This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Cache locale UTF8-ness lookups
[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
ab340fff
KW
1437#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1438#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1439
ab340fff 1440 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1441 * for those items dependent on it. This must be copied to a buffer before
1442 * switching back, as some systems destroy the buffer when setlocale() is
1443 * called */
f7416781
KW
1444
1445 LOCALE_LOCK;
1446
1447 if (toggle) {
c0d737a8
KW
1448 if ( ! PL_numeric_underlying
1449 && (item == PERL_RADIXCHAR || item == PERL_THOUSEP))
1450 {
837ce802 1451 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
f7416781
KW
1452 }
1453 else {
1454 toggle = FALSE;
1455 }
1456 }
1457
1458 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1459
1460 if (toggle) {
837ce802 1461 do_setlocale_c(LC_NUMERIC, "C");
f7416781
KW
1462 }
1463
1464 LOCALE_UNLOCK;
1465
ab340fff
KW
1466# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1467
1468 bool do_free = FALSE;
1469 locale_t cur = uselocale((locale_t) 0);
1470
1471 if (cur == LC_GLOBAL_LOCALE) {
1472 cur = duplocale(LC_GLOBAL_LOCALE);
1473 do_free = TRUE;
1474 }
1475
c0d737a8 1476 if (toggle) {
ab340fff
KW
1477 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1478 do_free = TRUE;
1479 }
1480
1481 save_to_buffer(nl_langinfo_l(item, cur),
1482 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1483 if (do_free) {
1484 freelocale(cur);
1485 }
1486
c1566110
KW
1487# endif
1488
1489 if (strEQ(PL_langinfo_buf, "")) {
1490 if (item == PERL_YESSTR) {
1491 return "yes";
1492 }
1493 if (item == PERL_NOSTR) {
1494 return "no";
1495 }
1496 }
1497
ab340fff
KW
1498 return PL_langinfo_buf;
1499
f7416781 1500#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
1501
1502 {
1503
f7416781
KW
1504# ifdef HAS_LOCALECONV
1505
43dd6b15 1506 const struct lconv* lc;
f7416781
KW
1507
1508# endif
1509# ifdef HAS_STRFTIME
1510
43dd6b15
KW
1511 struct tm tm;
1512 bool return_format = FALSE; /* Return the %format, not the value */
1513 const char * format;
f7416781
KW
1514
1515# endif
1516
43dd6b15
KW
1517 /* We copy the results to a per-thread buffer, even if not
1518 * multi-threaded. This is in part to simplify this code, and partly
1519 * because we need a buffer anyway for strftime(), and partly because a
1520 * call of localeconv() could otherwise wipe out the buffer, and the
1521 * programmer would not be expecting this, as this is a nl_langinfo()
1522 * substitute after all, so s/he might be thinking their localeconv()
1523 * is safe until another localeconv() call. */
f7416781 1524
43dd6b15
KW
1525 switch (item) {
1526 Size_t len;
1527 const char * retval;
f7416781 1528
43dd6b15
KW
1529 /* These 2 are unimplemented */
1530 case PERL_CODESET:
1531 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 1532
43dd6b15
KW
1533 default:
1534 return "";
f7416781 1535
43dd6b15
KW
1536 /* We use only an English set, since we don't know any more */
1537 case PERL_YESEXPR: return "^[+1yY]";
1538 case PERL_YESSTR: return "yes";
1539 case PERL_NOEXPR: return "^[-0nN]";
1540 case PERL_NOSTR: return "no";
f7416781
KW
1541
1542# ifdef HAS_LOCALECONV
1543
43dd6b15 1544 case PERL_CRNCYSTR:
f7416781 1545
43dd6b15 1546 LOCALE_LOCK;
f7416781 1547
43dd6b15
KW
1548 /* We don't bother with localeconv_l() because any system that
1549 * has it is likely to also have nl_langinfo() */
291a84fb 1550
43dd6b15
KW
1551 lc = localeconv();
1552 if ( ! lc
1553 || ! lc->currency_symbol
1554 || strEQ("", lc->currency_symbol))
1555 {
1556 LOCALE_UNLOCK;
1557 return "";
1558 }
f7416781 1559
43dd6b15
KW
1560 /* Leave the first spot empty to be filled in below */
1561 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1562 &PL_langinfo_bufsize, 1);
1563 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1564 { /* khw couldn't figure out how the localedef specifications
1565 would show that the $ should replace the radix; this is
1566 just a guess as to how it might work.*/
1567 *PL_langinfo_buf = '.';
1568 }
1569 else if (lc->p_cs_precedes) {
1570 *PL_langinfo_buf = '-';
1571 }
1572 else {
1573 *PL_langinfo_buf = '+';
1574 }
f7416781 1575
43dd6b15
KW
1576 LOCALE_UNLOCK;
1577 break;
f7416781 1578
43dd6b15
KW
1579 case PERL_RADIXCHAR:
1580 case PERL_THOUSEP:
f7416781 1581
43dd6b15 1582 LOCALE_LOCK;
f7416781 1583
43dd6b15
KW
1584 if (toggle) {
1585 if (! PL_numeric_underlying) {
1586 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
1587 }
1588 else {
1589 toggle = FALSE;
1590 }
c0d737a8 1591 }
f7416781 1592
43dd6b15
KW
1593 lc = localeconv();
1594 if (! lc) {
33394adc
KW
1595 retval = "";
1596 }
43dd6b15
KW
1597 else {
1598 retval = (item == PERL_RADIXCHAR)
1599 ? lc->decimal_point
1600 : lc->thousands_sep;
1601 if (! retval) {
1602 retval = "";
1603 }
1604 }
f7416781 1605
43dd6b15
KW
1606 save_to_buffer(retval, &PL_langinfo_buf,
1607 &PL_langinfo_bufsize, 0);
f7416781 1608
43dd6b15
KW
1609 if (toggle) {
1610 do_setlocale_c(LC_NUMERIC, "C");
1611 }
f7416781 1612
43dd6b15 1613 LOCALE_UNLOCK;
f7416781 1614
43dd6b15 1615 break;
f7416781
KW
1616
1617# endif
1618# ifdef HAS_STRFTIME
1619
43dd6b15
KW
1620 /* These are defined by C89, so we assume that strftime supports
1621 * them, and so are returned unconditionally; they may not be what
1622 * the locale actually says, but should give good enough results
1623 * for someone using them as formats (as opposed to trying to parse
1624 * them to figure out what the locale says). The other format
1625 * items are actually tested to verify they work on the platform */
1626 case PERL_D_FMT: return "%x";
1627 case PERL_T_FMT: return "%X";
1628 case PERL_D_T_FMT: return "%c";
1629
1630 /* These formats are only available in later strfmtime's */
1631 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1632 case PERL_T_FMT_AMPM:
1633
1634 /* The rest can be gotten from most versions of strftime(). */
1635 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1636 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1637 case PERL_ABDAY_7:
1638 case PERL_ALT_DIGITS:
1639 case PERL_AM_STR: case PERL_PM_STR:
1640 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1641 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1642 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1643 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1644 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1645 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1646 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1647 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1648 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
1649 case PERL_MON_12:
1650
1651 LOCALE_LOCK;
1652
1653 init_tm(&tm); /* Precaution against core dumps */
1654 tm.tm_sec = 30;
1655 tm.tm_min = 30;
1656 tm.tm_hour = 6;
1657 tm.tm_year = 2017 - 1900;
1658 tm.tm_wday = 0;
1659 tm.tm_mon = 0;
1660 switch (item) {
1661 default:
1662 LOCALE_UNLOCK;
1663 Perl_croak(aTHX_
1664 "panic: %s: %d: switch case: %d problem",
1665 __FILE__, __LINE__, item);
1666 NOT_REACHED; /* NOTREACHED */
1667
1668 case PERL_PM_STR: tm.tm_hour = 18;
1669 case PERL_AM_STR:
1670 format = "%p";
1671 break;
1672
1673 case PERL_ABDAY_7: tm.tm_wday++;
1674 case PERL_ABDAY_6: tm.tm_wday++;
1675 case PERL_ABDAY_5: tm.tm_wday++;
1676 case PERL_ABDAY_4: tm.tm_wday++;
1677 case PERL_ABDAY_3: tm.tm_wday++;
1678 case PERL_ABDAY_2: tm.tm_wday++;
1679 case PERL_ABDAY_1:
1680 format = "%a";
1681 break;
1682
1683 case PERL_DAY_7: tm.tm_wday++;
1684 case PERL_DAY_6: tm.tm_wday++;
1685 case PERL_DAY_5: tm.tm_wday++;
1686 case PERL_DAY_4: tm.tm_wday++;
1687 case PERL_DAY_3: tm.tm_wday++;
1688 case PERL_DAY_2: tm.tm_wday++;
1689 case PERL_DAY_1:
1690 format = "%A";
1691 break;
1692
1693 case PERL_ABMON_12: tm.tm_mon++;
1694 case PERL_ABMON_11: tm.tm_mon++;
1695 case PERL_ABMON_10: tm.tm_mon++;
1696 case PERL_ABMON_9: tm.tm_mon++;
1697 case PERL_ABMON_8: tm.tm_mon++;
1698 case PERL_ABMON_7: tm.tm_mon++;
1699 case PERL_ABMON_6: tm.tm_mon++;
1700 case PERL_ABMON_5: tm.tm_mon++;
1701 case PERL_ABMON_4: tm.tm_mon++;
1702 case PERL_ABMON_3: tm.tm_mon++;
1703 case PERL_ABMON_2: tm.tm_mon++;
1704 case PERL_ABMON_1:
1705 format = "%b";
1706 break;
1707
1708 case PERL_MON_12: tm.tm_mon++;
1709 case PERL_MON_11: tm.tm_mon++;
1710 case PERL_MON_10: tm.tm_mon++;
1711 case PERL_MON_9: tm.tm_mon++;
1712 case PERL_MON_8: tm.tm_mon++;
1713 case PERL_MON_7: tm.tm_mon++;
1714 case PERL_MON_6: tm.tm_mon++;
1715 case PERL_MON_5: tm.tm_mon++;
1716 case PERL_MON_4: tm.tm_mon++;
1717 case PERL_MON_3: tm.tm_mon++;
1718 case PERL_MON_2: tm.tm_mon++;
1719 case PERL_MON_1:
1720 format = "%B";
1721 break;
1722
1723 case PERL_T_FMT_AMPM:
1724 format = "%r";
1725 return_format = TRUE;
1726 break;
1727
1728 case PERL_ERA_D_FMT:
1729 format = "%Ex";
1730 return_format = TRUE;
1731 break;
1732
1733 case PERL_ERA_T_FMT:
1734 format = "%EX";
1735 return_format = TRUE;
1736 break;
1737
1738 case PERL_ERA_D_T_FMT:
1739 format = "%Ec";
1740 return_format = TRUE;
1741 break;
1742
1743 case PERL_ALT_DIGITS:
1744 tm.tm_wday = 0;
1745 format = "%Ow"; /* Find the alternate digit for 0 */
1746 break;
1747 }
f7416781 1748
43dd6b15
KW
1749 /* We can't use my_strftime() because it doesn't look at
1750 * tm_wday */
1751 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1752 format, &tm))
1753 {
1754 /* A zero return means one of:
1755 * a) there wasn't enough space in PL_langinfo_buf
1756 * b) the format, like a plain %p, returns empty
1757 * c) it was an illegal format, though some
1758 * implementations of strftime will just return the
1759 * illegal format as a plain character sequence.
1760 *
1761 * To quickly test for case 'b)', try again but precede
1762 * the format with a plain character. If that result is
1763 * still empty, the problem is either 'a)' or 'c)' */
1764
1765 Size_t format_size = strlen(format) + 1;
1766 Size_t mod_size = format_size + 1;
1767 char * mod_format;
1768 char * temp_result;
1769
1770 Newx(mod_format, mod_size, char);
1771 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 1772 *mod_format = ' ';
43dd6b15
KW
1773 my_strlcpy(mod_format + 1, format, mod_size);
1774 len = strftime(temp_result,
1775 PL_langinfo_bufsize,
1776 mod_format, &tm);
1777 Safefree(mod_format);
1778 Safefree(temp_result);
1779
1780 /* If 'len' is non-zero, it means that we had a case like
1781 * %p which means the current locale doesn't use a.m. or
1782 * p.m., and that is valid */
1783 if (len == 0) {
1784
1785 /* Here, still didn't work. If we get well beyond a
1786 * reasonable size, bail out to prevent an infinite
1787 * loop. */
1788
1789 if (PL_langinfo_bufsize > 100 * format_size) {
1790 *PL_langinfo_buf = '\0';
1791 }
1792 else {
1793 /* Double the buffer size to retry; Add 1 in case
1794 * original was 0, so we aren't stuck at 0. */
1795 PL_langinfo_bufsize *= 2;
1796 PL_langinfo_bufsize++;
1797 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1798 continue;
1799 }
1800 }
f7416781 1801
f7416781 1802 break;
43dd6b15 1803 }
f7416781 1804
43dd6b15
KW
1805 /* Here, we got a result.
1806 *
1807 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1808 * alternate format for wday 0. If the value is the same as
1809 * the normal 0, there isn't an alternate, so clear the buffer.
1810 * */
1811 if ( item == PERL_ALT_DIGITS
1812 && strEQ(PL_langinfo_buf, "0"))
1813 {
1814 *PL_langinfo_buf = '\0';
1815 }
f7416781 1816
43dd6b15
KW
1817 /* ALT_DIGITS is problematic. Experiments on it showed that
1818 * strftime() did not always work properly when going from
1819 * alt-9 to alt-10. Only a few locales have this item defined,
1820 * and in all of them on Linux that khw was able to find,
1821 * nl_langinfo() merely returned the alt-0 character, possibly
1822 * doubled. Most Unicode digits are in blocks of 10
1823 * consecutive code points, so that is sufficient information
1824 * for those scripts, as we can infer alt-1, alt-2, .... But
1825 * for a Japanese locale, a CJK ideographic 0 is returned, and
1826 * the CJK digits are not in code point order, so you can't
1827 * really infer anything. The localedef for this locale did
1828 * specify the succeeding digits, so that strftime() works
1829 * properly on them, without needing to infer anything. But
1830 * the nl_langinfo() return did not give sufficient information
1831 * for the caller to understand what's going on. So until
1832 * there is evidence that it should work differently, this
1833 * returns the alt-0 string for ALT_DIGITS.
1834 *
1835 * wday was chosen because its range is all a single digit.
1836 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 1837
43dd6b15 1838 LOCALE_UNLOCK;
f7416781 1839
43dd6b15
KW
1840 /* If to return the format, not the value, overwrite the buffer
1841 * with it. But some strftime()s will keep the original format
1842 * if illegal, so change those to "" */
1843 if (return_format) {
1844 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
1845 *PL_langinfo_buf = '\0';
1846 }
43dd6b15
KW
1847 else {
1848 save_to_buffer(format, &PL_langinfo_buf,
1849 &PL_langinfo_bufsize, 0);
f7416781
KW
1850 }
1851 }
1852
1853 break;
f7416781
KW
1854
1855# endif
1856
43dd6b15 1857 }
f7416781
KW
1858 }
1859
1860 return PL_langinfo_buf;
1861
1862#endif
1863
a4f00dcc 1864}
b385bb4d 1865
98994639
HS
1866/*
1867 * Initialize locale awareness.
1868 */
1869int
1870Perl_init_i18nl10n(pTHX_ int printwarn)
1871{
0e92a118
KW
1872 /* printwarn is
1873 *
1874 * 0 if not to output warning when setup locale is bad
1875 * 1 if to output warning based on value of PERL_BADLANG
1876 * >1 if to output regardless of PERL_BADLANG
1877 *
1878 * returns
98994639 1879 * 1 = set ok or not applicable,
0e92a118
KW
1880 * 0 = fallback to a locale of lower priority
1881 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1882 *
1883 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1884 * set, debugging information is output.
1885 *
1886 * This looks more complicated than it is, mainly due to the #ifdefs.
1887 *
1888 * We try to set LC_ALL to the value determined by the environment. If
1889 * there is no LC_ALL on this platform, we try the individual categories we
1890 * know about. If this works, we are done.
1891 *
1892 * But if it doesn't work, we have to do something else. We search the
1893 * environment variables ourselves instead of relying on the system to do
1894 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1895 * think there is one), and the ultimate fallback "C". This is all done in
1896 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
1897 * more complex. The 'trial_locales' array is initialized with just one
1898 * element; it causes the behavior described in the paragraph above this to
1899 * happen. If that fails, we add elements to 'trial_locales', and do extra
1900 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
1901 *
1902 * On Ultrix, the locale MUST come from the environment, so there is
1903 * preliminary code to set it. I (khw) am not sure that it is necessary,
1904 * and that this couldn't be folded into the loop, but barring any real
1905 * platforms to test on, it's staying as-is
1906 *
1907 * A slight complication is that in embedded Perls, the locale may already
1908 * be set-up, and we don't want to get it from the normal environment
1909 * variables. This is handled by having a special environment variable
1910 * indicate we're in this situation. We simply set setlocale's 2nd
1911 * parameter to be a NULL instead of "". That indicates to setlocale that
1912 * it is not to change anything, but to return the current value,
1913 * effectively initializing perl's db to what the locale already is.
1914 *
1915 * We play the same trick with NULL if a LC_ALL succeeds. We call
1916 * setlocale() on the individual categores with NULL to get their existing
1917 * values for our db, instead of trying to change them.
1918 * */
98994639 1919
0e92a118
KW
1920 int ok = 1;
1921
7d4bcc4a
KW
1922#ifndef USE_LOCALE
1923
1924 PERL_UNUSED_ARG(printwarn);
1925
1926#else /* USE_LOCALE */
7d4bcc4a
KW
1927# ifdef __GLIBC__
1928
175c4cf9 1929 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
1930
1931# endif
65ebb059 1932
ccd65d51
KW
1933 /* NULL uses the existing already set up locale */
1934 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
1935 ? NULL
1936 : "";
c3fcd832
KW
1937 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
1938 unsigned int trial_locales_count;
175c4cf9
KW
1939 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
1940 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 1941 bool setlocale_failure = FALSE;
65ebb059 1942 unsigned int i;
175c4cf9
KW
1943
1944 /* A later getenv() could zap this, so only use here */
1945 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
1946
1947 const bool locwarn = (printwarn > 1
e5f10d49
KW
1948 || ( printwarn
1949 && ( ! bad_lang_use_once
22ff3130 1950 || (
e5f10d49
KW
1951 /* disallow with "" or "0" */
1952 *bad_lang_use_once
1953 && strNE("0", bad_lang_use_once)))));
ea92aad8 1954
291a84fb 1955 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 1956 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
1957
1958 /* current locale for given category; should have been copied so aren't
1959 * volatile */
8de4332b 1960 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 1961
7d4bcc4a
KW
1962# ifdef WIN32
1963
6bce99ee
JH
1964 /* In some systems you can find out the system default locale
1965 * and use that as the fallback locale. */
7d4bcc4a
KW
1966# define SYSTEM_DEFAULT_LOCALE
1967# endif
1968# ifdef SYSTEM_DEFAULT_LOCALE
1969
65ebb059 1970 const char *system_default_locale = NULL;
98994639 1971
7d4bcc4a 1972# endif
948523db
KW
1973
1974# ifndef DEBUGGING
1975# define DEBUG_LOCALE_INIT(a,b,c)
1976# else
7d4bcc4a 1977
8298454c 1978 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
1979
1980# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
1981 STMT_START { \
1982 if (debug_initialization) { \
1983 PerlIO_printf(Perl_debug_log, \
1984 "%s:%d: %s\n", \
1985 __FILE__, __LINE__, \
a4f00dcc 1986 setlocale_debug_string(category, \
2fcc0ca9
KW
1987 locale, \
1988 result)); \
1989 } \
1990 } STMT_END
2fcc0ca9 1991
948523db
KW
1992/* Make sure the parallel arrays are properly set up */
1993# ifdef USE_LOCALE_NUMERIC
1994 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
1995 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
1996# endif
1997# ifdef USE_LOCALE_CTYPE
1998 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
1999 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
2000# endif
2001# ifdef USE_LOCALE_COLLATE
2002 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
2003 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
2004# endif
2005# ifdef USE_LOCALE_TIME
2006 assert(categories[LC_TIME_INDEX] == LC_TIME);
2007 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2008# endif
2009# ifdef USE_LOCALE_MESSAGES
2010 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2011 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
2012# endif
2013# ifdef USE_LOCALE_MONETARY
2014 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2015 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
2016# endif
9821811f
KW
2017# ifdef USE_LOCALE_ADDRESS
2018 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
2019 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
2020# endif
2021# ifdef USE_LOCALE_IDENTIFICATION
2022 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
2023 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
2024# endif
2025# ifdef USE_LOCALE_MEASUREMENT
2026 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
2027 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
2028# endif
2029# ifdef USE_LOCALE_PAPER
2030 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
2031 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
2032# endif
2033# ifdef USE_LOCALE_TELEPHONE
2034 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
2035 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
2036# endif
948523db
KW
2037# ifdef LC_ALL
2038 assert(categories[LC_ALL_INDEX] == LC_ALL);
2039 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
2040 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
2041# endif
2042# endif /* DEBUGGING */
47280b20
KW
2043
2044 /* Initialize the cache of the program's UTF-8ness for the always known
2045 * locales C and POSIX */
2046 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
2047 sizeof(PL_locale_utf8ness));
2048
b56c4436 2049# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
2050
2051 /*
2052 * Ultrix setlocale(..., "") fails if there are no environment
2053 * variables from which to get a locale name.
2054 */
2055
b56c4436
KW
2056# ifndef LC_ALL
2057# error Ultrix without LC_ALL not implemented
2058# else
7d4bcc4a 2059
b56c4436
KW
2060 {
2061 bool done = FALSE;
ea92aad8
KW
2062 if (lang) {
2063 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
2064 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
2065 if (sl_result[LC_ALL_INDEX])
2066 done = TRUE;
2067 else
e5f10d49 2068 setlocale_failure = TRUE;
ea92aad8
KW
2069 }
2070 if (! setlocale_failure) {
2071 const char * locale_param;
2072 for (i = 0; i < LC_ALL_INDEX; i++) {
2073 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
2074 ? setlocale_init
2075 : NULL;
2076 sl_result[i] = do_setlocale_r(categories[i], locale_param);
2077 if (! sl_result[i]) {
2078 setlocale_failure = TRUE;
2079 }
2080 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 2081 }
c835d6be 2082 }
7d4bcc4a
KW
2083 }
2084
2085# endif /* LC_ALL */
e5f10d49 2086# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 2087
65ebb059 2088 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
2089 * the list. Normally the loop is executed just once. But if setting the
2090 * locale fails, inside the loop we add fallback trials to the array and so
2091 * will execute the loop multiple times */
c3fcd832
KW
2092 trial_locales[0] = setlocale_init;
2093 trial_locales_count = 1;
7d4bcc4a 2094
65ebb059
KW
2095 for (i= 0; i < trial_locales_count; i++) {
2096 const char * trial_locale = trial_locales[i];
2097
2098 if (i > 0) {
2099
2100 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
2101 * when i==0, but I (khw) don't think that behavior makes much
2102 * sense */
2103 setlocale_failure = FALSE;
2104
7d4bcc4a 2105# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 2106# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 2107
65ebb059
KW
2108 /* On Windows machines, an entry of "" after the 0th means to use
2109 * the system default locale, which we now proceed to get. */
2110 if (strEQ(trial_locale, "")) {
2111 unsigned int j;
2112
2113 /* Note that this may change the locale, but we are going to do
2114 * that anyway just below */
837ce802 2115 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 2116 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 2117
7d4bcc4a 2118 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
2119 * try */
2120 if (! system_default_locale) {
2121 goto next_iteration;
2122 }
2123 for (j = 0; j < trial_locales_count; j++) {
2124 if (strEQ(system_default_locale, trial_locales[j])) {
2125 goto next_iteration;
2126 }
2127 }
2128
2129 trial_locale = system_default_locale;
2130 }
ec0202b5
KW
2131# else
2132# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
2133# endif
7d4bcc4a 2134# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
2135
2136 } /* For i > 0 */
65ebb059 2137
7d4bcc4a
KW
2138# ifdef LC_ALL
2139
948523db
KW
2140 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
2141 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
2142 if (! sl_result[LC_ALL_INDEX]) {
49c85077 2143 setlocale_failure = TRUE;
7cd8b568
KW
2144 }
2145 else {
2146 /* Since LC_ALL succeeded, it should have changed all the other
2147 * categories it can to its value; so we massage things so that the
2148 * setlocales below just return their category's current values.
2149 * This adequately handles the case in NetBSD where LC_COLLATE may
2150 * not be defined for a locale, and setting it individually will
7d4bcc4a 2151 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
2152 * the POSIX locale. */
2153 trial_locale = NULL;
2154 }
7d4bcc4a
KW
2155
2156# endif /* LC_ALL */
98994639 2157
e5f10d49
KW
2158 if (! setlocale_failure) {
2159 unsigned int j;
2160 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2161 curlocales[j]
2162 = savepv(do_setlocale_r(categories[j], trial_locale));
2163 if (! curlocales[j]) {
2164 setlocale_failure = TRUE;
2165 }
2166 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2167 }
c835d6be 2168
e5f10d49
KW
2169 if (! setlocale_failure) { /* All succeeded */
2170 break; /* Exit trial_locales loop */
49c85077 2171 }
65ebb059 2172 }
98994639 2173
49c85077
KW
2174 /* Here, something failed; will need to try a fallback. */
2175 ok = 0;
65ebb059 2176
49c85077
KW
2177 if (i == 0) {
2178 unsigned int j;
98994639 2179
65ebb059 2180 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2181
2182# ifdef LC_ALL
98994639 2183
49c85077
KW
2184 PerlIO_printf(Perl_error_log,
2185 "perl: warning: Setting locale failed.\n");
98994639 2186
7d4bcc4a 2187# else /* !LC_ALL */
98994639 2188
49c85077
KW
2189 PerlIO_printf(Perl_error_log,
2190 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2191
e5f10d49
KW
2192 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2193 if (! curlocales[j]) {
2194 PerlIO_printf(Perl_error_log, category_names[j]);
2195 }
2196 else {
2197 Safefree(curlocales[j]);
2198 }
2199 }
7d4bcc4a 2200
7d4bcc4a 2201# endif /* LC_ALL */
98994639 2202
49c85077
KW
2203 PerlIO_printf(Perl_error_log,
2204 "perl: warning: Please check that your locale settings:\n");
98994639 2205
7d4bcc4a
KW
2206# ifdef __GLIBC__
2207
49c85077
KW
2208 PerlIO_printf(Perl_error_log,
2209 "\tLANGUAGE = %c%s%c,\n",
2210 language ? '"' : '(',
2211 language ? language : "unset",
2212 language ? '"' : ')');
7d4bcc4a 2213# endif
98994639 2214
49c85077
KW
2215 PerlIO_printf(Perl_error_log,
2216 "\tLC_ALL = %c%s%c,\n",
2217 lc_all ? '"' : '(',
2218 lc_all ? lc_all : "unset",
2219 lc_all ? '"' : ')');
98994639 2220
7d4bcc4a
KW
2221# if defined(USE_ENVIRON_ARRAY)
2222
49c85077 2223 {
cd999af9 2224 char **e;
d5e32b93
KW
2225
2226 /* Look through the environment for any variables of the
2227 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2228 * already handled above. These are assumed to be locale
2229 * settings. Output them and their values. */
cd999af9 2230 for (e = environ; *e; e++) {
d5e32b93
KW
2231 const STRLEN prefix_len = sizeof("LC_") - 1;
2232 STRLEN uppers_len;
2233
cd999af9 2234 if ( strBEGINs(*e, "LC_")
c8b388b0 2235 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2236 && (uppers_len = strspn(*e + prefix_len,
2237 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2238 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2239 {
2240 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2241 (int) (prefix_len + uppers_len), *e,
2242 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2243 }
2244 }
49c85077 2245 }
7d4bcc4a
KW
2246
2247# else
2248
49c85077
KW
2249 PerlIO_printf(Perl_error_log,
2250 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2251
2252# endif
98994639 2253
49c85077
KW
2254 PerlIO_printf(Perl_error_log,
2255 "\tLANG = %c%s%c\n",
2256 lang ? '"' : '(',
2257 lang ? lang : "unset",
2258 lang ? '"' : ')');
98994639 2259
49c85077
KW
2260 PerlIO_printf(Perl_error_log,
2261 " are supported and installed on your system.\n");
2262 }
98994639 2263
65ebb059 2264 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2265 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2266 * usually change the upper bound of the loop we are in.
2267 *
2268 * Since the system's default way of setting the locale has not
2269 * found one that works, We use Perl's defined ordering: LC_ALL,
2270 * LANG, and the C locale. We don't try the same locale twice, so
2271 * don't add to the list if already there. (On POSIX systems, the
2272 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2273 * but there's no harm done by doing it explicitly.
2274 *
2275 * Note that this tries the LC_ALL environment variable even on
2276 * systems which have no LC_ALL locale setting. This may or may
2277 * not have been originally intentional, but there's no real need
2278 * to change the behavior. */
65ebb059
KW
2279 if (lc_all) {
2280 for (j = 0; j < trial_locales_count; j++) {
2281 if (strEQ(lc_all, trial_locales[j])) {
2282 goto done_lc_all;
2283 }
2284 }
2285 trial_locales[trial_locales_count++] = lc_all;
2286 }
2287 done_lc_all:
98994639 2288
65ebb059
KW
2289 if (lang) {
2290 for (j = 0; j < trial_locales_count; j++) {
2291 if (strEQ(lang, trial_locales[j])) {
2292 goto done_lang;
2293 }
2294 }
2295 trial_locales[trial_locales_count++] = lang;
2296 }
2297 done_lang:
2298
7d4bcc4a
KW
2299# if defined(WIN32) && defined(LC_ALL)
2300
65ebb059
KW
2301 /* For Windows, we also try the system default locale before "C".
2302 * (If there exists a Windows without LC_ALL we skip this because
2303 * it gets too complicated. For those, the "C" is the next
2304 * fallback possibility). The "" is the same as the 0th element of
2305 * the array, but the code at the loop above knows to treat it
2306 * differently when not the 0th */
2307 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2308
2309# endif
65ebb059
KW
2310
2311 for (j = 0; j < trial_locales_count; j++) {
2312 if (strEQ("C", trial_locales[j])) {
2313 goto done_C;
2314 }
2315 }
2316 trial_locales[trial_locales_count++] = "C";
98994639 2317
65ebb059
KW
2318 done_C: ;
2319 } /* end of first time through the loop */
98994639 2320
7d4bcc4a
KW
2321# ifdef WIN32
2322
65ebb059 2323 next_iteration: ;
7d4bcc4a
KW
2324
2325# endif
65ebb059
KW
2326
2327 } /* end of looping through the trial locales */
2328
2329 if (ok < 1) { /* If we tried to fallback */
2330 const char* msg;
2331 if (! setlocale_failure) { /* fallback succeeded */
2332 msg = "Falling back to";
2333 }
2334 else { /* fallback failed */
e5f10d49 2335 unsigned int j;
98994639 2336
65ebb059
KW
2337 /* We dropped off the end of the loop, so have to decrement i to
2338 * get back to the value the last time through */
2339 i--;
98994639 2340
65ebb059
KW
2341 ok = -1;
2342 msg = "Failed to fall back to";
2343
2344 /* To continue, we should use whatever values we've got */
7d4bcc4a 2345
e5f10d49
KW
2346 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2347 Safefree(curlocales[j]);
2348 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2349 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2350 }
65ebb059
KW
2351 }
2352
2353 if (locwarn) {
2354 const char * description;
2355 const char * name = "";
2356 if (strEQ(trial_locales[i], "C")) {
2357 description = "the standard locale";
2358 name = "C";
2359 }
7d4bcc4a
KW
2360
2361# ifdef SYSTEM_DEFAULT_LOCALE
2362
65ebb059
KW
2363 else if (strEQ(trial_locales[i], "")) {
2364 description = "the system default locale";
2365 if (system_default_locale) {
2366 name = system_default_locale;
2367 }
2368 }
7d4bcc4a
KW
2369
2370# endif /* SYSTEM_DEFAULT_LOCALE */
2371
65ebb059
KW
2372 else {
2373 description = "a fallback locale";
2374 name = trial_locales[i];
2375 }
2376 if (name && strNE(name, "")) {
2377 PerlIO_printf(Perl_error_log,
2378 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2379 }
2380 else {
2381 PerlIO_printf(Perl_error_log,
2382 "perl: warning: %s %s.\n", msg, description);
2383 }
2384 }
2385 } /* End of tried to fallback */
98994639 2386
e5f10d49
KW
2387 /* Done with finding the locales; update our records */
2388
7d4bcc4a
KW
2389# ifdef USE_LOCALE_CTYPE
2390
948523db 2391 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2392
e5f10d49 2393# endif
7d4bcc4a
KW
2394# ifdef USE_LOCALE_COLLATE
2395
948523db 2396 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2397
e5f10d49 2398# endif
7d4bcc4a
KW
2399# ifdef USE_LOCALE_NUMERIC
2400
948523db 2401 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2402
2403# endif
2404
948523db 2405 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20
KW
2406
2407# if defined(USE_ITHREADS)
2408
2409 /* This caches whether each category's locale is UTF-8 or not. This
2410 * may involve changing the locale. It is ok to do this at
2411 * initialization time before any threads have started, but not later.
2412 * Caching means that if the program heeds our dictate not to change
2413 * locales in threaded applications, this data will remain valid, and
2414 * it may get queried without changing locales. If the environment is
2415 * such that all categories have the same locale, this isn't needed, as
2416 * the code will not change the locale; but this handles the uncommon
2417 * case where the environment has disparate locales for the categories
2418 * */
2419 (void) _is_cur_LC_category_utf8(categories[i]);
2420
2421# endif
2422
e5f10d49
KW
2423 Safefree(curlocales[i]);
2424 }
b310b053 2425
7d4bcc4a
KW
2426# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2427
49c85077 2428 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
2429 * locale is UTF-8. The call to new_ctype() just above has already
2430 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
2431 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
2432 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
2433 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
2434 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 2435
a05d7ebb 2436 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2437 This is an alternative to using the -C command line switch
2438 (the -C if present will override this). */
2439 {
dd374669 2440 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2441 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2442 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2443 PL_utf8cache = -1;
b310b053
JH
2444 }
2445
7d4bcc4a 2446# endif
7d4bcc4a
KW
2447# ifdef __GLIBC__
2448
175c4cf9 2449 Safefree(language);
7d4bcc4a
KW
2450
2451# endif
175c4cf9
KW
2452
2453 Safefree(lc_all);
2454 Safefree(lang);
2455
e3305790 2456#endif /* USE_LOCALE */
2fcc0ca9 2457#ifdef DEBUGGING
7d4bcc4a 2458
2fcc0ca9 2459 /* So won't continue to output stuff */
27cdc72e 2460 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2461
2fcc0ca9
KW
2462#endif
2463
98994639
HS
2464 return ok;
2465}
2466
98994639
HS
2467#ifdef USE_LOCALE_COLLATE
2468
a4a439fb 2469char *
a4a439fb
KW
2470Perl__mem_collxfrm(pTHX_ const char *input_string,
2471 STRLEN len, /* Length of 'input_string' */
2472 STRLEN *xlen, /* Set to length of returned string
2473 (not including the collation index
2474 prefix) */
2475 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2476 )
98994639 2477{
a4a439fb
KW
2478
2479 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2480 * differences. First, it handles embedded NULs. Second, it allocates a bit
2481 * more memory than needed for the transformed data itself. The real
55e5378d 2482 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2483 * the length of that, and doesn't include the collation index size.
2484 * Please see sv_collxfrm() to see how this is used. */
2485
55e5378d
KW
2486#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2487
6696cfa7
KW
2488 char * s = (char *) input_string;
2489 STRLEN s_strlen = strlen(input_string);
79f120c8 2490 char *xbuf = NULL;
55e5378d 2491 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2492 STRLEN length_in_chars;
c664130f 2493 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2494
a4a439fb
KW
2495 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2496
2497 /* Must be NUL-terminated */
2498 assert(*(input_string + len) == '\0');
7918f24d 2499
79f120c8
KW
2500 /* If this locale has defective collation, skip */
2501 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2502 DEBUG_L(PerlIO_printf(Perl_debug_log,
2503 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2504 goto bad;
2505 }
2506
6696cfa7
KW
2507 /* Replace any embedded NULs with the control that sorts before any others.
2508 * This will give as good as possible results on strings that don't
2509 * otherwise contain that character, but otherwise there may be
2510 * less-than-perfect results with that character and NUL. This is
fdc080f3 2511 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2512 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2513 NUL */
6696cfa7
KW
2514 char * e = s + len;
2515 char * sans_nuls;
fdc080f3 2516 STRLEN sans_nuls_len;
94762aa0 2517 int try_non_controls;
afc4976f
KW
2518 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2519 making sure 2nd byte is NUL.
2520 */
2521 STRLEN this_replacement_len;
2522
1e4c9676
KW
2523 /* If we don't know what non-NUL control character sorts lowest for
2524 * this locale, find it */
f28f4d2a 2525 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2526 int j;
afc4976f 2527 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2528 includes the collation index
2529 prefixed. */
2530
91c0e2e0 2531 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2532
2533 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2534 * NUL, in which case we instead look for any character. Controls
2535 * are preferred because collation order is, in general, context
2536 * sensitive, with adjoining characters affecting the order, and
2537 * controls are less likely to have such interactions, allowing the
2538 * NUL-replacement to stand on its own. (Another way to look at it
2539 * is to imagine what would happen if the NUL were replaced by a
2540 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2541 for (try_non_controls = 0;
2542 try_non_controls < 2;
2543 try_non_controls++)
2544 {
d4ff9586
KW
2545 /* Look through all legal code points (NUL isn't) */
2546 for (j = 1; j < 256; j++) {
2547 char * x; /* j's xfrm plus collation index */
2548 STRLEN x_len; /* length of 'x' */
2549 STRLEN trial_len = 1;
736a4fed 2550 char cur_source[] = { '\0', '\0' };
d4ff9586 2551
736a4fed
KW
2552 /* Skip non-controls the first time through the loop. The
2553 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2554 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2555 ? ! isCNTRL_L1(j)
2556 : ! isCNTRL_LC(j))
2557 {
d4ff9586 2558 continue;
6696cfa7 2559 }
6696cfa7 2560
736a4fed
KW
2561 /* Create a 1-char string of the current code point */
2562 cur_source[0] = (char) j;
2563
d4ff9586
KW
2564 /* Then transform it */
2565 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2566 0 /* The string is not in UTF-8 */);
6696cfa7 2567
1e4c9676 2568 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2569 * */
2570 if (! x) {
2571 continue;
2572 }
6696cfa7 2573
d4ff9586
KW
2574 /* If this character's transformation is lower than
2575 * the current lowest, this one becomes the lowest */
2576 if ( cur_min_x == NULL
2577 || strLT(x + COLLXFRM_HDR_LEN,
2578 cur_min_x + COLLXFRM_HDR_LEN))
2579 {
f28f4d2a 2580 PL_strxfrm_NUL_replacement = j;
d4ff9586 2581 cur_min_x = x;
d4ff9586
KW
2582 }
2583 else {
2584 Safefree(x);
2585 }
1e4c9676 2586 } /* end of loop through all 255 characters */
6696cfa7 2587
1e4c9676 2588 /* Stop looking if found */
94762aa0
KW
2589 if (cur_min_x) {
2590 break;
2591 }
2592
2593 /* Unlikely, but possible, if there aren't any controls that
2594 * work in the locale, repeat the loop, looking for any
2595 * character that works */
2596 DEBUG_L(PerlIO_printf(Perl_debug_log,
2597 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2598 } /* End of loop to try first the controls, then any char */
6696cfa7 2599
94762aa0
KW
2600 if (! cur_min_x) {
2601 DEBUG_L(PerlIO_printf(Perl_debug_log,
2602 "_mem_collxfrm: Couldn't find any character to replace"
2603 " embedded NULs in locale %s with", PL_collation_name));
2604 goto bad;
58eebef2
KW
2605 }
2606
94762aa0
KW
2607 DEBUG_L(PerlIO_printf(Perl_debug_log,
2608 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2609 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2610
6696cfa7 2611 Safefree(cur_min_x);
1e4c9676 2612 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2613
2614 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2615 * UTF8-ness of the original */
f28f4d2a
KW
2616 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2617 this_replacement_char[0] =
2618 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2619 this_replacement_char[1] =
2620 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2621 this_replacement_len = 2;
2622 }
2623 else {
f28f4d2a 2624 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2625 /* this_replacement_char[1] = '\0' was done at initialization */
2626 this_replacement_len = 1;
6696cfa7
KW
2627 }
2628
2629 /* The worst case length for the replaced string would be if every
2630 * character in it is NUL. Multiply that by the length of each
2631 * replacement, and allow for a trailing NUL */
afc4976f 2632 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2633 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2634 *sans_nuls = '\0';
2635
6696cfa7
KW
2636 /* Replace each NUL with the lowest collating control. Loop until have
2637 * exhausted all the NULs */
2638 while (s + s_strlen < e) {
6069d6c5 2639 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2640
2641 /* Do the actual replacement */
6069d6c5 2642 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2643
2644 /* Move past the input NUL */
2645 s += s_strlen + 1;
2646 s_strlen = strlen(s);
2647 }
2648
2649 /* And add anything that trails the final NUL */
6069d6c5 2650 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2651
2652 /* Switch so below we transform this modified string */
2653 s = sans_nuls;
2654 len = strlen(s);
1e4c9676 2655 } /* End of replacing NULs */
6696cfa7 2656
a4a439fb
KW
2657 /* Make sure the UTF8ness of the string and locale match */
2658 if (utf8 != PL_in_utf8_COLLATE_locale) {
2659 const char * const t = s; /* Temporary so we can later find where the
2660 input was */
2661
2662 /* Here they don't match. Change the string's to be what the locale is
2663 * expecting */
2664
2665 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2666 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2667 utf8 = TRUE;
2668 }
2669 else { /* locale is not UTF-8; but input is; downgrade the input */
2670
2671 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2672
2673 /* If the downgrade was successful we are done, but if the input
2674 * contains things that require UTF-8 to represent, have to do
2675 * damage control ... */
2676 if (UNLIKELY(utf8)) {
2677
2678 /* What we do is construct a non-UTF-8 string with
2679 * 1) the characters representable by a single byte converted
2680 * to be so (if necessary);
2681 * 2) and the rest converted to collate the same as the
2682 * highest collating representable character. That makes
2683 * them collate at the end. This is similar to how we
2684 * handle embedded NULs, but we use the highest collating
2685 * code point instead of the smallest. Like the NUL case,
2686 * this isn't perfect, but is the best we can reasonably
2687 * do. Every above-255 code point will sort the same as
2688 * the highest-sorting 0-255 code point. If that code
2689 * point can combine in a sequence with some other code
2690 * points for weight calculations, us changing something to
2691 * be it can adversely affect the results. But in most
2692 * cases, it should work reasonably. And note that this is
2693 * really an illegal situation: using code points above 255
2694 * on a locale where only 0-255 are valid. If two strings
2695 * sort entirely equal, then the sort order for the
2696 * above-255 code points will be in code point order. */
2697
2698 utf8 = FALSE;
2699
2700 /* If we haven't calculated the code point with the maximum
2701 * collating order for this locale, do so now */
2702 if (! PL_strxfrm_max_cp) {
2703 int j;
2704
2705 /* The current transformed string that collates the
2706 * highest (except it also includes the prefixed collation
2707 * index. */
2708 char * cur_max_x = NULL;
2709
2710 /* Look through all legal code points (NUL isn't) */
2711 for (j = 1; j < 256; j++) {
2712 char * x;
2713 STRLEN x_len;
736a4fed 2714 char cur_source[] = { '\0', '\0' };
a4a439fb 2715
736a4fed
KW
2716 /* Create a 1-char string of the current code point */
2717 cur_source[0] = (char) j;
a4a439fb
KW
2718
2719 /* Then transform it */
2720 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2721
2722 /* If something went wrong (which it shouldn't), just
2723 * ignore this code point */
94762aa0 2724 if (! x) {
a4a439fb
KW
2725 continue;
2726 }
2727
2728 /* If this character's transformation is higher than
2729 * the current highest, this one becomes the highest */
2730 if ( cur_max_x == NULL
55e5378d
KW
2731 || strGT(x + COLLXFRM_HDR_LEN,
2732 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2733 {
2734 PL_strxfrm_max_cp = j;
2735 cur_max_x = x;
2736 }
2737 else {
2738 Safefree(x);
2739 }
2740 }
2741
94762aa0
KW
2742 if (! cur_max_x) {
2743 DEBUG_L(PerlIO_printf(Perl_debug_log,
2744 "_mem_collxfrm: Couldn't find any character to"
2745 " replace above-Latin1 chars in locale %s with",
2746 PL_collation_name));
2747 goto bad;
2748 }
2749
58eebef2
KW
2750 DEBUG_L(PerlIO_printf(Perl_debug_log,
2751 "_mem_collxfrm: highest 1-byte collating character"
2752 " in locale %s is 0x%02X\n",
2753 PL_collation_name,
2754 PL_strxfrm_max_cp));
58eebef2 2755
a4a439fb
KW
2756 Safefree(cur_max_x);
2757 }
2758
2759 /* Here we know which legal code point collates the highest.
2760 * We are ready to construct the non-UTF-8 string. The length
2761 * will be at least 1 byte smaller than the input string
2762 * (because we changed at least one 2-byte character into a
2763 * single byte), but that is eaten up by the trailing NUL */
2764 Newx(s, len, char);
2765
2766 {
2767 STRLEN i;
2768 STRLEN d= 0;
042d9e50 2769 char * e = (char *) t + len;
a4a439fb
KW
2770
2771 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2772 U8 cur_char = t[i];
2773 if (UTF8_IS_INVARIANT(cur_char)) {
2774 s[d++] = cur_char;
2775 }
042d9e50 2776 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2777 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2778 }
2779 else { /* Replace illegal cp with highest collating
2780 one */
2781 s[d++] = PL_strxfrm_max_cp;
2782 }
2783 }
2784 s[d++] = '\0';
2785 Renew(s, d, char); /* Free up unused space */
2786 }
2787 }
2788 }
2789
2790 /* Here, we have constructed a modified version of the input. It could
2791 * be that we already had a modified copy before we did this version.
2792 * If so, that copy is no longer needed */
2793 if (t != input_string) {
2794 Safefree(t);
2795 }
2796 }
2797
17f41037
KW
2798 length_in_chars = (utf8)
2799 ? utf8_length((U8 *) s, (U8 *) s + len)
2800 : len;
2801
59c018b9
KW
2802 /* The first element in the output is the collation id, used by
2803 * sv_collxfrm(); then comes the space for the transformed string. The
2804 * equation should give us a good estimate as to how much is needed */
55e5378d 2805 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2806 + PL_collxfrm_base
17f41037 2807 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2808 Newx(xbuf, xAlloc, char);
c7202dee
KW
2809 if (UNLIKELY(! xbuf)) {
2810 DEBUG_L(PerlIO_printf(Perl_debug_log,
2811 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2812 goto bad;
c7202dee 2813 }
98994639 2814
d35fca5f 2815 /* Store the collation id */
98994639 2816 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2817
2818 /* Then the transformation of the input. We loop until successful, or we
2819 * give up */
4ebeff16 2820 for (;;) {
1adab0a7 2821
55e5378d 2822 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2823
2824 /* If the transformed string occupies less space than we told strxfrm()
2825 * was available, it means it successfully transformed the whole
2826 * string. */
55e5378d 2827 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2828
1adab0a7
KW
2829 /* Some systems include a trailing NUL in the returned length.
2830 * Ignore it, using a loop in case multiple trailing NULs are
2831 * returned. */
2832 while ( (*xlen) > 0
2833 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2834 {
2835 (*xlen)--;
2836 }
2837
17f41037
KW
2838 /* If the first try didn't get it, it means our prediction was low.
2839 * Modify the coefficients so that we predict a larger value in any
2840 * future transformations */
2841 if (! first_time) {
2842 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2843 STRLEN computed_guess = PL_collxfrm_base
2844 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2845
2846 /* On zero-length input, just keep current slope instead of
2847 * dividing by 0 */
2848 const STRLEN new_m = (length_in_chars != 0)
2849 ? needed / length_in_chars
2850 : PL_collxfrm_mult;
17f41037
KW
2851
2852 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2853 "%s: %d: initial size of %zu bytes for a length "
2854 "%zu string was insufficient, %zu needed\n",
17f41037 2855 __FILE__, __LINE__,
b07929e4 2856 computed_guess, length_in_chars, needed));
17f41037
KW
2857
2858 /* If slope increased, use it, but discard this result for
2859 * length 1 strings, as we can't be sure that it's a real slope
2860 * change */
2861 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
2862
2863# ifdef DEBUGGING
2864
17f41037
KW
2865 STRLEN old_m = PL_collxfrm_mult;
2866 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
2867
2868# endif
2869
17f41037
KW
2870 PL_collxfrm_mult = new_m;
2871 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2872 computed_guess = PL_collxfrm_base
2873 + (PL_collxfrm_mult * length_in_chars);
2874 if (computed_guess < needed) {
2875 PL_collxfrm_base += needed - computed_guess;
2876 }
2877
2878 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2879 "%s: %d: slope is now %zu; was %zu, base "
2880 "is now %zu; was %zu\n",
17f41037 2881 __FILE__, __LINE__,
b07929e4
KW
2882 PL_collxfrm_mult, old_m,
2883 PL_collxfrm_base, old_b));
17f41037
KW
2884 }
2885 else { /* Slope didn't change, but 'b' did */
2886 const STRLEN new_b = needed
2887 - computed_guess
2888 + PL_collxfrm_base;
2889 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2890 "%s: %d: base is now %zu; was %zu\n",
17f41037 2891 __FILE__, __LINE__,
b07929e4 2892 new_b, PL_collxfrm_base));
17f41037
KW
2893 PL_collxfrm_base = new_b;
2894 }
2895 }
2896
4ebeff16
KW
2897 break;
2898 }
bb0f664e 2899
c7202dee
KW
2900 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2901 DEBUG_L(PerlIO_printf(Perl_debug_log,
2902 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2903 *xlen, PERL_INT_MAX));
4ebeff16 2904 goto bad;
c7202dee 2905 }
d35fca5f 2906
c664130f 2907 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
2908 * (usually not including the trailing NUL) when it fails due to not
2909 * enough space being provided. Assume that this is the case unless
2910 * it's been proven otherwise */
c664130f 2911 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 2912 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
2913 }
2914 else { /* Here, either:
2915 * 1) The strxfrm() has previously shown bad behavior; or
2916 * 2) It isn't the first time through the loop, which means
2917 * that the strxfrm() is now showing bad behavior, because
2918 * we gave it what it said was needed in the previous
2919 * iteration, and it came back saying it needed still more.
2920 * (Many versions of cygwin fit this. When the buffer size
2921 * isn't sufficient, they return the input size instead of
2922 * how much is needed.)
d4ff9586
KW
2923 * Increase the buffer size by a fixed percentage and try again.
2924 * */
6ddd902c 2925 xAlloc += (xAlloc / 4) + 1;
c664130f 2926 PL_strxfrm_is_behaved = FALSE;
c664130f 2927
7d4bcc4a
KW
2928# ifdef DEBUGGING
2929
58eebef2
KW
2930 if (DEBUG_Lv_TEST || debug_initialization) {
2931 PerlIO_printf(Perl_debug_log,
2932 "_mem_collxfrm required more space than previously calculated"
b07929e4 2933 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 2934 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 2935 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 2936 }
7d4bcc4a
KW
2937
2938# endif
2939
58eebef2 2940 }
c664130f 2941
4ebeff16 2942 Renew(xbuf, xAlloc, char);
c7202dee
KW
2943 if (UNLIKELY(! xbuf)) {
2944 DEBUG_L(PerlIO_printf(Perl_debug_log,
2945 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 2946 goto bad;
c7202dee 2947 }
c664130f
KW
2948
2949 first_time = FALSE;
4ebeff16 2950 }
98994639 2951
6696cfa7 2952
7d4bcc4a
KW
2953# ifdef DEBUGGING
2954
58eebef2 2955 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
2956
2957 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
2958 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
2959 PerlIO_printf(Perl_debug_log, "%s\n",
2960 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
2961 *xlen, 1));
58eebef2 2962 }
7d4bcc4a
KW
2963
2964# endif
58eebef2 2965
3c5f993e 2966 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 2967 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 2968
6696cfa7
KW
2969 if (s != input_string) {
2970 Safefree(s);
98994639
HS
2971 }
2972
98994639
HS
2973 return xbuf;
2974
2975 bad:
2976 Safefree(xbuf);
6696cfa7
KW
2977 if (s != input_string) {
2978 Safefree(s);
2979 }
98994639 2980 *xlen = 0;
7d4bcc4a
KW
2981
2982# ifdef DEBUGGING
2983
58eebef2 2984 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 2985 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 2986 }
7d4bcc4a
KW
2987
2988# endif
2989
98994639
HS
2990 return NULL;
2991}
2992
7d4bcc4a 2993# ifdef DEBUGGING
c7202dee 2994
4cbaac56 2995STATIC void
c7202dee
KW
2996S_print_collxfrm_input_and_return(pTHX_
2997 const char * const s,
2998 const char * const e,
2999 const STRLEN * const xlen,
3000 const bool is_utf8)
3001{
c7202dee
KW
3002
3003 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
3004
511e4ff7
DM
3005 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
3006 (UV)PL_collation_ix);
c7202dee 3007 if (xlen) {
08b6dc1d 3008 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
3009 }
3010 else {
3011 PerlIO_printf(Perl_debug_log, "NULL");
3012 }
3013 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
3014 PL_collation_name);
9c8a6dc2
KW
3015 print_bytes_for_locale(s, e, is_utf8);
3016
3017 PerlIO_printf(Perl_debug_log, "'\n");
3018}
3019
3020STATIC void
3021S_print_bytes_for_locale(pTHX_
3022 const char * const s,
3023 const char * const e,
3024 const bool is_utf8)
3025{
3026 const char * t = s;
3027 bool prev_was_printable = TRUE;
3028 bool first_time = TRUE;
3029
3030 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
3031
3032 while (t < e) {
3033 UV cp = (is_utf8)
3034 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
3035 : * (U8 *) t;
3036 if (isPRINT(cp)) {
3037 if (! prev_was_printable) {
3038 PerlIO_printf(Perl_debug_log, " ");
3039 }
3040 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
3041 prev_was_printable = TRUE;
3042 }
3043 else {
3044 if (! first_time) {
3045 PerlIO_printf(Perl_debug_log, " ");
3046 }
147e3846 3047 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
3048 prev_was_printable = FALSE;
3049 }
3050 t += (is_utf8) ? UTF8SKIP(t) : 1;
3051 first_time = FALSE;
3052 }
c7202dee
KW
3053}
3054
7d4bcc4a 3055# endif /* #ifdef DEBUGGING */
98994639 3056#endif /* USE_LOCALE_COLLATE */
58eebef2 3057
8ef6e574
KW
3058#ifdef USE_LOCALE
3059
c1284011
KW
3060bool
3061Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
3062{
3063 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
3064 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 3065 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
3066 * could give the wrong result. The result will very likely be correct for
3067 * languages that have commonly used non-ASCII characters, but for notably
3068 * English, it comes down to if the locale's name ends in something like
3069 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61 3070
47280b20 3071 /* Name of current locale corresponding to the input category */
8de4332b 3072 const char *save_input_locale = NULL;
47280b20
KW
3073
3074 bool is_utf8 = FALSE; /* The return value */
7d74bb61
KW
3075 STRLEN final_pos;
3076
47280b20
KW
3077 /* The variables below are for the cache of previous lookups using this
3078 * function. The cache is a C string, described at the definition for
3079 * 'C_and_POSIX_utf8ness'.
3080 *
3081 * The first part of the cache is fixed, for the C and POSIX locales. The
3082 * varying part starts just after them. */
3083 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
3084
3085 Size_t utf8ness_cache_size; /* Size of the varying portion */
3086 Size_t input_name_len; /* Length in bytes of save_input_locale */
3087 Size_t input_name_len_with_overhead; /* plus extra chars used to store
3088 the name in the cache */
3089 char * delimited; /* The name plus the delimiters used to store
3090 it in the cache */
3091 char * name_pos; /* position of 'delimited' in the cache, or 0
3092 if not there */
3093
3094
7d4bcc4a
KW
3095# ifdef LC_ALL
3096
7d74bb61 3097 assert(category != LC_ALL);
7d4bcc4a
KW
3098
3099# endif
7d74bb61 3100
47280b20 3101 /* Get the desired category's locale */
837ce802 3102 save_input_locale = do_setlocale_r(category, NULL);
7d74bb61 3103 if (! save_input_locale) {
47280b20
KW
3104 Perl_croak(aTHX_
3105 "panic: %s: %d: Could not find current locale for %s\n",
3106 __FILE__, __LINE__, category_name(category));
7d74bb61 3107 }
47280b20 3108
b07fffd1 3109 save_input_locale = stdize_locale(savepv(save_input_locale));
47280b20
KW
3110 DEBUG_L(PerlIO_printf(Perl_debug_log,
3111 "Current locale for %s is %s\n",
3112 category_name(category), save_input_locale));
3113
3114 input_name_len = strlen(save_input_locale);
3115
3116 /* In our cache, each name is accompanied by two delimiters and a single
3117 * utf8ness digit */
3118 input_name_len_with_overhead = input_name_len + 3;
3119
3120 /* Allocate and populate space for a copy of the name surrounded by the
3121 * delimiters */
3122 Newx(delimited, input_name_len_with_overhead, char);
3123 delimited[0] = UTF8NESS_SEP[0];
3124 Copy(save_input_locale, delimited + 1, input_name_len, char);
3125 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
3126 delimited[input_name_len+2] = '\0';
3127
3128 /* And see if that is in the cache */
3129 name_pos = instr(PL_locale_utf8ness, delimited);
3130 if (name_pos) {
3131 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
3132
3133# ifdef DEBUGGING
3134
3135 if (DEBUG_Lv_TEST || debug_initialization) {
3136 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
3137 save_input_locale, is_utf8);
3138 }
3139
3140# endif
3141
3142 /* And, if not already in that position, move it to the beginning of
3143 * the non-constant portion of the list, since it is the most recently
3144 * used. (We don't have to worry about overflow, since just moving
3145 * existing names around) */
3146 if (name_pos > utf8ness_cache) {
3147 Move(utf8ness_cache,
3148 utf8ness_cache + input_name_len_with_overhead,
3149 name_pos - utf8ness_cache, char);
3150 Copy(delimited,
3151 utf8ness_cache,
3152 input_name_len_with_overhead - 1, char);
3153 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3154 }
3155
3156 Safefree(delimited);
b07fffd1 3157 Safefree(save_input_locale);
47280b20 3158 return is_utf8;
7d74bb61
KW
3159 }
3160
47280b20
KW
3161 /* Here we don't have stored the utf8ness for the input locale. We have to
3162 * calculate it */
3163
7d4bcc4a 3164# if defined(USE_LOCALE_CTYPE) \
1d958db2 3165 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 3166
1d958db2 3167 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
3168
3169 char *save_ctype_locale = NULL;
7d74bb61 3170
119ee68b 3171 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
3172
3173 /* Get the current LC_CTYPE locale */
837ce802 3174 save_ctype_locale = do_setlocale_c(LC_CTYPE, NULL);
7d74bb61 3175 if (! save_ctype_locale) {
69014004
KW
3176 DEBUG_L(PerlIO_printf(Perl_debug_log,
3177 "Could not find current locale for LC_CTYPE\n"));
7d74bb61
KW
3178 goto cant_use_nllanginfo;
3179 }
4f72bb37 3180 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
3181
3182 /* If LC_CTYPE and the desired category use the same locale, this
3183 * means that finding the value for LC_CTYPE is the same as finding
3184 * the value for the desired category. Otherwise, switch LC_CTYPE
3185 * to the desired category's locale */
3186 if (strEQ(save_ctype_locale, save_input_locale)) {
3187 Safefree(save_ctype_locale);
3188 save_ctype_locale = NULL;
3189 }
837ce802 3190 else if (! do_setlocale_c(LC_CTYPE, save_input_locale)) {
69014004
KW
3191 DEBUG_L(PerlIO_printf(Perl_debug_log,
3192 "Could not change LC_CTYPE locale to %s\n",
3193 save_input_locale));
7d74bb61
KW
3194 Safefree(save_ctype_locale);
3195 goto cant_use_nllanginfo;
3196 }
3197 }
3198
69014004
KW
3199 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
3200 save_input_locale));
3201
7d74bb61 3202 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
3203 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
3204 * should give the correct results */
119ee68b 3205
7d4bcc4a
KW
3206# if defined(HAS_NL_LANGINFO) && defined(CODESET)
3207
291a84fb 3208 { /* The task is easiest if the platform has this POSIX 2001 function */
c70a3e68
KW
3209 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
3210 /* FALSE => already in dest locale */
119ee68b 3211
c70a3e68
KW
3212 DEBUG_L(PerlIO_printf(Perl_debug_log,
3213 "\tnllanginfo returned CODESET '%s'\n", codeset));
3214
3215 if (codeset && strNE(codeset, "")) {
1d958db2
KW
3216 /* If we switched LC_CTYPE, switch back */
3217 if (save_ctype_locale) {
837ce802 3218 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2
KW
3219 Safefree(save_ctype_locale);
3220 }
3221
abc1d81d
KW
3222 is_utf8 = ( ( strlen(codeset) == STRLENs("UTF-8")
3223 && foldEQ(codeset, STR_WITH_LEN("UTF-8")))
3224 || ( strlen(codeset) == STRLENs("UTF8")
3225 && foldEQ(codeset, STR_WITH_LEN("UTF8"))));
1d958db2 3226
69014004
KW
3227 DEBUG_L(PerlIO_printf(Perl_debug_log,
3228 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
3229 codeset, is_utf8));
47280b20 3230 goto finish_and_return;
1d958db2 3231 }
119ee68b
KW
3232 }
3233
7d4bcc4a
KW
3234# endif
3235# ifdef MB_CUR_MAX
1d958db2
KW
3236
3237 /* Here, either we don't have nl_langinfo, or it didn't return a
3238 * codeset. Try MB_CUR_MAX */
3239
119ee68b
KW
3240 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3241 * Unicode code point. Since UTF-8 is the only non-single byte
3242 * encoding we handle, we just say any such encoding is UTF-8, and if
3243 * turns out to be wrong, other things will fail */
5f7616bd 3244 is_utf8 = (unsigned) MB_CUR_MAX >= STRLENs(MAX_UNICODE_UTF8);
119ee68b 3245
69014004
KW
3246 DEBUG_L(PerlIO_printf(Perl_debug_log,
3247 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
3248 (int) MB_CUR_MAX, is_utf8));
3249
119ee68b
KW
3250 Safefree(save_input_locale);
3251
7d4bcc4a 3252# ifdef HAS_MBTOWC
119ee68b
KW
3253
3254 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
3255 * since they are both in the C99 standard. We can feed a known byte
3256 * string to the latter function, and check that it gives the expected
3257 * result */
3258 if (is_utf8) {
3259 wchar_t wc;
51fc4b19
KW
3260 int len;
3261
856b881c 3262 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 3263 errno = 0;
b1d4925c 3264 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
51fc4b19 3265
b1d4925c
KW
3266
3267 if ( len != STRLENs(REPLACEMENT_CHARACTER_UTF8)
5579633c 3268 || wc != (wchar_t) UNICODE_REPLACEMENT)
119ee68b
KW
3269 {
3270 is_utf8 = FALSE;
b1d4925c
KW
3271 DEBUG_L(PerlIO_printf(Perl_debug_log, "\replacement=U+%x\n",
3272 (unsigned int)wc));
69014004
KW
3273 DEBUG_L(PerlIO_printf(Perl_debug_log,
3274 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
51fc4b19 3275 len, errno));
119ee68b
KW
3276 }
3277 }
7d4bcc4a
KW
3278
3279# endif
119ee68b 3280
1d958db2
KW
3281 /* If we switched LC_CTYPE, switch back */
3282 if (save_ctype_locale) {
837ce802 3283 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2 3284 Safefree(save_ctype_locale);
119ee68b 3285 }
7d74bb61 3286
47280b20 3287 goto finish_and_return;
7d4bcc4a
KW
3288
3289# endif
3290
7d74bb61 3291 }
119ee68b 3292
7d74bb61
KW
3293 cant_use_nllanginfo:
3294
7d4bcc4a 3295# else /* nl_langinfo should work if available, so don't bother compiling this
0080c90a
KW
3296 fallback code. The final fallback of looking at the name is
3297 compiled, and will be executed if nl_langinfo fails */
7d74bb61 3298
97f4de96
KW
3299 /* nl_langinfo not available or failed somehow. Next try looking at the
3300 * currency symbol to see if it disambiguates things. Often that will be
3301 * in the native script, and if the symbol isn't in UTF-8, we know that the
3302 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
3303 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
3304 * */
fa9b773e 3305
7d4bcc4a
KW
3306# ifdef HAS_LOCALECONV
3307# ifdef USE_LOCALE_MONETARY
3308
fa9b773e
KW
3309 {
3310 char *save_monetary_locale = NULL;
fa9b773e 3311 bool only_ascii = FALSE;
13542a67 3312 struct lconv* lc;
fa9b773e 3313
97f4de96
KW
3314 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
3315 * the desired category, if it isn't that locale already */
3316
fa9b773e
KW
3317 if (category != LC_MONETARY) {
3318
837ce802 3319 save_monetary_locale = do_setlocale_c(LC_MONETARY, NULL);
fa9b773e 3320 if (! save_monetary_locale) {
69014004
KW
3321 DEBUG_L(PerlIO_printf(Perl_debug_log,
3322 "Could not find current locale for LC_MONETARY\n"));
fa9b773e
KW
3323 goto cant_use_monetary;
3324 }
4f72bb37 3325 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 3326
13542a67
KW
3327 if (strEQ(save_monetary_locale, save_input_locale)) {
3328 Safefree(save_monetary_locale);
3329 save_monetary_locale = NULL;
3330 }
837ce802 3331 else if (! do_setlocale_c(LC_MONETARY, save_input_locale)) {
59c234b4
KW
3332 DEBUG_L(PerlIO_printf(Perl_debug_log,
3333 "Could not change LC_MONETARY locale to %s\n",
3334 save_input_locale));
3335 Safefree(save_monetary_locale);
3336 goto cant_use_monetary;
fa9b773e
KW
3337 }
3338 }
3339
3340 /* Here the current LC_MONETARY is set to the locale of the category
3341 * whose information is desired. */
3342
13542a67
KW
3343 lc = localeconv();
3344 if (! lc
3345 || ! lc->currency_symbol
c5f058df 3346 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
3347 {
3348 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));
3349 only_ascii = TRUE;
3350 }
3351 else {
3352 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
3353 }
3354
3355 /* If we changed it, restore LC_MONETARY to its original locale */
3356 if (save_monetary_locale) {
837ce802 3357 do_setlocale_c(LC_MONETARY, save_monetary_locale);
fa9b773e
KW
3358 Safefree(save_monetary_locale);
3359 }
3360
13542a67 3361 if (! only_ascii) {
fa9b773e 3362
59c234b4
KW
3363 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3364 * otherwise assume the locale is UTF-8 if and only if the symbol
3365 * is non-ascii UTF-8. */
3366 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
3367 save_input_locale, is_utf8));
47280b20 3368 goto finish_and_return;
13542a67 3369 }
fa9b773e
KW
3370 }
3371 cant_use_monetary:
3372
7d4bcc4a
KW
3373# endif /* USE_LOCALE_MONETARY */
3374# endif /* HAS_LOCALECONV */
fa9b773e 3375
7d4bcc4a 3376# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e
KW
3377
3378/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3379 * the names of the months and weekdays, timezone, and am/pm indicator */
3380 {
3381 char *save_time_locale = NULL;
3382 int hour = 10;
3383 bool is_dst = FALSE;
3384 int dom = 1;
3385 int month = 0;
3386 int i;
3387 char * formatted_time;
3388
3389
3390 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
3391 * desired category, if it isn't that locale already */
3392
3393 if (category != LC_TIME) {
3394
837ce802 3395 save_time_locale = do_setlocale_c(LC_TIME, NULL);
15f7e74e
KW
3396 if (! save_time_locale) {
3397 DEBUG_L(PerlIO_printf(Perl_debug_log,
3398 "Could not find current locale for LC_TIME\n"));
3399 goto cant_use_time;
3400 }
3401 save_time_locale = stdize_locale(savepv(save_time_locale));
3402
3403 if (strEQ(save_time_locale, save_input_locale)) {
3404 Safefree(save_time_locale);
3405 save_time_locale = NULL;
3406 }
837ce802 3407 else if (! do_setlocale_c(LC_TIME, save_input_locale)) {
15f7e74e
KW
3408 DEBUG_L(PerlIO_printf(Perl_debug_log,
3409 "Could not change LC_TIME locale to %s\n",
3410 save_input_locale));
3411 Safefree(save_time_locale);
3412 goto cant_use_time;
3413 }
3414 }
3415
3416 /* Here the current LC_TIME is set to the locale of the category
3417 * whose information is desired. Look at all the days of the week and
9f10db87 3418 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
3419 * characters. The first such a one found will tell us if the locale
3420 * is UTF-8 or not */
3421
3422 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3423 formatted_time = my_strftime("%A %B %Z %p",
3ae5cd07 3424 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
c5f058df
KW
3425 if ( ! formatted_time
3426 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3427 {
15f7e74e
KW
3428
3429 /* Here, we didn't find a non-ASCII. Try the next time through
3430 * with the complemented dst and am/pm, and try with the next
3431 * weekday. After we have gotten all weekdays, try the next
3432 * month */
3433 is_dst = ! is_dst;
3434 hour = (hour + 12) % 24;
3435 dom++;
3436 if (i > 6) {
3437 month++;
3438 }
3439 continue;
3440 }
3441
3442 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3443 * false otherwise. But first, restore LC_TIME to its original
3444 * locale if we changed it */
3445 if (save_time_locale) {
837ce802 3446 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3447 Safefree(save_time_locale);
3448 }
3449
3450 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3451 save_input_locale,
3452 is_utf8_string((U8 *) formatted_time, 0)));
47280b20
KW
3453 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
3454 goto finish_and_return;
15f7e74e
KW
3455 }
3456
3457 /* Falling off the end of the loop indicates all the names were just
3458 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3459 * to its original locale */
3460 if (save_time_locale) {
837ce802 3461 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3462 Safefree(save_time_locale);
3463 }
3464 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));
3465 }
3466 cant_use_time:
3467
7d4bcc4a 3468# endif
15f7e74e 3469
7d4bcc4a 3470# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
3471
3472/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
3473 * on our dromedary test machine, which has over 700 locales. There, this
3474 * added no value to looking at the currency symbol and the time strings. I
3475 * left it in so as to avoid rewriting it if real-world experience indicates
3476 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
3477 * haven't found illegal utf8, we continue on and examine all the strerror()
3478 * messages on the platform for utf8ness. If all are ASCII, we still don't
3479 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
3480 * utf8ness. The reason this doesn't help much is that the messages may not
3481 * have been translated into the locale. The currency symbol and time strings
3482 * are much more likely to have been translated. */
3483 {
855aeb93 3484 int e;
5857e934 3485 bool non_ascii = FALSE;
855aeb93 3486 char *save_messages_locale = NULL;
5857e934 3487 const char * errmsg = NULL;
855aeb93 3488
5857e934
KW
3489 /* Like above, we set LC_MESSAGES to the locale of the desired
3490 * category, if it isn't that locale already */
855aeb93
JH
3491
3492 if (category != LC_MESSAGES) {
3493
837ce802 3494 save_messages_locale = do_setlocale_c(LC_MESSAGES, NULL);
855aeb93 3495 if (! save_messages_locale) {
5857e934
KW
3496 DEBUG_L(PerlIO_printf(Perl_debug_log,
3497 "Could not find current locale for LC_MESSAGES\n"));
855aeb93
JH
3498 goto cant_use_messages;
3499 }
5857e934 3500 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
3501
3502 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
3503 Safefree(save_messages_locale);
3504 save_messages_locale = NULL;
855aeb93 3505 }
837ce802 3506 else if (! do_setlocale_c(LC_MESSAGES, save_input_locale)) {
5857e934
KW
3507 DEBUG_L(PerlIO_printf(Perl_debug_log,
3508 "Could not change LC_MESSAGES locale to %s\n",
3509 save_input_locale));
855aeb93
JH
3510 Safefree(save_messages_locale);
3511 goto cant_use_messages;
3512 }
3513 }
3514
3515 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
3516 * whose information is desired. Look through all the messages. We
3517 * can't use Strerror() here because it may expand to code that
3518 * segfaults in miniperl */
855aeb93 3519
5857e934
KW
3520 for (e = 0; e <= sys_nerr; e++) {
3521 errno = 0;
3522 errmsg = sys_errlist[e];
3523 if (errno || !errmsg) {
855aeb93
JH
3524 break;
3525 }
5857e934 3526 errmsg = savepv(errmsg);
c5f058df 3527 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
3528 non_ascii = TRUE;
3529 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3530 break;
855aeb93
JH
3531 }
3532 }
5857e934 3533 Safefree(errmsg);
855aeb93
JH
3534
3535 /* And, if we changed it, restore LC_MESSAGES to its original locale */
3536 if (save_messages_locale) {
837ce802 3537 do_setlocale_c(LC_MESSAGES, save_messages_locale);
855aeb93
JH
3538 Safefree(save_messages_locale);
3539 }
3540
5857e934
KW
3541 if (non_ascii) {
3542
3543 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3544 * any non-ascii means it is one; otherwise we assume it isn't */
3545 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3546 save_input_locale,
3547 is_utf8));
47280b20 3548 goto finish_and_return;
5857e934 3549 }
855aeb93 3550
5857e934 3551 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
3552 }
3553 cant_use_messages:
3554
7d4bcc4a
KW
3555# endif
3556# endif /* the code that is compiled when no nl_langinfo */
0080c90a 3557
7d4bcc4a 3558# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3559 UTF-8 locale */
7d4bcc4a 3560
97f4de96
KW
3561 /* As a last resort, look at the locale name to see if it matches
3562 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3563 * return of setlocale(), is actually defined to be opaque, so we can't
3564 * really rely on the absence of various substrings in the name to indicate
3565 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3566 * be a UTF-8 locale. Similarly for the other common names */
3567
3568 final_pos = strlen(save_input_locale) - 1;
3569 if (final_pos >= 3) {
8de4332b 3570 const char *name = save_input_locale;
97f4de96
KW
3571
3572 /* Find next 'U' or 'u' and look from there */
3573 while ((name += strcspn(name, "Uu") + 1)
3574 <= save_input_locale + final_pos - 2)
3575 {
c72493e0 3576 if ( isALPHA_FOLD_NE(*name, 't')
305b8651 3577 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
3578 {
3579 continue;
3580 }
3581 name += 2;
3582 if (*(name) == '-') {
3583 if ((name > save_input_locale + final_pos - 1)) {
3584 break;
3585 }
3586 name++;
3587 }
3588 if (*(name) == '8') {
97f4de96
KW
3589 DEBUG_L(PerlIO_printf(Perl_debug_log,
3590 "Locale %s ends with UTF-8 in name\n",
3591 save_input_locale));
47280b20
KW
3592 is_utf8 = TRUE;
3593 goto finish_and_return;
97f4de96
KW
3594 }
3595 }
3596 DEBUG_L(PerlIO_printf(Perl_debug_log,
3597 "Locale %s doesn't end with UTF-8 in name\n",
3598 save_input_locale));
3599 }
3600
7d4bcc4a
KW
3601# endif
3602# ifdef WIN32
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
KW
3612
3613# endif
97f4de96
KW
3614
3615 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3616 * since we are about to return FALSE anyway, there is no point in doing
3617 * this extra work */
7d4bcc4a
KW
3618
3619# if 0
97f4de96
KW
3620 if (instr(save_input_locale, "8859")) {
3621 DEBUG_L(PerlIO_printf(Perl_debug_log,
3622 "Locale %s has 8859 in name, not UTF-8 locale\n",
3623 save_input_locale));
47280b20
KW
3624 is_utf8 = FALSE;
3625 goto finish_and_return;
97f4de96 3626 }
7d4bcc4a 3627# endif
97f4de96 3628
69014004
KW
3629 DEBUG_L(PerlIO_printf(Perl_debug_log,
3630 "Assuming locale %s is not a UTF-8 locale\n",
3631 save_input_locale));
47280b20
KW
3632 is_utf8 = FALSE;
3633
3634 finish_and_return:
3635
3636 /* Cache this result so we don't have to go through all this next time. */
3637 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
3638 - (utf8ness_cache - PL_locale_utf8ness);
3639
3640 /* But we can't save it if it is too large for the total space available */
3641 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
3642 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
3643
3644 /* Here it can fit, but we may need to clear out the oldest cached
3645 * result(s) to do so. Check */
3646 if (utf8ness_cache_len + input_name_len_with_overhead
3647 >= utf8ness_cache_size)
3648 {
3649 /* Here we have to clear something out to make room for this.
3650 * Start looking at the rightmost place where it could fit and find
3651 * the beginning of the entry that extends past that. */
3652 char * cutoff = (char *) my_memrchr(utf8ness_cache,
3653 UTF8NESS_SEP[0],
3654 utf8ness_cache_size
3655 - input_name_len_with_overhead);
3656
3657 assert(cutoff);
3658 assert(cutoff >= utf8ness_cache);
3659
3660 /* This and all subsequent entries must be removed */
3661 *cutoff = '\0';
3662 utf8ness_cache_len = strlen(utf8ness_cache);
3663 }
3664
3665 /* Make space for the new entry */
3666 Move(utf8ness_cache,
3667 utf8ness_cache + input_name_len_with_overhead,
3668 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
3669
3670 /* And insert it */
3671 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
3672 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3673
3674 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
3675 & (PERL_UINTMAX_T) ~1) != '0')
3676 {
3677 Perl_croak(aTHX_
3678 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%u,"
3679 " inserted_name=%s, its_len=%u\n",
3680 __FILE__, __LINE__,
3681 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
3682 delimited, input_name_len_with_overhead);
3683 }
3684 }
3685
3686# ifdef DEBUGGING
3687
3688 if (DEBUG_Lv_TEST || debug_initialization) {
3689 PerlIO_printf(Perl_debug_log,
3690 "PL_locale_utf8ness is now %s; returning %d\n",
3691 PL_locale_utf8ness, is_utf8);
3692 }
3693
3694# endif
3695
3696 Safefree(delimited);
fa9b773e 3697 Safefree(save_input_locale);
47280b20 3698 return is_utf8;
7d74bb61
KW
3699}
3700
8ef6e574 3701#endif
7d74bb61 3702
d6ded950
KW
3703
3704bool
3705Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3706{
1a4f13e1 3707 dVAR;
d6ded950
KW
3708 /* Internal function which returns if we are in the scope of a pragma that
3709 * enables the locale category 'category'. 'compiling' should indicate if
3710 * this is during the compilation phase (TRUE) or not (FALSE). */
3711
3712 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
3713
3714 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
3715 if (! categories || categories == &PL_sv_placeholder) {
3716 return FALSE;
3717 }
3718
3719 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
3720 * a valid unsigned */
3721 assert(category >= -1);
3722 return cBOOL(SvUV(categories) & (1U << (category + 1)));
3723}
3724
2c6ee1a7 3725char *
6ebbc862
KW
3726Perl_my_strerror(pTHX_ const int errnum)
3727{
3728 /* Returns a mortalized copy of the text of the error message associated
3729 * with 'errnum'. It uses the current locale's text unless the platform
3730 * doesn't have the LC_MESSAGES category or we are not being called from
3731 * within the scope of 'use locale'. In the former case, it uses whatever
3732 * strerror returns; in the latter case it uses the text from the C locale.
3733 *
3734 * The function just calls strerror(), but temporarily switches, if needed,
3735 * to the C locale */
3736
3737 char *errstr;
52770946 3738 dVAR;
6ebbc862 3739
52770946 3740#ifndef USE_LOCALE_MESSAGES
6ebbc862 3741
52770946
KW