(".");
3374
78f7c09f
KW
3375# endif
3376
e9bc6d6b
KW
3377# if defined(USE_POSIX_2008_LOCALE) && ! defined(HAS_QUERYLOCALE)
3378
3379 /* Initialize our records. If we have POSIX 2008, we have LC_ALL */
3380 do_setlocale_c(LC_ALL, my_setlocale(LC_ALL, NULL));
3381
3382# endif
b56c4436 3383# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
3384
3385 /*
3386 * Ultrix setlocale(..., "") fails if there are no environment
3387 * variables from which to get a locale name.
3388 */
3389
b56c4436
KW
3390# ifndef LC_ALL
3391# error Ultrix without LC_ALL not implemented
3392# else
7d4bcc4a 3393
b56c4436
KW
3394 {
3395 bool done = FALSE;
ea92aad8
KW
3396 if (lang) {
3397 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
3398 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
3399 if (sl_result[LC_ALL_INDEX])
3400 done = TRUE;
3401 else
e5f10d49 3402 setlocale_failure = TRUE;
ea92aad8
KW
3403 }
3404 if (! setlocale_failure) {
3405 const char * locale_param;
3406 for (i = 0; i < LC_ALL_INDEX; i++) {
3407 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
3408 ? setlocale_init
3409 : NULL;
3410 sl_result[i] = do_setlocale_r(categories[i], locale_param);
3411 if (! sl_result[i]) {
3412 setlocale_failure = TRUE;
3413 }
3414 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 3415 }
c835d6be 3416 }
7d4bcc4a
KW
3417 }
3418
3419# endif /* LC_ALL */
e5f10d49 3420# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 3421
65ebb059 3422 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
3423 * the list. Normally the loop is executed just once. But if setting the
3424 * locale fails, inside the loop we add fallback trials to the array and so
3425 * will execute the loop multiple times */
c3fcd832
KW
3426 trial_locales[0] = setlocale_init;
3427 trial_locales_count = 1;
7d4bcc4a 3428
65ebb059
KW
3429 for (i= 0; i < trial_locales_count; i++) {
3430 const char * trial_locale = trial_locales[i];
3431
3432 if (i > 0) {
3433
3434 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
3435 * when i==0, but I (khw) don't think that behavior makes much
3436 * sense */
3437 setlocale_failure = FALSE;
3438
7d4bcc4a 3439# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 3440# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 3441
65ebb059
KW
3442 /* On Windows machines, an entry of "" after the 0th means to use
3443 * the system default locale, which we now proceed to get. */
3444 if (strEQ(trial_locale, "")) {
3445 unsigned int j;
3446
3447 /* Note that this may change the locale, but we are going to do
3448 * that anyway just below */
837ce802 3449 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 3450 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 3451
7d4bcc4a 3452 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
3453 * try */
3454 if (! system_default_locale) {
3455 goto next_iteration;
3456 }
3457 for (j = 0; j < trial_locales_count; j++) {
3458 if (strEQ(system_default_locale, trial_locales[j])) {
3459 goto next_iteration;
3460 }
3461 }
3462
3463 trial_locale = system_default_locale;
3464 }
ec0202b5
KW
3465# else
3466# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
3467# endif
7d4bcc4a 3468# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
3469
3470 } /* For i > 0 */
65ebb059 3471
7d4bcc4a
KW
3472# ifdef LC_ALL
3473
948523db
KW
3474 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
3475 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
3476 if (! sl_result[LC_ALL_INDEX]) {
49c85077 3477 setlocale_failure = TRUE;
7cd8b568
KW
3478 }
3479 else {
3480 /* Since LC_ALL succeeded, it should have changed all the other
3481 * categories it can to its value; so we massage things so that the
3482 * setlocales below just return their category's current values.
3483 * This adequately handles the case in NetBSD where LC_COLLATE may
3484 * not be defined for a locale, and setting it individually will
7d4bcc4a 3485 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
3486 * the POSIX locale. */
3487 trial_locale = NULL;
3488 }
7d4bcc4a
KW
3489
3490# endif /* LC_ALL */
98994639 3491
e5f10d49
KW
3492 if (! setlocale_failure) {
3493 unsigned int j;
3494 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3495 curlocales[j]
3496 = savepv(do_setlocale_r(categories[j], trial_locale));
3497 if (! curlocales[j]) {
3498 setlocale_failure = TRUE;
3499 }
3500 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
3501 }
c835d6be 3502
e5f10d49
KW
3503 if (! setlocale_failure) { /* All succeeded */
3504 break; /* Exit trial_locales loop */
49c85077 3505 }
65ebb059 3506 }
98994639 3507
49c85077
KW
3508 /* Here, something failed; will need to try a fallback. */
3509 ok = 0;
65ebb059 3510
49c85077
KW
3511 if (i == 0) {
3512 unsigned int j;
98994639 3513
65ebb059 3514 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
3515
3516# ifdef LC_ALL
98994639 3517
49c85077
KW
3518 PerlIO_printf(Perl_error_log,
3519 "perl: warning: Setting locale failed.\n");
98994639 3520
7d4bcc4a 3521# else /* !LC_ALL */
98994639 3522
49c85077
KW
3523 PerlIO_printf(Perl_error_log,
3524 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 3525
e5f10d49
KW
3526 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3527 if (! curlocales[j]) {
3528 PerlIO_printf(Perl_error_log, category_names[j]);
3529 }
3530 else {
3531 Safefree(curlocales[j]);
3532 }
3533 }
7d4bcc4a 3534
7d4bcc4a 3535# endif /* LC_ALL */
98994639 3536
49c85077
KW
3537 PerlIO_printf(Perl_error_log,
3538 "perl: warning: Please check that your locale settings:\n");
98994639 3539
7d4bcc4a
KW
3540# ifdef __GLIBC__
3541
49c85077
KW
3542 PerlIO_printf(Perl_error_log,
3543 "\tLANGUAGE = %c%s%c,\n",
3544 language ? '"' : '(',
3545 language ? language : "unset",
3546 language ? '"' : ')');
7d4bcc4a 3547# endif
98994639 3548
49c85077
KW
3549 PerlIO_printf(Perl_error_log,
3550 "\tLC_ALL = %c%s%c,\n",
3551 lc_all ? '"' : '(',
3552 lc_all ? lc_all : "unset",
3553 lc_all ? '"' : ')');
98994639 3554
7d4bcc4a
KW
3555# if defined(USE_ENVIRON_ARRAY)
3556
49c85077 3557 {
cd999af9 3558 char **e;
d5e32b93
KW
3559
3560 /* Look through the environment for any variables of the
3561 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
3562 * already handled above. These are assumed to be locale
3563 * settings. Output them and their values. */
cd999af9 3564 for (e = environ; *e; e++) {
d5e32b93
KW
3565 const STRLEN prefix_len = sizeof("LC_") - 1;
3566 STRLEN uppers_len;
3567
cd999af9 3568 if ( strBEGINs(*e, "LC_")
c8b388b0 3569 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
3570 && (uppers_len = strspn(*e + prefix_len,
3571 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
3572 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
3573 {
3574 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
3575 (int) (prefix_len + uppers_len), *e,
3576 *e + prefix_len + uppers_len + 1);
cd999af9
KW
3577 }
3578 }
49c85077 3579 }
7d4bcc4a
KW
3580
3581# else
3582
49c85077
KW
3583 PerlIO_printf(Perl_error_log,
3584 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
3585
3586# endif
98994639 3587
49c85077
KW
3588 PerlIO_printf(Perl_error_log,
3589 "\tLANG = %c%s%c\n",
3590 lang ? '"' : '(',
3591 lang ? lang : "unset",
3592 lang ? '"' : ')');
98994639 3593
49c85077
KW
3594 PerlIO_printf(Perl_error_log,
3595 " are supported and installed on your system.\n");
3596 }
98994639 3597
65ebb059 3598 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 3599 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
3600 * usually change the upper bound of the loop we are in.
3601 *
3602 * Since the system's default way of setting the locale has not
3603 * found one that works, We use Perl's defined ordering: LC_ALL,
3604 * LANG, and the C locale. We don't try the same locale twice, so
3605 * don't add to the list if already there. (On POSIX systems, the
3606 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
3607 * but there's no harm done by doing it explicitly.
3608 *
3609 * Note that this tries the LC_ALL environment variable even on
3610 * systems which have no LC_ALL locale setting. This may or may
3611 * not have been originally intentional, but there's no real need
3612 * to change the behavior. */
65ebb059
KW
3613 if (lc_all) {
3614 for (j = 0; j < trial_locales_count; j++) {
3615 if (strEQ(lc_all, trial_locales[j])) {
3616 goto done_lc_all;
3617 }
3618 }
3619 trial_locales[trial_locales_count++] = lc_all;
3620 }
3621 done_lc_all:
98994639 3622
65ebb059
KW
3623 if (lang) {
3624 for (j = 0; j < trial_locales_count; j++) {
3625 if (strEQ(lang, trial_locales[j])) {
3626 goto done_lang;
3627 }
3628 }
3629 trial_locales[trial_locales_count++] = lang;
3630 }
3631 done_lang:
3632
7d4bcc4a
KW
3633# if defined(WIN32) && defined(LC_ALL)
3634
65ebb059
KW
3635 /* For Windows, we also try the system default locale before "C".
3636 * (If there exists a Windows without LC_ALL we skip this because
3637 * it gets too complicated. For those, the "C" is the next
3638 * fallback possibility). The "" is the same as the 0th element of
3639 * the array, but the code at the loop above knows to treat it
3640 * differently when not the 0th */
3641 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
3642
3643# endif
65ebb059
KW
3644
3645 for (j = 0; j < trial_locales_count; j++) {
3646 if (strEQ("C", trial_locales[j])) {
3647 goto done_C;
3648 }
3649 }
3650 trial_locales[trial_locales_count++] = "C";
98994639 3651
65ebb059
KW
3652 done_C: ;
3653 } /* end of first time through the loop */
98994639 3654
7d4bcc4a
KW
3655# ifdef WIN32
3656
65ebb059 3657 next_iteration: ;
7d4bcc4a
KW
3658
3659# endif
65ebb059
KW
3660
3661 } /* end of looping through the trial locales */
3662
3663 if (ok < 1) { /* If we tried to fallback */
3664 const char* msg;
3665 if (! setlocale_failure) { /* fallback succeeded */
3666 msg = "Falling back to";
3667 }
3668 else { /* fallback failed */
e5f10d49 3669 unsigned int j;
98994639 3670
65ebb059
KW
3671 /* We dropped off the end of the loop, so have to decrement i to
3672 * get back to the value the last time through */
3673 i--;
98994639 3674
65ebb059
KW
3675 ok = -1;
3676 msg = "Failed to fall back to";
3677
3678 /* To continue, we should use whatever values we've got */
7d4bcc4a 3679
e5f10d49
KW
3680 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
3681 Safefree(curlocales[j]);
3682 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
3683 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
3684 }
65ebb059
KW
3685 }
3686
3687 if (locwarn) {
3688 const char * description;
3689 const char * name = "";
3690 if (strEQ(trial_locales[i], "C")) {
3691 description = "the standard locale";
3692 name = "C";
3693 }
7d4bcc4a
KW
3694
3695# ifdef SYSTEM_DEFAULT_LOCALE
3696
65ebb059
KW
3697 else if (strEQ(trial_locales[i], "")) {
3698 description = "the system default locale";
3699 if (system_default_locale) {
3700 name = system_default_locale;
3701 }
3702 }
7d4bcc4a
KW
3703
3704# endif /* SYSTEM_DEFAULT_LOCALE */
3705
65ebb059
KW
3706 else {
3707 description = "a fallback locale";
3708 name = trial_locales[i];
3709 }
3710 if (name && strNE(name, "")) {
3711 PerlIO_printf(Perl_error_log,
3712 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
3713 }
3714 else {
3715 PerlIO_printf(Perl_error_log,
3716 "perl: warning: %s %s.\n", msg, description);
3717 }
3718 }
3719 } /* End of tried to fallback */
98994639 3720
e5f10d49
KW
3721 /* Done with finding the locales; update our records */
3722
7d4bcc4a
KW
3723# ifdef USE_LOCALE_CTYPE
3724
948523db 3725 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 3726
e5f10d49 3727# endif
7d4bcc4a
KW
3728# ifdef USE_LOCALE_COLLATE
3729
948523db 3730 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 3731
e5f10d49 3732# endif
7d4bcc4a
KW
3733# ifdef USE_LOCALE_NUMERIC
3734
948523db 3735 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
3736
3737# endif
3738
948523db 3739 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20 3740
e9bc6d6b 3741# if defined(USE_ITHREADS) && ! defined(USE_THREAD_SAFE_LOCALE)
47280b20
KW
3742
3743 /* This caches whether each category's locale is UTF-8 or not. This
3744 * may involve changing the locale. It is ok to do this at
e9bc6d6b
KW
3745 * initialization time before any threads have started, but not later
3746 * unless thread-safe operations are used.
47280b20
KW
3747 * Caching means that if the program heeds our dictate not to change
3748 * locales in threaded applications, this data will remain valid, and
9de6fe47
KW
3749 * it may get queried without having to change locales. If the
3750 * environment is such that all categories have the same locale, this
3751 * isn't needed, as the code will not change the locale; but this
3752 * handles the uncommon case where the environment has disparate
3753 * locales for the categories */
47280b20
KW
3754 (void) _is_cur_LC_category_utf8(categories[i]);
3755
3756# endif
3757
e5f10d49
KW
3758 Safefree(curlocales[i]);
3759 }
b310b053 3760
7d4bcc4a
KW
3761# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
3762
49c85077 3763 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
3764 * locale is UTF-8. The call to new_ctype() just above has already
3765 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
3766 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
3767 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
3768 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
3769 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 3770
a05d7ebb 3771 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
3772 This is an alternative to using the -C command line switch
3773 (the -C if present will override this). */
3774 {
dd374669 3775 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 3776 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
3777 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
3778 PL_utf8cache = -1;
b310b053
JH
3779 }
3780
7d4bcc4a 3781# endif
7d4bcc4a
KW
3782# ifdef __GLIBC__
3783
175c4cf9 3784 Safefree(language);
7d4bcc4a
KW
3785
3786# endif
175c4cf9
KW
3787
3788 Safefree(lc_all);
3789 Safefree(lang);
3790
e3305790 3791#endif /* USE_LOCALE */
2fcc0ca9 3792#ifdef DEBUGGING
7d4bcc4a 3793
2fcc0ca9 3794 /* So won't continue to output stuff */
27cdc72e 3795 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 3796
2fcc0ca9
KW
3797#endif
3798
98994639
HS
3799 return ok;
3800}
3801
98994639
HS
3802#ifdef USE_LOCALE_COLLATE
3803
a4a439fb 3804char *
a4a439fb
KW
3805Perl__mem_collxfrm(pTHX_ const char *input_string,
3806 STRLEN len, /* Length of 'input_string' */
3807 STRLEN *xlen, /* Set to length of returned string
3808 (not including the collation index
3809 prefix) */
3810 bool utf8 /* Is the input in UTF-8? */
6696cfa7 3811 )
98994639 3812{
a4a439fb
KW
3813
3814 /* _mem_collxfrm() is a bit like strxfrm() but with two important
3815 * differences. First, it handles embedded NULs. Second, it allocates a bit
3816 * more memory than needed for the transformed data itself. The real
55e5378d 3817 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
3818 * the length of that, and doesn't include the collation index size.
3819 * Please see sv_collxfrm() to see how this is used. */
3820
55e5378d
KW
3821#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
3822
6696cfa7
KW
3823 char * s = (char *) input_string;
3824 STRLEN s_strlen = strlen(input_string);
79f120c8 3825 char *xbuf = NULL;
55e5378d 3826 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 3827 STRLEN length_in_chars;
c664130f 3828 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 3829
a4a439fb
KW
3830 PERL_ARGS_ASSERT__MEM_COLLXFRM;
3831
3832 /* Must be NUL-terminated */
3833 assert(*(input_string + len) == '\0');
7918f24d 3834
79f120c8
KW
3835 /* If this locale has defective collation, skip */
3836 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
3837 DEBUG_L(PerlIO_printf(Perl_debug_log,
3838 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
3839 goto bad;
3840 }
3841
6696cfa7
KW
3842 /* Replace any embedded NULs with the control that sorts before any others.
3843 * This will give as good as possible results on strings that don't
3844 * otherwise contain that character, but otherwise there may be
3845 * less-than-perfect results with that character and NUL. This is
fdc080f3 3846 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
3847 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
3848 NUL */
6696cfa7
KW
3849 char * e = s + len;
3850 char * sans_nuls;
fdc080f3 3851 STRLEN sans_nuls_len;
94762aa0 3852 int try_non_controls;
afc4976f
KW
3853 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
3854 making sure 2nd byte is NUL.
3855 */
3856 STRLEN this_replacement_len;
3857
1e4c9676
KW
3858 /* If we don't know what non-NUL control character sorts lowest for
3859 * this locale, find it */
f28f4d2a 3860 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 3861 int j;
afc4976f 3862 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW