This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Change some "shouldn't happen" failures into panics
[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
87f8e8e7
KW
391 const char * radix = (use_locale)
392 ? my_nl_langinfo(PERL_RADIXCHAR, FALSE)
393 /* FALSE => already in dest locale */
394 : ".";
2213a3be 395
87f8e8e7 396 sv_setpv(PL_numeric_radix_sv, radix);
86799d2d 397
87f8e8e7
KW
398 /* If this is valid UTF-8 that isn't totally ASCII, and we are in
399 * a UTF-8 locale, then mark the radix as being in UTF-8 */
400 if (is_utf8_non_invariant_string((U8 *) SvPVX(PL_numeric_radix_sv),
7a393424 401 SvCUR(PL_numeric_radix_sv))
87f8e8e7
KW
402 && _is_cur_LC_category_utf8(LC_NUMERIC))
403 {
404 SvUTF8_on(PL_numeric_radix_sv);
405 }
86799d2d
KW
406
407# ifdef DEBUGGING
7d4bcc4a 408
2fcc0ca9
KW
409 if (DEBUG_L_TEST || debug_initialization) {
410 PerlIO_printf(Perl_debug_log, "Locale radix is '%s', ?UTF-8=%d\n",
3ca88433
KW
411 SvPVX(PL_numeric_radix_sv),
412 cBOOL(SvUTF8(PL_numeric_radix_sv)));
2fcc0ca9 413 }
69014004 414
86799d2d
KW
415# endif
416#endif /* USE_LOCALE_NUMERIC and can find the radix char */
7d4bcc4a 417
98994639
HS
418}
419
a39edc4c 420
98994639 421void
8772537c 422Perl_new_numeric(pTHX_ const char *newnum)
98994639 423{
7d4bcc4a
KW
424
425#ifndef USE_LOCALE_NUMERIC
426
427 PERL_UNUSED_ARG(newnum);
428
429#else
0d071d52 430
291a84fb 431 /* Called after each libc setlocale() call affecting LC_NUMERIC, to tell
0d071d52
KW
432 * core Perl this and that 'newnum' is the name of the new locale.
433 * It installs this locale as the current underlying default.
434 *
435 * The default locale and the C locale can be toggled between by use of the
5792c642
KW
436 * set_numeric_underlying() and set_numeric_standard() functions, which
437 * should probably not be called directly, but only via macros like
0d071d52
KW
438 * SET_NUMERIC_STANDARD() in perl.h.
439 *
440 * The toggling is necessary mainly so that a non-dot radix decimal point
441 * character can be output, while allowing internal calculations to use a
442 * dot.
443 *
444 * This sets several interpreter-level variables:
bb304765 445 * PL_numeric_name The underlying locale's name: a copy of 'newnum'
892e6465 446 * PL_numeric_underlying A boolean indicating if the toggled state is such
7738054c
KW
447 * that the current locale is the program's underlying
448 * locale
449 * PL_numeric_standard An int indicating if the toggled state is such
4c68b815
KW
450 * that the current locale is the C locale or
451 * indistinguishable from the C locale. If non-zero, it
452 * is in C; if > 1, it means it may not be toggled away
7738054c 453 * from C.
4c68b815
KW
454 * PL_numeric_underlying_is_standard A bool kept by this function
455 * indicating that the underlying locale and the standard
456 * C locale are indistinguishable for the purposes of
457 * LC_NUMERIC. This happens when both of the above two
458 * variables are true at the same time. (Toggling is a
459 * no-op under these circumstances.) This variable is
460 * used to avoid having to recalculate.
0d071d52
KW
461 * Any code changing the locale (outside this file) should use
462 * POSIX::setlocale, which calls this function. Therefore this function
463 * should be called directly only from this file and from
464 * POSIX::setlocale() */
465
b03f34cf 466 char *save_newnum;
98994639
HS
467
468 if (! newnum) {
43c5f42d
NC
469 Safefree(PL_numeric_name);
470 PL_numeric_name = NULL;
98994639 471 PL_numeric_standard = TRUE;
892e6465 472 PL_numeric_underlying = TRUE;
4c68b815 473 PL_numeric_underlying_is_standard = TRUE;
98994639
HS
474 return;
475 }
476
b03f34cf 477 save_newnum = stdize_locale(savepv(newnum));
892e6465 478 PL_numeric_underlying = TRUE;
4c68b815
KW
479 PL_numeric_standard = isNAME_C_OR_POSIX(save_newnum);
480
481 /* If its name isn't C nor POSIX, it could still be indistinguishable from
482 * them */
483 if (! PL_numeric_standard) {
484 PL_numeric_standard = cBOOL(strEQ(".", my_nl_langinfo(PERL_RADIXCHAR,
485 FALSE /* Don't toggle locale */ ))
486 && strEQ("", my_nl_langinfo(PERL_THOUSEP,
487 FALSE)));
488 }
abe1abcf 489
4c68b815 490 /* Save the new name if it isn't the same as the previous one, if any */
b03f34cf 491 if (! PL_numeric_name || strNE(PL_numeric_name, save_newnum)) {
98994639 492 Safefree(PL_numeric_name);
b03f34cf 493 PL_numeric_name = save_newnum;
b03f34cf 494 }
abe1abcf
KW
495 else {
496 Safefree(save_newnum);
497 }
4c28b29c 498
4c68b815
KW
499 PL_numeric_underlying_is_standard = PL_numeric_standard;
500
e1aa2579
KW
501# if defined(HAS_NEWLOCALE) && ! defined(NO_POSIX_2008_LOCALE)
502
503 PL_underlying_numeric_obj = newlocale(LC_NUMERIC_MASK,
504 PL_numeric_name,
505 PL_underlying_numeric_obj);
506
507#endif
508
4c68b815
KW
509 if (DEBUG_L_TEST || debug_initialization) {
510 PerlIO_printf(Perl_debug_log, "Called new_numeric with %s, PL_numeric_name=%s\n", newnum, PL_numeric_name);
511 }
512
4c28b29c
KW
513 /* Keep LC_NUMERIC in the C locale. This is for XS modules, so they don't
514 * have to worry about the radix being a non-dot. (Core operations that
515 * need the underlying locale change to it temporarily). */
516 set_numeric_standard();
517
98994639 518#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 519
98994639
HS
520}
521
522void
523Perl_set_numeric_standard(pTHX)
524{
7d4bcc4a 525
98994639 526#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 527
28c1bf33
KW
528 /* Toggle the LC_NUMERIC locale to C. Most code should use the macros like
529 * SET_NUMERIC_STANDARD() in perl.h instead of calling this directly. The
530 * macro avoids calling this routine if toggling isn't necessary according
531 * to our records (which could be wrong if some XS code has changed the
532 * locale behind our back) */
0d071d52 533
837ce802 534 do_setlocale_c(LC_NUMERIC, "C");
a9b8c0d8 535 PL_numeric_standard = TRUE;
4c68b815 536 PL_numeric_underlying = PL_numeric_underlying_is_standard;
86799d2d 537 set_numeric_radix(0);
7d4bcc4a
KW
538
539# ifdef DEBUGGING
540
2fcc0ca9
KW
541 if (DEBUG_L_TEST || debug_initialization) {
542 PerlIO_printf(Perl_debug_log,
58e4a467 543 "LC_NUMERIC locale now is standard C\n");
2fcc0ca9 544 }
98994639 545
7d4bcc4a 546# endif
98994639 547#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 548
98994639
HS
549}
550
551void
5792c642 552Perl_set_numeric_underlying(pTHX)
98994639 553{
7d4bcc4a 554
98994639 555#ifdef USE_LOCALE_NUMERIC
7d4bcc4a 556
28c1bf33 557 /* Toggle the LC_NUMERIC locale to the current underlying default. Most
7d4bcc4a
KW
558 * code should use the macros like SET_NUMERIC_UNDERLYING() in perl.h
559 * instead of calling this directly. The macro avoids calling this routine
560 * if toggling isn't necessary according to our records (which could be
561 * wrong if some XS code has changed the locale behind our back) */
a9b8c0d8 562
837ce802 563 do_setlocale_c(LC_NUMERIC, PL_numeric_name);
4c68b815 564 PL_numeric_standard = PL_numeric_underlying_is_standard;
892e6465 565 PL_numeric_underlying = TRUE;
3c62fd46 566 set_numeric_radix(! PL_numeric_standard);
7d4bcc4a
KW
567
568# ifdef DEBUGGING
569
2fcc0ca9
KW
570 if (DEBUG_L_TEST || debug_initialization) {
571 PerlIO_printf(Perl_debug_log,
58e4a467 572 "LC_NUMERIC locale now is %s\n",
2fcc0ca9
KW
573 PL_numeric_name);
574 }
98994639 575
7d4bcc4a 576# endif
98994639 577#endif /* USE_LOCALE_NUMERIC */
7d4bcc4a 578
98994639
HS
579}
580
581/*
582 * Set up for a new ctype locale.
583 */
a4f00dcc
KW
584STATIC void
585S_new_ctype(pTHX_ const char *newctype)
98994639 586{
7d4bcc4a
KW
587
588#ifndef USE_LOCALE_CTYPE
589
590 PERL_ARGS_ASSERT_NEW_CTYPE;
591 PERL_UNUSED_ARG(newctype);
592 PERL_UNUSED_CONTEXT;
593
594#else
0d071d52 595
291a84fb 596 /* Called after each libc setlocale() call affecting LC_CTYPE, to tell
0d071d52
KW
597 * core Perl this and that 'newctype' is the name of the new locale.
598 *
599 * This function sets up the folding arrays for all 256 bytes, assuming
600 * that tofold() is tolc() since fold case is not a concept in POSIX,
601 *
602 * Any code changing the locale (outside this file) should use
603 * POSIX::setlocale, which calls this function. Therefore this function
604 * should be called directly only from this file and from
605 * POSIX::setlocale() */
606
27da23d5 607 dVAR;
68067e4e 608 UV i;
98994639 609
7918f24d
NC
610 PERL_ARGS_ASSERT_NEW_CTYPE;
611
215c5139
KW
612 /* We will replace any bad locale warning with 1) nothing if the new one is
613 * ok; or 2) a new warning for the bad new locale */
614 if (PL_warn_locale) {
615 SvREFCNT_dec_NN(PL_warn_locale);
616 PL_warn_locale = NULL;
617 }
618
c1284011 619 PL_in_utf8_CTYPE_locale = _is_cur_LC_category_utf8(LC_CTYPE);
31f05a37
KW
620
621 /* A UTF-8 locale gets standard rules. But note that code still has to
622 * handle this specially because of the three problematic code points */
623 if (PL_in_utf8_CTYPE_locale) {
624 Copy(PL_fold_latin1, PL_fold_locale, 256, U8);
625 }
626 else {
8c6180a9
KW
627 /* Assume enough space for every character being bad. 4 spaces each
628 * for the 94 printable characters that are output like "'x' "; and 5
629 * spaces each for "'\\' ", "'\t' ", and "'\n' "; plus a terminating
630 * NUL */
631 char bad_chars_list[ (94 * 4) + (3 * 5) + 1 ];
632
cc9eaeb0
KW
633 /* Don't check for problems if we are suppressing the warnings */
634 bool check_for_problems = ckWARN_d(WARN_LOCALE)
635 || UNLIKELY(DEBUG_L_TEST);
8c6180a9
KW
636 bool multi_byte_locale = FALSE; /* Assume is a single-byte locale
637 to start */
638 unsigned int bad_count = 0; /* Count of bad characters */
639
baa60164 640 for (i = 0; i < 256; i++) {
5a8a8f21
KW
641 if (isupper(i))
642 PL_fold_locale[i] = (U8) tolower(i);
643 else if (islower(i))
644 PL_fold_locale[i] = (U8) toupper(i);
baa60164
KW
645 else
646 PL_fold_locale[i] = (U8) i;
8c6180a9
KW
647
648 /* If checking for locale problems, see if the native ASCII-range
649 * printables plus \n and \t are in their expected categories in
650 * the new locale. If not, this could mean big trouble, upending
651 * Perl's and most programs' assumptions, like having a
652 * metacharacter with special meaning become a \w. Fortunately,
653 * it's very rare to find locales that aren't supersets of ASCII
654 * nowadays. It isn't a problem for most controls to be changed
655 * into something else; we check only \n and \t, though perhaps \r
656 * could be an issue as well. */
7d4bcc4a 657 if ( check_for_problems
8c6180a9
KW
658 && (isGRAPH_A(i) || isBLANK_A(i) || i == '\n'))
659 {
adf947ba
KW
660 if ( cBOOL(isalnum(i)) != cBOOL(isALPHANUMERIC(i))
661 || cBOOL(isalpha(i)) != cBOOL(isALPHA_A(i))
662 || cBOOL(isdigit(i)) != cBOOL(isDIGIT_A(i))
663 || cBOOL(isgraph(i)) != cBOOL(isGRAPH_A(i))
664 || cBOOL(islower(i)) != cBOOL(isLOWER_A(i))
665 || cBOOL(isprint(i)) != cBOOL(isPRINT_A(i))
666 || cBOOL(ispunct(i)) != cBOOL(isPUNCT_A(i))
667 || cBOOL(isspace(i)) != cBOOL(isSPACE_A(i))
668 || cBOOL(isupper(i)) != cBOOL(isUPPER_A(i))
669 || cBOOL(isxdigit(i))!= cBOOL(isXDIGIT_A(i))
670 || tolower(i) != (int) toLOWER_A(i)
671 || toupper(i) != (int) toUPPER_A(i)
672 || (i == '\n' && ! isCNTRL_LC(i)))
8c6180a9
KW
673 {
674 if (bad_count) { /* Separate multiple entries with a
675 blank */
676 bad_chars_list[bad_count++] = ' ';
677 }
678 bad_chars_list[bad_count++] = '\'';
679 if (isPRINT_A(i)) {
680 bad_chars_list[bad_count++] = (char) i;
681 }
682 else {
683 bad_chars_list[bad_count++] = '\\';
684 if (i == '\n') {
685 bad_chars_list[bad_count++] = 'n';
686 }
687 else {
688 assert(i == '\t');
689 bad_chars_list[bad_count++] = 't';
690 }
691 }
692 bad_chars_list[bad_count++] = '\'';
693 bad_chars_list[bad_count] = '\0';
694 }
695 }
696 }
697
7d4bcc4a
KW
698# ifdef MB_CUR_MAX
699
8c6180a9 700 /* We only handle single-byte locales (outside of UTF-8 ones; so if
d35fca5f 701 * this locale requires more than one byte, there are going to be
8c6180a9 702 * problems. */
9c8a6dc2
KW
703 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
704 "%s:%d: check_for_problems=%d, MB_CUR_MAX=%d\n",
705 __FILE__, __LINE__, check_for_problems, (int) MB_CUR_MAX));
706
ba1a4362
KW
707 if (check_for_problems && MB_CUR_MAX > 1
708
709 /* Some platforms return MB_CUR_MAX > 1 for even the "C"
710 * locale. Just assume that the implementation for them (plus
711 * for POSIX) is correct and the > 1 value is spurious. (Since
712 * these are specially handled to never be considered UTF-8
713 * locales, as long as this is the only problem, everything
714 * should work fine */
715 && strNE(newctype, "C") && strNE(newctype, "POSIX"))
716 {
8c6180a9
KW
717 multi_byte_locale = TRUE;
718 }
7d4bcc4a
KW
719
720# endif
8c6180a9
KW
721
722 if (bad_count || multi_byte_locale) {
780fcc9f 723 PL_warn_locale = Perl_newSVpvf(aTHX_
8c6180a9 724 "Locale '%s' may not work well.%s%s%s\n",
780fcc9f 725 newctype,
8c6180a9
KW
726 (multi_byte_locale)
727 ? " Some characters in it are not recognized by"
728 " Perl."
729 : "",
730 (bad_count)
731 ? "\nThe following characters (and maybe others)"
732 " may not have the same meaning as the Perl"
733 " program expects:\n"
734 : "",
735 (bad_count)
736 ? bad_chars_list
737 : ""
738 );
cc9eaeb0 739 /* If we are actually in the scope of the locale or are debugging,
bddebb56
KW
740 * output the message now. If not in that scope, we save the
741 * message to be output at the first operation using this locale,
742 * if that actually happens. Most programs don't use locales, so
743 * they are immune to bad ones. */
cc9eaeb0 744 if (IN_LC(LC_CTYPE) || UNLIKELY(DEBUG_L_TEST)) {
780fcc9f
KW
745
746 /* We have to save 'newctype' because the setlocale() just
747 * below may destroy it. The next setlocale() further down
748 * should restore it properly so that the intermediate change
749 * here is transparent to this function's caller */
750 const char * const badlocale = savepv(newctype);
751
837ce802 752 do_setlocale_c(LC_CTYPE, "C");
780fcc9f
KW
753
754 /* The '0' below suppresses a bogus gcc compiler warning */
755 Perl_warner(aTHX_ packWARN(WARN_LOCALE), SvPVX(PL_warn_locale), 0);
bddebb56 756
837ce802 757 do_setlocale_c(LC_CTYPE, badlocale);
c0f3a893 758 Safefree(badlocale);
bddebb56
KW
759
760 if (IN_LC(LC_CTYPE)) {
761 SvREFCNT_dec_NN(PL_warn_locale);
762 PL_warn_locale = NULL;
763 }
780fcc9f 764 }
baa60164 765 }
31f05a37 766 }
98994639
HS
767
768#endif /* USE_LOCALE_CTYPE */
7d4bcc4a 769
98994639
HS
770}
771
98994639 772void
2726666d
KW
773Perl__warn_problematic_locale()
774{
2726666d
KW
775
776#ifdef USE_LOCALE_CTYPE
777
5f04a188
KW
778 dTHX;
779
780 /* Internal-to-core function that outputs the message in PL_warn_locale,
781 * and then NULLS it. Should be called only through the macro
782 * _CHECK_AND_WARN_PROBLEMATIC_LOCALE */
783
2726666d 784 if (PL_warn_locale) {
2726666d
KW
785 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
786 SvPVX(PL_warn_locale),
787 0 /* dummy to avoid compiler warning */ );
2726666d
KW
788 SvREFCNT_dec_NN(PL_warn_locale);
789 PL_warn_locale = NULL;
790 }
791
792#endif
793
794}
795
a4f00dcc
KW
796STATIC void
797S_new_collate(pTHX_ const char *newcoll)
98994639 798{
7d4bcc4a
KW
799
800#ifndef USE_LOCALE_COLLATE
801
802 PERL_UNUSED_ARG(newcoll);
803 PERL_UNUSED_CONTEXT;
804
805#else
0d071d52 806
291a84fb 807 /* Called after each libc setlocale() call affecting LC_COLLATE, to tell
0d071d52
KW
808 * core Perl this and that 'newcoll' is the name of the new locale.
809 *
d35fca5f
KW
810 * The design of locale collation is that every locale change is given an
811 * index 'PL_collation_ix'. The first time a string particpates in an
812 * operation that requires collation while locale collation is active, it
813 * is given PERL_MAGIC_collxfrm magic (via sv_collxfrm_flags()). That
814 * magic includes the collation index, and the transformation of the string
815 * by strxfrm(), q.v. That transformation is used when doing comparisons,
816 * instead of the string itself. If a string changes, the magic is
817 * cleared. The next time the locale changes, the index is incremented,
818 * and so we know during a comparison that the transformation is not
819 * necessarily still valid, and so is recomputed. Note that if the locale
820 * changes enough times, the index could wrap (a U32), and it is possible
821 * that a transformation would improperly be considered valid, leading to
822 * an unlikely bug */
0d071d52 823
98994639
HS
824 if (! newcoll) {
825 if (PL_collation_name) {
826 ++PL_collation_ix;
827 Safefree(PL_collation_name);
828 PL_collation_name = NULL;
829 }
830 PL_collation_standard = TRUE;
00bf60ca 831 is_standard_collation:
98994639
HS
832 PL_collxfrm_base = 0;
833 PL_collxfrm_mult = 2;
165a1c52 834 PL_in_utf8_COLLATE_locale = FALSE;
f28f4d2a 835 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 836 PL_strxfrm_max_cp = 0;
98994639
HS
837 return;
838 }
839
d35fca5f 840 /* If this is not the same locale as currently, set the new one up */
98994639
HS
841 if (! PL_collation_name || strNE(PL_collation_name, newcoll)) {
842 ++PL_collation_ix;
843 Safefree(PL_collation_name);
844 PL_collation_name = stdize_locale(savepv(newcoll));
a39edc4c 845 PL_collation_standard = isNAME_C_OR_POSIX(newcoll);
00bf60ca
KW
846 if (PL_collation_standard) {
847 goto is_standard_collation;
848 }
98994639 849
165a1c52 850 PL_in_utf8_COLLATE_locale = _is_cur_LC_category_utf8(LC_COLLATE);
f28f4d2a 851 PL_strxfrm_NUL_replacement = '\0';
a4a439fb 852 PL_strxfrm_max_cp = 0;
165a1c52 853
59c018b9
KW
854 /* A locale collation definition includes primary, secondary, tertiary,
855 * etc. weights for each character. To sort, the primary weights are
856 * used, and only if they compare equal, then the secondary weights are
857 * used, and only if they compare equal, then the tertiary, etc.
858 *
859 * strxfrm() works by taking the input string, say ABC, and creating an
860 * output transformed string consisting of first the primary weights,
861 * A¹B¹C¹ followed by the secondary ones, A²B²C²; and then the
862 * tertiary, etc, yielding A¹B¹C¹ A²B²C² A³B³C³ .... Some characters
863 * may not have weights at every level. In our example, let's say B
864 * doesn't have a tertiary weight, and A doesn't have a secondary
865 * weight. The constructed string is then going to be
866 * A¹B¹C¹ B²C² A³C³ ....
867 * This has the desired effect that strcmp() will look at the secondary
868 * or tertiary weights only if the strings compare equal at all higher
869 * priority weights. The spaces shown here, like in
c342d20e 870 * "A¹B¹C¹ A²B²C² "
59c018b9
KW
871 * are not just for readability. In the general case, these must
872 * actually be bytes, which we will call here 'separator weights'; and
873 * they must be smaller than any other weight value, but since these
874 * are C strings, only the terminating one can be a NUL (some
875 * implementations may include a non-NUL separator weight just before
876 * the NUL). Implementations tend to reserve 01 for the separator
877 * weights. They are needed so that a shorter string's secondary
878 * weights won't be misconstrued as primary weights of a longer string,
879 * etc. By making them smaller than any other weight, the shorter
880 * string will sort first. (Actually, if all secondary weights are
881 * smaller than all primary ones, there is no need for a separator
882 * weight between those two levels, etc.)
883 *
884 * The length of the transformed string is roughly a linear function of
885 * the input string. It's not exactly linear because some characters
886 * don't have weights at all levels. When we call strxfrm() we have to
887 * allocate some memory to hold the transformed string. The
888 * calculations below try to find coefficients 'm' and 'b' for this
889 * locale so that m*x + b equals how much space we need, given the size
890 * of the input string in 'x'. If we calculate too small, we increase
891 * the size as needed, and call strxfrm() again, but it is better to
892 * get it right the first time to avoid wasted expensive string
893 * transformations. */
894
98994639 895 {
79f120c8
KW
896 /* We use the string below to find how long the tranformation of it
897 * is. Almost all locales are supersets of ASCII, or at least the
898 * ASCII letters. We use all of them, half upper half lower,
899 * because if we used fewer, we might hit just the ones that are
900 * outliers in a particular locale. Most of the strings being
901 * collated will contain a preponderance of letters, and even if
902 * they are above-ASCII, they are likely to have the same number of
903 * weight levels as the ASCII ones. It turns out that digits tend
904 * to have fewer levels, and some punctuation has more, but those
905 * are relatively sparse in text, and khw believes this gives a
906 * reasonable result, but it could be changed if experience so
907 * dictates. */
908 const char longer[] = "ABCDEFGHIJKLMnopqrstuvwxyz";
909 char * x_longer; /* Transformed 'longer' */
910 Size_t x_len_longer; /* Length of 'x_longer' */
911
912 char * x_shorter; /* We also transform a substring of 'longer' */
913 Size_t x_len_shorter;
914
a4a439fb 915 /* _mem_collxfrm() is used get the transformation (though here we
79f120c8
KW
916 * are interested only in its length). It is used because it has
917 * the intelligence to handle all cases, but to work, it needs some
918 * values of 'm' and 'b' to get it started. For the purposes of
919 * this calculation we use a very conservative estimate of 'm' and
920 * 'b'. This assumes a weight can be multiple bytes, enough to
921 * hold any UV on the platform, and there are 5 levels, 4 weight
922 * bytes, and a trailing NUL. */
923 PL_collxfrm_base = 5;
924 PL_collxfrm_mult = 5 * sizeof(UV);
925
926 /* Find out how long the transformation really is */
a4a439fb
KW
927 x_longer = _mem_collxfrm(longer,
928 sizeof(longer) - 1,
929 &x_len_longer,
930
931 /* We avoid converting to UTF-8 in the
932 * called function by telling it the
933 * string is in UTF-8 if the locale is a
934 * UTF-8 one. Since the string passed
935 * here is invariant under UTF-8, we can
936 * claim it's UTF-8 even though it isn't.
937 * */
938 PL_in_utf8_COLLATE_locale);
79f120c8
KW
939 Safefree(x_longer);
940
941 /* Find out how long the transformation of a substring of 'longer'
942 * is. Together the lengths of these transformations are
943 * sufficient to calculate 'm' and 'b'. The substring is all of
944 * 'longer' except the first character. This minimizes the chances
945 * of being swayed by outliers */
a4a439fb 946 x_shorter = _mem_collxfrm(longer + 1,
79f120c8 947 sizeof(longer) - 2,
a4a439fb
KW
948 &x_len_shorter,
949 PL_in_utf8_COLLATE_locale);
79f120c8
KW
950 Safefree(x_shorter);
951
952 /* If the results are nonsensical for this simple test, the whole
953 * locale definition is suspect. Mark it so that locale collation
954 * is not active at all for it. XXX Should we warn? */
955 if ( x_len_shorter == 0
956 || x_len_longer == 0
957 || x_len_shorter >= x_len_longer)
958 {
959 PL_collxfrm_mult = 0;
960 PL_collxfrm_base = 0;
961 }
962 else {
963 SSize_t base; /* Temporary */
964
965 /* We have both: m * strlen(longer) + b = x_len_longer
966 * m * strlen(shorter) + b = x_len_shorter;
967 * subtracting yields:
968 * m * (strlen(longer) - strlen(shorter))
969 * = x_len_longer - x_len_shorter
970 * But we have set things up so that 'shorter' is 1 byte smaller
971 * than 'longer'. Hence:
972 * m = x_len_longer - x_len_shorter
973 *
974 * But if something went wrong, make sure the multiplier is at
975 * least 1.
976 */
977 if (x_len_longer > x_len_shorter) {
978 PL_collxfrm_mult = (STRLEN) x_len_longer - x_len_shorter;
979 }
980 else {
981 PL_collxfrm_mult = 1;
982 }
983
984 /* mx + b = len
985 * so: b = len - mx
986 * but in case something has gone wrong, make sure it is
987 * non-negative */
988 base = x_len_longer - PL_collxfrm_mult * (sizeof(longer) - 1);
989 if (base < 0) {
990 base = 0;
991 }
992
993 /* Add 1 for the trailing NUL */
994 PL_collxfrm_base = base + 1;
995 }
58eebef2 996
7d4bcc4a
KW
997# ifdef DEBUGGING
998
58eebef2
KW
999 if (DEBUG_L_TEST || debug_initialization) {
1000 PerlIO_printf(Perl_debug_log,
b07929e4
KW
1001 "%s:%d: ?UTF-8 locale=%d; x_len_shorter=%zu, "
1002 "x_len_longer=%zu,"
1003 " collate multipler=%zu, collate base=%zu\n",
58eebef2
KW
1004 __FILE__, __LINE__,
1005 PL_in_utf8_COLLATE_locale,
1006 x_len_shorter, x_len_longer,
1007 PL_collxfrm_mult, PL_collxfrm_base);
1008 }
7d4bcc4a
KW
1009# endif
1010
98994639
HS
1011 }
1012 }
1013
1014#endif /* USE_LOCALE_COLLATE */
7d4bcc4a 1015
98994639
HS
1016}
1017
d2b24094 1018#ifdef WIN32
b8cc575c 1019
a4f00dcc 1020STATIC char *
b8cc575c 1021S_win32_setlocale(pTHX_ int category, const char* locale)
b385bb4d
KW
1022{
1023 /* This, for Windows, emulates POSIX setlocale() behavior. There is no
7d4bcc4a
KW
1024 * difference between the two unless the input locale is "", which normally
1025 * means on Windows to get the machine default, which is set via the
1026 * computer's "Regional and Language Options" (or its current equivalent).
1027 * In POSIX, it instead means to find the locale from the user's
1028 * environment. This routine changes the Windows behavior to first look in
1029 * the environment, and, if anything is found, use that instead of going to
1030 * the machine default. If there is no environment override, the machine
1031 * default is used, by calling the real setlocale() with "".
1032 *
1033 * The POSIX behavior is to use the LC_ALL variable if set; otherwise to
1034 * use the particular category's variable if set; otherwise to use the LANG
1035 * variable. */
b385bb4d 1036
175c4cf9 1037 bool override_LC_ALL = FALSE;
89f7b9aa 1038 char * result;
e5f10d49 1039 unsigned int i;
89f7b9aa 1040
b385bb4d 1041 if (locale && strEQ(locale, "")) {
7d4bcc4a
KW
1042
1043# ifdef LC_ALL
1044
b385bb4d
KW
1045 locale = PerlEnv_getenv("LC_ALL");
1046 if (! locale) {
e5f10d49
KW
1047 if (category == LC_ALL) {
1048 override_LC_ALL = TRUE;
1049 }
1050 else {
7d4bcc4a
KW
1051
1052# endif
7d4bcc4a 1053
e5f10d49
KW
1054 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
1055 if (category == categories[i]) {
1056 locale = PerlEnv_getenv(category_names[i]);
1057 goto found_locale;
1058 }
1059 }
7d4bcc4a 1060
b385bb4d 1061 locale = PerlEnv_getenv("LANG");
481465ea 1062 if (! locale) {
b385bb4d
KW
1063 locale = "";
1064 }
e5f10d49
KW
1065
1066 found_locale: ;
7d4bcc4a
KW
1067
1068# ifdef LC_ALL
1069
e5f10d49 1070 }
b385bb4d 1071 }
7d4bcc4a
KW
1072
1073# endif
1074
b385bb4d
KW
1075 }
1076
89f7b9aa 1077 result = setlocale(category, locale);
bbc98134 1078 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n", __FILE__, __LINE__,
a4f00dcc 1079 setlocale_debug_string(category, locale, result)));
89f7b9aa 1080
481465ea 1081 if (! override_LC_ALL) {
89f7b9aa
KW
1082 return result;
1083 }
1084
dfd77d7a 1085 /* Here the input category was LC_ALL, and we have set it to what is in the
481465ea
KW
1086 * LANG variable or the system default if there is no LANG. But these have
1087 * lower priority than the other LC_foo variables, so override it for each
1088 * one that is set. (If they are set to "", it means to use the same thing
1089 * we just set LC_ALL to, so can skip) */
7d4bcc4a 1090
948523db 1091 for (i = 0; i < LC_ALL_INDEX; i++) {
e5f10d49
KW
1092 result = PerlEnv_getenv(category_names[i]);
1093 if (result && strNE(result, "")) {
1094 setlocale(categories[i], result);
1095 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1096 __FILE__, __LINE__,
1097 setlocale_debug_string(categories[i], result, "not captured")));
1098 }
89f7b9aa 1099 }
7d4bcc4a 1100
bbc98134
KW
1101 result = setlocale(LC_ALL, NULL);
1102 DEBUG_L(PerlIO_printf(Perl_debug_log, "%s:%d: %s\n",
1103 __FILE__, __LINE__,
a4f00dcc 1104 setlocale_debug_string(LC_ALL, NULL, result)));
89f7b9aa 1105
bbc98134 1106 return result;
b385bb4d
KW
1107}
1108
1109#endif
1110
a4f00dcc
KW
1111char *
1112Perl_setlocale(int category, const char * locale)
1113{
1114 /* This wraps POSIX::setlocale() */
1115
1116 char * retval;
1159483a 1117 char * newlocale;
a4f00dcc
KW
1118 dTHX;
1119
a4f00dcc
KW
1120#ifdef USE_LOCALE_NUMERIC
1121
291a84fb
KW
1122 /* A NULL locale means only query what the current one is. We have the
1123 * LC_NUMERIC name saved, because we are normally switched into the C
1124 * locale for it. For an LC_ALL query, switch back to get the correct
1125 * results. All other categories don't require special handling */
a4f00dcc
KW
1126 if (locale == NULL) {
1127 if (category == LC_NUMERIC) {
1128 return savepv(PL_numeric_name);
1129 }
1130
7d4bcc4a 1131# ifdef LC_ALL
a4f00dcc 1132
c0d737a8
KW
1133 else if (category == LC_ALL && ! PL_numeric_underlying) {
1134
a4f00dcc
KW
1135 SET_NUMERIC_UNDERLYING();
1136 }
1137
7d4bcc4a 1138# endif
a4f00dcc
KW
1139
1140 }
1141
1142#endif
1143
e5f10d49
KW
1144 /* Save retval since subsequent setlocale() calls may overwrite it. */
1145 retval = savepv(do_setlocale_r(category, locale));
a4f00dcc
KW
1146
1147 DEBUG_L(PerlIO_printf(Perl_debug_log,
1148 "%s:%d: %s\n", __FILE__, __LINE__,
1149 setlocale_debug_string(category, locale, retval)));
1150 if (! retval) {
1151 /* Should never happen that a query would return an error, but be
1152 * sure and reset to C locale */
1153 if (locale == 0) {
1154 SET_NUMERIC_STANDARD();
1155 }
7d4bcc4a 1156
a4f00dcc
KW
1157 return NULL;
1158 }
1159
a4f00dcc
KW
1160 /* If locale == NULL, we are just querying the state, but may have switched
1161 * to NUMERIC_UNDERLYING. Switch back before returning. */
1162 if (locale == NULL) {
1163 SET_NUMERIC_STANDARD();
1164 return retval;
1165 }
a4f00dcc 1166
1159483a
KW
1167 /* Now that have switched locales, we have to update our records to
1168 * correspond. */
a4f00dcc 1169
1159483a 1170 switch (category) {
a4f00dcc 1171
1159483a 1172#ifdef USE_LOCALE_CTYPE
a4f00dcc 1173
1159483a
KW
1174 case LC_CTYPE:
1175 new_ctype(retval);
1176 break;
a4f00dcc 1177
1159483a 1178#endif
a4f00dcc
KW
1179#ifdef USE_LOCALE_COLLATE
1180
1159483a
KW
1181 case LC_COLLATE:
1182 new_collate(retval);
1183 break;
a4f00dcc 1184
1159483a 1185#endif
a4f00dcc
KW
1186#ifdef USE_LOCALE_NUMERIC
1187
1159483a
KW
1188 case LC_NUMERIC:
1189 new_numeric(retval);
1190 break;
a4f00dcc 1191
1159483a
KW
1192#endif
1193#ifdef LC_ALL
a4f00dcc 1194
1159483a 1195 case LC_ALL:
a4f00dcc 1196
1159483a
KW
1197 /* LC_ALL updates all the things we care about. The values may not
1198 * be the same as 'retval', as the locale "" may have set things
1199 * individually */
a4f00dcc 1200
1159483a 1201# ifdef USE_LOCALE_CTYPE
a4f00dcc 1202
1159483a
KW
1203 newlocale = do_setlocale_c(LC_CTYPE, NULL);
1204 new_ctype(newlocale);
a4f00dcc 1205
1159483a
KW
1206# endif /* USE_LOCALE_CTYPE */
1207# ifdef USE_LOCALE_COLLATE
1208
1209 newlocale = do_setlocale_c(LC_COLLATE, NULL);
1210 new_collate(newlocale);
a4f00dcc 1211
7d4bcc4a 1212# endif
1159483a 1213# ifdef USE_LOCALE_NUMERIC
a4f00dcc 1214
1159483a
KW
1215 newlocale = do_setlocale_c(LC_NUMERIC, NULL);
1216 new_numeric(newlocale);
a4f00dcc 1217
1159483a
KW
1218# endif /* USE_LOCALE_NUMERIC */
1219#endif /* LC_ALL */
a4f00dcc 1220
1159483a
KW
1221 default:
1222 break;
a4f00dcc
KW
1223 }
1224
1225 return retval;
1226
f7416781
KW
1227
1228}
1229
1230PERL_STATIC_INLINE const char *
1231S_save_to_buffer(const char * string, char **buf, Size_t *buf_size, const Size_t offset)
1232{
1233 /* Copy the NUL-terminated 'string' to 'buf' + 'offset'. 'buf' has size 'buf_size',
1234 * growing it if necessary */
1235
1236 const Size_t string_size = strlen(string) + offset + 1;
1237
1238 PERL_ARGS_ASSERT_SAVE_TO_BUFFER;
1239
1240 if (*buf_size == 0) {
1241 Newx(*buf, string_size, char);
1242 *buf_size = string_size;
1243 }
1244 else if (string_size > *buf_size) {
1245 Renew(*buf, string_size, char);
1246 *buf_size = string_size;
1247 }
1248
1249 Copy(string, *buf + offset, string_size - offset, char);
1250 return *buf;
1251}
1252
1253/*
1254
1255=head1 Locale-related functions and macros
1256
1257=for apidoc Perl_langinfo
1258
7d4bcc4a 1259This is an (almost ª) drop-in replacement for the system C<L<nl_langinfo(3)>>,
f7416781
KW
1260taking the same C<item> parameter values, and returning the same information.
1261But it is more thread-safe than regular C<nl_langinfo()>, and hides the quirks
1262of Perl's locale handling from your code, and can be used on systems that lack
1263a native C<nl_langinfo>.
1264
1265Expanding on these:
1266
1267=over
1268
1269=item *
1270
1271It delivers the correct results for the C<RADIXCHAR> and C<THOUSESEP> items,
1272without you having to write extra code. The reason for the extra code would be
1273because these are from the C<LC_NUMERIC> locale category, which is normally
1274kept set to the C locale by Perl, no matter what the underlying locale is
1275supposed to be, and so to get the expected results, you have to temporarily
1276toggle into the underlying locale, and later toggle back. (You could use
1277plain C<nl_langinfo> and C<L</STORE_LC_NUMERIC_FORCE_TO_UNDERLYING>> for this
1278but then you wouldn't get the other advantages of C<Perl_langinfo()>; not
1279keeping C<LC_NUMERIC> in the C locale would break a lot of CPAN, which is
1280expecting the radix (decimal point) character to be a dot.)
1281
1282=item *
1283
1284Depending on C<item>, it works on systems that don't have C<nl_langinfo>, hence
1285makes your code more portable. Of the fifty-some possible items specified by
1286the POSIX 2008 standard,
1287L<http://pubs.opengroup.org/onlinepubs/9699919799/basedefs/langinfo.h.html>,
1288only two are completely unimplemented. It uses various techniques to recover
1289the other items, including calling C<L<localeconv(3)>>, and C<L<strftime(3)>>,
1290both of which are specified in C89, so should be always be available. Later
1291C<strftime()> versions have additional capabilities; C<""> is returned for
1292those not available on your system.
1293
1294The details for those items which may differ from what this emulation returns
1295and what a native C<nl_langinfo()> would return are:
1296
1297=over
1298
1299=item C<CODESET>
1300
1301=item C<ERA>
1302
1303Unimplemented, so returns C<"">.
1304
1305=item C<YESEXPR>
1306
c1566110
KW
1307=item C<YESSTR>
1308
f7416781
KW
1309=item C<NOEXPR>
1310
c1566110
KW
1311=item C<NOSTR>
1312
1313Only the values for English are returned. C<YESSTR> and C<NOSTR> have been
1314removed from POSIX 2008, and are retained for backwards compatibility. Your
1315platform's C<nl_langinfo> may not support them.
f7416781
KW
1316
1317=item C<D_FMT>
1318
1319Always evaluates to C<%x>, the locale's appropriate date representation.
1320
1321=item C<T_FMT>
1322
1323Always evaluates to C<%X>, the locale's appropriate time representation.
1324
1325=item C<D_T_FMT>
1326
1327Always evaluates to C<%c>, the locale's appropriate date and time
1328representation.
1329
1330=item C<CRNCYSTR>
1331
1332The return may be incorrect for those rare locales where the currency symbol
1333replaces the radix character.
1334Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1335to work differently.
1336
1337=item C<ALT_DIGITS>
1338
1339Currently this gives the same results as Linux does.
1340Send email to L<mailto:perlbug@perl.org> if you have examples of it needing
1341to work differently.
1342
1343=item C<ERA_D_FMT>
1344
1345=item C<ERA_T_FMT>
1346
1347=item C<ERA_D_T_FMT>
1348
1349=item C<T_FMT_AMPM>
1350
1351These are derived by using C<strftime()>, and not all versions of that function
1352know about them. C<""> is returned for these on such systems.
1353
1354=back
1355
1356When using C<Perl_langinfo> on systems that don't have a native
1357C<nl_langinfo()>, you must
1358
1359 #include "perl_langinfo.h"
1360
1361before the C<perl.h> C<#include>. You can replace your C<langinfo.h>
1362C<#include> with this one. (Doing it this way keeps out the symbols that plain
1363C<langinfo.h> imports into the namespace for code that doesn't need it.)
1364
1365You also should not use the bare C<langinfo.h> item names, but should preface
1366them with C<PERL_>, so use C<PERL_RADIXCHAR> instead of plain C<RADIXCHAR>.
1367The C<PERL_I<foo>> versions will also work for this function on systems that do
1368have a native C<nl_langinfo>.
1369
1370=item *
1371
1372It is thread-friendly, returning its result in a buffer that won't be
1373overwritten by another thread, so you don't have to code for that possibility.
1374The buffer can be overwritten by the next call to C<nl_langinfo> or
1375C<Perl_langinfo> in the same thread.
1376
1377=item *
1378
7d4bcc4a 1379ª It returns S<C<const char *>>, whereas plain C<nl_langinfo()> returns S<C<char
f7416781
KW
1380*>>, but you are (only by documentation) forbidden to write into the buffer.
1381By declaring this C<const>, the compiler enforces this restriction. The extra
1382C<const> is why this isn't an unequivocal drop-in replacement for
1383C<nl_langinfo>.
1384
1385=back
1386
1387The original impetus for C<Perl_langinfo()> was so that code that needs to
1388find out the current currency symbol, floating point radix character, or digit
1389grouping separator can use, on all systems, the simpler and more
1390thread-friendly C<nl_langinfo> API instead of C<L<localeconv(3)>> which is a
1391pain to make thread-friendly. For other fields returned by C<localeconv>, it
1392is better to use the methods given in L<perlcall> to call
1393L<C<POSIX::localeconv()>|POSIX/localeconv>, which is thread-friendly.
1394
1395=cut
1396
1397*/
1398
1399const char *
1400#ifdef HAS_NL_LANGINFO
1401Perl_langinfo(const nl_item item)
1402#else
1403Perl_langinfo(const int item)
1404#endif
1405{
f61748ac
KW
1406 return my_nl_langinfo(item, TRUE);
1407}
1408
1409const char *
1410#ifdef HAS_NL_LANGINFO
1411S_my_nl_langinfo(const nl_item item, bool toggle)
1412#else
1413S_my_nl_langinfo(const int item, bool toggle)
1414#endif
1415{
ae74815b 1416 dTHX;
f7416781 1417
5a854ab3
KW
1418 /* We only need to toggle into the underlying LC_NUMERIC locale for these
1419 * two items, and only if not already there */
1420 if (toggle && (( item != PERL_RADIXCHAR && item != PERL_THOUSEP)
1421 || PL_numeric_underlying))
1422 {
1423 toggle = FALSE;
1424 }
1425
ab340fff
KW
1426#if defined(HAS_NL_LANGINFO) /* nl_langinfo() is available. */
1427#if ! defined(HAS_POSIX_2008_LOCALE)
f7416781 1428
ab340fff 1429 /* Here, use plain nl_langinfo(), switching to the underlying LC_NUMERIC
ae74815b
KW
1430 * for those items dependent on it. This must be copied to a buffer before
1431 * switching back, as some systems destroy the buffer when setlocale() is
1432 * called */
f7416781 1433
038d3702
KW
1434 {
1435 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
1436
b2fee59c 1437 if (toggle) {
038d3702 1438 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
b2fee59c 1439 }
f7416781 1440
5acc4454
KW
1441 LOCALE_LOCK; /* Prevent interference from another thread executing
1442 this code section (the only call to nl_langinfo in
1443 the core) */
1444
b2fee59c
KW
1445 save_to_buffer(nl_langinfo(item), &PL_langinfo_buf,
1446 &PL_langinfo_bufsize, 0);
f7416781 1447
5acc4454
KW
1448 LOCALE_UNLOCK;
1449
b2fee59c 1450 if (toggle) {
038d3702 1451 RESTORE_LC_NUMERIC();
b2fee59c 1452 }
f7416781
KW
1453 }
1454
ab340fff
KW
1455# else /* Use nl_langinfo_l(), avoiding both a mutex and changing the locale */
1456
5a854ab3 1457 {
b2fee59c
KW
1458 bool do_free = FALSE;
1459 locale_t cur = uselocale((locale_t) 0);
ab340fff 1460
b2fee59c
KW
1461 if (cur == LC_GLOBAL_LOCALE) {
1462 cur = duplocale(LC_GLOBAL_LOCALE);
1463 do_free = TRUE;
1464 }
ab340fff 1465
b2fee59c 1466 if (toggle) {
e1aa2579
KW
1467 if (PL_underlying_numeric_obj) {
1468 cur = PL_underlying_numeric_obj;
1469 }
1470 else {
1471 cur = newlocale(LC_NUMERIC_MASK, PL_numeric_name, cur);
1472 do_free = TRUE;
1473 }
b2fee59c 1474 }
ab340fff 1475
b2fee59c
KW
1476 save_to_buffer(nl_langinfo_l(item, cur),
1477 &PL_langinfo_buf, &PL_langinfo_bufsize, 0);
1478 if (do_free) {
1479 freelocale(cur);
1480 }
5a854ab3 1481 }
ab340fff 1482
c1566110
KW
1483# endif
1484
1485 if (strEQ(PL_langinfo_buf, "")) {
1486 if (item == PERL_YESSTR) {
1487 return "yes";
1488 }
1489 if (item == PERL_NOSTR) {
1490 return "no";
1491 }
1492 }
1493
ab340fff
KW
1494 return PL_langinfo_buf;
1495
f7416781 1496#else /* Below, emulate nl_langinfo as best we can */
43dd6b15
KW
1497
1498 {
1499
f7416781
KW
1500# ifdef HAS_LOCALECONV
1501
43dd6b15 1502 const struct lconv* lc;
038d3702 1503 DECLARATION_FOR_LC_NUMERIC_MANIPULATION;
f7416781
KW
1504
1505# endif
1506# ifdef HAS_STRFTIME
1507
43dd6b15
KW
1508 struct tm tm;
1509 bool return_format = FALSE; /* Return the %format, not the value */
1510 const char * format;
f7416781
KW
1511
1512# endif
1513
43dd6b15
KW
1514 /* We copy the results to a per-thread buffer, even if not
1515 * multi-threaded. This is in part to simplify this code, and partly
1516 * because we need a buffer anyway for strftime(), and partly because a
1517 * call of localeconv() could otherwise wipe out the buffer, and the
1518 * programmer would not be expecting this, as this is a nl_langinfo()
1519 * substitute after all, so s/he might be thinking their localeconv()
1520 * is safe until another localeconv() call. */
f7416781 1521
43dd6b15
KW
1522 switch (item) {
1523 Size_t len;
1524 const char * retval;
f7416781 1525
43dd6b15
KW
1526 /* These 2 are unimplemented */
1527 case PERL_CODESET:
1528 case PERL_ERA: /* For use with strftime() %E modifier */
f7416781 1529
43dd6b15
KW
1530 default:
1531 return "";
f7416781 1532
43dd6b15
KW
1533 /* We use only an English set, since we don't know any more */
1534 case PERL_YESEXPR: return "^[+1yY]";
1535 case PERL_YESSTR: return "yes";
1536 case PERL_NOEXPR: return "^[-0nN]";
1537 case PERL_NOSTR: return "no";
f7416781
KW
1538
1539# ifdef HAS_LOCALECONV
1540
43dd6b15 1541 case PERL_CRNCYSTR:
f7416781 1542
43dd6b15
KW
1543 /* We don't bother with localeconv_l() because any system that
1544 * has it is likely to also have nl_langinfo() */
291a84fb 1545
5acc4454
KW
1546 LOCALE_LOCK; /* Prevent interference with other threads
1547 using localeconv() */
1548
43dd6b15
KW
1549 lc = localeconv();
1550 if ( ! lc
1551 || ! lc->currency_symbol
1552 || strEQ("", lc->currency_symbol))
1553 {
1554 LOCALE_UNLOCK;
1555 return "";
1556 }
f7416781 1557
43dd6b15
KW
1558 /* Leave the first spot empty to be filled in below */
1559 save_to_buffer(lc->currency_symbol, &PL_langinfo_buf,
1560 &PL_langinfo_bufsize, 1);
1561 if (lc->mon_decimal_point && strEQ(lc->mon_decimal_point, ""))
1562 { /* khw couldn't figure out how the localedef specifications
1563 would show that the $ should replace the radix; this is
1564 just a guess as to how it might work.*/
1565 *PL_langinfo_buf = '.';
1566 }
1567 else if (lc->p_cs_precedes) {
1568 *PL_langinfo_buf = '-';
1569 }
1570 else {
1571 *PL_langinfo_buf = '+';
1572 }
f7416781 1573
43dd6b15
KW
1574 LOCALE_UNLOCK;
1575 break;
f7416781 1576
43dd6b15
KW
1577 case PERL_RADIXCHAR:
1578 case PERL_THOUSEP:
f7416781 1579
43dd6b15 1580 if (toggle) {
038d3702 1581 STORE_LC_NUMERIC_FORCE_TO_UNDERLYING();
c0d737a8 1582 }
f7416781 1583
5acc4454
KW
1584 LOCALE_LOCK; /* Prevent interference with other threads
1585 using localeconv() */
1586
43dd6b15
KW
1587 lc = localeconv();
1588 if (! lc) {
33394adc
KW
1589 retval = "";
1590 }
43dd6b15
KW
1591 else {
1592 retval = (item == PERL_RADIXCHAR)
1593 ? lc->decimal_point
1594 : lc->thousands_sep;
1595 if (! retval) {
1596 retval = "";
1597 }
1598 }
f7416781 1599
43dd6b15
KW
1600 save_to_buffer(retval, &PL_langinfo_buf,
1601 &PL_langinfo_bufsize, 0);
f7416781 1602
5acc4454
KW
1603 LOCALE_UNLOCK;
1604
43dd6b15 1605 if (toggle) {
038d3702 1606 RESTORE_LC_NUMERIC();
43dd6b15 1607 }
f7416781 1608
43dd6b15 1609 break;
f7416781
KW
1610
1611# endif
1612# ifdef HAS_STRFTIME
1613
43dd6b15
KW
1614 /* These are defined by C89, so we assume that strftime supports
1615 * them, and so are returned unconditionally; they may not be what
1616 * the locale actually says, but should give good enough results
1617 * for someone using them as formats (as opposed to trying to parse
1618 * them to figure out what the locale says). The other format
1619 * items are actually tested to verify they work on the platform */
1620 case PERL_D_FMT: return "%x";
1621 case PERL_T_FMT: return "%X";
1622 case PERL_D_T_FMT: return "%c";
1623
1624 /* These formats are only available in later strfmtime's */
1625 case PERL_ERA_D_FMT: case PERL_ERA_T_FMT: case PERL_ERA_D_T_FMT:
1626 case PERL_T_FMT_AMPM:
1627
1628 /* The rest can be gotten from most versions of strftime(). */
1629 case PERL_ABDAY_1: case PERL_ABDAY_2: case PERL_ABDAY_3:
1630 case PERL_ABDAY_4: case PERL_ABDAY_5: case PERL_ABDAY_6:
1631 case PERL_ABDAY_7:
1632 case PERL_ALT_DIGITS:
1633 case PERL_AM_STR: case PERL_PM_STR:
1634 case PERL_ABMON_1: case PERL_ABMON_2: case PERL_ABMON_3:
1635 case PERL_ABMON_4: case PERL_ABMON_5: case PERL_ABMON_6:
1636 case PERL_ABMON_7: case PERL_ABMON_8: case PERL_ABMON_9:
1637 case PERL_ABMON_10: case PERL_ABMON_11: case PERL_ABMON_12:
1638 case PERL_DAY_1: case PERL_DAY_2: case PERL_DAY_3: case PERL_DAY_4:
1639 case PERL_DAY_5: case PERL_DAY_6: case PERL_DAY_7:
1640 case PERL_MON_1: case PERL_MON_2: case PERL_MON_3: case PERL_MON_4:
1641 case PERL_MON_5: case PERL_MON_6: case PERL_MON_7: case PERL_MON_8:
1642 case PERL_MON_9: case PERL_MON_10: case PERL_MON_11:
1643 case PERL_MON_12:
1644
1645 LOCALE_LOCK;
1646
1647 init_tm(&tm); /* Precaution against core dumps */
1648 tm.tm_sec = 30;
1649 tm.tm_min = 30;
1650 tm.tm_hour = 6;
1651 tm.tm_year = 2017 - 1900;
1652 tm.tm_wday = 0;
1653 tm.tm_mon = 0;
1654 switch (item) {
1655 default:
1656 LOCALE_UNLOCK;
1657 Perl_croak(aTHX_
1658 "panic: %s: %d: switch case: %d problem",
1659 __FILE__, __LINE__, item);
1660 NOT_REACHED; /* NOTREACHED */
1661
1662 case PERL_PM_STR: tm.tm_hour = 18;
1663 case PERL_AM_STR:
1664 format = "%p";
1665 break;
1666
1667 case PERL_ABDAY_7: tm.tm_wday++;
1668 case PERL_ABDAY_6: tm.tm_wday++;
1669 case PERL_ABDAY_5: tm.tm_wday++;
1670 case PERL_ABDAY_4: tm.tm_wday++;
1671 case PERL_ABDAY_3: tm.tm_wday++;
1672 case PERL_ABDAY_2: tm.tm_wday++;
1673 case PERL_ABDAY_1:
1674 format = "%a";
1675 break;
1676
1677 case PERL_DAY_7: tm.tm_wday++;
1678 case PERL_DAY_6: tm.tm_wday++;
1679 case PERL_DAY_5: tm.tm_wday++;
1680 case PERL_DAY_4: tm.tm_wday++;
1681 case PERL_DAY_3: tm.tm_wday++;
1682 case PERL_DAY_2: tm.tm_wday++;
1683 case PERL_DAY_1:
1684 format = "%A";
1685 break;
1686
1687 case PERL_ABMON_12: tm.tm_mon++;
1688 case PERL_ABMON_11: tm.tm_mon++;
1689 case PERL_ABMON_10: tm.tm_mon++;
1690 case PERL_ABMON_9: tm.tm_mon++;
1691 case PERL_ABMON_8: tm.tm_mon++;
1692 case PERL_ABMON_7: tm.tm_mon++;
1693 case PERL_ABMON_6: tm.tm_mon++;
1694 case PERL_ABMON_5: tm.tm_mon++;
1695 case PERL_ABMON_4: tm.tm_mon++;
1696 case PERL_ABMON_3: tm.tm_mon++;
1697 case PERL_ABMON_2: tm.tm_mon++;
1698 case PERL_ABMON_1:
1699 format = "%b";
1700 break;
1701
1702 case PERL_MON_12: tm.tm_mon++;
1703 case PERL_MON_11: tm.tm_mon++;
1704 case PERL_MON_10: tm.tm_mon++;
1705 case PERL_MON_9: tm.tm_mon++;
1706 case PERL_MON_8: tm.tm_mon++;
1707 case PERL_MON_7: tm.tm_mon++;
1708 case PERL_MON_6: tm.tm_mon++;
1709 case PERL_MON_5: tm.tm_mon++;
1710 case PERL_MON_4: tm.tm_mon++;
1711 case PERL_MON_3: tm.tm_mon++;
1712 case PERL_MON_2: tm.tm_mon++;
1713 case PERL_MON_1:
1714 format = "%B";
1715 break;
1716
1717 case PERL_T_FMT_AMPM:
1718 format = "%r";
1719 return_format = TRUE;
1720 break;
1721
1722 case PERL_ERA_D_FMT:
1723 format = "%Ex";
1724 return_format = TRUE;
1725 break;
1726
1727 case PERL_ERA_T_FMT:
1728 format = "%EX";
1729 return_format = TRUE;
1730 break;
1731
1732 case PERL_ERA_D_T_FMT:
1733 format = "%Ec";
1734 return_format = TRUE;
1735 break;
1736
1737 case PERL_ALT_DIGITS:
1738 tm.tm_wday = 0;
1739 format = "%Ow"; /* Find the alternate digit for 0 */
1740 break;
1741 }
f7416781 1742
43dd6b15
KW
1743 /* We can't use my_strftime() because it doesn't look at
1744 * tm_wday */
1745 while (0 == strftime(PL_langinfo_buf, PL_langinfo_bufsize,
1746 format, &tm))
1747 {
1748 /* A zero return means one of:
1749 * a) there wasn't enough space in PL_langinfo_buf
1750 * b) the format, like a plain %p, returns empty
1751 * c) it was an illegal format, though some
1752 * implementations of strftime will just return the
1753 * illegal format as a plain character sequence.
1754 *
1755 * To quickly test for case 'b)', try again but precede
1756 * the format with a plain character. If that result is
1757 * still empty, the problem is either 'a)' or 'c)' */
1758
1759 Size_t format_size = strlen(format) + 1;
1760 Size_t mod_size = format_size + 1;
1761 char * mod_format;
1762 char * temp_result;
1763
1764 Newx(mod_format, mod_size, char);
1765 Newx(temp_result, PL_langinfo_bufsize, char);
6873aa47 1766 *mod_format = ' ';
43dd6b15
KW
1767 my_strlcpy(mod_format + 1, format, mod_size);
1768 len = strftime(temp_result,
1769 PL_langinfo_bufsize,
1770 mod_format, &tm);
1771 Safefree(mod_format);
1772 Safefree(temp_result);
1773
1774 /* If 'len' is non-zero, it means that we had a case like
1775 * %p which means the current locale doesn't use a.m. or
1776 * p.m., and that is valid */
1777 if (len == 0) {
1778
1779 /* Here, still didn't work. If we get well beyond a
1780 * reasonable size, bail out to prevent an infinite
1781 * loop. */
1782
1783 if (PL_langinfo_bufsize > 100 * format_size) {
1784 *PL_langinfo_buf = '\0';
1785 }
1786 else {
1787 /* Double the buffer size to retry; Add 1 in case
1788 * original was 0, so we aren't stuck at 0. */
1789 PL_langinfo_bufsize *= 2;
1790 PL_langinfo_bufsize++;
1791 Renew(PL_langinfo_buf, PL_langinfo_bufsize, char);
1792 continue;
1793 }
1794 }
f7416781 1795
f7416781 1796 break;
43dd6b15 1797 }
f7416781 1798
43dd6b15
KW
1799 /* Here, we got a result.
1800 *
1801 * If the item is 'ALT_DIGITS', PL_langinfo_buf contains the
1802 * alternate format for wday 0. If the value is the same as
1803 * the normal 0, there isn't an alternate, so clear the buffer.
1804 * */
1805 if ( item == PERL_ALT_DIGITS
1806 && strEQ(PL_langinfo_buf, "0"))
1807 {
1808 *PL_langinfo_buf = '\0';
1809 }
f7416781 1810
43dd6b15
KW
1811 /* ALT_DIGITS is problematic. Experiments on it showed that
1812 * strftime() did not always work properly when going from
1813 * alt-9 to alt-10. Only a few locales have this item defined,
1814 * and in all of them on Linux that khw was able to find,
1815 * nl_langinfo() merely returned the alt-0 character, possibly
1816 * doubled. Most Unicode digits are in blocks of 10
1817 * consecutive code points, so that is sufficient information
1818 * for those scripts, as we can infer alt-1, alt-2, .... But
1819 * for a Japanese locale, a CJK ideographic 0 is returned, and
1820 * the CJK digits are not in code point order, so you can't
1821 * really infer anything. The localedef for this locale did
1822 * specify the succeeding digits, so that strftime() works
1823 * properly on them, without needing to infer anything. But
1824 * the nl_langinfo() return did not give sufficient information
1825 * for the caller to understand what's going on. So until
1826 * there is evidence that it should work differently, this
1827 * returns the alt-0 string for ALT_DIGITS.
1828 *
1829 * wday was chosen because its range is all a single digit.
1830 * Things like tm_sec have two digits as the minimum: '00' */
f7416781 1831
43dd6b15 1832 LOCALE_UNLOCK;
f7416781 1833
43dd6b15
KW
1834 /* If to return the format, not the value, overwrite the buffer
1835 * with it. But some strftime()s will keep the original format
1836 * if illegal, so change those to "" */
1837 if (return_format) {
1838 if (strEQ(PL_langinfo_buf, format)) {
f7416781
KW
1839 *PL_langinfo_buf = '\0';
1840 }
43dd6b15
KW
1841 else {
1842 save_to_buffer(format, &PL_langinfo_buf,
1843 &PL_langinfo_bufsize, 0);
f7416781
KW
1844 }
1845 }
1846
1847 break;
f7416781
KW
1848
1849# endif
1850
43dd6b15 1851 }
f7416781
KW
1852 }
1853
1854 return PL_langinfo_buf;
1855
1856#endif
1857
a4f00dcc 1858}
b385bb4d 1859
98994639
HS
1860/*
1861 * Initialize locale awareness.
1862 */
1863int
1864Perl_init_i18nl10n(pTHX_ int printwarn)
1865{
0e92a118
KW
1866 /* printwarn is
1867 *
1868 * 0 if not to output warning when setup locale is bad
1869 * 1 if to output warning based on value of PERL_BADLANG
1870 * >1 if to output regardless of PERL_BADLANG
1871 *
1872 * returns
98994639 1873 * 1 = set ok or not applicable,
0e92a118
KW
1874 * 0 = fallback to a locale of lower priority
1875 * -1 = fallback to all locales failed, not even to the C locale
6b058d42
KW
1876 *
1877 * Under -DDEBUGGING, if the environment variable PERL_DEBUG_LOCALE_INIT is
1878 * set, debugging information is output.
1879 *
1880 * This looks more complicated than it is, mainly due to the #ifdefs.
1881 *
1882 * We try to set LC_ALL to the value determined by the environment. If
1883 * there is no LC_ALL on this platform, we try the individual categories we
1884 * know about. If this works, we are done.
1885 *
1886 * But if it doesn't work, we have to do something else. We search the
1887 * environment variables ourselves instead of relying on the system to do
1888 * it. We look at, in order, LC_ALL, LANG, a system default locale (if we
1889 * think there is one), and the ultimate fallback "C". This is all done in
1890 * the same loop as above to avoid duplicating code, but it makes things
7d4bcc4a
KW
1891 * more complex. The 'trial_locales' array is initialized with just one
1892 * element; it causes the behavior described in the paragraph above this to
1893 * happen. If that fails, we add elements to 'trial_locales', and do extra
1894 * loop iterations to cause the behavior described in this paragraph.
6b058d42
KW
1895 *
1896 * On Ultrix, the locale MUST come from the environment, so there is
1897 * preliminary code to set it. I (khw) am not sure that it is necessary,
1898 * and that this couldn't be folded into the loop, but barring any real
1899 * platforms to test on, it's staying as-is
1900 *
1901 * A slight complication is that in embedded Perls, the locale may already
1902 * be set-up, and we don't want to get it from the normal environment
1903 * variables. This is handled by having a special environment variable
1904 * indicate we're in this situation. We simply set setlocale's 2nd
1905 * parameter to be a NULL instead of "". That indicates to setlocale that
1906 * it is not to change anything, but to return the current value,
1907 * effectively initializing perl's db to what the locale already is.
1908 *
1909 * We play the same trick with NULL if a LC_ALL succeeds. We call
1910 * setlocale() on the individual categores with NULL to get their existing
1911 * values for our db, instead of trying to change them.
1912 * */
98994639 1913
0e92a118
KW
1914 int ok = 1;
1915
7d4bcc4a
KW
1916#ifndef USE_LOCALE
1917
1918 PERL_UNUSED_ARG(printwarn);
1919
1920#else /* USE_LOCALE */
7d4bcc4a
KW
1921# ifdef __GLIBC__
1922
175c4cf9 1923 const char * const language = savepv(PerlEnv_getenv("LANGUAGE"));
7d4bcc4a
KW
1924
1925# endif
65ebb059 1926
ccd65d51
KW
1927 /* NULL uses the existing already set up locale */
1928 const char * const setlocale_init = (PerlEnv_getenv("PERL_SKIP_LOCALE_INIT"))
1929 ? NULL
1930 : "";
c3fcd832
KW
1931 const char* trial_locales[5]; /* 5 = 1 each for "", LC_ALL, LANG, "", C */
1932 unsigned int trial_locales_count;
175c4cf9
KW
1933 const char * const lc_all = savepv(PerlEnv_getenv("LC_ALL"));
1934 const char * const lang = savepv(PerlEnv_getenv("LANG"));
98994639 1935 bool setlocale_failure = FALSE;
65ebb059 1936 unsigned int i;
175c4cf9
KW
1937
1938 /* A later getenv() could zap this, so only use here */
1939 const char * const bad_lang_use_once = PerlEnv_getenv("PERL_BADLANG");
1940
1941 const bool locwarn = (printwarn > 1
e5f10d49
KW
1942 || ( printwarn
1943 && ( ! bad_lang_use_once
22ff3130 1944 || (
e5f10d49
KW
1945 /* disallow with "" or "0" */
1946 *bad_lang_use_once
1947 && strNE("0", bad_lang_use_once)))));
ea92aad8 1948
291a84fb 1949 /* setlocale() return vals; not copied so must be looked at immediately */
8de4332b 1950 const char * sl_result[NOMINAL_LC_ALL_INDEX + 1];
291a84fb
KW
1951
1952 /* current locale for given category; should have been copied so aren't
1953 * volatile */
8de4332b 1954 const char * curlocales[NOMINAL_LC_ALL_INDEX + 1];
291a84fb 1955
7d4bcc4a
KW
1956# ifdef WIN32
1957
6bce99ee
JH
1958 /* In some systems you can find out the system default locale
1959 * and use that as the fallback locale. */
7d4bcc4a
KW
1960# define SYSTEM_DEFAULT_LOCALE
1961# endif
1962# ifdef SYSTEM_DEFAULT_LOCALE
1963
65ebb059 1964 const char *system_default_locale = NULL;
98994639 1965
7d4bcc4a 1966# endif
948523db
KW
1967
1968# ifndef DEBUGGING
1969# define DEBUG_LOCALE_INIT(a,b,c)
1970# else
7d4bcc4a 1971
8298454c 1972 DEBUG_INITIALIZATION_set(cBOOL(PerlEnv_getenv("PERL_DEBUG_LOCALE_INIT")));
7d4bcc4a
KW
1973
1974# define DEBUG_LOCALE_INIT(category, locale, result) \
2fcc0ca9
KW
1975 STMT_START { \
1976 if (debug_initialization) { \
1977 PerlIO_printf(Perl_debug_log, \
1978 "%s:%d: %s\n", \
1979 __FILE__, __LINE__, \
a4f00dcc 1980 setlocale_debug_string(category, \
2fcc0ca9
KW
1981 locale, \
1982 result)); \
1983 } \
1984 } STMT_END
2fcc0ca9 1985
948523db
KW
1986/* Make sure the parallel arrays are properly set up */
1987# ifdef USE_LOCALE_NUMERIC
1988 assert(categories[LC_NUMERIC_INDEX] == LC_NUMERIC);
1989 assert(strEQ(category_names[LC_NUMERIC_INDEX], "LC_NUMERIC"));
1990# endif
1991# ifdef USE_LOCALE_CTYPE
1992 assert(categories[LC_CTYPE_INDEX] == LC_CTYPE);
1993 assert(strEQ(category_names[LC_CTYPE_INDEX], "LC_CTYPE"));
1994# endif
1995# ifdef USE_LOCALE_COLLATE
1996 assert(categories[LC_COLLATE_INDEX] == LC_COLLATE);
1997 assert(strEQ(category_names[LC_COLLATE_INDEX], "LC_COLLATE"));
1998# endif
1999# ifdef USE_LOCALE_TIME
2000 assert(categories[LC_TIME_INDEX] == LC_TIME);
2001 assert(strEQ(category_names[LC_TIME_INDEX], "LC_TIME"));
2002# endif
2003# ifdef USE_LOCALE_MESSAGES
2004 assert(categories[LC_MESSAGES_INDEX] == LC_MESSAGES);
2005 assert(strEQ(category_names[LC_MESSAGES_INDEX], "LC_MESSAGES"));
2006# endif
2007# ifdef USE_LOCALE_MONETARY
2008 assert(categories[LC_MONETARY_INDEX] == LC_MONETARY);
2009 assert(strEQ(category_names[LC_MONETARY_INDEX], "LC_MONETARY"));
2010# endif
9821811f
KW
2011# ifdef USE_LOCALE_ADDRESS
2012 assert(categories[LC_ADDRESS_INDEX] == LC_ADDRESS);
2013 assert(strEQ(category_names[LC_ADDRESS_INDEX], "LC_ADDRESS"));
2014# endif
2015# ifdef USE_LOCALE_IDENTIFICATION
2016 assert(categories[LC_IDENTIFICATION_INDEX] == LC_IDENTIFICATION);
2017 assert(strEQ(category_names[LC_IDENTIFICATION_INDEX], "LC_IDENTIFICATION"));
2018# endif
2019# ifdef USE_LOCALE_MEASUREMENT
2020 assert(categories[LC_MEASUREMENT_INDEX] == LC_MEASUREMENT);
2021 assert(strEQ(category_names[LC_MEASUREMENT_INDEX], "LC_MEASUREMENT"));
2022# endif
2023# ifdef USE_LOCALE_PAPER
2024 assert(categories[LC_PAPER_INDEX] == LC_PAPER);
2025 assert(strEQ(category_names[LC_PAPER_INDEX], "LC_PAPER"));
2026# endif
2027# ifdef USE_LOCALE_TELEPHONE
2028 assert(categories[LC_TELEPHONE_INDEX] == LC_TELEPHONE);
2029 assert(strEQ(category_names[LC_TELEPHONE_INDEX], "LC_TELEPHONE"));
2030# endif
948523db
KW
2031# ifdef LC_ALL
2032 assert(categories[LC_ALL_INDEX] == LC_ALL);
2033 assert(strEQ(category_names[LC_ALL_INDEX], "LC_ALL"));
2034 assert(NOMINAL_LC_ALL_INDEX == LC_ALL_INDEX);
2035# endif
2036# endif /* DEBUGGING */
47280b20
KW
2037
2038 /* Initialize the cache of the program's UTF-8ness for the always known
2039 * locales C and POSIX */
2040 my_strlcpy(PL_locale_utf8ness, C_and_POSIX_utf8ness,
2041 sizeof(PL_locale_utf8ness));
2042
3ca88433
KW
2043 PL_numeric_radix_sv = newSVpvs(".");
2044
b56c4436 2045# ifdef LOCALE_ENVIRON_REQUIRED
98994639
HS
2046
2047 /*
2048 * Ultrix setlocale(..., "") fails if there are no environment
2049 * variables from which to get a locale name.
2050 */
2051
b56c4436
KW
2052# ifndef LC_ALL
2053# error Ultrix without LC_ALL not implemented
2054# else
7d4bcc4a 2055
b56c4436
KW
2056 {
2057 bool done = FALSE;
ea92aad8
KW
2058 if (lang) {
2059 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, setlocale_init);
2060 DEBUG_LOCALE_INIT(LC_ALL, setlocale_init, sl_result[LC_ALL_INDEX]);
2061 if (sl_result[LC_ALL_INDEX])
2062 done = TRUE;
2063 else
e5f10d49 2064 setlocale_failure = TRUE;
ea92aad8
KW
2065 }
2066 if (! setlocale_failure) {
2067 const char * locale_param;
2068 for (i = 0; i < LC_ALL_INDEX; i++) {
2069 locale_param = (! done && (lang || PerlEnv_getenv(category_names[i])))
2070 ? setlocale_init
2071 : NULL;
2072 sl_result[i] = do_setlocale_r(categories[i], locale_param);
2073 if (! sl_result[i]) {
2074 setlocale_failure = TRUE;
2075 }
2076 DEBUG_LOCALE_INIT(categories[i], locale_param, sl_result[i]);
e5f10d49 2077 }
c835d6be 2078 }
7d4bcc4a
KW
2079 }
2080
2081# endif /* LC_ALL */
e5f10d49 2082# endif /* LOCALE_ENVIRON_REQUIRED */
98994639 2083
65ebb059 2084 /* We try each locale in the list until we get one that works, or exhaust
20a240df
KW
2085 * the list. Normally the loop is executed just once. But if setting the
2086 * locale fails, inside the loop we add fallback trials to the array and so
2087 * will execute the loop multiple times */
c3fcd832
KW
2088 trial_locales[0] = setlocale_init;
2089 trial_locales_count = 1;
7d4bcc4a 2090
65ebb059
KW
2091 for (i= 0; i < trial_locales_count; i++) {
2092 const char * trial_locale = trial_locales[i];
2093
2094 if (i > 0) {
2095
2096 /* XXX This is to preserve old behavior for LOCALE_ENVIRON_REQUIRED
2097 * when i==0, but I (khw) don't think that behavior makes much
2098 * sense */
2099 setlocale_failure = FALSE;
2100
7d4bcc4a 2101# ifdef SYSTEM_DEFAULT_LOCALE
291a84fb 2102# ifdef WIN32 /* Note that assumes Win32 has LC_ALL */
7d4bcc4a 2103
65ebb059
KW
2104 /* On Windows machines, an entry of "" after the 0th means to use
2105 * the system default locale, which we now proceed to get. */
2106 if (strEQ(trial_locale, "")) {
2107 unsigned int j;
2108
2109 /* Note that this may change the locale, but we are going to do
2110 * that anyway just below */
837ce802 2111 system_default_locale = do_setlocale_c(LC_ALL, "");
5d1187d1 2112 DEBUG_LOCALE_INIT(LC_ALL, "", system_default_locale);
65ebb059 2113
7d4bcc4a 2114 /* Skip if invalid or if it's already on the list of locales to
65ebb059
KW
2115 * try */
2116 if (! system_default_locale) {
2117 goto next_iteration;
2118 }
2119 for (j = 0; j < trial_locales_count; j++) {
2120 if (strEQ(system_default_locale, trial_locales[j])) {
2121 goto next_iteration;
2122 }
2123 }
2124
2125 trial_locale = system_default_locale;
2126 }
ec0202b5
KW
2127# else
2128# error SYSTEM_DEFAULT_LOCALE only implemented for Win32
2129# endif
7d4bcc4a 2130# endif /* SYSTEM_DEFAULT_LOCALE */
291a84fb
KW
2131
2132 } /* For i > 0 */
65ebb059 2133
7d4bcc4a
KW
2134# ifdef LC_ALL
2135
948523db
KW
2136 sl_result[LC_ALL_INDEX] = do_setlocale_c(LC_ALL, trial_locale);
2137 DEBUG_LOCALE_INIT(LC_ALL, trial_locale, sl_result[LC_ALL_INDEX]);
2138 if (! sl_result[LC_ALL_INDEX]) {
49c85077 2139 setlocale_failure = TRUE;
7cd8b568
KW
2140 }
2141 else {
2142 /* Since LC_ALL succeeded, it should have changed all the other
2143 * categories it can to its value; so we massage things so that the
2144 * setlocales below just return their category's current values.
2145 * This adequately handles the case in NetBSD where LC_COLLATE may
2146 * not be defined for a locale, and setting it individually will
7d4bcc4a 2147 * fail, whereas setting LC_ALL succeeds, leaving LC_COLLATE set to
7cd8b568
KW
2148 * the POSIX locale. */
2149 trial_locale = NULL;
2150 }
7d4bcc4a
KW
2151
2152# endif /* LC_ALL */
98994639 2153
e5f10d49
KW
2154 if (! setlocale_failure) {
2155 unsigned int j;
2156 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2157 curlocales[j]
2158 = savepv(do_setlocale_r(categories[j], trial_locale));
2159 if (! curlocales[j]) {
2160 setlocale_failure = TRUE;
2161 }
2162 DEBUG_LOCALE_INIT(categories[j], trial_locale, curlocales[j]);
2163 }
c835d6be 2164
e5f10d49
KW
2165 if (! setlocale_failure) { /* All succeeded */
2166 break; /* Exit trial_locales loop */
49c85077 2167 }
65ebb059 2168 }
98994639 2169
49c85077
KW
2170 /* Here, something failed; will need to try a fallback. */
2171 ok = 0;
65ebb059 2172
49c85077
KW
2173 if (i == 0) {
2174 unsigned int j;
98994639 2175
65ebb059 2176 if (locwarn) { /* Output failure info only on the first one */
7d4bcc4a
KW
2177
2178# ifdef LC_ALL
98994639 2179
49c85077
KW
2180 PerlIO_printf(Perl_error_log,
2181 "perl: warning: Setting locale failed.\n");
98994639 2182
7d4bcc4a 2183# else /* !LC_ALL */
98994639 2184
49c85077
KW
2185 PerlIO_printf(Perl_error_log,
2186 "perl: warning: Setting locale failed for the categories:\n\t");
7d4bcc4a 2187
e5f10d49
KW
2188 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2189 if (! curlocales[j]) {
2190 PerlIO_printf(Perl_error_log, category_names[j]);
2191 }
2192 else {
2193 Safefree(curlocales[j]);
2194 }
2195 }
7d4bcc4a 2196
7d4bcc4a 2197# endif /* LC_ALL */
98994639 2198
49c85077
KW
2199 PerlIO_printf(Perl_error_log,
2200 "perl: warning: Please check that your locale settings:\n");
98994639 2201
7d4bcc4a
KW
2202# ifdef __GLIBC__
2203
49c85077
KW
2204 PerlIO_printf(Perl_error_log,
2205 "\tLANGUAGE = %c%s%c,\n",
2206 language ? '"' : '(',
2207 language ? language : "unset",
2208 language ? '"' : ')');
7d4bcc4a 2209# endif
98994639 2210
49c85077
KW
2211 PerlIO_printf(Perl_error_log,
2212 "\tLC_ALL = %c%s%c,\n",
2213 lc_all ? '"' : '(',
2214 lc_all ? lc_all : "unset",
2215 lc_all ? '"' : ')');
98994639 2216
7d4bcc4a
KW
2217# if defined(USE_ENVIRON_ARRAY)
2218
49c85077 2219 {
cd999af9 2220 char **e;
d5e32b93
KW
2221
2222 /* Look through the environment for any variables of the
2223 * form qr/ ^ LC_ [A-Z]+ = /x, except LC_ALL which was
2224 * already handled above. These are assumed to be locale
2225 * settings. Output them and their values. */
cd999af9 2226 for (e = environ; *e; e++) {
d5e32b93
KW
2227 const STRLEN prefix_len = sizeof("LC_") - 1;
2228 STRLEN uppers_len;
2229
cd999af9 2230 if ( strBEGINs(*e, "LC_")
c8b388b0 2231 && ! strBEGINs(*e, "LC_ALL=")
d5e32b93
KW
2232 && (uppers_len = strspn(*e + prefix_len,
2233 "ABCDEFGHIJKLMNOPQRSTUVWXYZ"))
2234 && ((*e)[prefix_len + uppers_len] == '='))
cd999af9
KW
2235 {
2236 PerlIO_printf(Perl_error_log, "\t%.*s = \"%s\",\n",
d5e32b93
KW
2237 (int) (prefix_len + uppers_len), *e,
2238 *e + prefix_len + uppers_len + 1);
cd999af9
KW
2239 }
2240 }
49c85077 2241 }
7d4bcc4a
KW
2242
2243# else
2244
49c85077
KW
2245 PerlIO_printf(Perl_error_log,
2246 "\t(possibly more locale environment variables)\n");
7d4bcc4a
KW
2247
2248# endif
98994639 2249
49c85077
KW
2250 PerlIO_printf(Perl_error_log,
2251 "\tLANG = %c%s%c\n",
2252 lang ? '"' : '(',
2253 lang ? lang : "unset",
2254 lang ? '"' : ')');
98994639 2255
49c85077
KW
2256 PerlIO_printf(Perl_error_log,
2257 " are supported and installed on your system.\n");
2258 }
98994639 2259
65ebb059 2260 /* Calculate what fallback locales to try. We have avoided this
f6bab5f6 2261 * until we have to, because failure is quite unlikely. This will
65ebb059
KW
2262 * usually change the upper bound of the loop we are in.
2263 *
2264 * Since the system's default way of setting the locale has not
2265 * found one that works, We use Perl's defined ordering: LC_ALL,
2266 * LANG, and the C locale. We don't try the same locale twice, so
2267 * don't add to the list if already there. (On POSIX systems, the
2268 * LC_ALL element will likely be a repeat of the 0th element "",
6b058d42
KW
2269 * but there's no harm done by doing it explicitly.
2270 *
2271 * Note that this tries the LC_ALL environment variable even on
2272 * systems which have no LC_ALL locale setting. This may or may
2273 * not have been originally intentional, but there's no real need
2274 * to change the behavior. */
65ebb059
KW
2275 if (lc_all) {
2276 for (j = 0; j < trial_locales_count; j++) {
2277 if (strEQ(lc_all, trial_locales[j])) {
2278 goto done_lc_all;
2279 }
2280 }
2281 trial_locales[trial_locales_count++] = lc_all;
2282 }
2283 done_lc_all:
98994639 2284
65ebb059
KW
2285 if (lang) {
2286 for (j = 0; j < trial_locales_count; j++) {
2287 if (strEQ(lang, trial_locales[j])) {
2288 goto done_lang;
2289 }
2290 }
2291 trial_locales[trial_locales_count++] = lang;
2292 }
2293 done_lang:
2294
7d4bcc4a
KW
2295# if defined(WIN32) && defined(LC_ALL)
2296
65ebb059
KW
2297 /* For Windows, we also try the system default locale before "C".
2298 * (If there exists a Windows without LC_ALL we skip this because
2299 * it gets too complicated. For those, the "C" is the next
2300 * fallback possibility). The "" is the same as the 0th element of
2301 * the array, but the code at the loop above knows to treat it
2302 * differently when not the 0th */
2303 trial_locales[trial_locales_count++] = "";
7d4bcc4a
KW
2304
2305# endif
65ebb059
KW
2306
2307 for (j = 0; j < trial_locales_count; j++) {
2308 if (strEQ("C", trial_locales[j])) {
2309 goto done_C;
2310 }
2311 }
2312 trial_locales[trial_locales_count++] = "C";
98994639 2313
65ebb059
KW
2314 done_C: ;
2315 } /* end of first time through the loop */
98994639 2316
7d4bcc4a
KW
2317# ifdef WIN32
2318
65ebb059 2319 next_iteration: ;
7d4bcc4a
KW
2320
2321# endif
65ebb059
KW
2322
2323 } /* end of looping through the trial locales */
2324
2325 if (ok < 1) { /* If we tried to fallback */
2326 const char* msg;
2327 if (! setlocale_failure) { /* fallback succeeded */
2328 msg = "Falling back to";
2329 }
2330 else { /* fallback failed */
e5f10d49 2331 unsigned int j;
98994639 2332
65ebb059
KW
2333 /* We dropped off the end of the loop, so have to decrement i to
2334 * get back to the value the last time through */
2335 i--;
98994639 2336
65ebb059
KW
2337 ok = -1;
2338 msg = "Failed to fall back to";
2339
2340 /* To continue, we should use whatever values we've got */
7d4bcc4a 2341
e5f10d49
KW
2342 for (j = 0; j < NOMINAL_LC_ALL_INDEX; j++) {
2343 Safefree(curlocales[j]);
2344 curlocales[j] = savepv(do_setlocale_r(categories[j], NULL));
2345 DEBUG_LOCALE_INIT(categories[j], NULL, curlocales[j]);
2346 }
65ebb059
KW
2347 }
2348
2349 if (locwarn) {
2350 const char * description;
2351 const char * name = "";
2352 if (strEQ(trial_locales[i], "C")) {
2353 description = "the standard locale";
2354 name = "C";
2355 }
7d4bcc4a
KW
2356
2357# ifdef SYSTEM_DEFAULT_LOCALE
2358
65ebb059
KW
2359 else if (strEQ(trial_locales[i], "")) {
2360 description = "the system default locale";
2361 if (system_default_locale) {
2362 name = system_default_locale;
2363 }
2364 }
7d4bcc4a
KW
2365
2366# endif /* SYSTEM_DEFAULT_LOCALE */
2367
65ebb059
KW
2368 else {
2369 description = "a fallback locale";
2370 name = trial_locales[i];
2371 }
2372 if (name && strNE(name, "")) {
2373 PerlIO_printf(Perl_error_log,
2374 "perl: warning: %s %s (\"%s\").\n", msg, description, name);
2375 }
2376 else {
2377 PerlIO_printf(Perl_error_log,
2378 "perl: warning: %s %s.\n", msg, description);
2379 }
2380 }
2381 } /* End of tried to fallback */
98994639 2382
e5f10d49
KW
2383 /* Done with finding the locales; update our records */
2384
7d4bcc4a
KW
2385# ifdef USE_LOCALE_CTYPE
2386
948523db 2387 new_ctype(curlocales[LC_CTYPE_INDEX]);
98994639 2388
e5f10d49 2389# endif
7d4bcc4a
KW
2390# ifdef USE_LOCALE_COLLATE
2391
948523db 2392 new_collate(curlocales[LC_COLLATE_INDEX]);
98994639 2393
e5f10d49 2394# endif
7d4bcc4a
KW
2395# ifdef USE_LOCALE_NUMERIC
2396
948523db 2397 new_numeric(curlocales[LC_NUMERIC_INDEX]);
e5f10d49
KW
2398
2399# endif
2400
948523db 2401 for (i = 0; i < NOMINAL_LC_ALL_INDEX; i++) {
47280b20
KW
2402
2403# if defined(USE_ITHREADS)
2404
2405 /* This caches whether each category's locale is UTF-8 or not. This
2406 * may involve changing the locale. It is ok to do this at
2407 * initialization time before any threads have started, but not later.
2408 * Caching means that if the program heeds our dictate not to change
2409 * locales in threaded applications, this data will remain valid, and
2410 * it may get queried without changing locales. If the environment is
2411 * such that all categories have the same locale, this isn't needed, as
2412 * the code will not change the locale; but this handles the uncommon
2413 * case where the environment has disparate locales for the categories
2414 * */
2415 (void) _is_cur_LC_category_utf8(categories[i]);
2416
2417# endif
2418
e5f10d49
KW
2419 Safefree(curlocales[i]);
2420 }
b310b053 2421
7d4bcc4a
KW
2422# if defined(USE_PERLIO) && defined(USE_LOCALE_CTYPE)
2423
49c85077 2424 /* Set PL_utf8locale to TRUE if using PerlIO _and_ the current LC_CTYPE
50bf02bd
KW
2425 * locale is UTF-8. The call to new_ctype() just above has already
2426 * calculated the latter value and saved it in PL_in_utf8_CTYPE_locale. If
2427 * both PL_utf8locale and PL_unicode (set by -C or by $ENV{PERL_UNICODE})
2428 * are true, perl.c:S_parse_body() will turn on the PerlIO :utf8 layer on
2429 * STDIN, STDOUT, STDERR, _and_ the default open discipline. */
2430 PL_utf8locale = PL_in_utf8_CTYPE_locale;
49c85077 2431
a05d7ebb 2432 /* Set PL_unicode to $ENV{PERL_UNICODE} if using PerlIO.
fde18df1
JH
2433 This is an alternative to using the -C command line switch
2434 (the -C if present will override this). */
2435 {
dd374669 2436 const char *p = PerlEnv_getenv("PERL_UNICODE");
a05d7ebb 2437 PL_unicode = p ? parse_unicode_opts(&p) : 0;
5a22a2bb
NC
2438 if (PL_unicode & PERL_UNICODE_UTF8CACHEASSERT_FLAG)
2439 PL_utf8cache = -1;
b310b053
JH
2440 }
2441
7d4bcc4a 2442# endif
7d4bcc4a
KW
2443# ifdef __GLIBC__
2444
175c4cf9 2445 Safefree(language);
7d4bcc4a
KW
2446
2447# endif
175c4cf9
KW
2448
2449 Safefree(lc_all);
2450 Safefree(lang);
2451
e3305790 2452#endif /* USE_LOCALE */
2fcc0ca9 2453#ifdef DEBUGGING
7d4bcc4a 2454
2fcc0ca9 2455 /* So won't continue to output stuff */
27cdc72e 2456 DEBUG_INITIALIZATION_set(FALSE);
7d4bcc4a 2457
2fcc0ca9
KW
2458#endif
2459
98994639
HS
2460 return ok;
2461}
2462
98994639
HS
2463#ifdef USE_LOCALE_COLLATE
2464
a4a439fb 2465char *
a4a439fb
KW
2466Perl__mem_collxfrm(pTHX_ const char *input_string,
2467 STRLEN len, /* Length of 'input_string' */
2468 STRLEN *xlen, /* Set to length of returned string
2469 (not including the collation index
2470 prefix) */
2471 bool utf8 /* Is the input in UTF-8? */
6696cfa7 2472 )
98994639 2473{
a4a439fb
KW
2474
2475 /* _mem_collxfrm() is a bit like strxfrm() but with two important
2476 * differences. First, it handles embedded NULs. Second, it allocates a bit
2477 * more memory than needed for the transformed data itself. The real
55e5378d 2478 * transformed data begins at offset COLLXFRM_HDR_LEN. *xlen is set to
a4a439fb
KW
2479 * the length of that, and doesn't include the collation index size.
2480 * Please see sv_collxfrm() to see how this is used. */
2481
55e5378d
KW
2482#define COLLXFRM_HDR_LEN sizeof(PL_collation_ix)
2483
6696cfa7
KW
2484 char * s = (char *) input_string;
2485 STRLEN s_strlen = strlen(input_string);
79f120c8 2486 char *xbuf = NULL;
55e5378d 2487 STRLEN xAlloc; /* xalloc is a reserved word in VC */
17f41037 2488 STRLEN length_in_chars;
c664130f 2489 bool first_time = TRUE; /* Cleared after first loop iteration */
98994639 2490
a4a439fb
KW
2491 PERL_ARGS_ASSERT__MEM_COLLXFRM;
2492
2493 /* Must be NUL-terminated */
2494 assert(*(input_string + len) == '\0');
7918f24d 2495
79f120c8
KW
2496 /* If this locale has defective collation, skip */
2497 if (PL_collxfrm_base == 0 && PL_collxfrm_mult == 0) {
c7202dee
KW
2498 DEBUG_L(PerlIO_printf(Perl_debug_log,
2499 "_mem_collxfrm: locale's collation is defective\n"));
79f120c8
KW
2500 goto bad;
2501 }
2502
6696cfa7
KW
2503 /* Replace any embedded NULs with the control that sorts before any others.
2504 * This will give as good as possible results on strings that don't
2505 * otherwise contain that character, but otherwise there may be
2506 * less-than-perfect results with that character and NUL. This is
fdc080f3 2507 * unavoidable unless we replace strxfrm with our own implementation. */
fd43f63c
KW
2508 if (UNLIKELY(s_strlen < len)) { /* Only execute if there is an embedded
2509 NUL */
6696cfa7
KW
2510 char * e = s + len;
2511 char * sans_nuls;
fdc080f3 2512 STRLEN sans_nuls_len;
94762aa0 2513 int try_non_controls;
afc4976f
KW
2514 char this_replacement_char[] = "?\0"; /* Room for a two-byte string,
2515 making sure 2nd byte is NUL.
2516 */
2517 STRLEN this_replacement_len;
2518
1e4c9676
KW
2519 /* If we don't know what non-NUL control character sorts lowest for
2520 * this locale, find it */
f28f4d2a 2521 if (PL_strxfrm_NUL_replacement == '\0') {
6696cfa7 2522 int j;
afc4976f 2523 char * cur_min_x = NULL; /* The min_char's xfrm, (except it also
6696cfa7
KW
2524 includes the collation index
2525 prefixed. */
2526
91c0e2e0 2527 DEBUG_Lv(PerlIO_printf(Perl_debug_log, "Looking to replace NUL\n"));
94762aa0
KW
2528
2529 /* Unlikely, but it may be that no control will work to replace
1e4c9676
KW
2530 * NUL, in which case we instead look for any character. Controls
2531 * are preferred because collation order is, in general, context
2532 * sensitive, with adjoining characters affecting the order, and
2533 * controls are less likely to have such interactions, allowing the
2534 * NUL-replacement to stand on its own. (Another way to look at it
2535 * is to imagine what would happen if the NUL were replaced by a
2536 * combining character; it wouldn't work out all that well.) */
94762aa0
KW
2537 for (try_non_controls = 0;
2538 try_non_controls < 2;
2539 try_non_controls++)
2540 {
d4ff9586
KW
2541 /* Look through all legal code points (NUL isn't) */
2542 for (j = 1; j < 256; j++) {
2543 char * x; /* j's xfrm plus collation index */
2544 STRLEN x_len; /* length of 'x' */
2545 STRLEN trial_len = 1;
736a4fed 2546 char cur_source[] = { '\0', '\0' };
d4ff9586 2547
736a4fed
KW
2548 /* Skip non-controls the first time through the loop. The
2549 * controls in a UTF-8 locale are the L1 ones */
afc4976f
KW
2550 if (! try_non_controls && (PL_in_utf8_COLLATE_locale)
2551 ? ! isCNTRL_L1(j)
2552 : ! isCNTRL_LC(j))
2553 {
d4ff9586 2554 continue;
6696cfa7 2555 }
6696cfa7 2556
736a4fed
KW
2557 /* Create a 1-char string of the current code point */
2558 cur_source[0] = (char) j;
2559
d4ff9586
KW
2560 /* Then transform it */
2561 x = _mem_collxfrm(cur_source, trial_len, &x_len,
afc4976f 2562 0 /* The string is not in UTF-8 */);
6696cfa7 2563
1e4c9676 2564 /* Ignore any character that didn't successfully transform.
d4ff9586
KW
2565 * */
2566 if (! x) {
2567 continue;
2568 }
6696cfa7 2569
d4ff9586
KW
2570 /* If this character's transformation is lower than
2571 * the current lowest, this one becomes the lowest */
2572 if ( cur_min_x == NULL
2573 || strLT(x + COLLXFRM_HDR_LEN,
2574 cur_min_x + COLLXFRM_HDR_LEN))
2575 {
f28f4d2a 2576 PL_strxfrm_NUL_replacement = j;
d4ff9586 2577 cur_min_x = x;
d4ff9586
KW
2578 }
2579 else {
2580 Safefree(x);
2581 }
1e4c9676 2582 } /* end of loop through all 255 characters */
6696cfa7 2583
1e4c9676 2584 /* Stop looking if found */
94762aa0
KW
2585 if (cur_min_x) {
2586 break;
2587 }
2588
2589 /* Unlikely, but possible, if there aren't any controls that
2590 * work in the locale, repeat the loop, looking for any
2591 * character that works */
2592 DEBUG_L(PerlIO_printf(Perl_debug_log,
2593 "_mem_collxfrm: No control worked. Trying non-controls\n"));
1e4c9676 2594 } /* End of loop to try first the controls, then any char */
6696cfa7 2595
94762aa0
KW
2596 if (! cur_min_x) {
2597 DEBUG_L(PerlIO_printf(Perl_debug_log,
2598 "_mem_collxfrm: Couldn't find any character to replace"
2599 " embedded NULs in locale %s with", PL_collation_name));
2600 goto bad;
58eebef2
KW
2601 }
2602
94762aa0
KW
2603 DEBUG_L(PerlIO_printf(Perl_debug_log,
2604 "_mem_collxfrm: Replacing embedded NULs in locale %s with "
f28f4d2a 2605 "0x%02X\n", PL_collation_name, PL_strxfrm_NUL_replacement));
94762aa0 2606
6696cfa7 2607 Safefree(cur_min_x);
1e4c9676 2608 } /* End of determining the character that is to replace NULs */
afc4976f
KW
2609
2610 /* If the replacement is variant under UTF-8, it must match the
291a84fb 2611 * UTF8-ness of the original */
f28f4d2a
KW
2612 if ( ! UVCHR_IS_INVARIANT(PL_strxfrm_NUL_replacement) && utf8) {
2613 this_replacement_char[0] =
2614 UTF8_EIGHT_BIT_HI(PL_strxfrm_NUL_replacement);
2615 this_replacement_char[1] =
2616 UTF8_EIGHT_BIT_LO(PL_strxfrm_NUL_replacement);
afc4976f
KW
2617 this_replacement_len = 2;
2618 }
2619 else {
f28f4d2a 2620 this_replacement_char[0] = PL_strxfrm_NUL_replacement;
afc4976f
KW
2621 /* this_replacement_char[1] = '\0' was done at initialization */
2622 this_replacement_len = 1;
6696cfa7
KW
2623 }
2624
2625 /* The worst case length for the replaced string would be if every
2626 * character in it is NUL. Multiply that by the length of each
2627 * replacement, and allow for a trailing NUL */
afc4976f 2628 sans_nuls_len = (len * this_replacement_len) + 1;
fdc080f3 2629 Newx(sans_nuls, sans_nuls_len, char);
6696cfa7
KW
2630 *sans_nuls = '\0';
2631
6696cfa7
KW
2632 /* Replace each NUL with the lowest collating control. Loop until have
2633 * exhausted all the NULs */
2634 while (s + s_strlen < e) {
6069d6c5 2635 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2636
2637 /* Do the actual replacement */
6069d6c5 2638 my_strlcat(sans_nuls, this_replacement_char, sans_nuls_len);
6696cfa7
KW
2639
2640 /* Move past the input NUL */
2641 s += s_strlen + 1;
2642 s_strlen = strlen(s);
2643 }
2644
2645 /* And add anything that trails the final NUL */
6069d6c5 2646 my_strlcat(sans_nuls, s, sans_nuls_len);
6696cfa7
KW
2647
2648 /* Switch so below we transform this modified string */
2649 s = sans_nuls;
2650 len = strlen(s);
1e4c9676 2651 } /* End of replacing NULs */
6696cfa7 2652
a4a439fb
KW
2653 /* Make sure the UTF8ness of the string and locale match */
2654 if (utf8 != PL_in_utf8_COLLATE_locale) {
2655 const char * const t = s; /* Temporary so we can later find where the
2656 input was */
2657
2658 /* Here they don't match. Change the string's to be what the locale is
2659 * expecting */
2660
2661 if (! utf8) { /* locale is UTF-8, but input isn't; upgrade the input */
2662 s = (char *) bytes_to_utf8((const U8 *) s, &len);
2663 utf8 = TRUE;
2664 }
2665 else { /* locale is not UTF-8; but input is; downgrade the input */
2666
2667 s = (char *) bytes_from_utf8((const U8 *) s, &len, &utf8);
2668
2669 /* If the downgrade was successful we are done, but if the input
2670 * contains things that require UTF-8 to represent, have to do
2671 * damage control ... */
2672 if (UNLIKELY(utf8)) {
2673
2674 /* What we do is construct a non-UTF-8 string with
2675 * 1) the characters representable by a single byte converted
2676 * to be so (if necessary);
2677 * 2) and the rest converted to collate the same as the
2678 * highest collating representable character. That makes
2679 * them collate at the end. This is similar to how we
2680 * handle embedded NULs, but we use the highest collating
2681 * code point instead of the smallest. Like the NUL case,
2682 * this isn't perfect, but is the best we can reasonably
2683 * do. Every above-255 code point will sort the same as
2684 * the highest-sorting 0-255 code point. If that code
2685 * point can combine in a sequence with some other code
2686 * points for weight calculations, us changing something to
2687 * be it can adversely affect the results. But in most
2688 * cases, it should work reasonably. And note that this is
2689 * really an illegal situation: using code points above 255
2690 * on a locale where only 0-255 are valid. If two strings
2691 * sort entirely equal, then the sort order for the
2692 * above-255 code points will be in code point order. */
2693
2694 utf8 = FALSE;
2695
2696 /* If we haven't calculated the code point with the maximum
2697 * collating order for this locale, do so now */
2698 if (! PL_strxfrm_max_cp) {
2699 int j;
2700
2701 /* The current transformed string that collates the
2702 * highest (except it also includes the prefixed collation
2703 * index. */
2704 char * cur_max_x = NULL;
2705
2706 /* Look through all legal code points (NUL isn't) */
2707 for (j = 1; j < 256; j++) {
2708 char * x;
2709 STRLEN x_len;
736a4fed 2710 char cur_source[] = { '\0', '\0' };
a4a439fb 2711
736a4fed
KW
2712 /* Create a 1-char string of the current code point */
2713 cur_source[0] = (char) j;
a4a439fb
KW
2714
2715 /* Then transform it */
2716 x = _mem_collxfrm(cur_source, 1, &x_len, FALSE);
2717
2718 /* If something went wrong (which it shouldn't), just
2719 * ignore this code point */
94762aa0 2720 if (! x) {
a4a439fb
KW
2721 continue;
2722 }
2723
2724 /* If this character's transformation is higher than
2725 * the current highest, this one becomes the highest */
2726 if ( cur_max_x == NULL
55e5378d
KW
2727 || strGT(x + COLLXFRM_HDR_LEN,
2728 cur_max_x + COLLXFRM_HDR_LEN))
a4a439fb
KW
2729 {
2730 PL_strxfrm_max_cp = j;
2731 cur_max_x = x;
2732 }
2733 else {
2734 Safefree(x);
2735 }
2736 }
2737
94762aa0
KW
2738 if (! cur_max_x) {
2739 DEBUG_L(PerlIO_printf(Perl_debug_log,
2740 "_mem_collxfrm: Couldn't find any character to"
2741 " replace above-Latin1 chars in locale %s with",
2742 PL_collation_name));
2743 goto bad;
2744 }
2745
58eebef2
KW
2746 DEBUG_L(PerlIO_printf(Perl_debug_log,
2747 "_mem_collxfrm: highest 1-byte collating character"
2748 " in locale %s is 0x%02X\n",
2749 PL_collation_name,
2750 PL_strxfrm_max_cp));
58eebef2 2751
a4a439fb
KW
2752 Safefree(cur_max_x);
2753 }
2754
2755 /* Here we know which legal code point collates the highest.
2756 * We are ready to construct the non-UTF-8 string. The length
2757 * will be at least 1 byte smaller than the input string
2758 * (because we changed at least one 2-byte character into a
2759 * single byte), but that is eaten up by the trailing NUL */
2760 Newx(s, len, char);
2761
2762 {
2763 STRLEN i;
2764 STRLEN d= 0;
042d9e50 2765 char * e = (char *) t + len;
a4a439fb
KW
2766
2767 for (i = 0; i < len; i+= UTF8SKIP(t + i)) {
2768 U8 cur_char = t[i];
2769 if (UTF8_IS_INVARIANT(cur_char)) {
2770 s[d++] = cur_char;
2771 }
042d9e50 2772 else if (UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(t + i, e)) {
a4a439fb
KW
2773 s[d++] = EIGHT_BIT_UTF8_TO_NATIVE(cur_char, t[i+1]);
2774 }
2775 else { /* Replace illegal cp with highest collating
2776 one */
2777 s[d++] = PL_strxfrm_max_cp;
2778 }
2779 }
2780 s[d++] = '\0';
2781 Renew(s, d, char); /* Free up unused space */
2782 }
2783 }
2784 }
2785
2786 /* Here, we have constructed a modified version of the input. It could
2787 * be that we already had a modified copy before we did this version.
2788 * If so, that copy is no longer needed */
2789 if (t != input_string) {
2790 Safefree(t);
2791 }
2792 }
2793
17f41037
KW
2794 length_in_chars = (utf8)
2795 ? utf8_length((U8 *) s, (U8 *) s + len)
2796 : len;
2797
59c018b9
KW
2798 /* The first element in the output is the collation id, used by
2799 * sv_collxfrm(); then comes the space for the transformed string. The
2800 * equation should give us a good estimate as to how much is needed */
55e5378d 2801 xAlloc = COLLXFRM_HDR_LEN
a4a439fb 2802 + PL_collxfrm_base
17f41037 2803 + (PL_collxfrm_mult * length_in_chars);
a02a5408 2804 Newx(xbuf, xAlloc, char);
c7202dee
KW
2805 if (UNLIKELY(! xbuf)) {
2806 DEBUG_L(PerlIO_printf(Perl_debug_log,
2807 "_mem_collxfrm: Couldn't malloc %zu bytes\n", xAlloc));
98994639 2808 goto bad;
c7202dee 2809 }
98994639 2810
d35fca5f 2811 /* Store the collation id */
98994639 2812 *(U32*)xbuf = PL_collation_ix;
d35fca5f
KW
2813
2814 /* Then the transformation of the input. We loop until successful, or we
2815 * give up */
4ebeff16 2816 for (;;) {
1adab0a7 2817
55e5378d 2818 *xlen = strxfrm(xbuf + COLLXFRM_HDR_LEN, s, xAlloc - COLLXFRM_HDR_LEN);
4ebeff16
KW
2819
2820 /* If the transformed string occupies less space than we told strxfrm()
2821 * was available, it means it successfully transformed the whole
2822 * string. */
55e5378d 2823 if (*xlen < xAlloc - COLLXFRM_HDR_LEN) {
17f41037 2824
1adab0a7
KW
2825 /* Some systems include a trailing NUL in the returned length.
2826 * Ignore it, using a loop in case multiple trailing NULs are
2827 * returned. */
2828 while ( (*xlen) > 0
2829 && *(xbuf + COLLXFRM_HDR_LEN + (*xlen) - 1) == '\0')
2830 {
2831 (*xlen)--;
2832 }
2833
17f41037
KW
2834 /* If the first try didn't get it, it means our prediction was low.
2835 * Modify the coefficients so that we predict a larger value in any
2836 * future transformations */
2837 if (! first_time) {
2838 STRLEN needed = *xlen + 1; /* +1 For trailing NUL */
2839 STRLEN computed_guess = PL_collxfrm_base
2840 + (PL_collxfrm_mult * length_in_chars);
e1c30f0c
KW
2841
2842 /* On zero-length input, just keep current slope instead of
2843 * dividing by 0 */
2844 const STRLEN new_m = (length_in_chars != 0)
2845 ? needed / length_in_chars
2846 : PL_collxfrm_mult;
17f41037
KW
2847
2848 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2849 "%s: %d: initial size of %zu bytes for a length "
2850 "%zu string was insufficient, %zu needed\n",
17f41037 2851 __FILE__, __LINE__,
b07929e4 2852 computed_guess, length_in_chars, needed));
17f41037
KW
2853
2854 /* If slope increased, use it, but discard this result for
2855 * length 1 strings, as we can't be sure that it's a real slope
2856 * change */
2857 if (length_in_chars > 1 && new_m > PL_collxfrm_mult) {
7d4bcc4a
KW
2858
2859# ifdef DEBUGGING
2860
17f41037
KW
2861 STRLEN old_m = PL_collxfrm_mult;
2862 STRLEN old_b = PL_collxfrm_base;
7d4bcc4a
KW
2863
2864# endif
2865
17f41037
KW
2866 PL_collxfrm_mult = new_m;
2867 PL_collxfrm_base = 1; /* +1 For trailing NUL */
2868 computed_guess = PL_collxfrm_base
2869 + (PL_collxfrm_mult * length_in_chars);
2870 if (computed_guess < needed) {
2871 PL_collxfrm_base += needed - computed_guess;
2872 }
2873
2874 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4
KW
2875 "%s: %d: slope is now %zu; was %zu, base "
2876 "is now %zu; was %zu\n",
17f41037 2877 __FILE__, __LINE__,
b07929e4
KW
2878 PL_collxfrm_mult, old_m,
2879 PL_collxfrm_base, old_b));
17f41037
KW
2880 }
2881 else { /* Slope didn't change, but 'b' did */
2882 const STRLEN new_b = needed
2883 - computed_guess
2884 + PL_collxfrm_base;
2885 DEBUG_Lv(PerlIO_printf(Perl_debug_log,
b07929e4 2886 "%s: %d: base is now %zu; was %zu\n",
17f41037 2887 __FILE__, __LINE__,
b07929e4 2888 new_b, PL_collxfrm_base));
17f41037
KW
2889 PL_collxfrm_base = new_b;
2890 }
2891 }
2892
4ebeff16
KW
2893 break;
2894 }
bb0f664e 2895
c7202dee
KW
2896 if (UNLIKELY(*xlen >= PERL_INT_MAX)) {
2897 DEBUG_L(PerlIO_printf(Perl_debug_log,
2898 "_mem_collxfrm: Needed %zu bytes, max permissible is %u\n",
2899 *xlen, PERL_INT_MAX));
4ebeff16 2900 goto bad;
c7202dee 2901 }
d35fca5f 2902
c664130f 2903 /* A well-behaved strxfrm() returns exactly how much space it needs
1adab0a7
KW
2904 * (usually not including the trailing NUL) when it fails due to not
2905 * enough space being provided. Assume that this is the case unless
2906 * it's been proven otherwise */
c664130f 2907 if (LIKELY(PL_strxfrm_is_behaved) && first_time) {
55e5378d 2908 xAlloc = *xlen + COLLXFRM_HDR_LEN + 1;
c664130f
KW
2909 }
2910 else { /* Here, either:
2911 * 1) The strxfrm() has previously shown bad behavior; or
2912 * 2) It isn't the first time through the loop, which means
2913 * that the strxfrm() is now showing bad behavior, because
2914 * we gave it what it said was needed in the previous
2915 * iteration, and it came back saying it needed still more.
2916 * (Many versions of cygwin fit this. When the buffer size
2917 * isn't sufficient, they return the input size instead of
2918 * how much is needed.)
d4ff9586
KW
2919 * Increase the buffer size by a fixed percentage and try again.
2920 * */
6ddd902c 2921 xAlloc += (xAlloc / 4) + 1;
c664130f 2922 PL_strxfrm_is_behaved = FALSE;
c664130f 2923
7d4bcc4a
KW
2924# ifdef DEBUGGING
2925
58eebef2
KW
2926 if (DEBUG_Lv_TEST || debug_initialization) {
2927 PerlIO_printf(Perl_debug_log,
2928 "_mem_collxfrm required more space than previously calculated"
b07929e4 2929 " for locale %s, trying again with new guess=%d+%zu\n",
58eebef2 2930 PL_collation_name, (int) COLLXFRM_HDR_LEN,
b07929e4 2931 xAlloc - COLLXFRM_HDR_LEN);
58eebef2 2932 }
7d4bcc4a
KW
2933
2934# endif
2935
58eebef2 2936 }
c664130f 2937
4ebeff16 2938 Renew(xbuf, xAlloc, char);
c7202dee
KW
2939 if (UNLIKELY(! xbuf)) {
2940 DEBUG_L(PerlIO_printf(Perl_debug_log,
2941 "_mem_collxfrm: Couldn't realloc %zu bytes\n", xAlloc));
4ebeff16 2942 goto bad;
c7202dee 2943 }
c664130f
KW
2944
2945 first_time = FALSE;
4ebeff16 2946 }
98994639 2947
6696cfa7 2948
7d4bcc4a
KW
2949# ifdef DEBUGGING
2950
58eebef2 2951 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee
KW
2952
2953 print_collxfrm_input_and_return(s, s + len, xlen, utf8);
2954 PerlIO_printf(Perl_debug_log, "Its xfrm is:");
7e2f38b2
KW
2955 PerlIO_printf(Perl_debug_log, "%s\n",
2956 _byte_dump_string((U8 *) xbuf + COLLXFRM_HDR_LEN,
2957 *xlen, 1));
58eebef2 2958 }
7d4bcc4a
KW
2959
2960# endif
58eebef2 2961
3c5f993e 2962 /* Free up unneeded space; retain ehough for trailing NUL */
55e5378d 2963 Renew(xbuf, COLLXFRM_HDR_LEN + *xlen + 1, char);
98994639 2964
6696cfa7
KW
2965 if (s != input_string) {
2966 Safefree(s);
98994639
HS
2967 }
2968
98994639
HS
2969 return xbuf;
2970
2971 bad:
2972 Safefree(xbuf);
6696cfa7
KW
2973 if (s != input_string) {
2974 Safefree(s);
2975 }
98994639 2976 *xlen = 0;
7d4bcc4a
KW
2977
2978# ifdef DEBUGGING
2979
58eebef2 2980 if (DEBUG_Lv_TEST || debug_initialization) {
c7202dee 2981 print_collxfrm_input_and_return(s, s + len, NULL, utf8);
58eebef2 2982 }
7d4bcc4a
KW
2983
2984# endif
2985
98994639
HS
2986 return NULL;
2987}
2988
7d4bcc4a 2989# ifdef DEBUGGING
c7202dee 2990
4cbaac56 2991STATIC void
c7202dee
KW
2992S_print_collxfrm_input_and_return(pTHX_
2993 const char * const s,
2994 const char * const e,
2995 const STRLEN * const xlen,
2996 const bool is_utf8)
2997{
c7202dee
KW
2998
2999 PERL_ARGS_ASSERT_PRINT_COLLXFRM_INPUT_AND_RETURN;
3000
511e4ff7
DM
3001 PerlIO_printf(Perl_debug_log, "_mem_collxfrm[%" UVuf "]: returning ",
3002 (UV)PL_collation_ix);
c7202dee 3003 if (xlen) {
08b6dc1d 3004 PerlIO_printf(Perl_debug_log, "%zu", *xlen);
c7202dee
KW
3005 }
3006 else {
3007 PerlIO_printf(Perl_debug_log, "NULL");
3008 }
3009 PerlIO_printf(Perl_debug_log, " for locale '%s', string='",
3010 PL_collation_name);
9c8a6dc2
KW
3011 print_bytes_for_locale(s, e, is_utf8);
3012
3013 PerlIO_printf(Perl_debug_log, "'\n");
3014}
3015
3016STATIC void
3017S_print_bytes_for_locale(pTHX_
3018 const char * const s,
3019 const char * const e,
3020 const bool is_utf8)
3021{
3022 const char * t = s;
3023 bool prev_was_printable = TRUE;
3024 bool first_time = TRUE;
3025
3026 PERL_ARGS_ASSERT_PRINT_BYTES_FOR_LOCALE;
c7202dee
KW
3027
3028 while (t < e) {
3029 UV cp = (is_utf8)
3030 ? utf8_to_uvchr_buf((U8 *) t, e, NULL)
3031 : * (U8 *) t;
3032 if (isPRINT(cp)) {
3033 if (! prev_was_printable) {
3034 PerlIO_printf(Perl_debug_log, " ");
3035 }
3036 PerlIO_printf(Perl_debug_log, "%c", (U8) cp);
3037 prev_was_printable = TRUE;
3038 }
3039 else {
3040 if (! first_time) {
3041 PerlIO_printf(Perl_debug_log, " ");
3042 }
147e3846 3043 PerlIO_printf(Perl_debug_log, "%02" UVXf, cp);
c7202dee
KW
3044 prev_was_printable = FALSE;
3045 }
3046 t += (is_utf8) ? UTF8SKIP(t) : 1;
3047 first_time = FALSE;
3048 }
c7202dee
KW
3049}
3050
7d4bcc4a 3051# endif /* #ifdef DEBUGGING */
98994639 3052#endif /* USE_LOCALE_COLLATE */
58eebef2 3053
8ef6e574
KW
3054#ifdef USE_LOCALE
3055
c1284011
KW
3056bool
3057Perl__is_cur_LC_category_utf8(pTHX_ int category)
7d74bb61
KW
3058{
3059 /* Returns TRUE if the current locale for 'category' is UTF-8; FALSE
3060 * otherwise. 'category' may not be LC_ALL. If the platform doesn't have
119ee68b 3061 * nl_langinfo(), nor MB_CUR_MAX, this employs a heuristic, which hence
609548d2
KW
3062 * could give the wrong result. The result will very likely be correct for
3063 * languages that have commonly used non-ASCII characters, but for notably
3064 * English, it comes down to if the locale's name ends in something like
3065 * "UTF-8". It errs on the side of not being a UTF-8 locale. */
7d74bb61 3066
47280b20 3067 /* Name of current locale corresponding to the input category */
8de4332b 3068 const char *save_input_locale = NULL;
47280b20
KW
3069
3070 bool is_utf8 = FALSE; /* The return value */
7d74bb61
KW
3071 STRLEN final_pos;
3072
47280b20
KW
3073 /* The variables below are for the cache of previous lookups using this
3074 * function. The cache is a C string, described at the definition for
3075 * 'C_and_POSIX_utf8ness'.
3076 *
3077 * The first part of the cache is fixed, for the C and POSIX locales. The
3078 * varying part starts just after them. */
3079 char * utf8ness_cache = PL_locale_utf8ness + STRLENs(C_and_POSIX_utf8ness);
3080
3081 Size_t utf8ness_cache_size; /* Size of the varying portion */
3082 Size_t input_name_len; /* Length in bytes of save_input_locale */
3083 Size_t input_name_len_with_overhead; /* plus extra chars used to store
3084 the name in the cache */
3085 char * delimited; /* The name plus the delimiters used to store
3086 it in the cache */
3087 char * name_pos; /* position of 'delimited' in the cache, or 0
3088 if not there */
3089
3090
7d4bcc4a
KW
3091# ifdef LC_ALL
3092
7d74bb61 3093 assert(category != LC_ALL);
7d4bcc4a
KW
3094
3095# endif
7d74bb61 3096
47280b20 3097 /* Get the desired category's locale */
837ce802 3098 save_input_locale = do_setlocale_r(category, NULL);
7d74bb61 3099 if (! save_input_locale) {
47280b20 3100 Perl_croak(aTHX_
d707d779
KW
3101 "panic: %s: %d: Could not find current %s locale, errno=%d\n",
3102 __FILE__, __LINE__, category_name(category), errno);
7d74bb61 3103 }
47280b20 3104
b07fffd1 3105 save_input_locale = stdize_locale(savepv(save_input_locale));
47280b20
KW
3106 DEBUG_L(PerlIO_printf(Perl_debug_log,
3107 "Current locale for %s is %s\n",
3108 category_name(category), save_input_locale));
3109
3110 input_name_len = strlen(save_input_locale);
3111
3112 /* In our cache, each name is accompanied by two delimiters and a single
3113 * utf8ness digit */
3114 input_name_len_with_overhead = input_name_len + 3;
3115
3116 /* Allocate and populate space for a copy of the name surrounded by the
3117 * delimiters */
3118 Newx(delimited, input_name_len_with_overhead, char);
3119 delimited[0] = UTF8NESS_SEP[0];
3120 Copy(save_input_locale, delimited + 1, input_name_len, char);
3121 delimited[input_name_len+1] = UTF8NESS_PREFIX[0];
3122 delimited[input_name_len+2] = '\0';
3123
3124 /* And see if that is in the cache */
3125 name_pos = instr(PL_locale_utf8ness, delimited);
3126 if (name_pos) {
3127 is_utf8 = *(name_pos + input_name_len_with_overhead - 1) - '0';
3128
3129# ifdef DEBUGGING
3130
3131 if (DEBUG_Lv_TEST || debug_initialization) {
3132 PerlIO_printf(Perl_debug_log, "UTF8ness for locale %s=%d, \n",
3133 save_input_locale, is_utf8);
3134 }
3135
3136# endif
3137
3138 /* And, if not already in that position, move it to the beginning of
3139 * the non-constant portion of the list, since it is the most recently
3140 * used. (We don't have to worry about overflow, since just moving
3141 * existing names around) */
3142 if (name_pos > utf8ness_cache) {
3143 Move(utf8ness_cache,
3144 utf8ness_cache + input_name_len_with_overhead,
3145 name_pos - utf8ness_cache, char);
3146 Copy(delimited,
3147 utf8ness_cache,
3148 input_name_len_with_overhead - 1, char);
3149 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3150 }
3151
3152 Safefree(delimited);
b07fffd1 3153 Safefree(save_input_locale);
47280b20 3154 return is_utf8;
7d74bb61
KW
3155 }
3156
47280b20
KW
3157 /* Here we don't have stored the utf8ness for the input locale. We have to
3158 * calculate it */
3159
7d4bcc4a 3160# if defined(USE_LOCALE_CTYPE) \
1d958db2 3161 && (defined(MB_CUR_MAX) || (defined(HAS_NL_LANGINFO) && defined(CODESET)))
7d74bb61 3162
1d958db2 3163 { /* Next try nl_langinfo or MB_CUR_MAX if available */
7d74bb61
KW
3164
3165 char *save_ctype_locale = NULL;
7d74bb61 3166
119ee68b 3167 if (category != LC_CTYPE) { /* These work only on LC_CTYPE */
7d74bb61
KW
3168
3169 /* Get the current LC_CTYPE locale */
837ce802 3170 save_ctype_locale = do_setlocale_c(LC_CTYPE, NULL);
7d74bb61 3171 if (! save_ctype_locale) {
d707d779
KW
3172 Perl_croak(aTHX_
3173 "panic: %s: %d: Could not find current LC_CTYPE locale,"
3174 " errno=%d\n", __FILE__, __LINE__, errno);
7d74bb61 3175 }
4f72bb37 3176 save_ctype_locale = stdize_locale(savepv(save_ctype_locale));
7d74bb61
KW
3177
3178 /* If LC_CTYPE and the desired category use the same locale, this
3179 * means that finding the value for LC_CTYPE is the same as finding
3180 * the value for the desired category. Otherwise, switch LC_CTYPE
3181 * to the desired category's locale */
3182 if (strEQ(save_ctype_locale, save_input_locale)) {
3183 Safefree(save_ctype_locale);
3184 save_ctype_locale = NULL;
3185 }
837ce802 3186 else if (! do_setlocale_c(LC_CTYPE, save_input_locale)) {
7d74bb61 3187 Safefree(save_ctype_locale);
d707d779
KW
3188 Perl_croak(aTHX_
3189 "panic: %s: %d: Could not change LC_CTYPE locale to %s,"
3190 " errno=%d\n", __FILE__, __LINE__, save_input_locale, errno);
7d74bb61
KW
3191 }
3192 }
3193
69014004
KW
3194 DEBUG_L(PerlIO_printf(Perl_debug_log, "Current LC_CTYPE locale=%s\n",
3195 save_input_locale));
3196
7d74bb61 3197 /* Here the current LC_CTYPE is set to the locale of the category whose
1d958db2
KW
3198 * information is desired. This means that nl_langinfo() and MB_CUR_MAX
3199 * should give the correct results */
119ee68b 3200
7d4bcc4a
KW
3201# if defined(HAS_NL_LANGINFO) && defined(CODESET)
3202
291a84fb 3203 { /* The task is easiest if the platform has this POSIX 2001 function */
c70a3e68
KW
3204 const char *codeset = my_nl_langinfo(PERL_CODESET, FALSE);
3205 /* FALSE => already in dest locale */
119ee68b 3206
c70a3e68
KW
3207 DEBUG_L(PerlIO_printf(Perl_debug_log,
3208 "\tnllanginfo returned CODESET '%s'\n", codeset));
3209
3210 if (codeset && strNE(codeset, "")) {
1d958db2
KW
3211 /* If we switched LC_CTYPE, switch back */
3212 if (save_ctype_locale) {
837ce802 3213 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2
KW
3214 Safefree(save_ctype_locale);
3215 }
3216
1d075173
KW
3217 /* If the implementation of foldEQ() somehow were
3218 * to change to not go byte-by-byte, this could
3219 * read past end of string, as only one length is
3220 * checked. But currently, a premature NUL will
3221 * compare false, and it will stop there */
3222 is_utf8 = cBOOL( foldEQ(codeset, STR_WITH_LEN("UTF-8"))
3223 || foldEQ(codeset, STR_WITH_LEN("UTF8")));
1d958db2 3224
69014004
KW
3225 DEBUG_L(PerlIO_printf(Perl_debug_log,
3226 "\tnllanginfo returned CODESET '%s'; ?UTF8 locale=%d\n",
3227 codeset, is_utf8));
47280b20 3228 goto finish_and_return;
1d958db2 3229 }
119ee68b
KW
3230 }
3231
7d4bcc4a
KW
3232# endif
3233# ifdef MB_CUR_MAX
1d958db2
KW
3234
3235 /* Here, either we don't have nl_langinfo, or it didn't return a
3236 * codeset. Try MB_CUR_MAX */
3237
119ee68b
KW
3238 /* Standard UTF-8 needs at least 4 bytes to represent the maximum
3239 * Unicode code point. Since UTF-8 is the only non-single byte
3240 * encoding we handle, we just say any such encoding is UTF-8, and if
3241 * turns out to be wrong, other things will fail */
5f7616bd 3242 is_utf8 = (unsigned) MB_CUR_MAX >= STRLENs(MAX_UNICODE_UTF8);
119ee68b 3243
69014004
KW
3244 DEBUG_L(PerlIO_printf(Perl_debug_log,
3245 "\tMB_CUR_MAX=%d; ?UTF8 locale=%d\n",
3246 (int) MB_CUR_MAX, is_utf8));
3247
119ee68b
KW
3248 Safefree(save_input_locale);
3249
7d4bcc4a 3250# ifdef HAS_MBTOWC
119ee68b
KW
3251
3252 /* ... But, most system that have MB_CUR_MAX will also have mbtowc(),
3253 * since they are both in the C99 standard. We can feed a known byte
3254 * string to the latter function, and check that it gives the expected
3255 * result */
3256 if (is_utf8) {
3257 wchar_t wc;
51fc4b19
KW
3258 int len;
3259
856b881c 3260 PERL_UNUSED_RESULT(mbtowc(&wc, NULL, 0));/* Reset any shift state */
69014004 3261 errno = 0;
b1d4925c 3262 len = mbtowc(&wc, STR_WITH_LEN(REPLACEMENT_CHARACTER_UTF8));
51fc4b19 3263
b1d4925c
KW
3264
3265 if ( len != STRLENs(REPLACEMENT_CHARACTER_UTF8)
5579633c 3266 || wc != (wchar_t) UNICODE_REPLACEMENT)
119ee68b
KW
3267 {
3268 is_utf8 = FALSE;
b1d4925c
KW
3269 DEBUG_L(PerlIO_printf(Perl_debug_log, "\replacement=U+%x\n",
3270 (unsigned int)wc));
69014004
KW
3271 DEBUG_L(PerlIO_printf(Perl_debug_log,
3272 "\treturn from mbtowc=%d; errno=%d; ?UTF8 locale=0\n",
51fc4b19 3273 len, errno));
119ee68b
KW
3274 }
3275 }
7d4bcc4a
KW
3276
3277# endif
119ee68b 3278
1d958db2
KW
3279 /* If we switched LC_CTYPE, switch back */
3280 if (save_ctype_locale) {
837ce802 3281 do_setlocale_c(LC_CTYPE, save_ctype_locale);
1d958db2 3282 Safefree(save_ctype_locale);
119ee68b 3283 }
7d74bb61 3284
47280b20 3285 goto finish_and_return;
7d4bcc4a
KW
3286
3287# endif
3288
7d74bb61 3289 }
119ee68b 3290
7d4bcc4a 3291# else /* nl_langinfo should work if available, so don't bother compiling this
0080c90a
KW
3292 fallback code. The final fallback of looking at the name is
3293 compiled, and will be executed if nl_langinfo fails */
7d74bb61 3294
97f4de96
KW
3295 /* nl_langinfo not available or failed somehow. Next try looking at the
3296 * currency symbol to see if it disambiguates things. Often that will be
3297 * in the native script, and if the symbol isn't in UTF-8, we know that the
3298 * locale isn't. If it is non-ASCII UTF-8, we infer that the locale is
609548d2
KW
3299 * too, as the odds of a non-UTF8 string being valid UTF-8 are quite small
3300 * */
fa9b773e 3301
7d4bcc4a
KW
3302# ifdef HAS_LOCALECONV
3303# ifdef USE_LOCALE_MONETARY
3304
fa9b773e
KW
3305 {
3306 char *save_monetary_locale = NULL;
fa9b773e 3307 bool only_ascii = FALSE;
13542a67 3308 struct lconv* lc;
fa9b773e 3309
97f4de96
KW
3310 /* Like above for LC_CTYPE, we first set LC_MONETARY to the locale of
3311 * the desired category, if it isn't that locale already */
3312
fa9b773e
KW
3313 if (category != LC_MONETARY) {
3314
837ce802 3315 save_monetary_locale = do_setlocale_c(LC_MONETARY, NULL);
fa9b773e 3316 if (! save_monetary_locale) {
d707d779
KW
3317 Perl_croak(aTHX_
3318 "panic: %s: %d: Could not find current LC_MONETARY locale,"
3319 " errno=%d\n", __FILE__, __LINE__, errno);
fa9b773e 3320 }
4f72bb37 3321 save_monetary_locale = stdize_locale(savepv(save_monetary_locale));
fa9b773e 3322
13542a67
KW
3323 if (strEQ(save_monetary_locale, save_input_locale)) {
3324 Safefree(save_monetary_locale);
3325 save_monetary_locale = NULL;
3326 }
837ce802 3327 else if (! do_setlocale_c(LC_MONETARY, save_input_locale)) {
59c234b4 3328 Safefree(save_monetary_locale);
d707d779
KW
3329 Perl_croak(aTHX_
3330 "panic: %s: %d: Could not change LC_MONETARY locale to %s,"
3331 " errno=%d\n", __FILE__, __LINE__, save_input_locale, errno);
fa9b773e
KW
3332 }
3333 }
3334
3335 /* Here the current LC_MONETARY is set to the locale of the category
3336 * whose information is desired. */
3337
13542a67
KW
3338 lc = localeconv();
3339 if (! lc
3340 || ! lc->currency_symbol
c5f058df 3341 || is_utf8_invariant_string((U8 *) lc->currency_symbol, 0))
13542a67
KW
3342 {
3343 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));
3344 only_ascii = TRUE;
3345 }
3346 else {
3347 is_utf8 = is_utf8_string((U8 *) lc->currency_symbol, 0);
fa9b773e
KW
3348 }
3349
3350 /* If we changed it, restore LC_MONETARY to its original locale */
3351 if (save_monetary_locale) {
837ce802 3352 do_setlocale_c(LC_MONETARY, save_monetary_locale);
fa9b773e
KW
3353 Safefree(save_monetary_locale);
3354 }
3355
13542a67 3356 if (! only_ascii) {
fa9b773e 3357
59c234b4
KW
3358 /* It isn't a UTF-8 locale if the symbol is not legal UTF-8;
3359 * otherwise assume the locale is UTF-8 if and only if the symbol
3360 * is non-ascii UTF-8. */
3361 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?Currency symbol for %s is UTF-8=%d\n",
3362 save_input_locale, is_utf8));
47280b20 3363 goto finish_and_return;
13542a67 3364 }
fa9b773e 3365 }
fa9b773e 3366
7d4bcc4a
KW
3367# endif /* USE_LOCALE_MONETARY */
3368# endif /* HAS_LOCALECONV */
fa9b773e 3369
7d4bcc4a 3370# if defined(HAS_STRFTIME) && defined(USE_LOCALE_TIME)
15f7e74e
KW
3371
3372/* Still haven't found a non-ASCII string to disambiguate UTF-8 or not. Try
3373 * the names of the months and weekdays, timezone, and am/pm indicator */
3374 {
3375 char *save_time_locale = NULL;
3376 int hour = 10;
3377 bool is_dst = FALSE;
3378 int dom = 1;
3379 int month = 0;
3380 int i;
3381 char * formatted_time;
3382
3383
3384 /* Like above for LC_MONETARY, we set LC_TIME to the locale of the
3385 * desired category, if it isn't that locale already */
3386
3387 if (category != LC_TIME) {
3388
837ce802 3389 save_time_locale = do_setlocale_c(LC_TIME, NULL);
15f7e74e 3390 if (! save_time_locale) {
d707d779
KW
3391 Perl_croak(aTHX_
3392 "panic: %s: %d: Could not find current LC_TIME locale,"
3393 " errno=%d\n", __FILE__, __LINE__, errno);
15f7e74e
KW
3394 }
3395 save_time_locale = stdize_locale(savepv(save_time_locale));
3396
3397 if (strEQ(save_time_locale, save_input_locale)) {
3398 Safefree(save_time_locale);
3399 save_time_locale = NULL;
3400 }
837ce802 3401 else if (! do_setlocale_c(LC_TIME, save_input_locale)) {
15f7e74e 3402 Safefree(save_time_locale);
d707d779
KW
3403 Perl_croak(aTHX_
3404 "panic: %s: %d: Could not change LC_TIME locale to %s,"
3405 " errno=%d\n", __FILE__, __LINE__, save_input_locale, errno);
15f7e74e
KW
3406 }
3407 }
3408
3409 /* Here the current LC_TIME is set to the locale of the category
3410 * whose information is desired. Look at all the days of the week and
9f10db87 3411 * month names, and the timezone and am/pm indicator for UTF-8 variant
15f7e74e
KW
3412 * characters. The first such a one found will tell us if the locale
3413 * is UTF-8 or not */
3414
3415 for (i = 0; i < 7 + 12; i++) { /* 7 days; 12 months */
3416 formatted_time = my_strftime("%A %B %Z %p",
3ae5cd07 3417 0, 0, hour, dom, month, 2012 - 1900, 0, 0, is_dst);
c5f058df
KW
3418 if ( ! formatted_time
3419 || is_utf8_invariant_string((U8 *) formatted_time, 0))
3420 {
15f7e74e
KW
3421
3422 /* Here, we didn't find a non-ASCII. Try the next time through
3423 * with the complemented dst and am/pm, and try with the next
3424 * weekday. After we have gotten all weekdays, try the next
3425 * month */
3426 is_dst = ! is_dst;
3427 hour = (hour + 12) % 24;
3428 dom++;
3429 if (i > 6) {
3430 month++;
3431 }
3432 continue;
3433 }
3434
3435 /* Here, we have a non-ASCII. Return TRUE is it is valid UTF8;
3436 * false otherwise. But first, restore LC_TIME to its original
3437 * locale if we changed it */
3438 if (save_time_locale) {
837ce802 3439 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3440 Safefree(save_time_locale);
3441 }
3442
3443 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?time-related strings for %s are UTF-8=%d\n",
3444 save_input_locale,
3445 is_utf8_string((U8 *) formatted_time, 0)));
47280b20
KW
3446 is_utf8 = is_utf8_string((U8 *) formatted_time, 0);
3447 goto finish_and_return;
15f7e74e
KW
3448 }
3449
3450 /* Falling off the end of the loop indicates all the names were just
3451 * ASCII. Go on to the next test. If we changed it, restore LC_TIME
3452 * to its original locale */
3453 if (save_time_locale) {
837ce802 3454 do_setlocale_c(LC_TIME, save_time_locale);
15f7e74e
KW
3455 Safefree(save_time_locale);
3456 }
3457 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));
3458 }
15f7e74e 3459
7d4bcc4a 3460# endif
15f7e74e 3461
7d4bcc4a 3462# if 0 && defined(USE_LOCALE_MESSAGES) && defined(HAS_SYS_ERRLIST)
855aeb93
JH
3463
3464/* This code is ifdefd out because it was found to not be necessary in testing
5857e934
KW
3465 * on our dromedary test machine, which has over 700 locales. There, this
3466 * added no value to looking at the currency symbol and the time strings. I
3467 * left it in so as to avoid rewriting it if real-world experience indicates
3468 * that dromedary is an outlier. Essentially, instead of returning abpve if we
855aeb93
JH
3469 * haven't found illegal utf8, we continue on and examine all the strerror()
3470 * messages on the platform for utf8ness. If all are ASCII, we still don't
3471 * know the answer; but otherwise we have a pretty good indication of the
5857e934
KW
3472 * utf8ness. The reason this doesn't help much is that the messages may not
3473 * have been translated into the locale. The currency symbol and time strings
3474 * are much more likely to have been translated. */
3475 {
855aeb93 3476 int e;
5857e934 3477 bool non_ascii = FALSE;
855aeb93 3478 char *save_messages_locale = NULL;
5857e934 3479 const char * errmsg = NULL;
855aeb93 3480
5857e934
KW
3481 /* Like above, we set LC_MESSAGES to the locale of the desired
3482 * category, if it isn't that locale already */
855aeb93
JH
3483
3484 if (category != LC_MESSAGES) {
3485
837ce802 3486 save_messages_locale = do_setlocale_c(LC_MESSAGES, NULL);
855aeb93 3487 if (! save_messages_locale) {
d707d779
KW
3488 Perl_croak(aTHX_
3489 "panic: %s: %d: Could not find current LC_MESSAGES locale,"
3490 " errno=%d\n", __FILE__, __LINE__, errno);
855aeb93 3491 }
5857e934 3492 save_messages_locale = stdize_locale(savepv(save_messages_locale));
855aeb93
JH
3493
3494 if (strEQ(save_messages_locale, save_input_locale)) {
5857e934
KW
3495 Safefree(save_messages_locale);
3496 save_messages_locale = NULL;
855aeb93 3497 }
837ce802 3498 else if (! do_setlocale_c(LC_MESSAGES, save_input_locale)) {
855aeb93 3499 Safefree(save_messages_locale);
d707d779
KW
3500 Perl_croak(aTHX_
3501 "panic: %s: %d: Could not change LC_MESSAGES locale to %s,"
3502 " errno=%d\n", __FILE__, __LINE__, save_input_locale, errno);
855aeb93
JH
3503 }
3504 }
3505
3506 /* Here the current LC_MESSAGES is set to the locale of the category
5857e934
KW
3507 * whose information is desired. Look through all the messages. We
3508 * can't use Strerror() here because it may expand to code that
3509 * segfaults in miniperl */
855aeb93 3510
5857e934
KW
3511 for (e = 0; e <= sys_nerr; e++) {
3512 errno = 0;
3513 errmsg = sys_errlist[e];
3514 if (errno || !errmsg) {
855aeb93
JH
3515 break;
3516 }
5857e934 3517 errmsg = savepv(errmsg);
c5f058df 3518 if (! is_utf8_invariant_string((U8 *) errmsg, 0)) {
5857e934
KW
3519 non_ascii = TRUE;
3520 is_utf8 = is_utf8_string((U8 *) errmsg, 0);
3521 break;
855aeb93
JH
3522 }
3523 }
5857e934 3524 Safefree(errmsg);
855aeb93
JH
3525
3526 /* And, if we changed it, restore LC_MESSAGES to its original locale */
3527 if (save_messages_locale) {
837ce802 3528 do_setlocale_c(LC_MESSAGES, save_messages_locale);
855aeb93
JH
3529 Safefree(save_messages_locale);
3530 }
3531
5857e934
KW
3532 if (non_ascii) {
3533
3534 /* Any non-UTF-8 message means not a UTF-8 locale; if all are valid,
3535 * any non-ascii means it is one; otherwise we assume it isn't */
3536 DEBUG_L(PerlIO_printf(Perl_debug_log, "\t?error messages for %s are UTF-8=%d\n",
3537 save_input_locale,
3538 is_utf8));
47280b20 3539 goto finish_and_return;
5857e934 3540 }
855aeb93 3541
5857e934 3542 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 3543 }
855aeb93 3544
7d4bcc4a
KW
3545# endif
3546# endif /* the code that is compiled when no nl_langinfo */
0080c90a 3547
7d4bcc4a 3548# ifndef EBCDIC /* On os390, even if the name ends with "UTF-8', it isn't a
92c0a900 3549 UTF-8 locale */
7d4bcc4a 3550
97f4de96
KW
3551 /* As a last resort, look at the locale name to see if it matches
3552 * qr/UTF -? * 8 /ix, or some other common locale names. This "name", the
3553 * return of setlocale(), is actually defined to be opaque, so we can't
3554 * really rely on the absence of various substrings in the name to indicate
3555 * its UTF-8ness, but if it has UTF8 in the name, it is extremely likely to
3556 * be a UTF-8 locale. Similarly for the other common names */
3557
3558 final_pos = strlen(save_input_locale) - 1;
3559 if (final_pos >= 3) {
8de4332b 3560 const char *name = save_input_locale;
97f4de96
KW
3561
3562 /* Find next 'U' or 'u' and look from there */
3563 while ((name += strcspn(name, "Uu") + 1)
3564 <= save_input_locale + final_pos - 2)
3565 {
c72493e0 3566 if ( isALPHA_FOLD_NE(*name, 't')
305b8651 3567 || isALPHA_FOLD_NE(*(name + 1), 'f'))
97f4de96
KW
3568 {
3569 continue;
3570 }
3571 name += 2;
3572 if (*(name) == '-') {
3573 if ((name > save_input_locale + final_pos - 1)) {
3574 break;
3575 }
3576 name++;
3577 }
3578 if (*(name) == '8') {
97f4de96
KW
3579 DEBUG_L(PerlIO_printf(Perl_debug_log,
3580 "Locale %s ends with UTF-8 in name\n",
3581 save_input_locale));
47280b20
KW
3582 is_utf8 = TRUE;
3583 goto finish_and_return;
97f4de96
KW
3584 }
3585 }
3586 DEBUG_L(PerlIO_printf(Perl_debug_log,
3587 "Locale %s doesn't end with UTF-8 in name\n",
3588 save_input_locale));
3589 }
3590
4d8d465a 3591# ifdef WIN32
7d4bcc4a 3592
97f4de96 3593 /* http://msdn.microsoft.com/en-us/library/windows/desktop/dd317756.aspx */
8a0832a1 3594 if (memENDs(save_input_locale, final_pos, "65001")) {
97f4de96 3595 DEBUG_L(PerlIO_printf(Perl_debug_log,
8a0832a1 3596 "Locale %s ends with 65001 in name, is UTF-8 locale\n",
97f4de96 3597 save_input_locale));
47280b20
KW
3598 is_utf8 = TRUE;
3599 goto finish_and_return;
97f4de96 3600 }
7d4bcc4a 3601
4d8d465a
KW
3602# endif
3603# endif
97f4de96
KW
3604
3605 /* Other common encodings are the ISO 8859 series, which aren't UTF-8. But
3606 * since we are about to return FALSE anyway, there is no point in doing
3607 * this extra work */
7d4bcc4a
KW
3608
3609# if 0
97f4de96
KW
3610 if (instr(save_input_locale, "8859")) {
3611 DEBUG_L(PerlIO_printf(Perl_debug_log,
3612 "Locale %s has 8859 in name, not UTF-8 locale\n",
3613 save_input_locale));
47280b20
KW
3614 is_utf8 = FALSE;
3615 goto finish_and_return;
97f4de96 3616 }
7d4bcc4a 3617# endif
97f4de96 3618
69014004
KW
3619 DEBUG_L(PerlIO_printf(Perl_debug_log,
3620 "Assuming locale %s is not a UTF-8 locale\n",
3621 save_input_locale));
47280b20
KW
3622 is_utf8 = FALSE;
3623
3624 finish_and_return:
3625
3626 /* Cache this result so we don't have to go through all this next time. */
3627 utf8ness_cache_size = sizeof(PL_locale_utf8ness)
3628 - (utf8ness_cache - PL_locale_utf8ness);
3629
3630 /* But we can't save it if it is too large for the total space available */
3631 if (LIKELY(input_name_len_with_overhead < utf8ness_cache_size)) {
3632 Size_t utf8ness_cache_len = strlen(utf8ness_cache);
3633
3634 /* Here it can fit, but we may need to clear out the oldest cached
3635 * result(s) to do so. Check */
3636 if (utf8ness_cache_len + input_name_len_with_overhead
3637 >= utf8ness_cache_size)
3638 {
3639 /* Here we have to clear something out to make room for this.
3640 * Start looking at the rightmost place where it could fit and find
3641 * the beginning of the entry that extends past that. */
3642 char * cutoff = (char *) my_memrchr(utf8ness_cache,
3643 UTF8NESS_SEP[0],
3644 utf8ness_cache_size
3645 - input_name_len_with_overhead);
3646
3647 assert(cutoff);
3648 assert(cutoff >= utf8ness_cache);
3649
3650 /* This and all subsequent entries must be removed */
3651 *cutoff = '\0';
3652 utf8ness_cache_len = strlen(utf8ness_cache);
3653 }
3654
3655 /* Make space for the new entry */
3656 Move(utf8ness_cache,
3657 utf8ness_cache + input_name_len_with_overhead,
3658 utf8ness_cache_len + 1 /* Incl. trailing NUL */, char);
3659
3660 /* And insert it */
3661 Copy(delimited, utf8ness_cache, input_name_len_with_overhead - 1, char);
3662 utf8ness_cache[input_name_len_with_overhead - 1] = is_utf8 + '0';
3663
3664 if ((PL_locale_utf8ness[strlen(PL_locale_utf8ness)-1]
3665 & (PERL_UINTMAX_T) ~1) != '0')
3666 {
3667 Perl_croak(aTHX_
3668 "panic: %s: %d: Corrupt utf8ness_cache=%s\nlen=%u,"
3669 " inserted_name=%s, its_len=%u\n",
3670 __FILE__, __LINE__,
3671 PL_locale_utf8ness, strlen(PL_locale_utf8ness),
3672 delimited, input_name_len_with_overhead);
3673 }
3674 }
3675
3676# ifdef DEBUGGING
3677
3678 if (DEBUG_Lv_TEST || debug_initialization) {
3679 PerlIO_printf(Perl_debug_log,
3680 "PL_locale_utf8ness is now %s; returning %d\n",
3681 PL_locale_utf8ness, is_utf8);
3682 }
3683
3684# endif
3685
3686 Safefree(delimited);
fa9b773e 3687 Safefree(save_input_locale);
47280b20 3688 return is_utf8;
7d74bb61
KW
3689}
3690
8ef6e574 3691#endif
7d74bb61 3692
d6ded950
KW
3693
3694bool
3695Perl__is_in_locale_category(pTHX_ const bool compiling, const int category)
3696{
1a4f13e1 3697 dVAR;
d6ded950
KW
3698 /* Internal function which returns if we are in the scope of a pragma that
3699 * enables the locale category 'category'. 'compiling' should indicate if
3700 * this is during the compilation phase (TRUE) or not (FALSE). */
3701
3702 const COP * const cop = (compiling) ? &PL_compiling : PL_curcop;
3703
3704 SV *categories = cop_hints_fetch_pvs(cop, "locale", 0);
3705 if (! categories || categories == &PL_sv_placeholder) {
3706 return FALSE;
3707 }
3708
3709 /* The pseudo-category 'not_characters' is -1, so just add 1 to each to get
3710 * a valid unsigned */
3711 assert(category >= -1);
3712 return cBOOL(SvUV(categories) & (1U << (category + 1)));
3713}
3714
2c6ee1a7 3715char *
6ebbc862
KW
3716Perl_my_strerror(pTHX_ const int errnum)
3717{
3718 /* Returns a mortalized copy of the text of the error message associated
3719 * with 'errnum'. It uses the current locale's text unless the platform
3720 * doesn't have the LC_MESSAGES category or we are not being called from
3721 * within the scope of 'use locale'. In the former case, it uses whatever
3722 * strerror returns; in the latter case it uses the text from the C locale.
3723 *