This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
locale.c: Do common task in one place
[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'.
166f8a29
DM
35 */
36
98994639
HS
37#include "EXTERN.h"
38#define PERL_IN_LOCALE_C
f7416781 39#include "perl_langinfo.h"
98994639
HS
40#include "perl.h"
41
a4af207c
JH
42#include "reentr.h"
43
2fcc0ca9
KW
44/* If the environment says to, we can output debugging information during
45 * initialization. This is done before option parsing, and before any thread
46 * creation, so can be a file-level static */
5a4b0634
KW
47#if ! defined(DEBUGGING) || defined(PERL_GLOBAL_STRUCT)
48# define debug_initialization 0
49# define DEBUG_INITIALIZATION_set(v)
50#else
2fcc0ca9 51static bool debug_initialization = FALSE;
5a4b0634 52# define DEBUG_INITIALIZATION_set(v) (debug_initialization = v)
2fcc0ca9
KW
53#endif
54
291a84fb
KW
55/* strlen() of a literal string constant. We might want this more general,
56 * but using it in just this file for now. A problem with more generality is
57 * the compiler warnings about comparing unlike signs */
ff1b739b
KW
58#define STRLENs(s) (sizeof("" s "") - 1)
59
63e5b0d7
KW
60/* Is the C string input 'name' "C" or "POSIX"? If so, and 'name' is the
61 * return of setlocale(), then this is extremely likely to be the C or POSIX
62 * locale. However, the output of setlocale() is documented to be opaque, but
63 * the odds are extremely small that it would return these two strings for some
64 * other locale. Note that VMS in these two locales includes many non-ASCII
65 * characters as controls and punctuation (below are hex bytes):
66 * cntrl: 84-97 9B-9F
67 * punct: A1-A3 A5 A7-AB B0-B3 B5-B7 B9-BD BF-CF D1-DD DF-EF F1-FD
68 * Oddly, none there are listed as alphas, though some represent alphabetics
69 * http://www.nntp.perl.org/group/perl.perl5.porters/2013/02/msg198753.html */
70#define isNAME_C_OR_POSIX(name) \
71 ( (name) != NULL \
72 && (( *(name) == 'C' && (*(name + 1)) == '\0') \
73 || strEQ((name), "POSIX")))
74
8ef6e574
KW
75#ifdef USE_LOCALE
76
47280b20
KW
77/* This code keeps a LRU cache of the UTF-8ness of the locales it has so-far
78 * looked up. This is in the form of a C string: */
79
80#define UTF8NESS_SEP "\v"
81#define UTF8NESS_PREFIX "\f"
82
83/* So, the string looks like:
98994639 84 *
47280b20 85 * \vC\a0\vPOSIX\a0\vam_ET\a0\vaf_ZA.utf8\a1\ven_US.UTF-8\a1\0
98994639 86 *
47280b20
KW
87 * where the digit 0 after the \a indicates that the locale starting just
88 * after the preceding \v is not UTF-8, and the digit 1 mean it is. */
89
90STATIC_ASSERT_DECL(STRLENs(UTF8NESS_SEP) == 1);
91STATIC_ASSERT_DECL(STRLENs(UTF8NESS_PREFIX) == 1);
92
93#define C_and_POSIX_utf8ness UTF8NESS_SEP "C" UTF8NESS_PREFIX "0" \
94 UTF8NESS_SEP "POSIX" UTF8NESS_PREFIX "0"
95
96/* The cache is initialized to C_and_POSIX_utf8ness at start up. These are
97 * kept there always. The remining portion of the cache is LRU, with the
98 * oldest looked-up locale at the tail end */
99
98994639
HS
100STATIC char *
101S_stdize_locale(pTHX_ char *locs)
102{
47280b20
KW
103 /* Standardize the locale name from a string returned by 'setlocale',
104 * possibly modifying that string.
105 *
106 * The typical return value of setlocale() is either
107 * (1) "xx_YY" if the first argument of setlocale() is not LC_ALL
108 * (2) "xa_YY xb_YY ..." if the first argument of setlocale() is LC_ALL
109 * (the space-separated values represent the various sublocales,
110 * in some unspecified order). This is not handled by this function.
111 *
112 * In some platforms it has a form like "LC_SOMETHING=Lang_Country.866\n",
113 * which is harmful for further use of the string in setlocale(). This
114 * function removes the trailing new line and everything up through the '='
115 * */
116
7452cf6a 117 const char * const s = strchr(locs, '=');
98994639
HS
118 bool okay = TRUE;
119
7918f24d
NC
120 PERL_ARGS_ASSERT_STDIZE_LOCALE;
121
8772537c
AL
122 if (s) {
123 const char * const t = strchr(s, '.');
98994639 124 okay = FALSE;
8772537c
AL
125 if (t) {
126 const char * const u = strchr(t, '\n');
127 if (u && (u[1] == 0)) {
128 const STRLEN len = u - s;
129 Move(s + 1, locs, len, char);
130 locs[len] = 0;
131 okay = TRUE;
98994639
HS
132 }
133 }
134 }
135
136 if (!okay)
137 Perl_croak(aTHX_ "Can't fix broken locale name \"%s\"", locs);
138
139 return locs;
140}
141
e5f10d49
KW
142/* Two parallel arrays; first the locale categories Perl uses on this system;
143 * the second array is their names. These arrays are in mostly arbitrary
144 * order. */
145
146const int categories[] = {
147
148# ifdef USE_LOCALE_NUMERIC
149 LC_NUMERIC,
150# endif
151# ifdef USE_LOCALE_CTYPE
152 LC_CTYPE,
153# endif
154# ifdef USE_LOCALE_COLLATE
155 LC_COLLATE,
156# endif
157# ifdef USE_LOCALE_TIME
158 LC_TIME,
159# endif
160# ifdef USE_LOCALE_MESSAGES
161 LC_MESSAGES,
162# endif
163# ifdef USE_LOCALE_MONETARY
164 LC_MONETARY,
165# endif
9821811f
KW
166# ifdef USE_LOCALE_ADDRESS
167 LC_ADDRESS,
168# endif
169# ifdef USE_LOCALE_IDENTIFICATION
170 LC_IDENTIFICATION,
171# endif
172# ifdef USE_LOCALE_MEASUREMENT
173 LC_MEASUREMENT,
174# endif
175# ifdef USE_LOCALE_PAPER
176 LC_PAPER,
177# endif
178# ifdef USE_LOCALE_TELEPHONE
179 LC_TELEPHONE,
180# endif
e5f10d49
KW
181# ifdef LC_ALL
182 LC_ALL,
183# endif
184 -1 /* Placeholder because C doesn't allow a
185 trailing comma, and it would get complicated
186 with all the #ifdef's */
187};
188
189/* The top-most real element is LC_ALL */
190
191const char * category_names[] = {
192
193# ifdef USE_LOCALE_NUMERIC
194 "LC_NUMERIC",
195# endif
196# ifdef USE_LOCALE_CTYPE
197 "LC_CTYPE",
198# endif
199# ifdef USE_LOCALE_COLLATE
200 "LC_COLLATE",
201# endif
202# ifdef USE_LOCALE_TIME
203 "LC_TIME",
204# endif
205# ifdef USE_LOCALE_MESSAGES
206 "LC_MESSAGES",
207# endif
208# ifdef USE_LOCALE_MONETARY
209 "LC_MONETARY",
210# endif
9821811f
KW
211# ifdef USE_LOCALE_ADDRESS
212 "LC_ADDRESS",
213# endif
214# ifdef USE_LOCALE_IDENTIFICATION
215 "LC_IDENTIFICATION",
216# endif
217# ifdef USE_LOCALE_MEASUREMENT
218 "LC_MEASUREMENT",
219# endif
220# ifdef USE_LOCALE_PAPER
221 "LC_PAPER",
222# endif
223# ifdef USE_LOCALE_TELEPHONE
224 "LC_TELEPHONE",
225# endif
e5f10d49
KW
226# ifdef LC_ALL
227 "LC_ALL",
228# endif
229 NULL /* Placeholder */
230 };
231
232# ifdef LC_ALL
233
234 /* On systems with LC_ALL, it is kept in the highest index position. (-2
235 * to account for the final unused placeholder element.) */
236# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 2)
237
238# else
239
240 /* On systems without LC_ALL, we pretend it is there, one beyond the real
241 * top element, hence in the unused placeholder element. */
242# define NOMINAL_LC_ALL_INDEX (C_ARRAY_LENGTH(categories) - 1)
243
244# endif
245
246/* Pretending there is an LC_ALL element just above allows us to avoid most
247 * special cases. Most loops through these arrays in the code below are
248 * written like 'for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++)'. They will work
249 * on either type of system. But the code must be written to not access the
948523db
KW
250 * element at 'LC_ALL_INDEX' except on platforms that have it. This can be
251 * checked for at compile time by using the #define LC_ALL_INDEX which is only
252 * defined if we do have LC_ALL. */
e5f10d49 253
b09aaf40
KW
254STATIC const char *
255S_category_name(const int category)
256{
257 unsigned int i;
258
259#ifdef LC_ALL
260
261 if (category == LC_ALL) {
262 return "LC_ALL";
263 }
264
265#endif
266
267 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
268 if (category == categories[i]) {
269 return category_names[i];
270 }
271 }
272
273 {
274 const char suffix[] = " (unknown)";
275 int temp = category;
276 Size_t length = sizeof(suffix) + 1;
277 char * unknown;
278 dTHX;
279
280 if (temp < 0) {
281 length++;
282 temp = - temp;
283 }
284
285 /* Calculate the number of digits */
286 while (temp >= 10) {
287 temp /= 10;
288 length++;
289 }
290
291 Newx(unknown, length, char);
292 my_snprintf(unknown, length, "%d%s", category, suffix);
293 SAVEFREEPV(unknown);
294 return unknown;
295 }
296}
297
948523db
KW
298/* Now create LC_foo_INDEX #defines for just those categories on this system */
299# ifdef USE_LOCALE_NUMERIC
300# define LC_NUMERIC_INDEX 0
301# define _DUMMY_NUMERIC LC_NUMERIC_INDEX
302# else
303# define _DUMMY_NUMERIC -1
304# endif
305# ifdef USE_LOCALE_CTYPE
306# define LC_CTYPE_INDEX _DUMMY_NUMERIC + 1
307# define _DUMMY_CTYPE LC_CTYPE_INDEX
308# else
309# define _DUMMY_CTYPE _DUMMY_NUMERIC
310# endif
311# ifdef USE_LOCALE_COLLATE
312# define LC_COLLATE_INDEX _DUMMY_CTYPE + 1
313# define _DUMMY_COLLATE LC_COLLATE_INDEX
314# else
315# define _DUMMY_COLLATE _DUMMY_COLLATE
316# endif
317# ifdef USE_LOCALE_TIME
318# define LC_TIME_INDEX _DUMMY_COLLATE + 1
319# define _DUMMY_TIME LC_TIME_INDEX
320# else
321# define _DUMMY_TIME _DUMMY_COLLATE
322# endif
323# ifdef USE_LOCALE_MESSAGES
324# define LC_MESSAGES_INDEX _DUMMY_TIME + 1
325# define _DUMMY_MESSAGES LC_MESSAGES_INDEX
326# else
327# define _DUMMY_MESSAGES _DUMMY_TIME
328# endif
329# ifdef USE_LOCALE_MONETARY
330# define LC_MONETARY_INDEX _DUMMY_MESSAGES + 1
331# define _DUMMY_MONETARY LC_MONETARY_INDEX
332# else
333# define _DUMMY_MONETARY _DUMMY_MESSAGES
334# endif
9821811f
KW
335# ifdef USE_LOCALE_ADDRESS
336# define LC_ADDRESS_INDEX _DUMMY_MONETARY + 1
337# define _DUMMY_ADDRESS LC_ADDRESS_INDEX
338# else
339# define _DUMMY_ADDRESS _DUMMY_MONETARY
340# endif
341# ifdef USE_LOCALE_IDENTIFICATION
342# define LC_IDENTIFICATION_INDEX _DUMMY_ADDRESS + 1
343# define _DUMMY_IDENTIFICATION LC_IDENTIFICATION_INDEX
344# else
345# define _DUMMY_IDENTIFICATION _DUMMY_ADDRESS
346# endif
347# ifdef USE_LOCALE_MEASUREMENT
348# define LC_MEASUREMENT_INDEX _DUMMY_IDENTIFICATION + 1
349# define _DUMMY_MEASUREMENT LC_MEASUREMENT_INDEX
350# else
351# define _DUMMY_MEASUREMENT _DUMMY_IDENTIFICATION
352# endif
353# ifdef USE_LOCALE_PAPER
354# define LC_PAPER_INDEX _DUMMY_MEASUREMENT + 1
355# define _DUMMY_PAPER LC_PAPER_INDEX
356# else
357# define _DUMMY_PAPER _DUMMY_MEASUREMENT
358# endif
359# ifdef USE_LOCALE_TELEPHONE
360# define LC_TELEPHONE_INDEX _DUMMY_PAPER + 1
361# define _DUMMY_TELEPHONE LC_TELEPHONE_INDEX
362# else
363# define _DUMMY_TELEPHONE _DUMMY_PAPER
364# endif
948523db 365# ifdef LC_ALL
9821811f 366# define LC_ALL_INDEX _DUMMY_TELEPHONE + 1
948523db
KW
367# endif
368#endif /* ifdef USE_LOCALE */
8ef6e574 369
d2b24094
KW
370/* Windows requres a customized base-level setlocale() */
371# ifdef WIN32
372# define my_setlocale(cat, locale) win32_setlocale(cat, locale)
373# else
374# define my_setlocale(cat, locale) setlocale(cat, locale)
375# endif
376
837ce802
KW
377/* Just placeholders for now. "_c" is intended to be called when the category
378 * is a constant known at compile time; "_r", not known until run time */
d2b24094
KW
379# define do_setlocale_c(category, locale) my_setlocale(category, locale)
380# define do_setlocale_r(category, locale) my_setlocale(category, locale)
837ce802 381
a4f00dcc 382STATIC void
86799d2d 383S_set_numeric_radix(pTHX_ const bool use_locale)
98994639 384{
86799d2d
KW
385 /* If 'use_locale' is FALSE, set to use a dot for the radix character. If
386 * TRUE, use the radix character derived from the current locale */
7d4bcc4a 387
86799d2d
KW
388#if defined(USE_LOCALE_NUMERIC) && ( defined(HAS_LOCALECONV) \
389 || defined(HAS_NL_LANGINFO))
98994639 390
86799d2d
KW
391 /* We only set up the radix SV if we are to use a locale radix ... */
392 if (use_locale) {
393 const char * radix = my_nl_langinfo(PERL_RADIXCHAR, FALSE);
394 /* FALSE => already in dest locale */
86799d2d
KW
395 /* ... and the character being used isn't a dot */
396 if (strNE(radix, ".")) {
2213a3be
KW
397 const U8 * first_variant;
398
86799d2d
KW
399 if (PL_numeric_radix_sv) {
400 sv_setpv(PL_numeric_radix_sv, radix);
401 }
402 else {
403 PL_numeric_radix_sv = newSVpv(radix, 0);
404 }
405
2213a3be
KW
406 /* If there is a byte variant under UTF-8, and if the remainder of
407 * the string starting there is valid UTF-8, and we are in a UTF-8
408 * locale, then mark the radix as being in UTF-8 */
409 if ( ! is_utf8_invariant_string_loc(
410 (U8 *) SvPVX(PL_numeric_radix_sv),
411 SvCUR(PL_numeric_radix_sv),
412 &first_variant)
413 && is_utf8_string(first_variant,
414 SvCUR(PL_numeric_radix_sv)
415 - ((char *) first_variant
416 - SvPVX(PL_numeric_radix_sv)))
c1284011 417 && _is_cur_LC_category_utf8(LC_NUMERIC))
28acfe03 418 {
86799d2d 419 SvUTF8_on(PL_numeric_radix_sv);
28acfe03 420 }
86799d2d
KW
421 goto done;
422 }
98994639 423 }
69014004 424
86799d2d
KW
425 SvREFCNT_dec(PL_numeric_radix_sv);
426 PL_numeric_radix_sv = NULL;
427
428 done: ;
429
430# ifdef DEBUGGING
7d4bcc4a 431
2fcc0ca9
KW
432 if (DEBUG_L_TEST || debug_initialization) {
433 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
69014004 434 (PL_numeric_radix_sv)
37b7e435
KW
435 ? SvPVX(PL_numeric_radix_sv)
436 : "NULL",
437 (PL_numeric_radix_sv)
39eb7305 438 ? cBOOL(SvUTF8(PL_numeric_radix_sv))
2fcc0ca9
KW
439 : 0);
440 }
69014004 441
86799d2d
KW
442# endif
443#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 444
98994639
HS
445}
446
a39edc4c 447
98994639 448void
8772537c 449Perl_new_numeric(pTHX_ const char *newnum)
98994639 450{
7d4bcc4a
KW
451
452#ifndef USE_LOCALE_NUMERIC
453
454 PERL_UNUSED_ARG(newnum);
455
456#else
0d071d52 457
291a84fb 458 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
459 * core Perl this and that 'newnum' is the name of the new locale.
460 * It installs this locale as the current underlying default.
461 *
462 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
463 * set_numeric_underlying() and set_numeric_standard() functions, which
464 * should probably not be called directly, but only via macros like
0d071d52
KW
465 * SET_NUMERIC_STANDARD() in perl.h.
466 *
467 * The toggling is necessary mainly so that a non-dot radix decimal point
468 * character can be output, while allowing internal calculations to use a
469 * dot.
470 *
471 * This sets several interpreter-level variables:
bb304765 472 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 473 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
474 * that the current locale is the program's underlying
475 * locale
476 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
477 * that the current locale is the C locale or
478 * indistinguishable from the C locale. If non-zero, it
479 * is in C; if > 1, it means it may not be toggled away
7738054c 480 * from C.
4c68b815
KW
481 * PL_numeric_underlying_is_standard A bool kept by this function
482 * indicating that the underlying locale and the standard
483 * C locale are indistinguishable for the purposes of
484 * LC_NUMERIC. This happens when both of the above two
485 * variables are true at the same time. (Toggling is a
486 * no-op under these circumstances.) This variable is
487 * used to avoid having to recalculate.
0d071d52
KW
488 * Any code changing the locale (outside this file) should use
489 * POSIX::setlocale, which calls this function. Therefore this function
490 * should be called directly only from this file and from
491 * POSIX::setlocale() */
492
b03f34cf 493 char *save_newnum;
98994639
HS
494
495 if (! newnum) {
43c5f42d
NC
496 Safefree(PL_numeric_name);
497 PL_numeric_name = NULL;
98994639 498 PL_numeric_standard = TRUE;
892e6465 499 PL_numeric_underlying = TRUE;
4c68b815 500 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
501 return;
502 }
503
b03f34cf 504 save_newnum = stdize_locale(savepv(newnum));
892e6465 505 PL_numeric_underlying = TRUE;
4c68b815
KW
506 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
507
508 /* If its name isn't C nor POSIX, it could still be indistinguishable from
509 * them */
510 if (! PL_numeric_standard) {
511 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
512 FALSE /* Don't toggle locale */ ))
513 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
514 FALSE)));
515 }
abe1abcf 516
4c68b815 517 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 518 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 519 Safefree(PL_numeric_name);
b03f34cf 520 PL_numeric_name = save_newnum;
b03f34cf 521 }
abe1abcf
KW
522 else {
523 Safefree(save_newnum);
524 }
4c28b29c 525
4c68b815
KW
526 PL_numeric_underlying_is_standard = PL_numeric_standard;
527
528 if (DEBUG_L_TEST || debug_initialization) {
529 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
530 }
531
4c28b29c
KW
532 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
533 * have to worry about the radix being a non-dot. (Core operations that
534 * need the underlying locale change to it temporarily). */
535 set_numeric_standard();
536
98994639 537#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 538
98994639
HS
539}
540
541void
542Perl_set_numeric_standard(pTHX)
543{
7d4bcc4a 544
98994639 545#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 546
28c1bf33
KW
547 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
548 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
549 * macro avoids calling this routine if toggling isn't necessary according
550 * to our records (which could be wrong if some XS code has changed the
551 * locale behind our back) */
0d071d52 552
837ce802 553 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 554 PL_numeric_standard = TRUE;
4c68b815 555 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 556 set_numeric_radix(0);
7d4bcc4a
KW
557
558# ifdef DEBUGGING
559
2fcc0ca9
KW
560 if (DEBUG_L_TEST || debug_initialization) {
561 PerlIO_printf(Perl_debug_log,
58e4a467 562 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 563 }
98994639 564
7d4bcc4a 565# endif
98994639 566#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 567
98994639
HS
568}
569
570void
5792c642 571Perl_set_numeric_underlying(pTHX)
98994639 572{
7d4bcc4a 573
98994639 574#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 575
28c1bf33 576 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
577 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
578 * instead of calling this directly. The macro avoids calling this routine
579 * if toggling isn't necessary according to our records (which could be
580 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 581
837ce802 582 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 583 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 584 PL_numeric_underlying = TRUE;
3c62fd46 585 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
586
587# ifdef DEBUGGING
588
2fcc0ca9
KW
589 if (DEBUG_L_TEST || debug_initialization) {
590 PerlIO_printf(Perl_debug_log,
58e4a467 591 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
592 PL_numeric_name);
593 }
98994639 594
7d4bcc4a 595# endif
98994639 596#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 597
98994639
HS
598}
599
600/*
601 * Set up for a new ctype locale.
602 */
a4f00dcc
KW
603STATIC void
604S_new_ctype(pTHX_ const char *newctype)
98994639 605{
7d4bcc4a
KW
606
607#ifndef USE_LOCALE_CTYPE
608
609 PERL_ARGS_ASSERT_NEW_CTYPE;
610 PERL_UNUSED_ARG(newctype);
611 PERL_UNUSED_CONTEXT;
612
613#else
0d071d52 614
291a84fb 615 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
616 * core Perl this and that 'newctype' is the name of the new locale.
617 *
618 * This function sets up the folding arrays for all 256 bytes, assuming
619 * that tofold() is tolc() since fold case is not a concept in POSIX,
620 *
621 * Any code changing the locale (outside this file) should use
622 * POSIX::setlocale, which calls this function. Therefore this function
623 * should be called directly only from this file and from
624 * POSIX::setlocale() */
625
27da23d5 626 dVAR;
68067e4e 627 UV i;
98994639 628
7918f24d
NC
629 PERL_ARGS_ASSERT_NEW_CTYPE;
630
215c5139
KW
631 /* We will replace any bad locale warning with 1) nothing if the new one is
632 * ok; or 2) a new warning for the bad new locale */
633 if (PL_warn_locale) {
634 SvREFCNT_dec_NN(PL_warn_locale);
635 PL_warn_locale = NULL;
636 }
637
c1284011 638 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
639
640 /* A UTF-8 locale gets standard rules. But note that code still has to
641 * handle this specially because of the three problematic code points */
642 if (PL_in_utf8_CTYPE_locale) {
643 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
644 }
645 else {
8c6180a9
KW
646 /* Assume enough space for every character being bad. 4 spaces each
647 * for the 94 printable characters that are output like "'x' "; and 5
648 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
649 * NUL */
650 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
651
cc9eaeb0
KW
652 /* Don't check for problems if we are suppressing the warnings */
653 bool check_for_problems = ckWARN_d(WARN_LOCALE)
654 || UNLIKELY(DEBUG_L_TEST);
8c6180a9
KW
655 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
656 to start */
657 unsigned int bad_count = 0; /* Count of bad characters */
658
baa60164 659 for (i = 0; i < 256; i++) {
5a8a8f21
KW
660 if (isupper(i))
661 PL_fold_locale[i] = (U8) tolower(i);
662 else if (islower(i))
663 PL_fold_locale[i] = (U8) toupper(i);
baa60164
KW
664 else
665 PL_fold_locale[i] = (U8) i;
8c6180a9
KW
666
667 /* If checking for locale problems, see if the native ASCII-range
668 * printables plus \n and \t are in their expected categories in
669 * the new locale. If not, this could mean big trouble, upending
670 * Perl's and most programs' assumptions, like having a
671 * metacharacter with special meaning become a \w. Fortunately,
672 * it's very rare to find locales that aren't supersets of ASCII
673 * nowadays. It isn't a problem for most controls to be changed
674 * into something else; we check only \n and \t, though perhaps \r
675 * could be an issue as well. */
7d4bcc4a 676 if ( check_for_problems
8c6180a9
KW
677 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
678 {
adf947ba
KW
679 if ( cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC(i))
680 || cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i))
681 || cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i))
682 || cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i))
683 || cBOOL(islower(i)) != cBOOL(isLOWER_A(i))
684 || cBOOL(isprint(i)) != cBOOL(isPRINT_A(i))
685 || cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i))
686 || cBOOL(isspace(i)) != cBOOL(isSPACE_A(i))
687 || cBOOL(isupper(i)) != cBOOL(isUPPER_A(i))
688 || cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i))
689 || tolower(i) != (int) toLOWER_A(i)
690 || toupper(i) != (int) toUPPER_A(i)
691 || (i == '\n' && ! isCNTRL_LC(i)))
8c6180a9
KW
692 {
693 if (bad_count) { /* Separate multiple entries with a
694 blank */
695 bad_chars_list[bad_count++] = ' ';
696 }
697 bad_chars_list[bad_count++] = '\'';
698 if (isPRINT_A(i)) {
699 bad_chars_list[bad_count++] = (char) i;
700 }
701 else {
702 bad_chars_list[bad_count++] = '\\';
703 if (i == '\n') {
704 bad_chars_list[bad_count++] = 'n';
705 }
706 else {
707 assert(i == '\t');
708 bad_chars_list[bad_count++] = 't';
709 }
710 }
711 bad_chars_list[bad_count++] = '\'';
712 bad_chars_list[bad_count] = '\0';
713 }
714 }
715 }
716
7d4bcc4a
KW
717# ifdef MB_CUR_MAX
718
8c6180a9 719 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 720 * this locale requires more than one byte, there are going to be
8c6180a9 721 * problems. */
9c8a6dc2
KW
722 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
723 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
724 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
725
ba1a4362
KW
726 if (check_for_problems && MB_CUR_MAX > 1
727
728 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
729 * locale. Just assume that the implementation for them (plus
730 * for POSIX) is correct and the > 1 value is spurious. (Since
731 * these are specially handled to never be considered UTF-8
732 * locales, as long as this is the only problem, everything
733 * should work fine */
734 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
735 {
8c6180a9
KW
736 multi_byte_locale = TRUE;
737 }
7d4bcc4a
KW
738
739# endif
8c6180a9
KW
740
741 if (bad_count || multi_byte_locale) {
780fcc9f 742 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 743 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 744 newctype,
8c6180a9
KW
745 (multi_byte_locale)
746 ? " Some characters in it are not recognized by"
747 " Perl."
748 : "",
749 (bad_count)
750 ? "\nThe following characters (and maybe others)"
751 " may not have the same meaning as the Perl"
752 " program expects:\n"
753 : "",
754 (bad_count)
755 ? bad_chars_list
756 : ""
757 );
cc9eaeb0 758 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
759 * output the message now. If not in that scope, we save the
760 * message to be output at the first operation using this locale,
761 * if that actually happens. Most programs don't use locales, so
762 * they are immune to bad ones. */
cc9eaeb0 763 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f
KW
764
765 /* We have to save 'newctype' because the setlocale() just
766 * below may destroy it. The next setlocale() further down
767 * should restore it properly so that the intermediate change
768 * here is transparent to this function's caller */
769 const char * const badlocale = savepv(newctype);
770
837ce802 771 do_setlocale_c(LC_CTYPE, "C");
780fcc9f
KW
772
773 /* The '0' below suppresses a bogus gcc compiler warning */
774 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 775
837ce802 776 do_setlocale_c(LC_CTYPE, badlocale);
c0f3a893 777 Safefree(badlocale);
bddebb56
KW
778
779 if (IN_LC(LC_CTYPE)) {
780 SvREFCNT_dec_NN(PL_warn_locale);
781 PL_warn_locale = NULL;
782 }
780fcc9f 783 }
baa60164 784 }
31f05a37 785 }
98994639
HS
786
787#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 788
98994639
HS
789}
790
98994639 791void
2726666d
KW
792Perl__warn_problematic_locale()
793{
2726666d
KW
794
795#ifdef USE_LOCALE_CTYPE
796
5f04a188
KW
797 dTHX;
798
799 /* Internal-to-core function that outputs the message in PL_warn_locale,
800 * and then NULLS it. Should be called only through the macro
801 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
802
2726666d 803 if (PL_warn_locale) {
2726666d
KW
804 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
805 SvPVX(PL_warn_locale),
806 0 /* dummy to avoid compiler warning */ );
2726666d
KW
807 SvREFCNT_dec_NN(PL_warn_locale);
808 PL_warn_locale = NULL;
809 }
810
811#endif
812
813}
814
a4f00dcc
KW
815STATIC void
816S_new_collate(pTHX_ const char *newcoll)
98994639 817{
7d4bcc4a
KW
818
819#ifndef USE_LOCALE_COLLATE
820
821 PERL_UNUSED_ARG(newcoll);
822 PERL_UNUSED_CONTEXT;
823
824#else
0d071d52 825
291a84fb 826 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
827 * core Perl this and that 'newcoll' is the name of the new locale.
828 *
d35fca5f
KW
829 * The design of locale collation is that every locale change is given an
830 * index 'PL_collation_ix'. The first time a string particpates in an
831 * operation that requires collation while locale collation is active, it
832 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
833 * magic includes the collation index, and the transformation of the string
834 * by strxfrm(), q.v. That transformation is used when doing comparisons,
835 * instead of the string itself. If a string changes, the magic is
836 * cleared. The next time the locale changes, the index is incremented,
837 * and so we know during a comparison that the transformation is not
838 * necessarily still valid, and so is recomputed. Note that if the locale
839 * changes enough times, the index could wrap (a U32), and it is possible
840 * that a transformation would improperly be considered valid, leading to
841 * an unlikely bug */
0d071d52 842
98994639
HS
843 if (! newcoll) {
844 if (PL_collation_name) {
845 ++PL_collation_ix;
846 Safefree(PL_collation_name);
847 PL_collation_name = NULL;
848 }
849 PL_collation_standard = TRUE;
00bf60ca 850 is_standard_collation:
98994639
HS
851 PL_collxfrm_base = 0;
852 PL_collxfrm_mult = 2;
165a1c52 853 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 854 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 855 PL_strxfrm_max_cp = 0;
98994639
HS
856 return;
857 }
858
d35fca5f 859 /* If this is not the same locale as currently, set the new one up */
98994639
HS
860 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
861 ++PL_collation_ix;
862 Safefree(PL_collation_name);
863 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 864 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
865 if (PL_collation_standard) {
866 goto is_standard_collation;
867 }
98994639 868
165a1c52 869 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 870 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 871 PL_strxfrm_max_cp = 0;
165a1c52 872
59c018b9
KW
873 /* A locale collation definition includes primary, secondary, tertiary,
874 * etc. weights for each character. To sort, the primary weights are
875 * used, and only if they compare equal, then the secondary weights are
876 * used, and only if they compare equal, then the tertiary, etc.
877 *
878 * strxfrm() works by taking the input string, say ABC, and creating an
879 * output transformed string consisting of first the primary weights,
880 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
881 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
882 * may not have weights at every level. In our example, let's say B
883 * doesn't have a tertiary weight, and A doesn't have a secondary
884 * weight. The constructed string is then going to be
885 * A¹B¹C¹ B²C² A³C³ ....
886 * This has the desired effect that strcmp() will look at the secondary
887 * or tertiary weights only if the strings compare equal at all higher
888 * priority weights. The spaces shown here, like in
c342d20e 889 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
890 * are not just for readability. In the general case, these must
891 * actually be bytes, which we will call here 'separator weights'; and
892 * they must be smaller than any other weight value, but since these
893 * are C strings, only the terminating one can be a NUL (some
894 * implementations may include a non-NUL separator weight just before
895 * the NUL). Implementations tend to reserve 01 for the separator
896 * weights. They are needed so that a shorter string's secondary
897 * weights won't be misconstrued as primary weights of a longer string,
898 * etc. By making them smaller than any other weight, the shorter
899 * string will sort first. (Actually, if all secondary weights are
900 * smaller than all primary ones, there is no need for a separator
901 * weight between those two levels, etc.)
902 *
903 * The length of the transformed string is roughly a linear function of
904 * the input string. It's not exactly linear because some characters
905 * don't have weights at all levels. When we call strxfrm() we have to
906 * allocate some memory to hold the transformed string. The
907 * calculations below try to find coefficients 'm' and 'b' for this
908 * locale so that m*x + b equals how much space we need, given the size
909 * of the input string in 'x'. If we calculate too small, we increase
910 * the size as needed, and call strxfrm() again, but it is better to
911 * get it right the first time to avoid wasted expensive string
912 * transformations. */
913
98994639 914 {
79f120c8
KW
915 /* We use the string below to find how long the tranformation of it
916 * is. Almost all locales are supersets of ASCII, or at least the
917 * ASCII letters. We use all of them, half upper half lower,
918 * because if we used fewer, we might hit just the ones that are
919 * outliers in a particular locale. Most of the strings being
920 * collated will contain a preponderance of letters, and even if
921 * they are above-ASCII, they are likely to have the same number of
922 * weight levels as the ASCII ones. It turns out that digits tend
923 * to have fewer levels, and some punctuation has more, but those
924 * are relatively sparse in text, and khw believes this gives a
925 * reasonable result, but it could be changed if experience so
926 * dictates. */
927 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
928 char * x_longer; /* Transformed 'longer' */
929 Size_t x_len_longer; /* Length of 'x_longer' */
930
931 char * x_shorter; /* We also transform a substring of 'longer' */
932 Size_t x_len_shorter;
933
a4a439fb 934 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
935 * are interested only in its length). It is used because it has
936 * the intelligence to handle all cases, but to work, it needs some
937 * values of 'm' and 'b' to get it started. For the purposes of
938 * this calculation we use a very conservative estimate of 'm' and
939 * 'b'. This assumes a weight can be multiple bytes, enough to
940 * hold any UV on the platform, and there are 5 levels, 4 weight
941 * bytes, and a trailing NUL. */
942 PL_collxfrm_base = 5;
943 PL_collxfrm_mult = 5 * sizeof(UV);
944
945 /* Find out how long the transformation really is */
a4a439fb
KW
946 x_longer = _mem_collxfrm(longer,
947 sizeof(longer) - 1,
948 &x_len_longer,
949
950 /* We avoid converting to UTF-8 in the
951 * called function by telling it the
952 * string is in UTF-8 if the locale is a
953 * UTF-8 one. Since the string passed
954 * here is invariant under UTF-8, we can
955 * claim it's UTF-8 even though it isn't.
956 * */
957 PL_in_utf8_COLLATE_locale);
79f120c8
KW
958 Safefree(x_longer);
959
960 /* Find out how long the transformation of a substring of 'longer'
961 * is. Together the lengths of these transformations are
962 * sufficient to calculate 'm' and 'b'. The substring is all of
963 * 'longer' except the first character. This minimizes the chances
964 * of being swayed by outliers */
a4a439fb 965 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 966 sizeof(longer) - 2,
a4a439fb
KW
967 &x_len_shorter,
968 PL_in_utf8_COLLATE_locale);
79f120c8
KW
969 Safefree(x_shorter);
970
971 /* If the results are nonsensical for this simple test, the whole
972 * locale definition is suspect. Mark it so that locale collation
973 * is not active at all for it. XXX Should we warn? */
974 if ( x_len_shorter == 0
975 || x_len_longer == 0
976 || x_len_shorter >= x_len_longer)
977 {
978 PL_collxfrm_mult = 0;
979 PL_collxfrm_base = 0;
980 }
981 else {
982 SSize_t base; /* Temporary */
983
984 /* We have both: m * strlen(longer) + b = x_len_longer
985 * m * strlen(shorter) + b = x_len_shorter;
986 * subtracting yields:
987 * m * (strlen(longer) - strlen(shorter))
988 * = x_len_longer - x_len_shorter
989 * But we have set things up so that 'shorter' is 1 byte smaller
990 * than 'longer'. Hence:
991 * m = x_len_longer - x_len_shorter
992 *
993 * But if something went wrong, make sure the multiplier is at
994 * least 1.
995 */
996 if (x_len_longer > x_len_shorter) {
997 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
998 }
999 else {
1000 PL_collxfrm_mult = 1;
1001 }
1002
1003 /* mx + b = len
1004 * so: b = len - mx
1005 * but in case something has gone wrong, make sure it is
1006 * non-negative */
1007 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
1008 if (base < 0) {
1009 base = 0;
1010 }
1011
1012 /* Add 1 for the trailing NUL */
1013 PL_collxfrm_base = base + 1;
1014 }
58eebef2 1015
7d4bcc4a
KW
1016# ifdef DEBUGGING
1017
58eebef2
KW
1018 if (DEBUG_L_TEST || debug_initialization) {
1019 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1020 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1021 "x_len_longer=%zu,"
1022 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1023 __FILE__, __LINE__,
1024 PL_in_utf8_COLLATE_locale,
1025 x_len_shorter, x_len_longer,
1026 PL_collxfrm_mult, PL_collxfrm_base);
1027 }
7d4bcc4a
KW
1028# endif
1029
98994639
HS
1030 }
1031 }
1032
1033#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1034
98994639
HS
1035}
1036
d2b24094 1037#ifdef WIN32
b8cc575c 1038
a4f00dcc 1039STATIC char *
b8cc575c 1040S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1041{
1042 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1043 * difference between the two unless the input locale is "", which normally
1044 * means on Windows to get the machine default, which is set via the
1045 * computer's "Regional and Language Options" (or its current equivalent).
1046 * In POSIX, it instead means to find the locale from the user's
1047 * environment. This routine changes the Windows behavior to first look in
1048 * the environment, and, if anything is found, use that instead of going to
1049 * the machine default. If there is no environment override, the machine
1050 * default is used, by calling the real setlocale() with "".
1051 *
1052 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1053 * use the particular category's variable if set; otherwise to use the LANG
1054 * variable. */
b385bb4d 1055
175c4cf9 1056 bool override_LC_ALL = FALSE;
89f7b9aa 1057 char * result;
e5f10d49 1058 unsigned int i;
89f7b9aa 1059
b385bb4d 1060 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1061
1062# ifdef LC_ALL
1063
b385bb4d
KW
1064 locale = PerlEnv_getenv("LC_ALL");
1065 if (! locale) {
e5f10d49
KW
1066 if (category == LC_ALL) {
1067 override_LC_ALL = TRUE;
1068 }
1069 else {
7d4bcc4a
KW
1070
1071# endif
7d4bcc4a 1072
e5f10d49
KW
1073 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1074 if (category == categories[i]) {
1075 locale = PerlEnv_getenv(category_names[i]);
1076 goto found_locale;
1077 }
1078 }
7d4bcc4a 1079
b385bb4d 1080 locale = PerlEnv_getenv("LANG");
481465ea 1081 if (! locale) {
b385bb4d
KW
1082 locale = "";
1083 }
e5f10d49
KW
1084
1085 found_locale: ;
7d4bcc4a
KW
1086
1087# ifdef LC_ALL
1088
e5f10d49 1089 }
b385bb4d 1090 }
7d4bcc4a
KW
1091
1092# endif
1093
b385bb4d
KW
1094 }
1095
89f7b9aa 1096 result = setlocale(category, locale);
bbc98134 1097 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
a4f00dcc 1098 setlocale_debug_string(category, locale, result)));
89f7b9aa 1099
481465ea 1100 if (! override_LC_ALL) {
89f7b9aa
KW
1101 return result;
1102 }
1103
dfd77d7a 1104 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
1105 * LANG variable or the system default if there is no LANG. But these have
1106 * lower priority than the other LC_foo variables, so override it for each
1107 * one that is set. (If they are set to "", it means to use the same thing
1108 * we just set LC_ALL to, so can skip) */
7d4bcc4a 1109
948523db 1110 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
1111 result = PerlEnv_getenv(category_names[i]);
1112 if (result && strNE(result, "")) {
1113 setlocale(categories[i], result);
1114 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1115 __FILE__, __LINE__,
1116 setlocale_debug_string(categories[i], result, "not captured")));
1117 }
89f7b9aa 1118 }
7d4bcc4a 1119
bbc98134
KW
1120 result = setlocale(LC_ALL, NULL);
1121 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1122 __FILE__, __LINE__,
a4f00dcc 1123 setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 1124
bbc98134 1125 return result;
b385bb4d
KW
1126}
1127
1128#endif
1129
a4f00dcc
KW
1130char *
1131Perl_setlocale(int category, const char * locale)
1132{
1133 /* This wraps POSIX::setlocale() */
1134
1135 char * retval;
1159483a 1136 char * newlocale;
a4f00dcc
KW
1137 dTHX;
1138
a4f00dcc
KW
1139#ifdef USE_LOCALE_NUMERIC
1140
291a84fb
KW
1141 /* A NULL locale means only query what the current one is. We have the
1142 * LC_NUMERIC name saved, because we are normally switched into the C
1143 * locale for it. For an LC_ALL query, switch back to get the correct
1144 * results. All other categories don't require special handling */
a4f00dcc
KW
1145 if (locale == NULL) {
1146 if (category == LC_NUMERIC) {
1147 return savepv(PL_numeric_name);
1148 }
1149
7d4bcc4a 1150# ifdef LC_ALL
a4f00dcc 1151
c0d737a8
KW
1152 else if (category == LC_ALL && ! PL_numeric_underlying) {
1153
a4f00dcc
KW
1154 SET_NUMERIC_UNDERLYING();
1155 }
1156
7d4bcc4a 1157# endif
a4f00dcc
KW
1158
1159 }
1160
1161#endif
1162
e5f10d49
KW
1163 /* Save retval since subsequent setlocale() calls may overwrite it. */
1164 retval = savepv(do_setlocale_r(category, locale));
a4f00dcc
KW
1165
1166 DEBUG_L(PerlIO_printf(Perl_debug_log,
1167 "%s:%d: %s\n", __FILE__, __LINE__,
1168 setlocale_debug_string(category, locale, retval)));
1169 if (! retval) {
1170 /* Should never happen that a query would return an error, but be
1171 * sure and reset to C locale */
1172 if (locale == 0) {
1173 SET_NUMERIC_STANDARD();
1174 }
7d4bcc4a 1175
a4f00dcc
KW
1176 return NULL;
1177 }
1178
a4f00dcc
KW
1179 /* If locale == NULL, we are just querying the state, but may have switched
1180 * to NUMERIC_UNDERLYING. Switch back before returning. */
1181 if (locale == NULL) {
1182 SET_NUMERIC_STANDARD();
1183 return retval;
1184 }
a4f00dcc 1185
1159483a
KW
1186 /* Now that have switched locales, we have to update our records to
1187 * correspond. */
a4f00dcc 1188
1159483a 1189 switch (category) {
a4f00dcc 1190
1159483a 1191#ifdef USE_LOCALE_CTYPE
a4f00dcc 1192
1159483a
KW
1193 case LC_CTYPE:
1194 new_ctype(retval);
1195 break;
a4f00dcc 1196
1159483a 1197#endif
a4f00dcc
KW
1198#ifdef USE_LOCALE_COLLATE
1199
1159483a
KW
1200 case LC_COLLATE:
1201 new_collate(retval);
1202 break;
a4f00dcc 1203
1159483a 1204#endif
a4f00dcc
KW
1205#ifdef USE_LOCALE_NUMERIC
1206
1159483a
KW
1207 case LC_NUMERIC:
1208 new_numeric(retval);
1209 break;
a4f00dcc 1210
1159483a
KW
1211#endif
1212#ifdef LC_ALL
a4f00dcc 1213
1159483a 1214 case LC_ALL:
a4f00dcc 1215
1159483a
KW
1216 /* LC_ALL updates all the things we care about. The values may not
1217 * be the same as 'retval', as the locale "" may have set things
1218 * individually */
a4f00dcc 1219
1159483a 1220# ifdef USE_LOCALE_CTYPE
a4f00dcc 1221
1159483a
KW
1222 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1223 new_ctype(newlocale);
a4f00dcc 1224
1159483a
KW
1225# endif /* USE_LOCALE_CTYPE */
1226# ifdef USE_LOCALE_COLLATE
1227
1228 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1229 new_collate(newlocale);
a4f00dcc 1230
7d4bcc4a 1231# endif
1159483a 1232# ifdef USE_LOCALE_NUMERIC
a4f00dcc 1233
1159483a
KW
1234 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1235 new_numeric(newlocale);
a4f00dcc 1236
1159483a
KW
1237# endif /* USE_LOCALE_NUMERIC */
1238#endif /* LC_ALL */
a4f00dcc 1239
1159483a
KW
1240 default:
1241 break;
a4f00dcc
KW
1242 }
1243
1244 return retval;
1245
f7416781
KW
1246
1247}
1248
1249PERL_STATIC_INLINE const char *
1250S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1251{
1252 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1253 * growing it if necessary */
1254
1255 const Size_t string_size = strlen(string) + offset + 1;
1256
1257 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1258
1259 if (*buf_size == 0) {
1260 Newx(*buf, string_size, char);
1261 *buf_size = string_size;
1262 }
1263 else if (string_size > *buf_size) {
1264 Renew(*buf, string_size, char);
1265 *buf_size = string_size;
1266 }
1267
1268 Copy(string, *buf + offset, string_size - offset, char);
1269 return *buf;
1270}
1271
1272/*
1273
1274=head1 Locale-related functions and macros
1275
1276=for apidoc Perl_langinfo
1277
7d4bcc4a 1278This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
1279taking the same C<item> parameter values, and returning the same information.
1280But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1281of Perl's locale handling from your code, and can be used on systems that lack
1282a native C<nl_langinfo>.
1283
1284Expanding on these:
1285
1286=over
1287
1288=item *
1289
1290It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1291without you having to write extra code. The reason for the extra code would be
1292because these are from the C<LC_NUMERIC> locale category, which is normally
1293kept set to the C locale by Perl, no matter what the underlying locale is
1294supposed to be, and so to get the expected results, you have to temporarily
1295toggle into the underlying locale, and later toggle back. (You could use
1296plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1297but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1298keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1299expecting the radix (decimal point) character to be a dot.)
1300
1301=item *
1302
1303Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1304makes your code more portable. Of the fifty-some possible items specified by
1305the POSIX 2008 standard,
1306L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1307only two are completely unimplemented. It uses various techniques to recover
1308the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1309both of which are specified in C89, so should be always be available. Later
1310C<strftime()> versions have additional capabilities; C<""> is returned for
1311those not available on your system.
1312
1313The details for those items which may differ from what this emulation returns
1314and what a native C<nl_langinfo()> would return are:
1315
1316=over
1317
1318=item C<CODESET>
1319
1320=item C<ERA>
1321
1322Unimplemented, so returns C<"">.
1323
1324=item C<YESEXPR>
1325
c1566110
KW
1326=item C<YESSTR>
1327
f7416781
KW
1328=item C<NOEXPR>
1329
c1566110
KW
1330=item C<NOSTR>
1331
1332Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
1333removed from POSIX 2008, and are retained for backwards compatibility. Your
1334platform's C<nl_langinfo> may not support them.
f7416781
KW
1335
1336=item C<D_FMT>
1337
1338Always evaluates to C<%x>, the locale's appropriate date representation.
1339
1340=item C<T_FMT>
1341
1342Always evaluates to C<%X>, the locale's appropriate time representation.
1343
1344=item C<D_T_FMT>
1345
1346Always evaluates to C<%c>, the locale's appropriate date and time
1347representation.
1348
1349=item C<CRNCYSTR>
1350
1351The return may be incorrect for those rare locales where the currency symbol
1352replaces the radix character.
1353Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1354to work differently.
1355
1356=item C<ALT_DIGITS>
1357
1358Currently this gives the same results as Linux does.
1359Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1360to work differently.
1361
1362=item C<ERA_D_FMT>
1363
1364=item C<ERA_T_FMT>
1365
1366=item C<ERA_D_T_FMT>
1367
1368=item C<T_FMT_AMPM>
1369
1370These are derived by using C<strftime()>, and not all versions of that function
1371know about them. C<""> is returned for these on such systems.
1372
1373=back
1374
1375When using C<Perl_langinfo> on systems that don't have a native
1376C<nl_langinfo()>, you must
1377
1378 #include "perl_langinfo.h"
1379
1380before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1381C<#include> with this one. (Doing it this way keeps out the symbols that plain
1382C<langinfo.h> imports into the namespace for code that doesn't need it.)
1383
1384You also should not use the bare C<langinfo.h> item names, but should preface
1385them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1386The C<PERL_I<foo>> versions will also work for this function on systems that do
1387have a native C<nl_langinfo>.
1388
1389=item *
1390
1391It is thread-friendly, returning its result in a buffer that won't be
1392overwritten by another thread, so you don't have to code for that possibility.
1393The buffer can be overwritten by the next call to C<nl_langinfo> or
1394C<Perl_langinfo> in the same thread.
1395
1396=item *
1397
7d4bcc4a 1398ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
1399*>>, but you are (only by documentation) forbidden to write into the buffer.
1400By declaring this C<const>, the compiler enforces this restriction. The extra
1401C<const> is why this isn't an unequivocal drop-in replacement for
1402C<nl_langinfo>.
1403
1404=back
1405
1406The original impetus for C<Perl_langinfo()> was so that code that needs to
1407find out the current currency symbol, floating point radix character, or digit
1408grouping separator can use, on all systems, the simpler and more
1409thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1410pain to make thread-friendly. For other fields returned by C<localeconv>, it
1411is better to use the methods given in L<perlcall> to call
1412L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1413
1414=cut
1415
1416*/
1417
1418const char *
1419#ifdef HAS_NL_LANGINFO
1420Perl_langinfo(const nl_item item)
1421#else
1422Perl_langinfo(const int item)
1423#endif
1424{
f61748ac
KW
1425 return my_nl_langinfo(item, TRUE);
1426}
1427
1428const char *
1429#ifdef HAS_NL_LANGINFO
1430S_my_nl_langinfo(const nl_item item, bool toggle)
1431#else
1432S_my_nl_langinfo(const int item, bool toggle)
1433#endif
1434{
ae74815b 1435 dTHX;
f7416781 1436
5a854ab3
KW
1437 /* We only need to toggle into the underlying LC_NUMERIC locale for these
1438 * two items, and only if not already there */
1439 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
1440 || PL_numeric_underlying))
1441 {
1442 toggle = FALSE;
1443 }
1444
ab340fff
KW
1445#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1446#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1447
ab340fff 1448 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1449 * for those items dependent on it. This must be copied to a buffer before
1450 * switching back, as some systems destroy the buffer when setlocale() is
1451 * called */
f7416781
KW
1452
1453 LOCALE_LOCK;
1454
1455 if (toggle) {
837ce802 1456 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
f7416781
KW
1457 }
1458
1459 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1460
1461 if (toggle) {
837ce802 1462 do_setlocale_c(LC_NUMERIC, "C");
f7416781
KW
1463 }
1464
1465 LOCALE_UNLOCK;
1466
ab340fff
KW
1467# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1468
5a854ab3 1469 {
ab340fff
KW
1470 bool do_free = FALSE;
1471 locale_t cur = uselocale((locale_t) 0);
1472
1473 if (cur == LC_GLOBAL_LOCALE) {
1474 cur = duplocale(LC_GLOBAL_LOCALE);
1475 do_free = TRUE;
1476 }
1477
c0d737a8 1478 if (toggle) {
ab340fff
KW
1479 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1480 do_free = TRUE;
1481 }
1482
1483 save_to_buffer(nl_langinfo_l(item, cur),
1484 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1485 if (do_free) {
1486 freelocale(cur);
1487 }
5a854ab3 1488 }
ab340fff 1489
c1566110
KW
1490# endif
1491
1492 if (strEQ(PL_langinfo_buf, "")) {
1493 if (item == PERL_YESSTR) {
1494 return "yes";
1495 }
1496 if (item == PERL_NOSTR) {
1497 return "no";
1498 }
1499 }
1500
ab340fff
KW
1501 return PL_langinfo_buf;
1502
f7416781 1503#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
1504
1505 {
1506
f7416781
KW
1507# ifdef HAS_LOCALECONV
1508
43dd6b15 1509 const struct lconv* lc;
f7416781
KW
1510
1511# endif
1512# ifdef HAS_STRFTIME
1513
43dd6b15
KW
1514 struct tm tm;
1515 bool return_format = FALSE; /* Return the %format, not the value */
1516 const char * format;
f7416781
KW
1517
1518# endif
1519
43dd6b15
KW
1520 /* We copy the results to a per-thread buffer, even if not
1521 * multi-threaded. This is in part to simplify this code, and partly
1522 * because we need a buffer anyway for strftime(), and partly because a
1523 * call of localeconv() could otherwise wipe out the buffer, and the
1524 * programmer would not be expecting this, as this is a nl_langinfo()
1525 * substitute after all, so s/he might be thinking their localeconv()
1526 * is safe until another localeconv() call. */
f7416781 1527
43dd6b15
KW
1528 switch (item) {
1529 Size_t len;
1530 const char * retval;
f7416781 1531
43dd6b15
KW
1532 /* These 2 are unimplemented */
1533 case PERL_CODESET:
1534 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 1535
43dd6b15
KW
1536 default:
1537 return "";
f7416781 1538
43dd6b15
KW
1539 /* We use only an English set, since we don't know any more */
1540 case PERL_YESEXPR: return "^[+1yY]";
1541 case PERL_YESSTR: return "yes";
1542 case PERL_NOEXPR: return "^[-0nN]";
1543 case PERL_NOSTR: return "no";
f7416781
KW
1544
1545# ifdef HAS_LOCALECONV
1546
43dd6b15 1547 case PERL_CRNCYSTR:
f7416781 1548
43dd6b15 1549 LOCALE_LOCK;
f7416781 1550
43dd6b15
KW
1551 /* We don't bother with localeconv_l() because any system that
1552 * has it is likely to also have nl_langinfo() */
291a84fb 1553
43dd6b15
KW
1554 lc = localeconv();
1555 if ( ! lc
1556 || ! lc->currency_symbol
1557 || strEQ("", lc->currency_symbol))
1558 {
1559 LOCALE_UNLOCK;
1560 return "";
1561 }
f7416781 1562
43dd6b15
KW
1563 /* Leave the first spot empty to be filled in below */
1564 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1565 &PL_langinfo_bufsize, 1);
1566 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1567 { /* khw couldn't figure out how the localedef specifications
1568 would show that the $ should replace the radix; this is
1569 just a guess as to how it might work.*/
1570 *PL_langinfo_buf = '.';
1571 }
1572 else if (lc->p_cs_precedes) {
1573 *PL_langinfo_buf = '-';
1574 }
1575 else {
1576 *PL_langinfo_buf = '+';
1577 }
f7416781 1578
43dd6b15
KW
1579 LOCALE_UNLOCK;
1580 break;
f7416781 1581
43dd6b15
KW
1582 case PERL_RADIXCHAR:
1583 case PERL_THOUSEP:
f7416781 1584
43dd6b15 1585 LOCALE_LOCK;
f7416781 1586
43dd6b15 1587 if (toggle) {
43dd6b15 1588 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
c0d737a8 1589 }
f7416781 1590
43dd6b15
KW
1591 lc = localeconv();
1592 if (! lc) {
33394adc
KW
1593 retval = "";
1594 }
43dd6b15
KW
1595 else {
1596 retval = (item == PERL_RADIXCHAR)
1597 ? lc->decimal_point
1598 : lc->thousands_sep;
1599 if (! retval) {
1600 retval = "";
1601 }
1602 }
f7416781 1603
43dd6b15
KW
1604 save_to_buffer(retval, &PL_langinfo_buf,
1605 &PL_langinfo_bufsize, 0);
f7416781 1606
43dd6b15
KW
1607 if (toggle) {
1608 do_setlocale_c(LC_NUMERIC, "C");
1609 }
f7416781 1610
43dd6b15 1611 LOCALE_UNLOCK;
f7416781 1612
43dd6b15 1613 break;
f7416781
KW
1614
1615# endif
1616# ifdef HAS_STRFTIME
1617
43dd6b15
KW
1618 /* These are defined by C89, so we assume that strftime supports
1619 * them, and so are returned unconditionally; they may not be what
1620 * the locale actually says, but should give good enough results
1621 * for someone using them as formats (as opposed to trying to parse
1622 * them to figure out what the locale says). The other format
1623 * items are actually tested to verify they work on the platform */
1624 case PERL_D_FMT: return "%x";
1625 case PERL_T_FMT: return "%X";
1626 case PERL_D_T_FMT: return "%c";
1627
1628 /* These formats are only available in later strfmtime's */
1629 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1630 case PERL_T_FMT_AMPM:
1631
1632 /* The rest can be gotten from most versions of strftime(). */
1633 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1634 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1635 case PERL_ABDAY_7:
1636 case PERL_ALT_DIGITS:
1637 case PERL_AM_STR: case PERL_PM_STR:
1638 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1639 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1640 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1641 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1642 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1643 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1644 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1645 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1646 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
1647 case PERL_MON_12:
1648
1649 LOCALE_LOCK;
1650
1651 init_tm(&tm); /* Precaution against core dumps */
1652 tm.tm_sec = 30;
1653 tm.tm_min = 30;
1654 tm.tm_hour = 6;
1655 tm.tm_year = 2017 - 1900;
1656 tm.tm_wday = 0;
1657 tm.tm_mon = 0;
1658 switch (item) {
1659 default:
1660 LOCALE_UNLOCK;
1661 Perl_croak(aTHX_
1662 "panic: %s: %d: switch case: %d problem",
1663 __FILE__, __LINE__, item);
1664 NOT_REACHED; /* NOTREACHED */
1665
1666 case PERL_PM_STR: tm.tm_hour = 18;
1667 case PERL_AM_STR:
1668 format = "%p";
1669 break;
1670
1671 case PERL_ABDAY_7: tm.tm_wday++;
1672 case PERL_ABDAY_6: tm.tm_wday++;
1673 case PERL_ABDAY_5: tm.tm_wday++;
1674 case PERL_ABDAY_4: tm.tm_wday++;
1675 case PERL_ABDAY_3: tm.tm_wday++;
1676 case PERL_ABDAY_2: tm.tm_wday++;
1677 case PERL_ABDAY_1:
1678 format = "%a";
1679 break;
1680
1681 case PERL_DAY_7: tm.tm_wday++;
1682 case PERL_DAY_6: tm.tm_wday++;
1683 case PERL_DAY_5: tm.tm_wday++;
1684 case PERL_DAY_4: tm.tm_wday++;
1685 case PERL_DAY_3: tm.tm_wday++;
1686 case PERL_DAY_2: tm.tm_wday++;
1687 case PERL_DAY_1:
1688 format = "%A";
1689 break;
1690
1691 case PERL_ABMON_12: tm.tm_mon++;
1692 case PERL_ABMON_11: tm.tm_mon++;
1693 case PERL_ABMON_10: tm.tm_mon++;
1694 case PERL_ABMON_9: tm.tm_mon++;
1695 case PERL_ABMON_8: tm.tm_mon++;
1696 case PERL_ABMON_7: tm.tm_mon++;
1697 case PERL_ABMON_6: tm.tm_mon++;
1698 case PERL_ABMON_5: tm.tm_mon++;
1699 case PERL_ABMON_4: tm.tm_mon++;
1700 case PERL_ABMON_3: tm.tm_mon++;
1701 case PERL_ABMON_2: tm.tm_mon++;
1702 case PERL_ABMON_1:
1703 format = "%b";
1704 break;
1705
1706 case PERL_MON_12: tm.tm_mon++;
1707 case PERL_MON_11: tm.tm_mon++;
1708 case PERL_MON_10: tm.tm_mon++;
1709 case PERL_MON_9: tm.tm_mon++;
1710 case PERL_MON_8: tm.tm_mon++;
1711 case PERL_MON_7: tm.tm_mon++;
1712 case PERL_MON_6: tm.tm_mon++;
1713 case PERL_MON_5: tm.tm_mon++;
1714 case PERL_MON_4: tm.tm_mon++;
1715 case PERL_MON_3: tm.tm_mon++;
1716 case PERL_MON_2: tm.tm_mon++;
1717 case PERL_MON_1:
1718 format = "%B";
1719 break;
1720
1721 case PERL_T_FMT_AMPM:
1722 format = "%r";
1723 return_format = TRUE;
1724 break;
1725
1726 case PERL_ERA_D_FMT:
1727 format = "%Ex";
1728 return_format = TRUE;
1729 break;
1730
1731 case PERL_ERA_T_FMT:
1732 format = "%EX";
1733 return_format = TRUE;
1734 break;
1735
1736 case PERL_ERA_D_T_FMT:
1737 format = "%Ec";
1738 return_format = TRUE;
1739 break;
1740
1741 case PERL_ALT_DIGITS:
1742 tm.tm_wday = 0;
1743 format = "%Ow"; /* Find the alternate digit for 0 */
1744 break;
1745 }
f7416781 1746
43dd6b15
KW
1747 /* We can't use my_strftime() because it doesn't look at
1748 * tm_wday */
1749 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1750 format, &tm))
1751 {
1752 /* A zero return means one of:
1753 * a) there wasn't enough space in PL_langinfo_buf
1754 * b) the format, like a plain %p, returns empty
1755 * c) it was an illegal format, though some
1756 * implementations of strftime will just return the
1757 * illegal format as a plain character sequence.
1758 *
1759 * To quickly test for case 'b)', try again but precede
1760 * the format with a plain character. If that result is
1761 * still empty, the problem is either 'a)' or 'c)' */
1762
1763 Size_t format_size = strlen(format) + 1;
1764 Size_t mod_size = format_size + 1;
1765 char * mod_format;
1766 char * temp_result;
1767
1768 Newx(mod_format, mod_size, char);
1769 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 1770 *mod_format = ' ';
43dd6b15
KW
1771 my_strlcpy(mod_format + 1, format, mod_size);
1772 len = strftime(temp_result,
1773 PL_langinfo_bufsize,
1774 mod_format, &tm);
1775 Safefree(mod_format);
1776 Safefree(temp_result);
1777
1778 /* If 'len' is non-zero, it means that we had a case like
1779 * %p which means the current locale doesn't use a.m. or
1780 * p.m., and that is valid */
1781 if (len == 0) {
1782
1783 /* Here, still didn't work. If we get well beyond a
1784 * reasonable size, bail out to prevent an infinite
1785 * loop. */
1786
1787 if (PL_langinfo_bufsize > 100 * format_size) {
1788 *PL_langinfo_buf = '\0';
1789 }
1790 else {
1791 /* Double the buffer size to retry; Add 1 in case
1792 * original was 0, so we aren't stuck at 0. */
1793 PL_langinfo_bufsize *= 2;
1794 PL_langinfo_bufsize++;
1795 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1796 continue;
1797 }
1798 }
f7416781 1799
f7416781 1800 break;
43dd6b15 1801 }
f7416781 1802
43dd6b15
KW
1803 /* Here, we got a result.
1804 *
1805 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1806 * alternate format for wday 0. If the value is the same as
1807 * the normal 0, there isn't an alternate, so clear the buffer.
1808 * */
1809 if ( item == PERL_ALT_DIGITS
1810 && strEQ(PL_langinfo_buf, "0"))
1811 {
1812 *PL_langinfo_buf = '\0';
1813 }
f7416781 1814
43dd6b15
KW
1815 /* ALT_DIGITS is problematic. Experiments on it showed that
1816 * strftime() did not always work properly when going from
1817 * alt-9 to alt-10. Only a few locales have this item defined,
1818 * and in all of them on Linux that khw was able to find,
1819 * nl_langinfo() merely returned the alt-0 character, possibly
1820 * doubled. Most Unicode digits are in blocks of 10
1821 * consecutive code points, so that is sufficient information
1822 * for those scripts, as we can infer alt-1, alt-2, .... But
1823 * for a Japanese locale, a CJK ideographic 0 is returned, and
1824 * the CJK digits are not in code point order, so you can't
1825 * really infer anything. The localedef for this locale did
1826 * specify the succeeding digits, so that strftime() works
1827 * properly on them, without needing to infer anything. But
1828 * the nl_langinfo() return did not give sufficient information
1829 * for the caller to understand what's going on. So until
1830 * there is evidence that it should work differently, this
1831 * returns the alt-0 string for ALT_DIGITS.
1832 *
1833 * wday was chosen because its range is all a single digit.
1834 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 1835
43dd6b15 1836 LOCALE_UNLOCK;
f7416781 1837
43dd6b15
KW
1838 /* If to return the format, not the value, overwrite the buffer
1839 * with it. But some strftime()s will keep the original format
1840 * if illegal, so change those to "" */
1841 if (return_format) {
1842 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
1843 *PL_langinfo_buf = '\0';
1844 }
43dd6b15
KW
1845 else {
1846 save_to_buffer(format, &PL_langinfo_buf,
1847 &PL_langinfo_bufsize, 0);
f7416781
KW
1848 }
1849 }
1850
1851 break;
f7416781
KW
1852
1853# endif
1854
43dd6b15 1855 }
f7416781
KW
1856 }
1857
1858 return PL_langinfo_buf;
1859
1860#endif
1861
a4f00dcc 1862}
b385bb4d 1863
98994639
HS
1864/*
1865 * Initialize locale awareness.
1866 */
1867int
1868Perl_init_i18nl10n(pTHX_ int printwarn)
1869{
0e92a118
KW
1870 /* printwarn is
1871 *
1872 * 0 if not to output warning when setup locale is bad
1873 * 1 if to output warning based on value of PERL_BADLANG
1874 * >1 if to output regardless of PERL_BADLANG
1875 *
1876 * returns
98994639 1877 * 1 = set ok or not applicable,
0e92a118
KW
1878 * 0 = fallback to a locale of lower priority
1879 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1880 *
1881 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1882 * set, debugging information is output.
1883 *
1884 * This looks more complicated than it is, mainly due to the #ifdefs.
1885 *
1886 * We try to set LC_ALL to the value determined by the environment. If
1887 * there is no LC_ALL on this platform, we try the individual categories we
1888 * know about. If this works, we are done.
1889 *
1890 * But if it doesn't work, we have to do something else. We search the
1891 * environment variables ourselves instead of relying on the system to do
1892 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1893 * think there is one), and the ultimate fallback "C". This is all done in
1894 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
1895 * more complex. The 'trial_locales' array is initialized with just one
1896 * element; it causes the behavior described in the paragraph above this to
1897 * happen. If that fails, we add elements to 'trial_locales', and do extra
1898 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
1899 *
1900 * On Ultrix, the locale MUST come from the environment, so there is
1901 * preliminary code to set it. I (khw) am not sure that it is necessary,
1902 * and that this couldn't be folded into the loop, but barring any real
1903 * platforms to test on, it's staying as-is
1904 *
1905 * A slight complication is that in embedded Perls, the locale may already
1906 * be set-up, and we don't want to get it from the normal environment
1907 * variables. This is handled by having a special environment variable
1908 * indicate we're in this situation. We simply set setlocale's 2nd
1909 * parameter to be a NULL instead of "". That indicates to setlocale that
1910 * it is not to change anything, but to return the current value,
1911 * effectively initializing perl's db to what the locale already is.
1912 *
1913 * We play the same trick with NULL if a LC_ALL succeeds. We call
1914 * setlocale() on the individual categores with NULL to get their existing
1915 * values for our db, instead of trying to change them.
1916 * */
98994639 1917
0e92a118
KW
1918 int ok = 1;
1919
7d4bcc4a
KW
1920#ifndef USE_LOCALE
1921
1922 PERL_UNUSED_ARG(printwarn);
1923
1924#else /* USE_LOCALE */
7d4bcc4a
KW
1925# ifdef __GLIBC__
1926
175c4cf9 1927 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
1928
1929# endif
65ebb059 1930
ccd65d51
KW
1931 /* NULL uses the existing already set up locale */
1932 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
1933 ? NULL
1934 : "";
c3fcd832
KW
1935 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
1936 unsigned int trial_locales_count;
175c4cf9
KW
1937 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
1938 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 1939 bool setlocale_failure = FALSE;
65ebb059 1940 unsigned int i;
175c4cf9
KW
1941
1942 /* A later getenv() could zap this, so only use here */
1943 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
1944
1945 const bool locwarn = (printwarn > 1
e5f10d49
KW
1946 || ( printwarn
1947 && ( ! bad_lang_use_once
22ff3130 1948 || (
e5f10d49
KW
1949 /* disallow with "" or "0" */
1950 *bad_lang_use_once
1951 && strNE("0", bad_lang_use_once)))));
ea92aad8 1952
291a84fb 1953 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 1954 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
1955
1956 /* current locale for given category; should have been copied so aren't
1957 * volatile */
8de4332b 1958 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 1959
7d4bcc4a
KW
1960# ifdef WIN32
1961
6bce99ee
JH
1962 /* In some systems you can find out the system default locale
1963 * and use that as the fallback locale. */
7d4bcc4a
KW
1964# define SYSTEM_DEFAULT_LOCALE
1965# endif
1966# ifdef SYSTEM_DEFAULT_LOCALE
1967
65ebb059 1968 const char *system_default_locale = NULL;
98994639 1969
7d4bcc4a 1970# endif
948523db
KW
1971
1972# ifndef DEBUGGING
1973# define DEBUG_LOCALE_INIT(a,b,c)
1974# else
7d4bcc4a 1975
8298454c 1976 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
1977
1978# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
1979 STMT_START { \
1980 if (debug_initialization) { \
1981 PerlIO_printf(Perl_debug_log, \
1982 "%s:%d: %s\n", \
1983 __FILE__, __LINE__, \
a4f00dcc 1984 setlocale_debug_string(category, \
2fcc0ca9
KW
1985 locale, \
1986 result)); \
1987 } \
1988 } STMT_END
2fcc0ca9 1989
948523db
KW
1990/* Make sure the parallel arrays are properly set up */
1991# ifdef USE_LOCALE_NUMERIC
1992 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
1993 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
1994# endif
1995# ifdef USE_LOCALE_CTYPE
1996 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
1997 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
1998# endif
1999# ifdef USE_LOCALE_COLLATE
2000 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2001 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
2002# endif
2003# ifdef USE_LOCALE_TIME
2004 assert(categories[LC_TIME_INDEX] == LC_TIME);
2005 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2006# endif
2007# ifdef USE_LOCALE_MESSAGES
2008 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2009 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
2010# endif
2011# ifdef USE_LOCALE_MONETARY
2012 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2013 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
2014# endif
9821811f
KW
2015# ifdef USE_LOCALE_ADDRESS
2016 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
2017 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
2018# endif
2019# ifdef USE_LOCALE_IDENTIFICATION
2020 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
2021 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
2022# endif
2023# ifdef USE_LOCALE_MEASUREMENT
2024 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
2025 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
2026# endif
2027# ifdef USE_LOCALE_PAPER
2028 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
2029 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
2030# endif
2031# ifdef USE_LOCALE_TELEPHONE
2032 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
2033 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
2034# endif
948523db
KW
2035# ifdef LC_ALL
2036 assert(categories[LC_ALL_INDEX] == LC_ALL);
2037 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
2038 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
2039# endif
2040# endif /* DEBUGGING */
47280b20
KW
2041
2042 /* Initialize the cache of the program's UTF-8ness for the always known
2043 * locales C and POSIX */
2044 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
2045 sizeof(PL_locale_utf8ness));
2046
b56c4436 2047# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
2048
2049 /*
2050 * Ultrix setlocale(..., "") fails if there are no environment
2051 * variables from which to get a locale name.
2052 */
2053
b56c4436
KW
2054# ifndef LC_ALL
2055# error Ultrix without LC_ALL not implemented
2056# else
7d4bcc4a 2057
b56c4436
KW
2058 {
2059 bool done = FALSE;
ea92aad8
KW
2060 if (lang) {
2061 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
2062 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
2063 if (sl_result[LC_ALL_INDEX])
2064 done = TRUE;
2065 else
e5f10d49 2066 setlocale_failure = TRUE;
ea92aad8
KW
2067 }
2068 if (! setlocale_failure) {
2069 const char * locale_param;
2070 for (i = 0; i < LC_ALL_INDEX; i++) {
2071 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
2072 ? setlocale_init
2073 : NULL;
2074 sl_result[i] = do_setlocale_r(categories[i], locale_param);
2075 if (! sl_result[i]) {
2076 setlocale_failure = TRUE;
2077 }
2078 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 2079 }
c835d6be 2080 }
7d4bcc4a
KW
2081 }
2082
2083# endif /* LC_ALL */
e5f10d49 2084# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 2085
65ebb059 2086 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
2087 * the list. Normally the loop is executed just once. But if setting the
2088 * locale fails, inside the loop we add fallback trials to the array and so
2089 * will execute the loop multiple times */
c3fcd832
KW
2090 trial_locales[0] = setlocale_init;
2091 trial_locales_count = 1;
7d4bcc4a 2092
65ebb059
KW
2093 for (i= 0; i < trial_locales_count; i++) {
2094 const char * trial_locale = trial_locales[i];
2095
2096 if (i > 0) {
2097
2098 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
2099 * when i==0, but I (khw) don't think that behavior makes much
2100 * sense */
2101 setlocale_failure = FALSE;
2102
7d4bcc4a 2103# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 2104# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 2105
65ebb059
KW
2106 /* On Windows machines, an entry of "" after the 0th means to use
2107 * the system default locale, which we now proceed to get. */
2108 if (strEQ(trial_locale, "")) {
2109 unsigned int j;
2110
2111 /* Note that this may change the locale, but we are going to do
2112 * that anyway just below */
837ce802 2113 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 2114 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 2115
7d4bcc4a 2116 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
2117 * try */
2118 if (! system_default_locale) {
2119 goto next_iteration;
2120 }
2121 for (j = 0; j < trial_locales_count; j++) {
2122 if (strEQ(system_default_locale, trial_locales[j])) {
2123 goto next_iteration;
2124 }
2125 }
2126
2127 trial_locale = system_default_locale;
2128 }
ec0202b5
KW
2129# else
2130# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
2131# endif
7d4bcc4a 2132# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
2133
2134 } /* For i > 0 */
65ebb059 2135
7d4bcc4a
KW
2136# ifdef LC_ALL
2137
948523db
KW
2138 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
2139 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
2140 if (! sl_result[LC_ALL_INDEX]) {
49c85077 2141 setlocale_failure = TRUE;
7cd8b568
KW
2142 }
2143 else {
2144 /* Since LC_ALL succeeded, it should have changed all the other
2145 * categories it can to its value; so we massage things so that the
2146 * setlocales below just return their category's current values.
2147 * This adequately handles the case in NetBSD where LC_COLLATE may
2148 * not be defined for a locale, and setting it individually will
7d4bcc4a 2149 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
2150 * the POSIX locale. */
2151 trial_locale = NULL;
2152 }
7d4bcc4a
KW
2153
2154# endif /* LC_ALL */
98994639 2155
e5f10d49
KW
2156 if (! setlocale_failure) {
2157 unsigned int j;
2158 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2159 curlocales[j]
2160 = savepv(do_setlocale_r(categories[j], trial_locale));
2161 if (! curlocales[j]) {
2162 setlocale_failure = TRUE;
2163 }
2164 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2165 }
c835d6be 2166
e5f10d49
KW
2167 if (! setlocale_failure) { /* All succeeded */
2168 break; /* Exit trial_locales loop */
49c85077 2169 }
65ebb059 2170 }
98994639 2171
49c85077
KW
2172 /* Here, something failed; will need to try a fallback. */
2173 ok = 0;
65ebb059 2174
49c85077
KW
2175 if (i == 0) {
2176 unsigned int j;
98994639 2177
65ebb059 2178 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2179
2180# ifdef LC_ALL
98994639 2181
49c85077
KW
2182 PerlIO_printf(Perl_error_log,
2183 "perl: warning: Setting locale failed.\n");
98994639 2184
7d4bcc4a 2185# else /* !LC_ALL */
98994639 2186
49c85077
KW
2187 PerlIO_printf(Perl_error_log,
2188 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2189
e5f10d49
KW
2190 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2191 if (! curlocales[j]) {
2192 PerlIO_printf(Perl_error_log, category_names[j]);
2193 }
2194 else {
2195 Safefree(curlocales[j]);
2196 }
2197 }
7d4bcc4a 2198
7d4bcc4a 2199# endif /* LC_ALL */
98994639 2200
49c85077
KW
2201 PerlIO_printf(Perl_error_log,
2202 "perl: warning: Please check that your locale settings:\n");
98994639 2203
7d4bcc4a
KW
2204# ifdef __GLIBC__
2205
49c85077
KW
2206 PerlIO_printf(Perl_error_log,
2207 "\tLANGUAGE = %c%s%c,\n",
2208 language ? '"' : '(',
2209 language ? language : "unset",
2210 language ? '"' : ')');
7d4bcc4a 2211# endif
98994639 2212
49c85077
KW
2213 PerlIO_printf(Perl_error_log,
2214 "\tLC_ALL = %c%s%c,\n",
2215 lc_all ? '"' : '(',
2216 lc_all ? lc_all : "unset",
2217 lc_all ? '"' : ')');
98994639 2218
7d4bcc4a
KW
2219# if defined(USE_ENVIRON_ARRAY)
2220
49c85077 2221 {
cd999af9 2222 char **e;
d5e32b93
KW
2223
2224 /* Look through the environment for any variables of the
2225 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2226 * already handled above. These are assumed to be locale
2227 * settings. Output them and their values. */
cd999af9 2228 for (e = environ; *e; e++) {
d5e32b93
KW
2229 const STRLEN prefix_len = sizeof("LC_") - 1;
2230 STRLEN uppers_len;
2231
cd999af9 2232 if ( strBEGINs(*e, "LC_")
c8b388b0 2233 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2234 && (uppers_len = strspn(*e + prefix_len,
2235 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2236 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2237 {
2238 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2239 (int) (prefix_len + uppers_len), *e,
2240 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2241 }
2242 }
49c85077 2243 }
7d4bcc4a
KW
2244
2245# else
2246
49c85077
KW
2247 PerlIO_printf(Perl_error_log,
2248 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2249
2250# endif
98994639 2251
49c85077
KW
2252 PerlIO_printf(Perl_error_log,
2253 "\tLANG = %c%s%c\n",
2254 lang ? '"' : '(',
2255 lang ? lang : "unset",
2256 lang ? '"' : ')');
98994639 2257
49c85077
KW
2258 PerlIO_printf(Perl_error_log,
2259 " are supported and installed on your system.\n");
2260 }
98994639 2261
65ebb059 2262 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2263 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2264 * usually change the upper bound of the loop we are in.
2265 *
2266 * Since the system's default way of setting the locale has not
2267 * found one that works, We use Perl's defined ordering: LC_ALL,
2268 * LANG, and the C locale. We don't try the same locale twice, so
2269 * don't add to the list if already there. (On POSIX systems, the
2270 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2271 * but there's no harm done by doing it explicitly.
2272 *
2273 * Note that this tries the LC_ALL environment variable even on
2274 * systems which have no LC_ALL locale setting. This may or may
2275 * not have been originally intentional, but there's no real need
2276 * to change the behavior. */
65ebb059
KW
2277 if (lc_all) {
2278 for (j = 0; j < trial_locales_count; j++) {
2279 if (strEQ(lc_all, trial_locales[j])) {
2280 goto done_lc_all;
2281 }
2282 }
2283 trial_locales[trial_locales_count++] = lc_all;
2284 }
2285 done_lc_all:
98994639 2286
65ebb059
KW
2287 if (lang) {
2288 for (j = 0; j < trial_locales_count; j++) {
2289 if (strEQ(lang, trial_locales[j])) {
2290 goto done_lang;
2291 }
2292 }
2293 trial_locales[trial_locales_count++] = lang;
2294 }
2295 done_lang:
2296
7d4bcc4a
KW
2297# if defined(WIN32) && defined(LC_ALL)
2298
65ebb059
KW
2299 /* For Windows, we also try the system default locale before "C".
2300 * (If there exists a Windows without LC_ALL we skip this because
2301 * it gets too complicated. For those, the "C" is the next
2302 * fallback possibility). The "" is the same as the 0th element of
2303 * the array, but the code at the loop above knows to treat it
2304 * differently when not the 0th */
2305 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2306
2307# endif
65ebb059
KW
2308
2309 for (j = 0; j < trial_locales_count; j++) {
2310 if (strEQ("C", trial_locales[j])) {
2311 goto done_C;
2312 }
2313 }
2314 trial_locales[trial_locales_count++] = "C";
98994639 2315
65ebb059
KW
2316 done_C: ;
2317 } /* end of first time through the loop */
98994639 2318
7d4bcc4a
KW
2319# ifdef WIN32
2320
65ebb059 2321 next_iteration: ;
7d4bcc4a
KW
2322
2323# endif
65ebb059
KW
2324
2325 } /* end of looping through the trial locales */
2326
2327 if (ok < 1) { /* If we tried to fallback */
2328 const char* msg;
2329 if (! setlocale_failure) { /* fallback succeeded */
2330 msg = "Falling back to";
2331 }
2332 else { /* fallback failed */
e5f10d49 2333 unsigned int j;
98994639 2334
65ebb059
KW
2335 /* We dropped off the end of the loop, so have to decrement i to
2336 * get back to the value the last time through */
2337 i--;
98994639 2338
65ebb059
KW
2339 ok = -1;
2340 msg = "Failed to fall back to";
2341
2342 /* To continue, we should use whatever values we've got */
7d4bcc4a 2343
e5f10d49
KW
2344 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2345 Safefree(curlocales[j]);
2346 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2347 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2348 }
65ebb059
KW
2349 }
2350
2351 if (locwarn) {
2352 const char * description;
2353 const char * name = "";
2354 if (strEQ(trial_locales[i], "C")) {
2355 description = "the standard locale";
2356 name = "C";
2357 }
7d4bcc4a
KW
2358
2359# ifdef SYSTEM_DEFAULT_LOCALE
2360
65ebb059
KW
2361 else if (strEQ(trial_locales[i], "")) {
2362 description = "the system default locale";
2363 if (system_default_locale) {
2364 name = system_default_locale;
2365 }
2366 }
7d4bcc4a
KW
2367
2368# endif /* SYSTEM_DEFAULT_LOCALE */
2369
65ebb059
KW
2370 else {
2371 description = "a fallback locale";
2372 name = trial_locales[i];
2373 }
2374 if (name && strNE(name, "")) {
2375 PerlIO_printf(Perl_error_log,
2376 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2377 }
2378 else {
2379 PerlIO_printf(Perl_error_log,
2380 "perl: warning: %s %s.\n", msg, description);
2381 }
2382 }
2383 } /* End of tried to fallback */
98994639 2384
e5f10d49
KW
2385 /* Done with finding the locales; update our records */
2386
7d4bcc4a
KW
2387# ifdef USE_LOCALE_CTYPE
2388
948523db 2389 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2390
e5f10d49 2391# endif
7d4bcc4a
KW
2392# ifdef USE_LOCALE_COLLATE
2393
948523db 2394 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2395
e5f10d49 2396# endif
7d4bcc4a
KW
2397# ifdef USE_LOCALE_NUMERIC
2398
948523db 2399 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2400
2401# endif
2402
948523db 2403 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20
KW
2404
2405# if defined(USE_ITHREADS)
2406
2407 /* This caches whether each category's locale is UTF-8 or not. This
2408 * may involve changing the locale. It is ok to do this at
2409 * initialization time before any threads have started, but not later.
2410 * Caching means that if the program heeds our dictate not to change
2411 * locales in threaded applications, this data will remain valid, and
2412 * it may get queried without changing locales. If the environment is
2413 * such that all categories have the same locale, this isn't needed, as
2414 * the code will not change the locale; but this handles the uncommon
2415 * case where the environment has disparate locales for the categories
2416 * */
2417 (void) _is_cur_LC_category_utf8(categories[i]);
2418
2419# endif
2420
e5f10d49
KW
2421 Safefree(curlocales[i]);
2422 }
b310b053 2423
7d4bcc4a
KW
2424# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2425
49c85077 2426 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
2427 * locale is UTF-8. The call to new_ctype() just above has already
2428 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
2429 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
2430 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
2431 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
2432 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 2433
a05d7ebb 2434 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2435 This is an alternative to using the -C command line switch
2436 (the -C if present will override this). */
2437 {
dd374669 2438 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2439 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2440 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2441 PL_utf8cache = -1;
b310b053
JH
2442 }
2443
7d4bcc4a 2444# endif
7d4bcc4a
KW
2445# ifdef __GLIBC__
2446
175c4cf9 2447 Safefree(language);
7d4bcc4a
KW
2448
2449# endif
175c4cf9
KW
2450
2451 Safefree(lc_all);
2452 Safefree(lang);
2453
e3305790 2454#endif /* USE_LOCALE */
2fcc0ca9 2455#ifdef DEBUGGING
7d4bcc4a 2456
2fcc0ca9 2457 /* So won't continue to output stuff */
27cdc72e 2458 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2459
2fcc0ca9
KW
2460#endif
2461
98994639
HS
2462 return ok;
2463}
2464
98994639
HS
2465#ifdef USE_LOCALE_COLLATE
2466
a4a439fb 2467char *
a4a439fb
KW
2468Perl__mem_collxfrm(pTHX_ const char *input_string,
2469 STRLEN len, /* Length of 'input_string' */
2470 STRLEN *xlen, /* Set to length of returned string
2471 (not including the collation index
2472 prefix) */
2473 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2474 )
98994639 2475{
a4a439fb
KW
2476
2477 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2478 * differences. First, it handles embedded NULs. Second, it allocates a bit
2479 * more memory than needed for the transformed data itself. The real
55e5378d 2480 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2481 * the length of that, and doesn't include the collation index size.
2482 * Please see sv_collxfrm() to see how this is used. */
2483
55e5378d
KW
2484#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2485
6696cfa7
KW
2486 char * s = (char *) input_string;
2487 STRLEN s_strlen = strlen(input_string);
79f120c8 2488 char *xbuf = NULL;
55e5378d 2489 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2490 STRLEN length_in_chars;
c664130f 2491 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2492
a4a439fb
KW
2493 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2494
2495 /* Must be NUL-terminated */
2496 assert(*(input_string + len) == '\0');
7918f24d 2497
79f120c8
KW
2498 /* If this locale has defective collation, skip */
2499 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2500 DEBUG_L(PerlIO_printf(Perl_debug_log,
2501 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2502 goto bad;
2503 }
2504
6696cfa7
KW
2505 /* Replace any embedded NULs with the control that sorts before any others.
2506 * This will give as good as possible results on strings that don't
2507 * otherwise contain that character, but otherwise there may be
2508 * less-than-perfect results with that character and NUL. This is
fdc080f3 2509 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2510 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2511 NUL */
6696cfa7
KW
2512 char * e = s + len;
2513 char * sans_nuls;
fdc080f3 2514 STRLEN sans_nuls_len;
94762aa0 2515 int try_non_controls;
afc4976f
KW
2516 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2517 making sure 2nd byte is NUL.
2518 */
2519 STRLEN this_replacement_len;
2520
1e4c9676
KW
2521 /* If we don't know what non-NUL control character sorts lowest for
2522 * this locale, find it */
f28f4d2a 2523 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2524 int j;
afc4976f 2525 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2526 includes the collation index
2527 prefixed. */
2528
91c0e2e0 2529 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2530
2531 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2532 * NUL, in which case we instead look for any character. Controls
2533 * are preferred because collation order is, in general, context
2534 * sensitive, with adjoining characters affecting the order, and
2535 * controls are less likely to have such interactions, allowing the
2536 * NUL-replacement to stand on its own. (Another way to look at it
2537 * is to imagine what would happen if the NUL were replaced by a
2538 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2539 for (try_non_controls = 0;
2540 try_non_controls < 2;
2541 try_non_controls++)
2542 {
d4ff9586
KW
2543 /* Look through all legal code points (NUL isn't) */
2544 for (j = 1; j < 256; j++) {
2545 char * x; /* j's xfrm plus collation index */
2546 STRLEN x_len; /* length of 'x' */
2547 STRLEN trial_len = 1;
736a4fed 2548 char cur_source[] = { '\0', '\0' };
d4ff9586 2549
736a4fed
KW
2550 /* Skip non-controls the first time through the loop. The
2551 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2552 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2553 ? ! isCNTRL_L1(j)
2554 : ! isCNTRL_LC(j))
2555 {
d4ff9586 2556 continue;
6696cfa7 2557 }
6696cfa7 2558
736a4fed
KW
2559 /* Create a 1-char string of the current code point */
2560 cur_source[0] = (char) j;
2561
d4ff9586
KW
2562 /* Then transform it */
2563 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2564 0 /* The string is not in UTF-8 */);
6696cfa7 2565
1e4c9676 2566 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2567 * */
2568 if (! x) {
2569 continue;
2570 }
6696cfa7 2571
d4ff9586
KW
2572 /* If this character's transformation is lower than
2573 * the current lowest, this one becomes the lowest */
2574 if ( cur_min_x == NULL
2575 || strLT(x + COLLXFRM_HDR_LEN,
2576 cur_min_x + COLLXFRM_HDR_LEN))
2577 {
f28f4d2a 2578 PL_strxfrm_NUL_replacement = j;
d4ff9586 2579 cur_min_x = x;
d4ff9586
KW
2580 }
2581 else {
2582 Safefree(x);
2583 }
1e4c9676 2584 } /* end of loop through all 255 characters */
6696cfa7 2585
1e4c9676 2586 /* Stop looking if found */
94762aa0
KW
2587 if (cur_min_x) {
2588 break;
2589 }
2590
2591 /* Unlikely, but possible, if there aren't any controls that
2592 * work in the locale, repeat the loop, looking for any
2593 * character that works */
2594 DEBUG_L(PerlIO_printf(Perl_debug_log,
2595 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2596 } /* End of loop to try first the controls, then any char */
6696cfa7 2597
94762aa0
KW
2598 if (! cur_min_x) {
2599 DEBUG_L(PerlIO_printf(Perl_debug_log,
2600 "_mem_collxfrm: Couldn't find any character to replace"
2601 " embedded NULs in locale %s with", PL_collation_name));
2602 goto bad;
58eebef2
KW
2603 }
2604
94762aa0
KW
2605 DEBUG_L(PerlIO_printf(Perl_debug_log,
2606 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2607 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2608
6696cfa7 2609 Safefree(cur_min_x);
1e4c9676 2610 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2611
2612 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2613 * UTF8-ness of the original */
f28f4d2a
KW
2614 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2615 this_replacement_char[0] =
2616 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2617 this_replacement_char[1] =
2618 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2619 this_replacement_len = 2;
2620 }
2621 else {
f28f4d2a 2622 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2623 /* this_replacement_char[1] = '\0' was done at initialization */
2624 this_replacement_len = 1;
6696cfa7
KW
2625 }
2626
2627 /* The worst case length for the replaced string would be if every
2628 * character in it is NUL. Multiply that by the length of each
2629 * replacement, and allow for a trailing NUL */
afc4976f 2630 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2631 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2632 *sans_nuls = '\0';
2633
6696cfa7
KW
2634 /* Replace each NUL with the lowest collating control. Loop until have
2635 * exhausted all the NULs */
2636 while (s + s_strlen < e) {
6069d6c5 2637 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2638
2639 /* Do the actual replacement */
6069d6c5 2640 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2641
2642 /* Move past the input NUL */
2643 s += s_strlen + 1;
2644 s_strlen = strlen(s);
2645 }
2646
2647 /* And add anything that trails the final NUL */
6069d6c5 2648 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2649
2650 /* Switch so below we transform this modified string */
2651 s = sans_nuls;
2652 len = strlen(s);
1e4c9676 2653 } /* End of replacing NULs */
6696cfa7 2654
a4a439fb
KW
2655 /* Make sure the UTF8ness of the string and locale match */
2656 if (utf8 != PL_in_utf8_COLLATE_locale) {
2657 const char * const t = s; /* Temporary so we can later find where the
2658 input was */
2659
2660 /* Here they don't match. Change the string's to be what the locale is
2661 * expecting */
2662
2663 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2664 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2665 utf8 = TRUE;
2666 }
2667 else { /* locale is not UTF-8; but input is; downgrade the input */
2668
2669 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2670
2671 /* If the downgrade was successful we are done, but if the input
2672 * contains things that require UTF-8 to represent, have to do
2673 * damage control ... */
2674 if (UNLIKELY(utf8)) {
2675
2676 /* What we do is construct a non-UTF-8 string with
2677 * 1) the characters representable by a single byte converted
2678 * to be so (if necessary);
2679 * 2) and the rest converted to collate the same as the
2680 * highest collating representable character. That makes
2681 * them collate at the end. This is similar to how we
2682 * handle embedded NULs, but we use the highest collating
2683 * code point instead of the smallest. Like the NUL case,
2684 * this isn't perfect, but is the best we can reasonably
2685 * do. Every above-255 code point will sort the same as
2686 * the highest-sorting 0-255 code point. If that code
2687 * point can combine in a sequence with some other code
2688 * points for weight calculations, us changing something to
2689 * be it can adversely affect the results. But in most
2690 * cases, it should work reasonably. And note that this is
2691 * really an illegal situation: using code points above 255
2692 * on a locale where only 0-255 are valid. If two strings
2693 * sort entirely equal, then the sort order for the
2694 * above-255 code points will be in code point order. */
2695
2696 utf8 = FALSE;
2697
2698 /* If we haven't calculated the code point with the maximum
2699 * collating order for this locale, do so now */
2700 if (! PL_strxfrm_max_cp) {
2701 int j;
2702
2703 /* The current transformed string that collates the
2704 * highest (except it also includes the prefixed collation
2705 * index. */
2706 char * cur_max_x = NULL;
2707
2708 /* Look through all legal code points (NUL isn't) */
2709 for (j = 1; j < 256; j++) {
2710 char * x;
2711 STRLEN x_len;
736a4fed 2712 char cur_source[] = { '\0', '\0' };
a4a439fb 2713
736a4fed
KW
2714 /* Create a 1-char string of the current code point */
2715 cur_source[0] = (char) j;
a4a439fb
KW
2716
2717 /* Then transform it */
2718 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2719
2720 /* If something went wrong (which it shouldn't), just
2721 * ignore this code point */
94762aa0 2722 if (! x) {
a4a439fb
KW
2723 continue;
2724 }
2725
2726 /* If this character's transformation is higher than
2727 * the current highest, this one becomes the highest */
2728 if ( cur_max_x == NULL
55e5378d
KW
2729 || strGT(x + COLLXFRM_HDR_LEN,
2730 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2731 {
2732 PL_strxfrm_max_cp = j;
2733 cur_max_x = x;
2734 }
2735 else {
2736 Safefree(x);
2737 }
2738 }
2739
94762aa0
KW
2740 if (! cur_max_x) {
2741 DEBUG_L(PerlIO_printf(Perl_debug_log,
2742 "_mem_collxfrm: Couldn't find any character to"
2743 " replace above-Latin1 chars in locale %s with",
2744 PL_collation_name));
2745 goto bad;
2746 }
2747
58eebef2
KW
2748 DEBUG_L(PerlIO_printf(Perl_debug_log,
2749 "_mem_collxfrm: highest 1-byte collating character"
2750 " in locale %s is 0x%02X\n",
2751 PL_collation_name,
2752 PL_strxfrm_max_cp));
58eebef2 2753
a4a439fb
KW
2754 Safefree(cur_max_x);
2755 }
2756
2757 /* Here we know which legal code point collates the highest.
2758 * We are ready to construct the non-UTF-8 string. The length
2759 * will be at least 1 byte smaller than the input string
2760 * (because we changed at least one 2-byte character into a
2761 * single byte), but that is eaten up by the trailing NUL */
2762 Newx(s, len, char);
2763
2764 {
2765 STRLEN i;
2766 STRLEN d= 0;
042d9e50 2767 char * e = (char *) t + len;
a4a439fb
KW
2768
2769 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2770 U8 cur_char = t[i];
2771 if (UTF8_IS_INVARIANT(cur_char)) {
2772 s[d++] = cur_char;
2773 }
042d9e50 2774 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2775 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2776 }
2777 else { /* Replace illegal cp with highest collating
2778 one */
2779 s[d++] = PL_strxfrm_max_cp;
2780 }
2781 }
2782 s[d++] = '\0';
2783 Renew(s, d, char); /* Free up unused space */
2784 }
2785 }
2786 }
2787
2788 /* Here, we have constructed a modified version of the input. It could
2789 * be that we already had a modified copy before we did this version.
2790 * If so, that copy is no longer needed */
2791 if (t != input_string) {
2792 Safefree(t);
2793 }
2794 }
2795
17f41037
KW
2796 length_in_chars = (utf8)
2797 ? utf8_length((U8 *) s, (U8 *) s + len)
2798 : len;
2799
59c018b9
KW
2800 /* The first element in the output is the collation id, used by
2801 * sv_collxfrm(); then comes the space for the transformed string. The
2802 * equation should give us a good estimate as to how much is needed */
55e5378d 2803 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2804 + PL_collxfrm_base
17f41037 2805 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2806 Newx(xbuf, xAlloc, char);
c7202dee
KW
2807 if (UNLIKELY(! xbuf)) {
2808 DEBUG_L(PerlIO_printf(Perl_debug_log,
2809 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2810 goto bad;
c7202dee 2811 }
98994639 2812
d35fca5f 2813 /* Store the collation id */
98994639 2814 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2815
2816 /* Then the transformation of the input. We loop until successful, or we
2817 * give up */
4ebeff16 2818 for (;;) {
1adab0a7 2819
55e5378d 2820 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2821
2822 /* If the transformed string occupies less space than we told strxfrm()
2823 * was available, it means it successfully transformed the whole
2824 * string. */
55e5378d 2825 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2826
1adab0a7
KW
2827 /* Some systems include a trailing NUL in the returned length.
2828 * Ignore it, using a loop in case multiple trailing NULs are
2829 * returned. */
2830 while ( (*xlen) > 0
2831 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2832 {
2833 (*xlen)--;
2834 }
2835
17f41037
KW
2836 /* If the first try didn't get it, it means our prediction was low.
2837 * Modify the coefficients so that we predict a larger value in any
2838 * future transformations */
2839 if (! first_time) {
2840 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2841 STRLEN computed_guess = PL_collxfrm_base
2842 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2843
2844 /* On zero-length input, just keep current slope instead of
2845 * dividing by 0 */
2846 const STRLEN new_m = (length_in_chars != 0)
2847 ? needed / length_in_chars
2848 : PL_collxfrm_mult;
17f41037
KW
2849
2850 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2851 "%s: %d: initial size of %zu bytes for a length "
2852 "%zu string was insufficient, %zu needed\n",
17f41037 2853 __FILE__, __LINE__,
b07929e4 2854 computed_guess, length_in_chars, needed));
17f41037
KW
2855
2856 /* If slope increased, use it, but discard this result for
2857 * length 1 strings, as we can't be sure that it's a real slope
2858 * change */
2859 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
2860
2861# ifdef DEBUGGING
2862
17f41037
KW
2863 STRLEN old_m = PL_collxfrm_mult;
2864 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
2865
2866# endif
2867
17f41037
KW
2868 PL_collxfrm_mult = new_m;
2869 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2870 computed_guess = PL_collxfrm_base
2871 + (PL_collxfrm_mult * length_in_chars);
2872 if (computed_guess < needed) {
2873 PL_collxfrm_base += needed - computed_guess;
2874 }
2875
2876 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2877 "%s: %d: slope is now %zu; was %zu, base "
2878 "is now %zu; was %zu\n",
17f41037 2879 __FILE__, __LINE__,
b07929e4
KW
2880 PL_collxfrm_mult, old_m,
2881 PL_collxfrm_base, old_b));
17f41037
KW
2882 }
2883 else { /* Slope didn't change, but 'b' did */
2884 const STRLEN new_b = needed
2885 - computed_guess
2886 + PL_collxfrm_base;
2887 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2888 "%s: %d: base is now %zu; was %zu\n",
17f41037 2889 __FILE__, __LINE__,
b07929e4 2890 new_b, PL_collxfrm_base));
17f41037
KW
2891 PL_collxfrm_base = new_b;
2892 }
2893 }
2894
4ebeff16
KW
2895 break;
2896 }
bb0f664e 2897
c7202dee
KW
2898 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2899 DEBUG_L(PerlIO_printf(Perl_debug_log,
2900 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2901 *xlen, PERL_INT_MAX));
4ebeff16 2902 goto bad;
c7202dee 2903 }
d35fca5f 2904
c664130f 2905 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
2906 * (usually not including the trailing NUL) when it fails due to not
2907 * enough space being provided. Assume that this is the case unless
2908 * it's been proven otherwise */
c664130f 2909 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 2910 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
2911 }
2912 else { /* Here, either:
2913 * 1) The strxfrm() has previously shown bad behavior; or
2914 * 2) It isn't the first time through the loop, which means
2915 * that the strxfrm() is now showing bad behavior, because
2916 * we gave it what it said was needed in the previous
2917 * iteration, and it came back saying it needed still more.
2918 * (Many versions of cygwin fit this. When the buffer size
2919 * isn't sufficient, they return the input size instead of
2920 * how much is needed.)
d4ff9586
KW
2921 * Increase the buffer size by a fixed percentage and try again.
2922 * */
6ddd902c 2923 xAlloc += (xAlloc / 4) + 1;
c664130f 2924 PL_strxfrm_is_behaved = FALSE;
c664130f 2925
7d4bcc4a
KW
2926# ifdef DEBUGGING
2927
58eebef2
KW
2928 if (DEBUG_Lv_TEST || debug_initialization) {
2929 PerlIO_printf(Perl_debug_log,
2930 "_mem_collxfrm required more space than previously calculated"
b07929e4 2931 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 2932 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 2933 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 2934 }
7d4bcc4a
KW
2935
2936# endif
2937
58eebef2 2938 }
c664130f 2939
4ebeff16 2940 Renew(xbuf, xAlloc, char);
c7202dee
KW
2941 if (UNLIKELY(! xbuf)) {
2942 DEBUG_L(PerlIO_printf(Perl_debug_log,
2943 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 2944 goto bad;
c7202dee 2945 }
c664130f
KW
2946
2947 first_time = FALSE;
4ebeff16 2948 }
98994639 2949
6696cfa7 2950
7d4bcc4a
KW
2951# ifdef DEBUGGING
2952
58eebef2 2953 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
2954
2955 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
2956 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
2957 PerlIO_printf(Perl_debug_log, "%s\n",
2958 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
2959 *xlen, 1));
58eebef2 2960 }
7d4bcc4a
KW
2961
2962# endif
58eebef2 2963
3c5f993e 2964 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 2965 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 2966
6696cfa7
KW
2967 if (s != input_string) {
2968 Safefree(s);
98994639
HS
2969 }
2970
98994639
HS
2971 return xbuf;
2972
2973 bad:
2974 Safefree(xbuf);
6696cfa7
KW
2975 if (s != input_string) {
2976 Safefree(s);
2977 }
98994639 2978 *xlen = 0;
7d4bcc4a
KW
2979
2980# ifdef DEBUGGING
2981
58eebef2 2982 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 2983 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 2984 }
7d4bcc4a
KW
2985
2986# endif
2987
98994639
HS
2988 return NULL;
2989}
2990
7d4bcc4a 2991# ifdef DEBUGGING
c7202dee 2992
4cbaac56 2993STATIC void
c7202dee
KW
2994S_print_collxfrm_input_and_return(pTHX_
2995 const char * const s,
2996 const char * const e,
2997 const STRLEN * const xlen,
2998 const bool is_utf8)
2999{
c7202dee
KW
3000
3001 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
3002
511e4ff7
DM
3003 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
3004 (UV)PL_collation_ix);
c7202dee 3005 if (xlen) {
08b6dc1d 3006 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
3007 }
3008 else {
3009 PerlIO_printf(Perl_debug_log, "NULL");
3010 }
3011 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
3012 PL_collation_name);
9c8a6dc2
KW
3013 print_bytes_for_locale(s, e, is_utf8);
3014
3015 PerlIO_printf(Perl_debug_log, "'\n");
3016}
3017
3018STATIC void
3019S_print_bytes_for_locale(pTHX_
3020 const char * const s,
3021 const char * const e,
3022 const bool is_utf8)
3023{
3024 const char * t = s;
3025 bool prev_was_printable = TRUE;
3026 bool first_time = TRUE;
3027
3028 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
3029
3030 while (t < e) {
3031 UV cp = (is_utf8)
3032 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
3033 : * (U8 *) t;
3034 if (isPRINT(cp)) {
3035 if (! prev_was_printable) {
3036 PerlIO_printf(Perl_debug_log, " ");
3037 }
3038 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
3039 prev_was_printable = TRUE;
3040 }
3041 else {
3042 if (! first_time) {
3043 PerlIO_printf(Perl_debug_log, " ");
3044 }
147e3846 3045 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
3046 prev_was_printable = FALSE;
3047 }
3048 t += (is_utf8) ? UTF8SKIP(t) : 1;
3049 first_time = FALSE;
3050 }
c7202dee
KW
3051}
3052
7d4bcc4a 3053# endif /* #ifdef DEBUGGING */
98994639 3054#endif /* USE_LOCALE_COLLATE */
58eebef2 3055
8ef6e574
KW
3056#ifdef USE_LOCALE
3057
c1284011
KW
3058bool
3059Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
3060{
3061 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
3062 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 3063 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
3064 * could give the wrong result. The result will very likely be correct for
3065 * languages that have commonly used non-ASCII characters, but for notably
3066 * English, it comes down to if the locale's name ends in something like
3067 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61 3068
47280b20 3069 /* Name of current locale corresponding to the input category */
8de4332b 3070 const char *save_input_locale = NULL;
47280b20
KW
3071
3072 bool is_utf8 = FALSE; /* The return value */
7d74bb61
KW
3073 STRLEN final_pos;
3074
47280b20
KW
3075 /* The variables below are for the cache of previous lookups using this
3076 * function. The cache is a C string, described at the definition for
3077 * 'C_and_POSIX_utf8ness'.
3078 *
3079 * The first part of the cache is fixed, for the C and POSIX locales. The
3080 * varying part starts just after them. */
3081 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
3082
3083 Size_t utf8ness_cache_size; /* Size of the varying portion */
3084 Size_t input_name_len; /* Length in bytes of save_input_locale */
3085 Size_t input_name_len_with_overhead; /* plus extra chars used to store
3086 the name in the cache */
3087 char * delimited; /* The name plus the delimiters used to store
3088 it in the cache */
3089 char * name_pos; /* position of 'delimited' in the cache, or 0
3090 if not there */
3091
3092
7d4bcc4a
KW
3093# ifdef LC_ALL
3094
7d74bb61 3095 assert(category != LC_ALL);
7d4bcc4a
KW
3096
3097# endif
7d74bb61 3098
47280b20 3099 /* Get the desired category's locale */
837ce802 3100 save_input_locale = do_setlocale_r(category, NULL);
7d74bb61 3101 if (! save_input_locale) {
47280b20
KW
3102 Perl_croak(aTHX_
3103 "panic: %s: %d: Could not find current locale for %s\n",
3104 __FILE__, __LINE__, category_name(category));
7d74bb61 3105 }
47280b20 3106
b07fffd1 3107 save_input_locale = stdize_locale(savepv(save_input_locale));
47280b20
KW
3108 DEBUG_L(PerlIO_printf(Perl_debug_log,
3109 "Current locale for %s is %s\n",
3110 category_name(category), save_input_locale));
3111
3112 input_name_len = strlen(save_input_locale);
3113
3114 /* In our cache, each name is accompanied by two delimiters and a single
3115 * utf8ness digit */
3116 input_name_len_with_overhead = input_name_len + 3;
3117
3118 /* Allocate and populate space for a copy of the name surrounded by the
3119 * delimiters */
3120 Newx(delimited, input_name_len_with_overhead, char);
3121 delimited[0] = UTF8NESS_SEP[0];
3122 Copy(save_input_locale, delimited + 1, input_name_len, char);
3123 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
3124 delimited[input_name_len+2] = '\0';
3125
3126 /* And see if that is in the cache */
3127 name_pos = instr(PL_locale_utf8ness, delimited);
3128 if (name_pos) {
3129 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
3130
3131# ifdef DEBUGGING
3132
3133 if (DEBUG_Lv_TEST || debug_initialization) {
3134 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
3135 save_input_locale, is_utf8);
3136 }
3137
3138# endif
3139
3140 /* And, if not already in that position, move it to the beginning of
3141 * the non-constant portion of the list, since it is the most recently
3142 * used. (We don't have to worry about overflow, since just moving
3143 * existing names around) */
3144 if (name_pos > utf8ness_cache) {
3145 Move(utf8ness_cache,
3146 utf8ness_cache + input_name_len_with_overhead,
3147 name_pos - utf8ness_cache, char);
3148 Copy(delimited,
3149 utf8ness_cache,
3150 input_name_len_with_overhead - 1, char);
3151 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3152 }
3153
3154 Safefree(delimited);
b07fffd1 3155 Safefree(save_input_locale);
47280b20 3156 return is_utf8;
7d74bb61
KW
3157 }
3158
47280b20
KW
3159 /* Here we don't have stored the utf8ness for the input locale. We have to
3160 * calculate it */
3161
7d4bcc4a 3162# if defined(USE_LOCALE_CTYPE) \
1d958db2 3163 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 3164
1d958db2 3165 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
3166
3167 char *save_ctype_locale = NULL;
7d74bb61 3168
119ee68b 3169 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
3170
3171 /* Get the current LC_CTYPE locale */
837ce802 3172 save_ctype_locale = do_setlocale_c(LC_CTYPE, NULL);
7d74bb61 3173 if (! save_ctype_locale) {
69014004
KW
3174 DEBUG_L(PerlIO_printf(Perl_debug_log,
3175 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
3176 goto cant_use_nllanginfo;
3177 }
4f72bb37 3178 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
3179
3180 /* If LC_CTYPE and the desired category use the same locale, this
3181 * means that finding the value for LC_CTYPE is the same as finding
3182 * the value for the desired category. Otherwise, switch LC_CTYPE
3183 * to the desired category's locale */
3184 if (strEQ(save_ctype_locale, save_input_locale)) {
3185 Safefree(save_ctype_locale);
3186 save_ctype_locale = NULL;
3187 }
837ce802 3188 else if (! do_setlocale_c(LC_CTYPE, save_input_locale)) {
69014004
KW
3189 DEBUG_L(PerlIO_printf(Perl_debug_log,
3190 "Could not change LC_CTYPE locale to %s\n",
3191 save_input_locale));
7d74bb61
KW
3192 Safefree(save_ctype_locale);
3193 goto cant_use_nllanginfo;
3194 }
3195 }
3196
69014004
KW
3197 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
3198 save_input_locale));
3199
7d74bb61 3200 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
3201 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
3202 * should give the correct results */
119ee68b 3203
7d4bcc4a
KW
3204# if defined(HAS_NL_LANGINFO) && defined(CODESET)
3205
291a84fb 3206 { /* The task is easiest if the platform has this POSIX 2001 function */
c70a3e68
KW
3207 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
3208 /* FALSE => already in dest locale */
119ee68b 3209
c70a3e68
KW
3210 DEBUG_L(PerlIO_printf(Perl_debug_log,
3211 "\tnllanginfo returned CODESET '%s'\n", codeset));
3212
3213 if (codeset && strNE(codeset, "")) {
1d958db2
KW
3214 /* If we switched LC_CTYPE, switch back */
3215 if (save_ctype_locale) {
837ce802 3216 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2
KW
3217 Safefree(save_ctype_locale);
3218 }
3219
1d075173
KW
3220 /* If the implementation of foldEQ() somehow were
3221 * to change to not go byte-by-byte, this could
3222 * read past end of string, as only one length is
3223 * checked. But currently, a premature NUL will
3224 * compare false, and it will stop there */
3225 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
3226 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
1d958db2 3227
69014004
KW
3228 DEBUG_L(PerlIO_printf(Perl_debug_log,
3229 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
3230 codeset, is_utf8));
47280b20 3231 goto finish_and_return;
1d958db2 3232 }
119ee68b
KW
3233 }
3234
7d4bcc4a
KW
3235# endif
3236# ifdef MB_CUR_MAX
1d958db2
KW
3237
3238 /* Here, either we don't have nl_langinfo, or it didn't return a
3239 * codeset. Try MB_CUR_MAX */
3240
119ee68b
KW
3241 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3242 * Unicode code point. Since UTF-8 is the only non-single byte
3243 * encoding we handle, we just say any such encoding is UTF-8, and if
3244 * turns out to be wrong, other things will fail */
5f7616bd 3245 is_utf8 = (unsigned) MB_CUR_MAX >= STRLENs(MAX_UNICODE_UTF8);
119ee68b 3246
69014004
KW
3247 DEBUG_L(PerlIO_printf(Perl_debug_log,
3248 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
3249 (int) MB_CUR_MAX, is_utf8));
3250
119ee68b
KW
3251 Safefree(save_input_locale);
3252
7d4bcc4a 3253# ifdef HAS_MBTOWC
119ee68b
KW
3254
3255 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
3256 * since they are both in the C99 standard. We can feed a known byte
3257 * string to the latter function, and check that it gives the expected
3258 * result */
3259 if (is_utf8) {
3260 wchar_t wc;
51fc4b19
KW
3261 int len;
3262
856b881c 3263 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 3264 errno = 0;
b1d4925c 3265 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
51fc4b19 3266
b1d4925c
KW
3267
3268 if ( len != STRLENs(REPLACEMENT_CHARACTER_UTF8)
5579633c 3269 || wc != (wchar_t) UNICODE_REPLACEMENT)
119ee68b
KW
3270 {
3271 is_utf8 = FALSE;
b1d4925c
KW
3272 DEBUG_L(PerlIO_printf(Perl_debug_log, "\replacement=U+%x\n",
3273 (unsigned int)wc));
69014004
KW
3274 DEBUG_L(PerlIO_printf(Perl_debug_log,
3275 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
51fc4b19 3276 len, errno));
119ee68b
KW
3277 }
3278 }
7d4bcc4a
KW
3279
3280# endif
119ee68b 3281
1d958db2
KW
3282 /* If we switched LC_CTYPE, switch back */
3283 if (save_ctype_locale) {
837ce802 3284 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2 3285 Safefree(save_ctype_locale);
119ee68b 3286 }
7d74bb61 3287
47280b20 3288 goto finish_and_return;
7d4bcc4a
KW
3289
3290# endif
3291
7d74bb61 3292 }
119ee68b 3293
7d74bb61
KW
3294 cant_use_nllanginfo:
3295
7d4bcc4a 3296# else /* nl_langinfo should work if available, so don't bother compiling this
0080c90a
KW
3297 fallback code. The final fallback of looking at the name is
3298 compiled, and will be executed if nl_langinfo fails */
7d74bb61 3299
97f4de96
KW
3300 /* nl_langinfo not available or failed somehow. Next try looking at the
3301 * currency symbol to see if it disambiguates things. Often that will be
3302 * in the native script, and if the symbol isn't in UTF-8, we know that the
3303 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
3304 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
3305 * */
fa9b773e 3306
7d4bcc4a
KW
3307# ifdef HAS_LOCALECONV
3308# ifdef USE_LOCALE_MONETARY
3309
fa9b773e
KW
3310 {
3311 char *save_monetary_locale = NULL;
fa9b773e 3312 bool only_ascii = FALSE;
13542a67 3313 struct lconv* lc;
fa9b773e 3314
97f4de96
KW
3315 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
3316 * the desired category, if it isn't that locale already */
3317
fa9b773e
KW
3318 if (category != LC_MONETARY) {
3319
837ce802 3320 save_monetary_locale = do_setlocale_c(LC_MONETARY, NULL);
fa9b773e 3321 if (! save_monetary_locale) {
69014004
KW
3322 DEBUG_L(PerlIO_printf(Perl_debug_log,
3323 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
3324 goto cant_use_monetary;
3325 }
4f72bb37 3326 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 3327
13542a67
KW
3328 if (strEQ(save_monetary_locale, save_input_locale)) {
3329 Safefree(save_monetary_locale);
3330 save_monetary_locale = NULL;
3331 }
837ce802 3332 else if (! do_setlocale_c(LC_MONETARY, save_input_locale)) {
59c234b4
KW
3333 DEBUG_L(PerlIO_printf(Perl_debug_log,
3334 "Could not change LC_MONETARY locale to %s\n",
3335 save_input_locale));
3336 Safefree(save_monetary_locale);
3337 goto cant_use_monetary;
fa9b773e
KW
3338 }
3339 }
3340
3341 /* Here the current LC_MONETARY is set to the locale of the category
3342 * whose information is desired. */
3343
13542a67
KW
3344 lc = localeconv();
3345 if (! lc
3346 || ! lc->currency_symbol
c5f058df 3347 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
3348 {
3349 DEBUG_L(PerlIO_printf(Perl_debug_log, "Couldn't get currency symbol for %s, or contains only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
3350 only_ascii = TRUE;
3351 }
3352 else {
3353 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
3354 }
3355
3356 /* If we changed it, restore LC_MONETARY to its original locale */
3357 if (save_monetary_locale) {
837ce802 3358 do_setlocale_c(LC_MONETARY, save_monetary_locale);
fa9b773e
KW
3359 Safefree(save_monetary_locale);
3360 }
3361
13542a67 3362 if (! only_ascii) {
fa9b773e 3363
59c234b4
KW
3364 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3365 * otherwise assume the locale is UTF-8 if and only if the symbol
3366 * is non-ascii UTF-8. */
3367 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
3368 save_input_locale, is_utf8));
47280b20 3369 goto finish_and_return;
13542a67 3370 }
fa9b773e
KW
3371 }
3372 cant_use_monetary:
3373
7d4bcc4a
KW
3374# endif /* USE_LOCALE_MONETARY */
3375# endif /* HAS_LOCALECONV */
fa9b773e 3376
7d4bcc4a 3377# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e
KW
3378
3379/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3380 * the names of the months and weekdays, timezone, and am/pm indicator */
3381 {
3382 char *save_time_locale = NULL;
3383 int hour = 10;
3384 bool is_dst = FALSE;
3385 int dom = 1;
3386 int month = 0;
3387 int i;
3388 char * formatted_time;
3389
3390
3391 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
3392 * desired category, if it isn't that locale already */
3393
3394 if (category != LC_TIME) {
3395
837ce802 3396 save_time_locale = do_setlocale_c(LC_TIME, NULL);
15f7e74e
KW
3397 if (! save_time_locale) {
3398 DEBUG_L(PerlIO_printf(Perl_debug_log,
3399 "Could not find current locale for LC_TIME\n"));
3400 goto cant_use_time;
3401 }
3402 save_time_locale = stdize_locale(savepv(save_time_locale));
3403
3404 if (strEQ(save_time_locale, save_input_locale)) {
3405 Safefree(save_time_locale);
3406 save_time_locale = NULL;
3407 }
837ce802 3408 else if (! do_setlocale_c(LC_TIME, save_input_locale)) {
15f7e74e
KW
3409 DEBUG_L(PerlIO_printf(Perl_debug_log,
3410 "Could not change LC_TIME locale to %s\n",
3411 save_input_locale));
3412 Safefree(save_time_locale);
3413 goto cant_use_time;
3414 }
3415 }
3416
3417 /* Here the current LC_TIME is set to the locale of the category
3418 * whose information is desired. Look at all the days of the week and
9f10db87 3419 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
3420 * characters. The first such a one found will tell us if the locale
3421 * is UTF-8 or not */
3422
3423 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3424 formatted_time = my_strftime("%A %B %Z %p",
3ae5cd07 3425 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
c5f058df
KW
3426 if ( ! formatted_time
3427 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3428 {
15f7e74e
KW
3429
3430 /* Here, we didn't find a non-ASCII. Try the next time through
3431 * with the complemented dst and am/pm, and try with the next
3432 * weekday. After we have gotten all weekdays, try the next
3433 * month */
3434 is_dst = ! is_dst;
3435 hour = (hour + 12) % 24;
3436 dom++;
3437 if (i > 6) {
3438 month++;
3439 }
3440 continue;
3441 }
3442
3443 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3444 * false otherwise. But first, restore LC_TIME to its original
3445 * locale if we changed it */
3446 if (save_time_locale) {
837ce802 3447 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3448 Safefree(save_time_locale);
3449 }
3450
3451 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3452 save_input_locale,
3453 is_utf8_string((U8 *) formatted_time, 0)));
47280b20
KW
3454 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
3455 goto finish_and_return;
15f7e74e
KW
3456 }
3457
3458 /* Falling off the end of the loop indicates all the names were just
3459 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3460 * to its original locale */
3461 if (save_time_locale) {
837ce802 3462 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3463 Safefree(save_time_locale);
3464 }
3465 DEBUG_L(PerlIO_printf(Perl_debug_log, "All time-related words for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
3466 }
3467 cant_use_time:
3468
7d4bcc4a 3469# endif
15f7e74e 3470
7d4bcc4a 3471# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
3472
3473/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
3474 * on our dromedary test machine, which has over 700 locales. There, this
3475 * added no value to looking at the currency symbol and the time strings. I
3476 * left it in so as to avoid rewriting it if real-world experience indicates
3477 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
3478 * haven't found illegal utf8, we continue on and examine all the strerror()
3479 * messages on the platform for utf8ness. If all are ASCII, we still don't
3480 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
3481 * utf8ness. The reason this doesn't help much is that the messages may not
3482 * have been translated into the locale. The currency symbol and time strings
3483 * are much more likely to have been translated. */
3484 {
855aeb93 3485 int e;
5857e934 3486 bool non_ascii = FALSE;
855aeb93 3487 char *save_messages_locale = NULL;
5857e934 3488 const char * errmsg = NULL;
855aeb93 3489
5857e934
KW
3490 /* Like above, we set LC_MESSAGES to the locale of the desired
3491 * category, if it isn't that locale already */
855aeb93
JH
3492
3493 if (category != LC_MESSAGES) {
3494
837ce802 3495 save_messages_locale = do_setlocale_c(LC_MESSAGES, NULL);
855aeb93 3496 if (! save_messages_locale) {
5857e934
KW
3497 DEBUG_L(PerlIO_printf(Perl_debug_log,
3498 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
3499 goto cant_use_messages;
3500 }
5857e934 3501 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
3502
3503 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
3504 Safefree(save_messages_locale);
3505 save_messages_locale = NULL;
855aeb93 3506 }
837ce802 3507 else if (! do_setlocale_c(LC_MESSAGES, save_input_locale)) {
5857e934
KW
3508 DEBUG_L(PerlIO_printf(Perl_debug_log,
3509 "Could not change LC_MESSAGES locale to %s\n",
3510 save_input_locale));
855aeb93
JH
3511 Safefree(save_messages_locale);
3512 goto cant_use_messages;
3513 }
3514 }
3515
3516 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
3517 * whose information is desired. Look through all the messages. We
3518 * can't use Strerror() here because it may expand to code that
3519 * segfaults in miniperl */
855aeb93 3520
5857e934
KW
3521 for (e = 0; e <= sys_nerr; e++) {
3522 errno = 0;
3523 errmsg = sys_errlist[e];
3524 if (errno || !errmsg) {
855aeb93
JH
3525 break;
3526 }
5857e934 3527 errmsg = savepv(errmsg);
c5f058df 3528 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
3529 non_ascii = TRUE;
3530 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3531 break;
855aeb93
JH
3532 }
3533 }
5857e934 3534 Safefree(errmsg);
855aeb93
JH
3535
3536 /* And, if we changed it, restore LC_MESSAGES to its original locale */
3537 if (save_messages_locale) {
837ce802 3538 do_setlocale_c(LC_MESSAGES, save_messages_locale);
855aeb93
JH
3539 Safefree(save_messages_locale);
3540 }
3541
5857e934
KW
3542 if (non_ascii) {
3543
3544 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3545 * any non-ascii means it is one; otherwise we assume it isn't */
3546 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3547 save_input_locale,
3548 is_utf8));
47280b20 3549 goto finish_and_return;
5857e934 3550 }
855aeb93 3551
5857e934 3552 DEBUG_L(PerlIO_printf(Perl_debug_log, "All error messages for %s contain only ASCII; can't use for determining if UTF-8 locale\n", save_input_locale));
855aeb93
JH
3553 }
3554 cant_use_messages:
3555
7d4bcc4a
KW
3556# endif
3557# endif /* the code that is compiled when no nl_langinfo */
0080c90a 3558
7d4bcc4a 3559# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3560 UTF-8 locale */
7d4bcc4a 3561
97f4de96
KW
3562 /* As a last resort, look at the locale name to see if it matches
3563 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3564 * return of setlocale(), is actually defined to be opaque, so we can't
3565 * really rely on the absence of various substrings in the name to indicate
3566 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3567 * be a UTF-8 locale. Similarly for the other common names */
3568
3569 final_pos = strlen(save_input_locale) - 1;
3570 if (final_pos >= 3) {
8de4332b 3571 const char *name = save_input_locale;
97f4de96
KW
3572
3573 /* Find next 'U' or 'u' and look from there */
3574 while ((name += strcspn(name, "Uu") + 1)
3575 <= save_input_locale + final_pos - 2)
3576 {
c72493e0 3577 if ( isALPHA_FOLD_NE(*name, 't')
305b8651 3578 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
3579 {
3580 continue;
3581 }
3582 name += 2;
3583 if (*(name) == '-') {
3584 if ((name > save_input_locale + final_pos - 1)) {
3585 break;
3586 }
3587 name++;
3588 }
3589 if (*(name) == '8') {
97f4de96
KW
3590 DEBUG_L(PerlIO_printf(Perl_debug_log,
3591 "Locale %s ends with UTF-8 in name\n",
3592 save_input_locale));
47280b20
KW
3593 is_utf8 = TRUE;
3594 goto finish_and_return;
97f4de96
KW
3595 }
3596 }
3597 DEBUG_L(PerlIO_printf(Perl_debug_log,
3598 "Locale %s doesn't end with UTF-8 in name\n",
3599 save_input_locale));
3600 }
3601
4d8d465a 3602# ifdef WIN32
7d4bcc4a 3603
97f4de96 3604 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
8a0832a1 3605 if (memENDs(save_input_locale, final_pos, "65001")) {
97f4de96 3606 DEBUG_L(PerlIO_printf(Perl_debug_log,
8a0832a1 3607 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
97f4de96 3608 save_input_locale));
47280b20
KW
3609 is_utf8 = TRUE;
3610 goto finish_and_return;
97f4de96 3611 }
7d4bcc4a 3612
4d8d465a
KW
3613# endif
3614# endif
97f4de96
KW
3615
3616 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3617 * since we are about to return FALSE anyway, there is no point in doing
3618 * this extra work */
7d4bcc4a
KW
3619
3620# if 0
97f4de96
KW
3621 if (instr(save_input_locale, "8859")) {
3622 DEBUG_L(PerlIO_printf(Perl_debug_log,
3623 "Locale %s has 8859 in name, not UTF-8 locale\n",
3624 save_input_locale));
47280b20
KW
3625 is_utf8 = FALSE;
3626 goto finish_and_return;
97f4de96 3627 }
7d4bcc4a 3628# endif
97f4de96 3629
69014004
KW
3630 DEBUG_L(PerlIO_printf(Perl_debug_log,
3631 "Assuming locale %s is not a UTF-8 locale\n",
3632 save_input_locale));
47280b20
KW
3633 is_utf8 = FALSE;
3634
3635 finish_and_return:
3636
3637 /* Cache this result so we don't have to go through all this next time. */
3638 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
3639 - (utf8ness_cache - PL_locale_utf8ness);
3640
3641 /* But we can't save it if it is too large for the total space available */
3642 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
3643 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
3644
3645 /* Here it can fit, but we may need to clear out the oldest cached
3646 * result(s) to do so. Check */
3647 if (utf8ness_cache_len + input_name_len_with_overhead
3648 >= utf8ness_cache_size)
3649 {
3650 /* Here we have to clear something out to make room for this.
3651 * Start looking at the rightmost place where it could fit and find
3652 * the beginning of the entry that extends past that. */
3653 char * cutoff = (char *) my_memrchr(utf8ness_cache,
3654 UTF8NESS_SEP[0],
3655 utf8ness_cache_size
3656 - input_name_len_with_overhead);
3657
3658 assert(cutoff);
3659 assert(cutoff >= utf8ness_cache);
3660
3661 /* This and all subsequent entries must be removed */
3662 *cutoff = '\0';
3663 utf8ness_cache_len = strlen(utf8ness_cache);
3664 }
3665
3666 /* Make space for the new entry */
3667 Move(utf8ness_cache,
3668 utf8ness_cache + input_name_len_with_overhead,
3669 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
3670
3671 /* And insert it */
3672 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
3673 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3674
3675 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
3676 & (PERL_UINTMAX_T) ~1) != '0')
3677 {
3678 Perl_croak(aTHX_
3679 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%u,"
3680 " inserted_name=%s, its_len=%u\n",
3681 __FILE__, __LINE__,
3682 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
3683 delimited, input_name_len_with_overhead);
3684 }
3685 }
3686
3687# ifdef DEBUGGING
3688
3689 if (DEBUG_Lv_TEST || debug_initialization) {
3690 PerlIO_printf(Perl_debug_log,
3691 "PL_locale_utf8ness is now %s; returning %d\n",
3692 PL_locale_utf8ness, is_utf8);
3693 }
3694
3695# endif
3696
3697 Safefree(delimited);
fa9b773e 3698 Safefree(save_input_locale);
47280b20 3699 return is_utf8;
7d74bb61
KW
3700}
3701
8ef6e574 3702#endif
7d74bb61 3703
d6ded950
KW
3704
3705bool
3706Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3707{
1a4f13e1 3708 dVAR;
d6ded950
KW
3709 /* Internal function which returns if we are in the scope of a pragma that
3710 * enables the locale category 'category'. 'compiling' should indicate if
3711 * this is during the compilation phase (TRUE) or not (FALSE). */
3712
3713 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
3714
3715 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
3716 if (! categories || categories == &PL_sv_placeholder) {
3717 return FALSE;
3718 }
3719
3720 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
3721 * a valid unsigned */
3722 assert(category >= -1);
3723 return cBOOL(SvUV(categories) & (1U << (category + 1)));
3724}
3725
2c6ee1a7 3726char *
6ebbc862
KW
3727Perl_my_strerror(pTHX_ const int errnum)
3728{
3729 /* Returns a mortalized copy of the text of the error message associated
3730 * with 'errnum'. It uses the current locale's text unless the platform
3731 * doesn't have the LC_MESSAGES category or we are not being called from
3732 * within the scope of 'use locale'. In the former case, it uses whatever
3733 * strerror returns; in the latter case it uses the text from the C locale.
3734 *
3735 * The function just calls strerror(), but temporarily switches, if needed,
3736 * to the C locale */
3737
3738 char *errstr;
52770946 3739 dVAR;